+ Guide · Immunoassays

A few molecules in a drop.

How antibodies, labels and analyzers turn a protein present at a few nanograms per liter into a number a physician can act on, why each new way of reading a bound antibody moved that limit lower, and what a reagent and its analyzer have to demonstrate before they are sold in the US, the EU and India.
+ The central question

How little of one protein can a test find in a drop of blood, and what sets that floor: the antibody, the label, or the instrument that reads it?

Starts at:school chemistry and biology: a protein is a folded chain with a shape, and a concentration is an amount in a volume. No immunology, optics, statistics or regulatory background is assumed.Ends at:working out from a package insert or a datasheet what limits how little a test can detect and how precise it is, telling a real result from an interference, and planning the evidence a reagent and its analyzer need together. Built around:the high-sensitivity troponin assays that let an emergency department rule out a heart attack within an hour, from a Swiss study published in 2009 to the first US clearance in January 2017; and a second case, the first blood tests cleared to aid the diagnosis of Alzheimer's disease.
+ Opening

Arrivals in the first three hours.

Each of the 786 patients had come to an emergency department with chest pain, or with another symptom that might have meant a heart attack, and each gave a tube of blood soon after arriving. The study that collected those tubes was coordinated from University Hospital Basel in Switzerland and enrolled patients between April 2006 and April 2008. The blood was not used to decide anyone's care. The plasma was frozen at minus 80 °C and measured later, in two batches in a core laboratory, by staff who did not know which patients had been ill.

The 718 samples that could be run on all five tests were measured for cardiac troponin, a protein of heart muscle that leaks into the blood when heart cells die. One test was the standard troponin T assay in routine use at the time. The other four were newer and more sensitive, made by three companies. Two cardiologists decided who had in fact had a heart attack. They worked independently through each patient's records from arrival to 60 days later, without seeing the results of the four new tests. They counted 123 patients, 17 percent of the 718.

The researchers then asked how well each test separated the patients who had had a heart attack from those who had not, using the single sample taken on arrival. They scored the separation as an area under a curve, a number that is 1.0 when a test separates two groups perfectly and 0.5 when it does no better than tossing a coin. Across all patients the standard assay scored 0.90 and the four sensitive assays 0.95 to 0.96. For patients who had arrived within three hours of the start of their pain, the gap opened: the standard assay scored 0.76, and the sensitive assays 0.92 to 0.94.

The sensitive assays had not found a new marker. They measured cardiac troponin, T or I, released by the same damaged muscle, at concentrations that the standard assay could not tell apart from none. The paper appeared in the New England Journal of Medicine on 27 August 2009, beside a German study of 1,818 patients at Mainz, Koblenz and Hamburg that reached the same conclusion with a sensitive troponin I assay.

What followed took years. The sensitive troponin T assay was launched in Britain and Ireland in late 2009. In 2015 the European Society of Cardiology recommended ruling a heart attack in or out from two samples taken an hour apart, with cut-offs set separately for each assay. In the United States the first such assay was cleared by the Food and Drug Administration on 18 January 2017, and the agency's review drew part of its clinical evidence from the same 718 patients of the Basel-coordinated study. The cleared test carried the name Troponin T Gen 5. A trade report stated that the agency had asked its maker to call it "next generation" rather than "high-sensitivity". Between the old assay and the new, the protein, the patients and even the antibodies' target stayed the same. What changed was how few molecules of troponin a test could tell apart from none.

+ Before we start

Before we start.

A test that could not see troponin in the first three hours left physicians waiting for a second sample, and a test that could see it changed the decision made on arrival. The difference lay below a few nanograms per liter, a concentration at which a drop of blood holds a few million troponin molecules among roughly ten million billion molecules of albumin alone. That raises the question this guide answers: how little of one protein can a test find in a drop of blood, and what sets that floor: the antibody, the label, or the instrument that reads it?

Some words first. An assay is a procedure that measures how much of a particular substance a sample contains. An immunoassay is an assay in which the part that recognizes the substance is an antibody, a protein the immune system makes to bind one molecular shape. The antibody's binding cannot be seen, so the antibody carries, or calls in, a label: an enzyme that makes color, a molecule that gives off light, a particle that shows as a colored line. Most immunoassays then wash: they rinse away everything that did not bind, while the surface that holds the antibodies keeps what did, so that only label held in place by the target protein is left to be measured.

The short answer runs as follows. The antibody decides what is measured, and how much of it can be caught in the minutes a test allows. The label and the way it is read decide how much signal each caught molecule gives. Color from an enzyme multiplied that signal. Light from a chemical reaction, fluorescence read after a delay and light switched on at an electrode each raised it further and cut the light that reaches the detector for other reasons. Between them they took the floor for troponin from about a hundred nanograms per liter in the early 1990s to a few nanograms per liter. Past that point the detector is rarely the limit. The floor becomes the label that stays behind when there is no target to hold it, plus the scatter in counting very few molecules. Assays that count labels one at a time instead of measuring a glow lower the floor again, but every label that sticks where it should not still counts as signal. The analyzer's washing, timing, temperature and calibration decide whether that floor holds from run to run and from laboratory to laboratory, which is what made a sensitive assay usable for a decision made within an hour.

Under all of this sit two physical facts. Binding is a balance, not a permanent bond: at every moment some antibody lets go of its target and some other antibody catches it, so the fraction bound depends on concentration and on time. And the label that reports a bond cannot tell the bond it was meant to report from any other surface or molecule it has stuck to. Every chapter that follows is about one or the other of these two facts.

How to read this guide

The 26 chapters form one sequence in eight parts. The first five parts build the answer: how an antibody catches its target, four ways of making a caught molecule visible, the analyzer that runs those chemistries as a machine, the strip that runs them with no machine at all, and the assays that count single molecules. The last three read results, survey the field as of October 2026, and set out what a test system must show regulators. Roche's high-sensitivity troponin T assay, the sensitive troponin T assay of the opening story, recurs throughout as the guide's running case, and the blood tests for Alzheimer's disease as a second case. Each chapter opens with the question left by the one before, and each works one example with real numbers. The illustrations, called plates, are numbered on their own, so any one can be cited by itself. Parts I to VIII hold Chapters 1 to 5, 6 to 9, 10 to 14, 15 and 16, 17 and 18, 19 to 21, 22, and 23 to 26. Four kinds of box recur:

Insight or solved example

Blue edge. Carries the structural point of a section, or a calculation worked once with real values and stated assumptions.

Caution

Orange edge. Names a common misreading, a trap, or the limit of a claim.

In practice

Green edge. Maps the idea onto the reader's own work: specifying or evaluating an analyzer, a reagent or a strip, reading a result, or planning the evidence for a submission. With the summaries and the Lessons at the back, these boxes are the shortest route through the guide.

+ What this chapter established
  • Each chapter closes with what it established, in four lines.

The mathematics stays at arithmetic: multiplying and dividing, percentages, ratios and powers of ten, as in 10⁹ for a billion. Concentrations appear in two forms, as a mass in a volume (nanograms per liter) and as a count of molecules, and Chapter 1 shows how to move between them. Binding, counting statistics and calibration curves are described in words and worked once with numbers; the plates draw the curves. Where a figure is a company's own claim about its product, the text says so. The guide uses US spelling.

Inside the plates, color is a legend and never decoration. The protein being measured is violet; antibodies and other binders are teal; the label, and the color, light or current it produces, is gold; whatever makes the label give its signal, such as exciting light, a voltage or a substrate, is red; and anything that binds or reads as signal without the target is magenta. The measured readout and every number computed from it are always blue, and one orange mark in each plate points at the detail that matters most. A key strip under every plate lists only the colors that plate uses.

The color key
THE THINGS THE PLATES TELL APART MEASUREMENT AND CONSTRUCTION ANALYTE the protein being measured: troponin, pTau217, hCG ANTIBODY capture and detection antibodies and other binders LABEL AND SIGNAL the tag, and the color, light or current it produces TRIGGER what makes the label speak: light in, voltage, substrate INTERFERENT anything that binds or reads as signal without the analyte READOUT the measured signal, and the path from signal to result FOCAL DETAIL the one detail that matters: a single orange mark per plate GRAY instruments, parts, axes and construction DASHED GRAY not detected, not measured, or not available ANALYTE ANTIBODY READOUT FOCAL DETAIL each plate's key strip lists only the colors it uses
Key — The same color means the same thing in every plate: violet for the protein being measured, teal for antibodies and other binders, gold for the label and the signal it gives, red for what triggers that signal, magenta for anything that reads as signal without the target, and blue for the measured readout.
The limits of this guide

This is a reference for understanding, not a laboratory protocol, a regulatory opinion or clinical advice. Products, clearances, guidelines, standards and regulatory positions are stated as of October 2026 and will date. Where a figure comes from a company about its own product, or from a paper whose authors work for that company, the text says so. Named companies and products are examples of a principle, not recommendations. Cost, time to market and make-or-buy decisions are outside the guide; Chapters 23 to 26 describe the evidence and the routes to market that bear on them.

+ Part I · Binding

What an antibody catches, and how fast.

A tube in which an antibody and its target are mixed looks exactly like the same tube without them: no color appears, nothing settles, and no instrument pointed at the liquid can tell that a single bond has formed. The five chapters of this part work out what happens out of sight. They put a number on how little protein there is to find, describe the antibody that finds it, show why catching takes time, set out the two ways of turning a catch into a measurement, and end on the signal that appears when there is nothing to catch at all, which is where every immunoassay's floor begins.

01 — The floor

A few nanograms in a liter.

+ The questionHow much troponin is in a healthy person's blood, and how does that compare with everything else in it?

Four million molecules in a sample

Five nanograms per liter is the limit of detection that Roche's European label gives, as a 2017 trade report records, for its high-sensitivity troponin T assay, the sensitive test of the opening story. A limit of detection is roughly the lowest concentration that a test can tell apart from none, and the last section of this chapter defines it exactly. A nanogram is a billionth of a gram, so 5 ng/L means that a liter of plasma, the liquid part of blood, holds five billionths of a gram of the protein. A 2021 review by scientists at an antibody supplier reports cardiac troponin T and troponin I in healthy people at about 1 to 50 ng/L, with men's values 1.2 to 2.4 times women's and values rising with age, particularly after 60. Many healthy women sit below 5 ng/L. A test that has to tell a slightly raised value from a normal one therefore works at the very bottom of that range.

A mass per liter hides how few molecules that is. Every protein has a molecular weight, the mass of one mole of it, and a mole is about 6.02 × 10²³ molecules. Human cardiac troponin T is a chain of 298 amino acids weighing 35,924 grams per mole, according to the UniProt protein database. Dividing a mass by the molecular weight gives moles, and multiplying by 6.02 × 10²³ gives molecules. The solved example below follows that route for a sample of 50 microliters, a little more than a drop. That volume is a round example chosen here, not taken from any one test's instructions.

Solved example: troponin T molecules in 50 µL

A liter at 5 ng/L holds 5 nanograms, and 50 µL is a twenty-thousandth of a liter, so the sample holds 5 ÷ 20,000 = 0.00025 nanogram, or 0.25 picogram. Dividing by 35,924 grams per mole gives about 7 attomoles, where an attomole is a billionth of a billionth of a mole. One attomole is about 602,000 molecules, so 7 attomoles is about 4.2 million molecules. The US 99th-percentile limit of the same test, defined below as the upper edge of healthy values, is 19 ng/L, and at that concentration the count is about 16 million. As a concentration, 5 ng/L of troponin T is about 0.14 picomolar: 0.14 trillionths of a mole per liter.

Four million sounds like plenty. It is not, for two reasons the rest of the guide develops. The molecules are spread through the whole sample, and only those that meet an antibody in the minutes the test allows can be counted. And the instrument never detects a troponin molecule itself. It detects a label that the antibody brings, and every label that sticks in the wrong place is counted as well.

A needle in forty grams of albumin

The troponin shares its liquid with far more of everything else. Albumin, the most abundant protein in plasma, makes up about half of all plasma protein at 35 to 50 grams per liter. Immunoglobulin G, the most common class of antibody in the blood, adds roughly another 12 grams per liter. Fats, sugars, salts and hundreds of other proteins share the same volume.

Side by side, the amounts of albumin and troponin differ by more than any everyday comparison can convey. At 40 g/L of albumin and 5 ng/L of troponin T, there are 8 billion grams of albumin for every gram of troponin. Albumin molecules are heavier, 66,438 grams per mole, so the gap in molecules is a little smaller, about 4.3 billion to one; the same 50 µL that holds 4.2 million troponin molecules holds about 18 million billion albumin molecules. A test that mistook one albumin molecule in a billion for troponin would count about four false molecules for every real one.

01 — Forty grams against five nanograms
100 g/L 1 g/L 10 mg/L 100 µg/L 1 µg/L 10 ng/L 0.1 ng/L CONCENTRATION IN PLASMA (EACH TICK: A FACTOR OF 100) albumin 35–50 g/L IgG about 12 g/L cardiac troponin, healthy: about 1–50 ng/L pTau217: about 0.1–0.7 ng/L (= pg/mL) 5 ng/L the floor the opening scene needed 8 billion grams of albumin per gram of troponin T in 50 µL at 5 ng/L: about 4.2 million troponin T molecules and about 18 million billion albumin molecules
AnalyteFocal detail
Plate 01 — A liter of plasma carries about 40 grams of albumin and, in a healthy adult, a few nanograms of troponin: about ten factors of ten apart. The tests in this guide work at the bottom of that axis, where a 50 µL sample holds a few million target molecules.

The second case in this guide sits lower still. In May 2025 the FDA cleared the first blood test to aid the diagnosis of Alzheimer's disease. It measures phosphorylated tau 217 together with a second protein fragment, beta-amyloid 1-42 (Chapter 21). Phosphorylated tau 217 is a fragment of the tau protein that carries a phosphate group at amino acid 217, and it circulates at fractions of a picogram per milliliter. A picogram per milliliter equals a nanogram per liter, so 0.5 pg/mL of tau is a tenth of the 5 ng/L of troponin T by mass, and by number of molecules it is closer to a thirteenth: about 260,000 molecules in the 40 µL that one cleared test uses. That count treats every molecule as full-length tau, which is an assumption, because the forms measured are fragments.

Measuring without seeing

No balance weighs a quarter of a picogram, and no microscope picks one protein out of a liquid crowded with billions of similar ones. Every immunoassay therefore does three things in sequence. It catches the target selectively, using antibodies that bind its shape and little else. It makes each caught molecule produce a signal large enough to measure, by attaching a label that gives color, light or current. And it separates that signal from everything else that could produce one: label that never bound, label that bound to the wrong thing, light from the sample itself, and the electronic noise of the detector.

Each step can fail on its own. An antibody that binds too slowly catches only part of the target in the nine minutes of the fastest version of Roche's troponin T assay (Chapter 3). A label that gives too little signal per molecule leaves a few caught molecules lost in the detector's noise. And a separation that leaves some unbound label behind creates a signal with no troponin at all. The guide takes the steps in order: catching in Part I, signal in Part II, and the machinery of separation and measurement in Part III.

What "the floor" means

A test's floor is not one number but three, defined by how they are measured. The limit of blank is the highest result expected from samples that contain none of the target, usually set so that only 5 percent of blank results exceed it. The limit of detection is the lowest concentration that reliably gives a result above the limit of blank. The limit of quantitation is the lowest concentration measured with an agreed precision, often taken as a typical scatter of repeated results no larger than 20 percent of their average. Chapter 5 shows where each comes from, and Chapter 23 how many measurements it takes to establish them.

The FDA cleared the US version of Roche's troponin T assay in 2017 for use on two analyzers. An assay is cleared when the FDA finds it substantially equivalent to one already legally sold, through a submission called a 510(k), and a decision summary is the FDA's public summary of the evidence it reviewed. For this assay, the decision summary gives a limit of blank of 2.5 to 3 ng/L, a limit of detection of 3 to 5 ng/L depending on the analyzer, and a limit of quantitation of 6 ng/L. Its 99th-percentile upper reference limit, the value that only one healthy person in a hundred exceeds, was 19 ng/L overall, 14 ng/L for women and 22 ng/L for men, from 1,301 healthy adults. The European Society of Cardiology's 2020 guidelines place the cut-off for ruling out a heart attack with this assay lower still, below 5 ng/L.

02 — Where Roche's troponin T assay can see
measured precisely, still normal a change here is visible before anything is 'raised' 0 5 10 15 20 25 99th percentile, women 14 overall 19 men 22 limit of blank 2.5–3 limit of detection 3–5 limit of quantitation 6 ESC 2020 rule-out cut-off for this chemistry: below 5 ng/L TROPONIN T, ng/L (US LABEL, FDA 2017) LoB, LoD, LoQ by analyzer: cobas e 601 and e 411
AnalyteReadoutFocal detail
Plate 02 — For the US version of Roche's troponin T assay, the limits of blank, detection and quantitation sit between 2.5 and 6 ng/L, below the 99th-percentile limits of 14 ng/L for women and 22 ng/L for men. A floor below the "raised" limit lets a test see a normal value change between two samples.
The same chemistry can carry two measuring ranges

The 2010 validation of Roche's troponin T assay gave a measuring range from 3 to 10,000 ng/L, starting at the limit of blank; the authors of that study included the maker's staff. In the US clearance the range is 6 to 10,000 ng/L, starting at the limit of quantitation. Nothing in the chemistry differs; the definitions do. A datasheet's lowest number is only comparable with another's when both state which limit they report.

Where these limits sit against the 99th percentile matters, as the plate shows. A test whose limit of detection sat above the 99th percentile could only say "high" or "not high". A test that measures well below the 99th percentile can see a value change between two samples taken an hour apart, even when both are normal, and that change is what the one-hour rule-out in Chapter 20 uses. The floor is therefore more than a figure in a datasheet. It decides which clinical questions a test can answer at all.

Write the target in molecules before choosing a method

A specification that says "measure troponin at 5 ng/L" hides the design problem. Convert it to molecules in the sample volume the method will use, then to molecules that a capture step can reach in the time allowed, then to signal per molecule. Each conversion names a subsystem and a requirement: sample volume and handling, antibody and incubation, label and detector. Chapter 10 follows that chain through an analyzer.

+ What this chapter established
  • Healthy people carry cardiac troponin at about 1 to 50 ng/L, and many healthy women sit below 5 ng/L.
  • At 5 ng/L a 50 µL sample holds about 4.2 million troponin T molecules among about 18 million billion albumin molecules.
  • Every immunoassay must catch the target, give each caught molecule a signal, and separate that signal from everything else.
  • A test's floor is three defined limits, of blank, detection and quantitation, and comparing tests means comparing like definitions.
02 — The antibody

A protein that recognizes a shape.

+ The questionWhat is an antibody, and how can one of them pick a single protein out of thousands?

A Y-shaped protein with two tips

Nearly every immunoassay in this guide uses one kind of antibody, immunoglobulin G. An immunoglobulin G molecule weighs about 146,000 grams per mole and is built from four protein chains: two identical heavy chains of about 50,000 grams per mole each and two identical light chains of about 25,000. The chains fold into a Y. Each of the two arms is formed by one light chain and the upper half of one heavy chain, and the arms are the fragments that bind antigen, the substance an antibody recognizes. The stem, joined to the arms by a flexible hinge, is the part the immune system uses to signal what the antibody has caught. Measured by how it moves through water, the whole molecule behaves like a sphere about 13 nanometers across.

The binding happens at the tip of each arm. There, six loops of the protein chain, three from the heavy chain and three from the light, come together to form a small surface whose shape and chemistry vary from one antibody to the next. The rest of the molecule is nearly the same in every antibody of its class. An immune system can make enormous numbers of different tips because the genes for those loops are assembled from pieces and then mutated as the immune response matures, so that cells making tighter binders are favored.

03 — Two binding sites on one Y-shaped protein
target protein epitope: about 20 amino acids, about 790 Ų binding happens over a patch about 3 nm across heavy chain, about 50,000 g/mol light chain, about 25,000 g/mol light chain, about 25,000 g/mol hinge stem (Fc) arm (Fab) binding site: six loops about 13 nm across in solution schematic
AnalyteAntibodyFocal detail
Plate 03 — An IgG antibody, about 146,000 grams per mole and about 13 nm across in solution, binds through a small surface at the tip of each arm. A typical epitope is about 20 amino acids over about 790 square ångströms, so a protein offers room for more than one antibody at a time.

For an immunoassay the important consequence is that one antibody carries two identical binding sites, and that the binding site is small compared with the antibody itself. The surface of the target that the antibody touches, its epitope, is only a patch of the target, so a protein the size of troponin offers room for several different antibodies to bind at once, each at its own epitope. Chapter 4 shows why that room makes possible the format that can detect the smallest amounts.

Recognition is shape and chemistry over a small patch

A 2008 survey of 53 crystal structures of antibodies bound to protein targets found that a typical epitope contains about 20 amino acids of the target and covers about 790 square ångströms. An ångström is a tenth of a nanometer, so that is an area about 3 nanometers on a side. Three in four epitopes fell between 15 and 25 amino acids and between 600 and 1,000 square ångströms. Most epitopes on proteins are not a continuous stretch of the chain but amino acids, or residues, from different parts of it, brought together by the way the protein folds. And only a fifth to a third of the residues in the antibody's loops actually touch the target.

Binding over such a patch is the sum of many weak contacts: hydrogen bonds, attractions between opposite charges, and the close packing of surfaces that fit. Each contact alone would break apart in an instant at body temperature. Together, when the shapes match, they hold the two molecules for minutes to hours, and a mismatch of a few atoms removes enough contacts to weaken the hold by a large factor. That sharp loss is the basis of an antibody's analytical specificity, meaning which shapes it binds and which it does not. Recognition is not a lock that admits one key but a fit whose strength falls off quickly as the shape departs from the one the antibody was selected against.

The strength of the fit is measured by the dissociation constant: the concentration of target at which half of the antibody's sites are occupied at equilibrium, when binding and release balance. Smaller is tighter. In 1980 the company Hybritech filed the founding patent for two-antibody assays built from single-clone antibodies, which Chapter 4 describes. The patent required each antibody to bind with an affinity, or binding strength, equivalent to a dissociation constant of 10 nanomolar or less. A 2010 analysis of single-molecule assays assumed antibodies a thousand times tighter, at 10 picomolar. Chapter 3 shows what such numbers mean for how much target an assay can catch.

One clone, one epitope

An animal immunized with a protein makes many different antibodies against it, each from a different line of antibody-producing cells and each binding its own epitope. Serum is the liquid left when blood clots, and serum collected from the animal holds this mixture, a polyclonal antibody. The mixture binds strongly because its antibodies hold the target at many sites, but no two animals make the same mixture, and a supply ends when the animal's serum runs out.

Working at the Medical Research Council's Laboratory of Molecular Biology in Cambridge, Georges Köhler and César Milstein fused antibody-producing cells from an immunized mouse with tumor cells that grow without limit. Each fused cell line, a hybridoma, produced one antibody indefinitely. Nature received their paper on 14 May 1975 and published it on 7 August, and in 1984 Köhler and Milstein received shares of the Nobel Prize in Physiology or Medicine for the work. A monoclonal antibody binds one epitope, and the same cell line can supply the same antibody for decades. That consistency, more than any gain in strength, is what turned immunoassays into products whose results could be compared from one reagent lot to the next.

Choosing what the antibody sees

Troponin shows how much rests on the choice of epitope. The troponin complex that controls muscle contraction has three proteins, T, I and C, and both heart and skeletal muscle make forms of T and I. The heart's forms differ from the skeletal ones, notably at one end of the chain; human cardiac troponin T has 298 amino acids and cardiac troponin I has 210. An assay for heart damage must bind the cardiac forms and ignore the skeletal ones, which injured muscle elsewhere in the body releases.

The first generation of the troponin T assay did not fully manage that: one of its antibodies cross-reacted with skeletal troponin T, binding it in addition to the cardiac form. The second generation, described in 1997, replaced that antibody with a different one specific for the cardiac form. Three of the 1997 paper's nine authors worked at the Boehringer Research Center in Tutzing. The 1997 paper reported no cross-reactivity with purified skeletal troponin T at 1,000 micrograms per liter, two hundred thousand times the 5 ng/L of Chapter 1, and no false results in 43 marathon runners.

04 — One antibody pair, several circulating forms
first hours after injury: mostly the complex later: more fragments 1 THREE-PROTEIN COMPLEX (T, I, C) 2 T RELEASED FROM THE COMPLEX 3 FRAGMENT CUT THROUGH ONE EPITOPE T I C capture detection T T a form that lost one epitope is invisible to this pair COUNTED COUNTED NOT COUNTED forms over time: after a 2021 review schematic; proportions vary between studies
AnalyteAntibodyLabel and signalReadoutFocal detail
Plate 04 — Troponin leaves damaged heart muscle largely as a three-protein complex and later also as smaller complexes and fragments. A two-antibody assay counts a molecule only when both of its epitopes are intact, so two assays can count different subsets of the same troponin.

The cardiac forms also do not circulate as one molecule. A 2021 review whose authors work at an antibody supplier describes how the circulating forms change over time. In the first hours after a heart attack, much of the released troponin travels as the three-protein complex; later, more of it circulates as smaller complexes and fragments cut by enzymes. Troponin I is mostly bound to troponin C in the blood, and an international task force on cardiac biomarkers has advised that troponin assays should recognize every circulating form equally. In practice each pair of antibodies recognizes a subset, defined by where its two epitopes sit on the chain and whether those regions survive cutting.

Two antibodies against troponin may not see the same troponin

Assays that both report "troponin I" can bind different epitopes, and a fragment that keeps one assay's epitopes may have lost the other's. The two assays then measure overlapping but different populations of molecules, and each converts its signal into a concentration against a different calibration material. Chapter 14 shows how far apart their numbers can be.

Roche's troponin T assay uses two monoclonal antibodies against cardiac troponin T. One is made in mouse and the other is a mouse and human hybrid, and the FDA's decision summary lists each at 2.5 milligrams per liter in its reagent. Those amounts make the antibody far more plentiful than the target.

Solved example: antibody against target

The decision summary gives no molecular weight, so assume the 146,000 grams per mole given at the start of this chapter for immunoglobulin G, or IgG. Then 2.5 milligrams per liter is 0.0025 ÷ 146,000 moles per liter, about 17 billionths of a mole per liter, or 17 nanomolar. Troponin T at 5 ng/L is 0.14 picomolar (Chapter 1). Dividing 17 nanomolar by 0.14 picomolar gives about 120,000. Each troponin molecule is surrounded by about 120,000 antibody molecules of each kind, so the target never runs short of binders; what limits the catch is how fast the two meet, the subject of Chapter 3.

Specify epitopes, not only the target

A reagent specification that names "anti-troponin I" is incomplete. It should name the epitopes, the circulating forms each antibody pair recognizes, the cross-reactivity panel covering skeletal forms, fragments and complexes, and the antibody's source and clone, since a change of clone is a change of the test. Chapter 24 shows how regulators treat such a change.

+ What this chapter established
  • An IgG antibody is a 146,000 g/mol Y-shaped protein with two identical binding sites, one at each arm's tip.
  • Binding is the sum of many weak contacts over an epitope of about 20 amino acids, so a small change in shape weakens it sharply.
  • Monoclonal antibodies bind one epitope and come from one cell line, which made reagent lots consistent.
  • For troponin the choice of epitopes decides which circulating forms are seen and whether skeletal muscle troponin is ignored.
03 — Binding takes time

Hold and let go.

+ The questionIf an antibody grips its target so tightly, why does the time an assay allows change how much of the target it catches?

Binding is a balance

Placed in a small tube with an excess of antibody, each troponin molecule of Chapter 1 is close to thousands of binding sites, the fit is good, and the bond, once made, lasts for minutes or hours. It seems the target should be caught almost at once and kept for good. Neither is true. The reason governs every incubation, the period in which sample and reagents are left to react, and every wash and every protocol in the analyzers of Part III.

Two molecules bind only when diffusion brings them together in the right orientation, and a bound pair comes apart whenever the jostling of the surrounding water happens to break enough of its weak contacts at once. Both events happen all the time. The rate of binding is the association rate constant: for proteins meeting proteins, antibodies included, it is typically about a million per molar per second, a hundred to a thousand times slower than the limit that pure diffusion would allow, because most collisions arrive at the wrong angle. The rate of coming apart is the dissociation rate constant: for good antibodies, between about one in a thousand and one in a hundred thousand per second.

The dissociation rate constant divided by the association rate constant is the dissociation constant of Chapter 2. An antibody that binds at a million per molar per second and lets go at one in ten thousand per second therefore has a dissociation constant of 0.1 nanomolar. The dissociation rate also gives the lifetime of a single bond. Half of a set of bound pairs comes apart in about 0.7 divided by the dissociation rate constant: about 12 minutes at one in a thousand per second, about two hours at one in ten thousand and about 19 hours at one in a hundred thousand. Binding is never permanent; it is a balance that shifts with concentration and time.

How much is caught at equilibrium

When the target is scarce and the antibody plentiful, the fraction of target bound at equilibrium depends only on the antibody concentration measured against the dissociation constant. It equals the antibody concentration divided by the sum of the antibody concentration and the dissociation constant. Roger Ekins spent much of his career on the limits of immunoassays. He worked out what this rule gives in numbers: antibody at a hundredth of the dissociation constant binds about 1 percent of the target, at a tenth about 9 percent, at the dissociation constant itself about half, and at twenty times the dissociation constant more than 90 percent.

In a sandwich assay, the format of Chapter 4, two antibodies hold each target molecule between them. Sandwich assays are designed to sit at the top of that range: they use far more antibody than the dissociation constant, so that almost all the target can be caught. Roche's troponin T assay carries its antibodies at about 17 nanomolar in the reagent, the figure worked out in Chapter 2, and at several-fold less once sample, reagents and beads are mixed. For an antibody with a dissociation constant of 0.1 nanomolar, even a few nanomolar is many times the constant, and more than 95 percent of the troponin would be bound at equilibrium. The question is how long equilibrium takes to arrive.

How long equilibrium takes

With the antibody in excess, the approach to equilibrium follows a simple rule. The observed rate is the association rate constant times the antibody concentration, plus the dissociation rate constant, and half of the remaining distance to equilibrium is covered in 0.7 divided by that rate. After five such half-times the binding is about 97 percent complete, a rule of thumb from a 2020 guide to measuring binding. The more antibody, the faster the approach, which is a second reason for the large excess.

Solved example: minutes to equilibrium

Take an association rate of one million per molar per second and a dissociation rate of one in ten thousand per second. With antibody at 1 nanomolar, the association rate constant times the concentration is 0.001 per second, and adding the dissociation rate of 0.0001 gives an observed rate of 0.0011 per second. Half-way to equilibrium then takes 0.7 ÷ 0.0011, about 630 seconds or 10.5 minutes, and 97 percent takes five times as long, about 53 minutes. With antibody at 17 nanomolar, the concentration in the reagent of Roche's assay before mixing, the observed rate is 0.017 per second, the half-time is about 41 seconds, and 97 percent arrives in about three and a half minutes. Seventeen times more antibody makes binding in a well-mixed solution about fifteen times faster. Once the antibody is diluted several-fold in the reaction, binding takes several times longer, which is why the minutes of a protocol are budgeted with care.

05 — More antibody, faster capture
0 % 25 % 50 % 75 % 100 % 0 10 20 30 40 50 60 FRACTION OF TARGET BOUND TIME, min 9 min 18 min equilibrium at 1 nM: 91 % ANTIBODY 17 nM (ROCHE'S REAGENT) ANTIBODY 1 nM half-time 10.5 min at 9 min, 1 nM antibody has caught about 41 % computed: association 10⁶ per molar per second, release 1 in 10,000 per second
AnalyteFocal detail
Plate 05 — With an association rate of a million per molar per second, antibody at 17 nanomolar binds almost all of the target within about three minutes, while antibody at 1 nanomolar needs most of an hour. Excess antibody is bought for speed as much as for completeness.

That calculation assumes that every antibody and every troponin molecule are free in a well-mixed liquid. In many immunoassays they are not: one antibody is fixed to a surface, the wall of a plastic well or the surface of a bead, and the target has to travel to it. Then the time the target takes to reach the surface starts to matter as well.

Distance is time

How fast a molecule spreads through water by diffusion is measured by its diffusion coefficient. For a protein the size of troponin T it is about 0.0000007 square centimeters per second, a figure scaled from the measured value for albumin. The distance a molecule wanders grows only with the square root of time: about 0.7 millimeter in an hour, and four times that distance takes sixteen times as long. In a standard plastic well holding 100 microliters, the liquid stands about 3.2 millimeters deep. A troponin molecule near the top of an unstirred well needs most of a day to reach antibodies coating the bottom, so in an hour's incubation the antibody on the walls catches target only from a thin layer of the liquid next to it.

06 — Distance is time
ANTIBODY ON THE FLOOR OF A WELL CAPTURE BEADS SPREAD THROUGH THE LIQUID 100 µL, about 3.2 mm deep from the top: most of a day (about 19 h) an hour's diffusion reaches only this layer reached in about 1 hour: about 0.7 mm beads a few micrometers across, everywhere in the sample diffusion of a troponin-sized protein: about 0.0000007 cm² per second; distance grows with the square root of time
AnalyteAntibodyFocal detail
Plate 06 — A protein the size of troponin diffuses about 0.7 mm in an hour, so antibody coating the floor of a 3.2 mm-deep well samples only a thin layer of the sample in that time. Beads spread through the liquid bring the capture surface to the target instead.

Laboratory assays in plates make up for slow diffusion with long incubations, shaking and warmth. Analyzers shorten the distance instead. In Roche's troponin T assay, the two antibodies find the troponin while all three molecules are dissolved, which is fast because the antibody concentration is high. Beads coated with streptavidin then collect the complexes. Streptavidin is a bacterial protein that binds biotin, a small vitamin, almost irreversibly, and one of the two antibodies carries biotin. Depending on the analyzer, the beads are added after the first incubation or are present from the start. The beads, a few micrometers across, are spread through the whole reaction volume, so no molecule is ever more than a short distance from one. A 2008 analysis of surface sensors made the general point: transport of the target to the surface "can play as critical a role as the chemical reaction itself".

A complex can come apart during the wash

Time works in both directions. A bound pair with a dissociation rate of one in a thousand per second loses about 6 percent of its target in each minute of washing, because the wash removes free target and nothing rebinds. Antibodies for washed assays are therefore chosen for slow release as much as for tight binding, and wash steps are kept short.

Roche's analyzers show the trade-off between time and floor in their protocols. Each test on them takes 9, 18 or 27 minutes, and the troponin T assay comes in a 9-minute version for emergencies and an 18-minute version for routine work. The International Federation of Clinical Chemistry compiled a table of figures supplied by the makers. In that table the 18-minute version on the cobas e 601 analyzer family had a limit of detection of 2.05 ng/L and the 9-minute version 2.85 ng/L, while on the newer cobas e 801 both were listed at 3 ng/L. These are the figures Roche supplied for the Federation's 2019 table, and they are lower than the 5 ng/L of the assay's European label in Chapter 1. Shorter incubations can raise the floor, by an amount that depends on the whole system.

Budget the minutes

An assay protocol is a time budget split between binding, capture, washing and reading. When a project shortens it, the first questions are how close each step was to equilibrium, which step was limited by diffusion rather than chemistry, and how much target a wash of the new length will strip. Answering them with measured binding curves, not with the old protocol's habits, saves a failed verification later (Chapter 23).

+ What this chapter established
  • Binding and release both happen continuously; for good antibodies, release half-times run from minutes to many hours.
  • With antibody far above the dissociation constant, nearly all the target is bound at equilibrium, and more antibody gets there faster.
  • When an antibody sits on a surface, diffusion over millimeters can take hours, so analyzers bind in solution and spread capture beads through the sample.
  • Shorter protocols can raise the floor, and washes strip some target, so incubation times are a design trade, not a habit.
04 — Two formats

Sandwich and competition.

+ The questionWhy are large proteins measured between two antibodies, while small molecules are measured by competing for one?

A binder nobody wanted to call an antibody

At the Radioisotope Service of the Veterans Administration Hospital in the Bronx, New York, Solomon Berson and Rosalyn Yalow injected insulin labeled with radioactive iodine into patients and followed how fast it left the blood. In people who had been treated with insulin for diabetes, the labeled hormone disappeared much more slowly than in people who had never received it. Electrophoresis showed why. The method separates proteins by how they move in an electric field, and in the plasma it showed the labeled insulin bound to a globulin, one class of blood protein. The binder appeared only in patients who had been injected with insulin, and it was present in every patient treated for three months or more.

In the body of their paper, the two researchers concluded that the binder met a textbook definition of an antibody, and they called it an "insulin-transporting antibody". The immunologists of the time did not accept that insulin could provoke antibodies at all. Science rejected the paper. The Journal of Clinical Investigation also rejected it at first, and accepted it on 17 October 1955 only after the title was changed to speak of an "insulin binding globulin". Yalow reproduced the rejection letter in her Nobel lecture of 8 December 1977. Berson had died in 1972. Yalow received half of the 1977 Nobel Prize in Physiology or Medicine, for the development of radioimmunoassays of peptide hormones.

Competition: measuring what the sample displaces

The binding that the editors doubted became a measuring method, the radioimmunoassay, an immunoassay whose label is radioactive. In a letter to Nature published on 21 November 1959, and in a full paper in July 1960, Yalow and Berson used serum from guinea pigs immunized with beef insulin to measure insulin in human plasma. Their mixtures held a fixed, small amount of antibody, a trace of insulin labeled with radioactive iodine at about 0.05 to 0.15 nanogram per milliliter, and the patient's plasma diluted tenfold. Over four days at 4 °C, the unlabeled insulin from the plasma and the labeled insulin competed for the same limited binding sites. The researchers then separated bound from free insulin on strips of paper, by a combination of chromatography and electrophoresis, and counted the radioactivity in each.

The more insulin the plasma contained, the less of the labeled insulin found a site, and the lower the ratio of bound to free label. Comparing that ratio with the ratios given by known amounts of insulin gave the patient's concentration. This method is a competitive immunoassay: the label is a labeled copy of the target, the antibody is scarce on purpose, and the signal falls as the concentration rises.

Solved example: what "less than one microunit" meant

The 1960 paper reported that the method could detect less than one microunit of insulin, and about 0.1 to 0.2 microunit with more dilute antiserum, enough to measure insulin in 10 to 20 microliters of plasma. In that paper, fasting adults averaged 21 microunits per milliliter. The unit of insulin was first defined by biological activity, and one microunit is about 40 picograms. With insulin weighing about 5,808 grams per mole, 40 picograms is about 6.9 femtomoles, where a femtomole is a thousand attomoles. That is about 4 billion molecules. A fasting sample of that size held about 0.2 to 0.4 microunits, one to two billion molecules, so the first radioimmunoassay worked with a few hundred times the four million troponin molecules of Chapter 1.

07 — Insulin from plasma pushes labeled insulin off the antibody
1 NO INSULIN IN SAMPLE 2 SOME INSULIN 3 MORE INSULIN labeled insulin (radioactive iodine) antibody, scarce on purpose displaced label, free insulin from plasma most label free most sites hold plasma insulin fixed antibody, a trace of labeled insulin (about 0.05–0.15 ng/mL), plasma diluted tenfold; 4 days at 4 °C separated on paper: chromatography plus electrophoresis bound free radioactivity counted in each spot BOUND ÷ FREE LABEL INSULIN IN SAMPLE schematic less label bound means more insulin
AnalyteAntibodyLabel and signalReadoutFocal detail
Plate 07 — In the radioimmunoassay of 1959 and 1960, insulin from plasma competed with a trace of radioactive insulin for a scarce antibody over four days at 4 °C. Bound and free label were separated on paper, and a falling ratio of bound to free label meant more insulin.

Competition has a structural weakness. When the sample holds no target at all, the bound signal is at its largest, since the labeled copies meet no competition for the limited sites. A small amount of target shows up as a small decrease in a large signal, and a small decrease is hard to tell from the ordinary scatter of a large count. A competitive assay can be made to detect smaller amounts by using less antibody and less label, but only up to the point where the count becomes too small to measure precisely.

A competitive assay reports more target as less signal

In a competitive assay the calibration curve, the curve that converts signal into concentration, slopes downward. A fault that lowers the signal, such as a weak label or a lost reagent, therefore reads as a high result. In a sandwich the same fault reads as a low result. Interpreting an unexpected result starts with knowing which way the curve runs.

The sandwich: two antibodies, two epitopes

The alternative puts the label on the antibody instead of on a copy of the target. In 1968 L. E. M. Miles and C. N. Hales at the University of Cambridge described immunoassays with labeled antibodies, and the form that dominates today followed: a sandwich, or two-site, assay. One antibody, the capture antibody, is fixed to a solid surface or carries a tag that a coated surface binds. A second antibody, carrying the label, binds a different epitope on the same target. Only a target molecule bound by both holds the label to the surface, so the signal rises with concentration and, ideally, starts from zero.

Monoclonal antibodies made the sandwich practical. In August 1980 Hybritech filed a patent claiming two-site assays built from two monoclonal antibodies against different sites, each binding with an affinity of at least about 10⁸ liters per mole, which is equivalent to a dissociation constant of about 10 nanomolar or less. The patent was granted on 8 March 1983. Polyclonal reagents could be used for sandwiches too, but they had to be purified laboriously to make a capture and a detection antibody that did not compete for the same sites.

A sandwich needs two epitopes far enough apart for two antibodies, each about 13 nanometers across, to bind at once. Proteins offer that room. Small molecules, with masses below about 1,000 grams per mole, do not. That is why drugs, steroids and thyroid hormones of that size are still measured by competition. A 1998 patent from the immunoassay maker Chiron Diagnostics put the difference in numbers. By its account, the most sensitive sandwich assays then in use detected several million molecules of thyroid-stimulating hormone per test cuvette, and the most sensitive competitive assays, for digoxin and the thyroid hormone T3, detected several billion: a gap of about a thousandfold.

08 — Competition starts high; a sandwich starts near zero
COMPETITIVE labeled target sample target scarce antibody SIGNAL TARGET CONCENTRATION (LOG) schematic zero target: maximum signal SANDWICH (TWO-SITE) label detection antibody target capture antibody SIGNAL TARGET CONCENTRATION (LOG) schematic background: label stuck without target zero target: only the background a little target is a large relative change most sensitive in the late 1990s: competitive, several billion molecules per test; sandwich, several million (a maker's patent)
AnalyteAntibodyLabel and signalInterferentReadoutFocal detail
Plate 08 — A competitive assay measures how much labeled target the sample displaces, so a little target is a small dent in a large signal. A sandwich holds label only through the target, so its signal rises from near zero; by the late 1990s the best sandwich assays detected about a thousand times fewer molecules.

The troponin T assay's format, and how it began

Troponin T, a protein of 298 amino acids, was already measured by sandwich in its first commercial assay. In 1992 Hugo Katus and colleagues, together with staff of the manufacturer Boehringer Mannheim, described that company's troponin T immunoassay. Boehringer Mannheim's diagnostics business is now Roche Diagnostics. It used two monoclonal antibodies: a capture antibody carrying biotin, held by streptavidin coated on the inside of a plastic tube, and a detection antibody carrying an enzyme. It ran in 90 minutes at room temperature and measured from 0.1 to 15 micrograms per liter.

The version of Roche's troponin T assay cleared in the US in 2017 keeps the same architecture. Its capture antibody still carries biotin, its beads are coated with streptavidin, and its second antibody still carries the label, now a ruthenium compound read by the method of Chapter 9. It runs in 9 minutes and quantifies from 6 ng/L. The lower end of the 1992 range, 0.1 micrograms per liter, is 100 ng/L. In 25 years, then, the bottom of the reported range fell about seventeenfold, to the 6 ng/L limit of quantitation where the US range starts, while the time fell tenfold, with the same two-antibody format throughout.

One property of the one-step sandwich needs a name here because later chapters return to it. In a one-step sandwich both antibodies meet the sample together. At very high target concentrations, free target molecules can then occupy the capture and the labeled antibodies separately, so fewer complexes carry both, and the signal falls instead of rising. This high-dose hook can make an enormous concentration read as a modest one. Chapter 13 shows how analyzers detect it, and Chapter 19 what it does to a patient's result.

Choose the format from the size of the molecule

For a target below about 1,000 grams per mole, plan a competitive assay and budget for its weaker floor. For a protein, look for two antibodies against well-separated epitopes that survive the forms in circulation, and decide early between a one-step design, which is faster but can hook, and a two-step design with a wash between capture and detection.

+ What this chapter established
  • Berson and Yalow found insulin antibodies in 1955 and turned them into the first radioimmunoassay of insulin in plasma in 1959 and 1960.
  • A competitive assay measures how much labeled target the sample displaces, so its signal falls as concentration rises and starts from a maximum.
  • A sandwich assay needs two antibodies on separate epitopes; its signal rises from near zero, and by one maker's 1998 count the best sandwich assays detected about a thousand times fewer molecules than the best competitive assays.
  • Roche's 2017 troponin T assay keeps the 1992 architecture of two antibodies, biotin and streptavidin, with a lower floor and a tenth of the time.
05 — Background

What stays without being asked.

+ The questionWhen a sample holds none of the target protein, why does the instrument still read a signal, and why is that reading the real floor?

The blank that is not blank

The Institute of Isotopes in Budapest makes a sandwich assay kit for prostate-specific antigen, a protein made by the prostate gland, with a radioactive label on its detection antibody. The kit's instructions print an example calibration, the signals measured for standards of known concentration. The current version of the instructions dates from 2020. Each tube receives 25 microliters of sample and 100 microliters of labeled antibody, and the labeled antibody gives 392,944 counts of radioactivity a minute. After the incubation and two washes, the tube with a standard containing no antigen at all still holds 292 counts. The tube with 0.1 nanogram per milliliter holds 630. The difference made by the antigen, 338 counts, is about the same size as the signal from nothing.

The signal of the zero standard is not a fault of that kit. Every immunoassay has such a signal, called its background, and has to keep it low. A label attached to an antibody gives its signal whether or not the antibody holds its target. It reports only its position, and some labeled antibodies end up on the surface without any target to hold them there. The 292 counts include the counter's own background, the counts it records with no sample in it. They are 0.074 percent of the label added: a small fraction, but a fraction of a very large number. And at the bottom of the range, the signal from label held by the target is no larger than this residue.

Separating bound from free: the wash

Every sandwich and competitive assay described so far has a step in which bound label is separated from free label. Yalow and Berson did it on paper strips. Most assays since have done it with a solid phase, a surface to which the capture antibody is fixed: the inside of a plastic tube or well, or the surface of beads. Once the target has bound to the capture antibody and the labeled antibody has bound to the target, the liquid is drawn off and replaced with a clean buffer, once or several times. That rinse is the wash. What the surface holds stays; what floats free leaves with the liquid.

Washing is a dilution, not a filter. Each rinse removes most of the free label that remains, and the residue after several rinses depends on how much liquid clings to the surface between them, how long each takes and how well the liquid mixes. Beads allow one more method: a magnet holds them against the wall of the vessel, or against an electrode, while the liquid around them is exchanged. Lateral flow test strips wash by flow, with the sample stream itself carrying unbound label past the capture line (Chapter 15). A few designs avoid washing altogether by making the label give a different signal when it sits next to its partner (Chapter 8).

09 — What a wash removes, and what it cannot
1 CAPTURE target caught other proteins float 2 ADD LABELED ANTIBODY labeled antibody binds one stuck, no target 3 WASH magnet liquid exchanged, bead held 4 READ SIGNAL label held without target survives every wash schematic; most washes repeat step 3 several times
AnalyteAntibodyLabel and signalInterferentReadoutFocal detail
Plate 09 — A wash exchanges the liquid while the surface or a magnet holds the solid phase, so free label leaves and bound label stays. Label stuck to the surface without any target stays too, and it is what the instrument reads from a blank.

Why label sticks where it should not

A wash removes most free label, never all of it, and none of the label that something holds. In a blank, two things leave label behind. Labeled antibody adsorbs directly onto the plastic or the bead, because proteins stick to most surfaces. And a small amount of label stays in the liquid film that clings after the last rinse. In a patient's sample a third source can join them. Some molecules in a sample can hold labeled antibody to the capture antibody with no target at all, and one example is a human antibody that recognizes the animal antibodies used as reagents. Such errors arise only in particular samples, and Chapter 19 treats them.

Assay makers counter adsorption with blockers: inert proteins such as bovine serum albumin or milk proteins, at a few percent, coat any surface the capture antibody left bare, and mild detergents such as Tween 20, at a tenth of a percent or so, discourage loose adsorption. Interfering human antibodies are countered by adding animal immunoglobulin, which they bind in place of the reagent antibodies. Labeled antibody sticking where no target holds it is called nonspecific adsorption. A 1998 patent from Chiron Diagnostics stated the general principle: reducing the nonspecific adsorption of the labeled antibody "to the solid phase and the reaction vessel" improves a labeled-antibody assay's analytical sensitivity, its ability to detect small amounts.

The consequence for design is the central claim of this guide's answer. Making each label brighter multiplies the signal from target-bound label and the signal from wrongly bound label by the same factor. The ratio between them does not change. A brighter label lowers the floor only while some other source of noise is larger than the zero signal. Such other sources include light reaching the detector from elsewhere and the detector's own electrical noise. Once the zero signal dominates, only less wrongly bound label, or a steadier count of it, lowers the floor further.

A brighter label raises the background too

A datasheet's claim of a "more sensitive" label means something only when the datasheet also gives the blank it was measured against.

The limit of blank, and the floor it sets

The zero signal is not one number but a spread. Measured many times, blank samples give a range of results, because the counts scatter, the wash leaves slightly different residues and the instrument drifts. A result just above the average blank therefore does not prove that any target was present. The Clinical and Laboratory Standards Institute writes consensus standards for clinical laboratories. Its guideline on detection limits is EP17. The second edition, published in 2012 and reaffirmed in 2017, was still current in October 2026, and the FDA has recognized it since January 2013. EP17 defines the limit of blank as the result that blanks exceed only 5 percent of the time: in its simplest form, the mean of the blanks plus 1.645 times their standard deviation, the usual measure of how widely results scatter around their mean. The limit of detection is then the concentration whose results exceed the limit of blank 95 percent of the time, found by adding 1.645 times the scatter of low-level samples.

Solved example: from blanks to a limit of detection

Suppose, for illustration, that blank samples give a mean of 1.0 ng/L with a standard deviation of 0.6 ng/L. The limit of blank is 1.0 plus 1.645 times 0.6, which is 2.0 ng/L. If samples with a little troponin scatter with a standard deviation of 0.8 ng/L, the limit of detection is 2.0 plus 1.645 times 0.8, about 3.3 ng/L. Halving the scatter of the blank, by washing more evenly or counting more light, would bring the limit of blank to 1.5 ng/L and the limit of detection down with it, without touching the antibody or the label.

10 — From blank scatter to a limit of detection
0 1 2 3 4 5 6 RESULT, ng/L illustrative values from the solved example; normal curves computed limit of blank 2.0 = 1.0 + 1.645 × 0.6 limit of detection 3.3 = 2.0 + 1.645 × 0.8 results from blank samples results from a sample at the limit of detection 5 % 5 % above this, a blank is unlikely
InterferentReadoutFocal detail
Plate 10 — Blank results scatter, so the limit of blank is set where only 5 percent of them fall above it, and the limit of detection where 95 percent of results from a low sample exceed the limit of blank. With the solved example's numbers they are 2.0 and 3.3 ng/L.

For Roche's troponin T assay, the FDA's 2017 decision summary describes how the limit of blank was set. For each reagent lot, blank samples were run over at least three days on two analyzers of each type to give 60 results. The limit of blank came out at 2.5 ng/L on one type of analyzer and 3 ng/L on the other. The limits of detection were 3 and 5 ng/L on the same two types. The limit of quantitation, 6 ng/L, was set where the standard deviation of repeated results fell to 20 percent of their mean.

How much the definition matters is plain in the isotope kit's instructions. They give an older-style "sensitivity" of 0.02 ng/mL, calculated as the mean of the zero standard plus two standard deviations from 20 repeats. They also give an EP17 limit of blank of 0.035, a limit of detection of 0.09 and a limit of quantitation of 0.12 ng/mL. All four figures describe the same tubes.

Report the floor with its recipe

A detection claim is only comparable when it states which of the three limits it is, how many blank and low samples went into it, over how many days, lots and instruments, and which statistical rule set it. Chapter 23 gives the designs regulators expect; inside a development project, the same recipe lets two prototypes be compared fairly.

+ What this chapter established
  • A sample with none of the target still gives a signal, from labeled antibody held by the surface or left in the liquid after rinsing; a patient's own antibodies can add errors of their own.
  • Washing separates bound from free label by repeated dilution, with beads held by magnets and strips washed by flow.
  • A brighter label lifts target signal and background together, so past a point only less wrongly bound label or a steadier count lowers the floor.
  • The limit of blank is set by the scatter of blank results, and the limits of detection and quantitation build on it.

+ Part II · Making a bond visible

Four ways to read a label.

In 1971 two groups independently published ways to replace the radioactive label of the first immunoassays with an enzyme, a catalyst that keeps working for as long as it has something to work on. One group worked at Stockholm University and the other in the Netherlands. Their idea opened a sequence of labels, each giving more signal per bound molecule against less unwanted light. This part follows four of them in order: an enzyme read by the color it makes, molecules that give off light in a chemical reaction, labels that glow long after a flash of exciting light, and a label switched on by a voltage at an electrode, the method of Roche's troponin T assay. By the end of this part, the floor of Chapter 5 can be stated exactly: a label's light measured against everything else that reaches the detector.

06 — Absorbance

Color from an enzyme.

+ The questionHow does an enzyme turn a few bound antibodies into a color a photometer can read, and where does color run out?

An enzyme as an amplifier

Horseradish peroxidase is the enzyme most often attached to antibodies in color-based immunoassays. A single molecule of it can convert a few hundred to a few thousand molecules of a colorless substrate, the molecule the enzyme acts on, into colored product every second. Measurements put the rate at about 240 per second near neutral pH and room temperature, while figures more than ten times higher are often quoted for the acidic, warmer conditions in which the enzyme works fastest. Over a half-hour incubation, even the lower rate lets each enzyme molecule make some 400,000 colored molecules.

That multiplication is why enzymes replaced radioactivity. A radioactive iodine atom on an antibody gives at most one decay, and an iodine-125 label decays slowly, so in a counting time of a minute only a small fraction of the labels on a tube decay at all. Radioactive labels also needed shielding, licenses and disposal, and they lost activity on the shelf. An enzyme label, by contrast, is stable and safe, and it keeps adding signal for as long as substrate remains. The signal of each bound antibody is no longer one event but an amount of colored product that grows with time.

The amplification has a cost. The enzyme makes product at the same rate whether it is attached to an antibody that holds a target or to one stuck on the plastic. Enzyme amplification therefore raises the signal from wrongly bound label exactly as much as the signal from target. That is the argument of Chapter 5, met here in its first practical form.

The first ELISA

The enzyme-linked immunosorbent assay, ELISA, was first worked out by Eva Engvall, then a graduate student of Peter Perlmann at Stockholm University. In a retrospective published in 2010, she recalled obtaining the first calibration curve early in 1970. The assay was a competitive one, adapted from a standard radioimmunoassay, with antibodies coupled to particles of cellulose; bound and free label were separated by repeated centrifugation and washing. The enzyme was alkaline phosphatase. It was chosen because its substrate dissolved readily and could reveal small amounts of the enzyme, and because the product could be measured. The target was rabbit immunoglobulin G. "We were continually asked why anyone would want to measure rabbit IgG!" she wrote.

The paper appeared in 1971 in the journal Immunochemistry. In the Netherlands, B. K. van Weemen and A. H. W. M. Schuurs published an enzyme immunoassay the same year, and a 2005 history describes the two methods as developed independently and simultaneously. Most laboratories now picture an ELISA as a sandwich assay in the wells of a plastic plate. That form came later, as plates, monoclonal antibodies and plate readers arrived. The principle did not change: catch, wash, add substrate, and measure the color.

11 — Catch, wash, add substrate, measure the color
1 CATCH other proteins target capture 2 WASH unbound: washed away 3 LABEL, WASH AGAIN enzyme, e.g. peroxidase stuck without target 4 SUBSTRATE AND COLOR substrate lamp detector ABSORBANCE it makes color at the same rate 1971: Engvall and Perlmann; van Weemen and Schuurs (both competitive at first) peroxidase: about 240 products per second near neutral pH, room temperature half an hour: some 400,000 colored molecules per enzyme
AnalyteAntibodyLabel and signalTriggerInterferentReadoutFocal detail
Plate 11 — An enzyme label multiplies each bound antibody into hundreds of thousands of colored molecules over an incubation. It multiplies the label stuck on the plastic just as much, so the gain over radioactivity comes with the background of Chapter 5.

Reading color: absorbance

A colored solution absorbs part of the light shone through it. A photometer measures how much light of one wavelength passes through the well and how much passed through a blank, and reports the absorbance, a measure of the loss in which each whole unit means a tenfold drop in light. A plate reader is a photometer that holds many wells at once. An absorbance of 1 means a tenth of the light got through, 2 means a hundredth, and 3 a thousandth. Absorbance is proportional to the concentration of the colored molecule, to the length of liquid the light crosses, and to how strongly that molecule absorbs at that wavelength, its molar absorptivity.

The common peroxidase substrate, tetramethylbenzidine, turns blue as the enzyme works, and yellow when the reaction is stopped with acid. Its yellow product absorbs at 450 nanometers with a molar absorptivity of about 59,000 per molar per centimeter, high for a small molecule. The product of alkaline phosphatase, para-nitrophenol, absorbs at about 400 nanometers with 18,380 per molar per centimeter, a value measured in 1980 carefully enough for the compound to serve as a reference material. Plate readers made for this work resolve a thousandth of an absorbance unit, and their makers specify that readings repeat within about half a percent plus 0.005 units.

Solved example: how many enzymes make a visible color

Take a reader's repeatability of 0.005 absorbance units as the smallest change it can trust, and assume a 200-microliter well in which light crosses about 0.6 centimeter of liquid. That depth is an assumption, because it depends on the plate. The colored product then has to reach 0.005 divided by the product of 59,000 and 0.6, about 0.14 micromolar, or roughly 17 million million molecules in the well. Assume one enzyme per captured molecule and 1 million product molecules per enzyme over the incubation, a round figure of the same order as the 400,000 that the section above gives for half an hour. That takes about 17 million enzyme molecules, some 28 attomoles. A troponin sample of Chapter 1 holds 4.2 million target molecules in 50 microliters, fewer than the enzymes this estimate needs.

Laboratory sandwich ELISAs reach low concentrations by putting more sample in, waiting longer and catching more of the target. One maker states that its commercial kit for interleukin 6, an inflammatory protein, reached a sensitivity of 0.7 picogram per milliliter with 100 microliters of sample in an assay that took four and a half hours. Sensitivity here means the lowest concentration the kit can detect. A picogram per milliliter equals a nanogram per liter, so the maker's figure is 0.7 ng/L, lower than the 5 ng/L of Roche's troponin T assay in Chapter 1, although the two figures are for different proteins and may not be defined the same way. Analyzers must answer in minutes, so they cannot lower their floor that way.

Where color runs out

Color fails at both ends. At the low end, a trace of product changes the light leaving the well by a small fraction of a large beam. The reader has to measure that beam twice, through the well and through a blank, and subtract; any flicker of the lamp, any scratch on the plastic and any bubble enters both measurements and survives the subtraction as noise. The enzyme stuck on the plastic makes its own color in the meantime. At the high end, absorbance stops following concentration. Readers report up to 4 absorbance units, but their makers specify linearity only to about 2 or 3, because very little light gets through a dark well and stray light inside the reader starts to dominate.

12 — Color between two walls
0 1 2 3 4 0.01 0.1 1 10 100 1,000 COLORED PRODUCT, µM (YELLOW TMB PRODUCT, 0.6 cm PATH) ABSORBANCE measured curve: schematic reader repeatability about 0.005: the trace of color is lost here proportional (ideal) measured: reader with 0.1 % stray light 0.14 µM ≈ 17 million million molecules in 200 µL 1 = 10 % of light through 2 = 1 % 3 = 0.1 % readers display up to 4; makers specify linearity to about 2–3 stray light takes over
ReadoutFocal detail
Plate 12 — Absorbance follows concentration only between two walls: at the bottom, a trace of color is a tiny dip in a bright beam, and above about 2 to 3 units stray light in the reader takes over. The solved example needs about 0.14 µM of product to clear the lower wall.
Readers lose linearity before their display stops

A plate reader that shows 3.8 does not mean twice the product of a well showing 1.9. Above about 2 absorbance units the reading compresses, so high calibrators and high samples crowd together and a dilution is needed to place them. Checking the reader's specified linear range, not its display range, belongs in setting up any color assay.

Between a noise floor near 0.005 and a ceiling near 2 lies a factor of about 400, less than three factors of ten. A sample whose concentration falls outside that window must be diluted and run again, or read at a different time. The light-emitting labels of the next three chapters improve both ends at once, because they measure light where there should be darkness instead of a dip in a bright beam.

Set the stop time against the top calibrator

In a color assay the enzyme keeps working until the reaction is stopped, so the same plate can be too pale or too dark depending on minutes. Fix the substrate time so that the highest calibrator stays inside the reader's linear range and the lowest stays clearly above the blank, and record that window as part of the method, not as a habit of whoever runs it.

+ What this chapter established
  • An enzyme label turns each bound antibody into hundreds of thousands of colored molecules, and began replacing radioactivity from 1971.
  • The first ELISA was a competitive assay on cellulose particles with alkaline phosphatase, measuring rabbit IgG.
  • Absorbance measures a small dip in a bright beam, so at the low end lamp and plate noise and stuck enzyme set the floor.
  • Readers stay linear only to about 2 or 3 absorbance units, giving color a working range of less than three factors of ten.
07 — Chemiluminescence

Light in the dark.

+ The questionWhy does light from a chemical reaction reach lower concentrations than color does?

A measurement with no lamp

An absorbance reader cannot work in the dark. It needs a lamp, and it reports the target as a small shadow: a slight dimming of the lamp's light as it passes through the sample. If instead the label itself makes the light, and the detector sits in a light-tight box, the measurement works on a different principle. With no label present, the detector sees nothing but its own small electrical noise. Every photon that arrives was made by a reaction in the tube, so the instrument counts upward from near zero instead of subtracting one large reading from another.

A label that makes its own light relies on chemiluminescence: a chemical reaction whose product is formed in an excited state, holding more energy than its stable form, and releases that extra energy as light. Fireflies and some deep-sea animals make light this way with enzymes; immunoassays do it with small organic molecules. Each such label often gives more signal than a color label, but the gain over color does not come from that. It comes from measuring the signal against darkness. A shadow of 0.005 absorbance units is a change of about 1 percent in a bright beam, whereas a light of a few hundred photons a second is unmistakable to a detector that registers a few dozen in the dark.

The label still has to be attached to an antibody and separated from free label by washing, and it still adds to the signal when it sticks where it should not. What changes is how small a quantity of label the instrument can see, and over how wide a range. Two families of chemistry dominate analyzers today: labels that flash once, and enzymes that keep glowing.

Flash: one label, one burst

In 1983 Ian Weeks and colleagues described acridinium esters, small organic molecules that give off light when they react, as labels for immunoassays. They attached up to about 2.8 labels to each monoclonal antibody, and the paper reported that the ester could be detected at about 0.8 attomole, which is roughly 480,000 label molecules. Their two-site assay incubated for 30 minutes and read each sample in 5 seconds.

The light comes when the ester meets hydrogen peroxide in an alkaline solution. In the scheme used on one maker's analyzers, the reagents are added in two steps: an acidic peroxide solution and then sodium hydroxide, with a detergent that speeds and strengthens the emission. Chemists at Siemens Healthcare Diagnostics, a company that uses these labels, published a study in 2014. They found that with the detergent at least 95 percent of the light from four esters arrived within 1.2 to 5.3 seconds, against up to a minute without it. Each acridinium molecule reacts once and gives at most one photon, so the whole flash must be caught as it happens: the trigger solutions that start the reaction are injected in front of the detector, and the light is counted in a few seconds.

Relative light units belong to one instrument

Analyzers report chemiluminescence in relative light units, a count scaled by each instrument's optics and electronics. In the 2014 Siemens study, proteins joined to acridinium labels gave about 4.3 to 6.1 × 10¹⁹ light units per mole of label on one laboratory luminometer. That is about one unit for every 10,000 to 14,000 label molecules. The number links label to signal only on that instrument; it is not a photon count and does not transfer between analyzers.

Two makers' analyzers use acridinium labels, according to FDA decision summaries and the makers' own pages. Abbott's ARCHITECT and Alinity analyzers use acridinium-labeled antibodies on paramagnetic microparticles, tiny beads that are drawn to a magnet but do not stay magnetized, in a method the company calls chemiluminescent microparticle immunoassay. Siemens Healthineers' Atellica and ADVIA Centaur analyzers use acridinium esters too.

Glow: enzymes that make light

The other family keeps the enzyme label of Chapter 6 and changes its substrate. In 1983 researchers in Birmingham reported that firefly luciferin multiplied the light from luminol oxidized by horseradish peroxidase. In 1985 the same group found phenol compounds that raised that light more than a thousandfold and kept it glowing for several minutes. Light-producing substrates for alkaline phosphatase followed in the late 1980s, molecules called 1,2-dioxetanes: the enzyme removes a phosphate group, and the unstable remainder breaks apart and emits light. A 1991 review gave the detection limit for alkaline phosphatase with such a substrate as 0.001 attomole, about 600 enzyme molecules, with emission lasting more than 30 minutes.

As in Chapter 6, the enzyme works as an amplifier, but now of light instead of color. Each enzyme molecule keeps converting substrate molecules for minutes, each of which can give a photon, so a few hundred enzymes can be seen. Fujirebio's Lumipulse analyzers use alkaline phosphatase with the dioxetane substrate AMPPD, whose light peaks at 477 nanometers. The FDA's 2025 decision summary describes that chemistry for the company's blood test for Alzheimer's disease, one of the tests in the guide's second case. QuidelOrtho's VITROS analyzers measure peroxidase with luminol.

Solved example: flash against glow, in molecules

The detection limit in the 1983 acridinium paper works out to about 480,000 label molecules. The limit in the 1991 review, for alkaline phosphatase with a dioxetane substrate, works out to about 600 enzyme molecules. The ratio is about 800: an amplifying enzyme can be seen at far fewer copies than a label that reacts once. The comparison is of labels, not of assays. Both figures were measured with label alone, and in an assay both labels sit on the same wrongly bound antibodies of Chapter 5, so an enzyme also amplifies the light from wrongly bound label.

13 — One burst, or a glow that keeps building
FLASH: ACRIDINIUM ESTER LIGHT, RELATIVE 0 2 4 6 8 10 TIME AFTER TRIGGER, S acidic peroxide, then sodium hydroxide counted: the whole burst each ester reacts once with detergent, ≥95 % of the light within 1.2–5.3 s (Siemens chemists, 2014) GLOW: ENZYME + DIOXETANE LIGHT, RELATIVE 0 10 20 30 TIME AFTER SUBSTRATE, min substrate added READ each enzyme keeps turning substrate over emission lasts more than 30 min (1991 review) 100 1,000 10,000 100,000 1,000,000 LABEL MOLECULES DETECTED (LOG) acridinium ester, 1983: about 480,000 labels alkaline phosphatase + dioxetane, 1991: about 600 enzymes about 800 times an amplifying label is seen at far fewer copies schematic curve shapes; detection figures measured with label alone
Label and signalTriggerReadoutFocal detail
Plate 13 — A flash label gives at most one photon per molecule within seconds of its trigger; an enzyme with a light-producing substrate keeps glowing for half an hour. With label alone, the enzyme is seen at about 600 copies against about 480,000 for the ester.

Glow has costs of its own. The light rises over minutes, so the analyzer must wait before reading, and that wait sets a minimum time per test. A 2019 poster by Beckman Coulter and its substrate subsidiary states that the older substrate for the company's Access analyzers needed about 6.3 minutes to generate its signal. Beckman's DxI 9000 analyzer uses a newer acridan-based substrate that gives its light in about a minute, and the time to the first troponin result fell from 17 to 11 minutes, by the company's account.

What light changes, and what it does not

Measuring light against darkness widens the range as well as lowering the floor. A detector that counts single photons can register a few dozen per second in the dark and millions per second before it saturates and can count no faster, a span of five or more factors of ten against less than three for absorbance (Chapter 13). A sample can then be measured from near the floor to far above the clinical decision point without dilution, which saves reruns and reagent.

14 — A shadow in a bright beam, or light in the dark
ABSORBANCE lamp colored product detector blank: 100 sample: about 98.9 0.005 absorbance: about a 1 % dip in a large reading CHEMILUMINESCENCE detector emitted light light-tight box signal: a few hundred per second dark: a few dozen counts per second counting up from near zero absorbance: less than 3 factors of ten photon counting: 5 or more factors of ten (Chapter 13) schematic; log scale, one tick per factor of ten
Label and signalTriggerReadoutFocal detail
Plate 14 — Absorbance measures a target as a small shadow in a bright beam; chemiluminescence counts light against darkness. The same small amount of label becomes a large relative signal, and the measurable range widens from under three factors of ten to five or more.

What light does not change is the label that stays behind without a target. An acridinium ester stuck to the wall of a cuvette, the small vessel in which the reaction is read, flashes as brightly as one on a sandwich. An enzyme on the plastic glows as long as one on a bead. Chemiluminescence removed the lamp, the scratches and the bubbles from the noise of Chapter 6, so the floor that remains is much closer to the one set by wrongly bound label. The next two chapters describe two more ways to read a label against near-darkness. In Chapter 8 a lamp remains, and the reading waits until the scattered light and the sample's own glow have faded; in Chapter 9 the light is made at an electrode, so label that is not held against the electrode contributes little light.

Match the read to the light's timing

A flash label needs its trigger injected in front of the detector and the whole burst captured in seconds; a glow needs an incubation before reading and a stable temperature while the light builds. The choice fixes the analyzer's mechanics, its cycle time and where its detector sits, so it belongs in the first architecture decisions, together with the label (Chapter 10).

+ What this chapter established
  • Chemiluminescence measures light made by the label against darkness, not a shadow in a lamp's beam.
  • Acridinium esters flash once per label within a few seconds of an alkaline peroxide trigger.
  • Enzymes with light-producing substrates glow for minutes; alkaline phosphatase with a dioxetane can be seen at a few hundred copies.
  • Light widens the measuring range and removes lamp noise, but label stuck without target still shines.
08 — Fluorescence

Waiting for the glow.

+ The questionFluorescence works well enough to count single cells in a flow cytometer; why must an immunoassay wait before reading it?

Light in, light out, and light everywhere

A europium ion can be held in an organic chelate, a complex in which an organic molecule, the ligand, binds the metal ion at several points at once. The ligand absorbs ultraviolet light near 340 nanometers and passes the energy to the ion, which gives back red light near 615 nanometers. The ion keeps giving that light back for close to a millisecond after the exciting flash has ended. That long emission, unusual among labels, is the subject of this chapter, because it solved a problem that ordinary fluorescence could not.

Fluorescence is the obvious way to read a label. A dye absorbs light of one color and returns light of a longer wavelength, and filters separate the two, as the flow cytometry guide in this library explains in detail. But an immunoassay reads its label in a plastic well or cuvette full of serum and buffer, not in a narrow stream passing a focused laser, as a flow cytometer does. The exciting light scatters off the plastic and the liquid. Serum contains proteins and other molecules that fluoresce on their own. And some of the exciting light leaks through any filter. All of that unwanted light arrives together with the label's light, from the moment the lamp fires.

Timing can separate the label's light from the unwanted light. The unwanted fluorescence and the scattered light die away in about 10 nanoseconds, according to Revvity, which makes the DELFIA system described below. Ordinary fluorescent dyes fade just as fast, so their light and the background arrive together and cannot be told apart by timing. A label whose glow lasts tens of thousands of times longer can be read after the background has gone, and reading a label after such a delay is what time-resolved fluorescence means.

A label that glows for a millisecond

The lanthanides, a row of metals at the bottom of the periodic table, give off light when their electrons jump between energy levels, and those jumps are slow by atomic standards. Europium, samarium and terbium emit light in sharp, narrow bands of wavelength when they are held in a molecule that absorbs light and passes the energy on. Their decay times run from tens of microseconds to about a millisecond, where the decay time is how long the emission takes to fall to about 37 percent of its starting strength. In 1983 E. Soini and H. Kojola described a time-resolved fluorometer for such labels, built around an ordinary xenon flash lamp. Its one-second measurements could detect as little as the radioactive methods then in use.

Revvity now sells a time-resolved method called DELFIA, and its user guide gives the timing: an excitation flash, a wait of 400 microseconds, a counting window of 400 microseconds, and a thousand such cycles a second. Before reading, an acidic enhancement solution releases the europium from the label and binds it into a new, brightly fluorescent complex. The maker says that this step multiplies the fluorescence by one to ten million times.

Solved example: what a 400-microsecond wait keeps

Take a europium decay time of 730 microseconds and a background decay time of 10 nanoseconds. After a 400-microsecond wait, the europium has given out about 42 percent of its light and keeps about 58 percent; the window from 400 to 800 microseconds collects about 24 percent of its total. The background, in the same 400 microseconds, has passed 40,000 of its decay times, and what remains of it is far too small to measure. The wait gives up about two fifths of the label's light and almost all of the background's, and closing the window at 800 microseconds leaves about a quarter of the label's total counted.

15 — Wait, then count
WAIT 400 µs COUNT 400 µs 1 1 in 10 1 in 100 1 in 1,000 0 200 400 600 800 1,000 LIGHT, RELATIVE (LOG) TIME AFTER THE FLASH, µs excitation flash scatter and serum fluorescence: about 10 ns europium decay 730 µs only the label's light is left europium keeps 58 % of its light after 400 µs; the window collects about 24 % of the total 1,000 cycles a second (DELFIA user guide)
Label and signalTriggerInterferentReadoutFocal detail
Plate 15 — Background fluorescence and scattered light die away in about 10 nanoseconds; a europium label glows for close to a millisecond. Waiting 400 µs before counting gives up three quarters of the label's light and almost all of the background.
A delayed read gives up most of the label's light

Waiting 400 microseconds and then counting for 400 collects only about a quarter of a europium label's emission, and labels that decay in microseconds would lose almost all of theirs. Time-resolved fluorescence works only with long-lived labels and only because the background it rejects is far larger than the light it discards.

The gain is the same in kind as chemiluminescence's: the detector reads its label against near-darkness. But the darkness here is made by timing, with a lamp in the instrument. The optical background therefore comes from the lamp's own afterglow, the plate's phosphorescence and the detector's dark counts. Phosphorescence is a slow glow that some materials give off after light has fallen on them, and dark counts are the counts a detector registers when no light reaches it (Chapter 13). In an assay, labeled antibody stuck without target and stray europium usually set the floor. The DELFIA guide gives typical plate backgrounds of 100 to 1,000 counts, depending on the plate.

Reading without washing: energy transfer

Long-lived labels allowed a further step, which Chapter 5 mentioned as one of the few ways around the wash: telling bound label from free label without washing at all. In 1993 a chemist at the French company CIS bio international described a europium label for this purpose. The europium is held in a cage-like molecule called a cryptate and attached to one antibody as an energy donor. A second antibody carries an acceptor molecule. When both antibodies bind the same target molecule, donor and acceptor sit a few nanometers apart, and the europium passes its energy to the acceptor, which then emits at its own wavelength. Because the acceptor receives its energy from the long-lived europium, its light is delayed as the europium's is. Free antibodies are scattered through the liquid, too far apart for the transfer.

Thermo Fisher Scientific's B·R·A·H·M·S KRYPTOR analyzers use this method under the name TRACE. The company's white paper gives the sequence: a laser pulse at 337 nanometers excites the europium, the instrument waits 50 microseconds, and it then measures light at 665 nanometers from the acceptor and at 620 nanometers from the europium itself. The ratio of the two signals corrects for a cloudy or colored sample, and the paper states that no washing or separation step is needed. The 1993 paper reported that its assay measured the hormone prolactin down to 0.3 microgram per liter in human serum.

16 — Light from the acceptor only when two antibodies meet
BOTH ANTIBODIES ON ONE TARGET target energy transfer: a few nm transfer only when they sit nanometers apart laser 337 nm 620 nm 665 nm europium cryptate (donor) acceptor FREE ANTIBODIES, FAR APART 620 nm donor still emits no transfer: no 665 nm measure after a 50 µs wait: ratio 665 / 620 nm corrects for cloudy or colored samples 1993: prolactin to 0.3 µg/L in serum no washing or separation (maker's white paper)
AnalyteAntibodyLabel and signalTriggerReadoutFocal detail
Plate 16 — In TRACE the label is a pair: a long-lived europium donor and an acceptor that lights only when both antibodies hold the same molecule. Reading the ratio of the two wavelengths after a short wait lets the analyzer skip the wash.

Outside the central laboratory

Long-lived labels suit small instruments, because the timing that rejects background depends less on a dark, precisely built optical chamber than on a fast switch between flash and count. The FDA cleared Radiometer's AQT90 FLEX analyzer in January 2012 with a test for myoglobin, cleared a CK-MB test for it under a separate decision, and in 2017 cleared its myoglobin and CK-MB test kits. The analyzer reads a europium chelate by time-resolved fluorescence, counting single photons, in a test cup that holds the antibodies, according to the FDA's decision summary. The maker lists troponin among its tests and gives results in 11 to 21 minutes.

A fluorescence reader can also measure an enzyme label instead of a lanthanide. bioMérieux's VIDAS analyzers use what the company calls an enzyme-linked fluorescent assay: an enzyme on the detection antibody turns a substrate into a fluorescent product, so the amplification of Chapter 6 is read as light rather than color. The company lists a high-sensitivity troponin I test on these analyzers that takes 20 minutes and covers 1.5 to 40,000 pg/mL, a span of more than four factors of ten.

Fluorescent beads put fluorescence to a different use. Authors from Luminex described the company's bead arrays in 1997. In these arrays, tiny beads are color-coded so that dozens of assays run in one sample and are read in a flow cytometer, a method the flow cytometry guide in this library covers.

Set the gate from two decay times

The gate of a time-resolved reader is its delay and counting window, and both are specified from the label's decay time and the background's. The delay must be long compared with the background's decay time and short compared with the label's, and the counting window must collect enough of the label's light to make the count precise. A homogeneous energy-transfer assay, one run without any wash, adds a third requirement, a ratio channel that corrects for sample color and turbidity, or cloudiness. The timing and the ratio channel both belong in the reader's design inputs, the requirements that its design must meet.

+ What this chapter established
  • Ordinary fluorescence in serum competes with scattered light and the sample's own glow, which die away in about 10 nanoseconds.
  • Europium and other lanthanide labels glow for tens of microseconds to about a millisecond, so a delayed read keeps the label and drops the background.
  • Energy transfer between two labeled antibodies on one target lets an assay tell bound from free without washing.
  • The delayed read discards most of the label's light, so the method depends on long-lived labels and stable timing.
09 — Electrochemiluminescence

Light switched on at an electrode.

+ The questionWhat does starting the light with a voltage gain over starting it with a reagent?

A label the electrode re-arms

On 12 June 1996 the FDA cleared an assay for the pregnancy hormone hCG that Boehringer Mannheim had submitted. It was one of the first tests for an analyzer the company called Elecsys. Its label was a metal complex that gives off light only while an electric voltage drives a reaction at an electrode, a method called electrochemiluminescence. The technology came from IGEN, a company in Maryland, which had licensed it in 1992 to the business that Roche later acquired. Twenty years on, the same method reads the label of Roche's high-sensitivity troponin T assay, the assay of this guide's opening story.

The label is a ruthenium atom held by three ring-shaped organic molecules. The complex weighs about a thousand grams per mole, small enough that several can be attached to one antibody. In the measuring solution the label sits with a large excess of a second chemical, tripropylamine, called the coreactant. When the electrode is raised to a positive voltage, it takes electrons from the tripropylamine, which breaks into short-lived, highly reactive fragments. A 2020 study describes the sequence that follows: one kind of fragment hands the ruthenium complex an extra electron, and another then takes an electron away, which leaves the complex in an excited state. The complex relaxes by emitting a photon of orange-red light near 620 nanometers and returns to exactly the form it started in.

The label's return to its starting form is the point of the method. An acridinium ester reacts once and is spent. A ruthenium label can be excited again after it emits, as long as the electrode keeps the voltage on and fresh coreactant reaches it. University chemists carried out the 2020 study of the mechanism with researchers from Roche Diagnostics and Hitachi High-Tech. The study set out the reaction steps and applied 1.4 volts to drive them. Roche's own patent on the measuring method describes the label reaction as preferably circular, so that one label emits "a plurality of photons".

The measuring cell

The light is made in a small flow cell, a chamber that liquids are pumped through, not in the reaction vessel. A patent by Roche Diagnostics and Hitachi High-Technologies describes the arrangement. The reaction mixture holds streptavidin-coated magnetic beads, typically 2 to 3 micrometers across, and is drawn into the cell over a working electrode, the electrode whose voltage drives the reaction. A permanent magnet under the electrode pulls the beads, with the sandwiches they carry, onto the electrode surface. A buffer containing the coreactant then flows through, washing away unbound label. The voltage is applied, and a photomultiplier tube records the light through the electrode assembly for about two seconds. A photomultiplier is a light detector sensitive enough to register single photons (Chapter 13). The magnet is then moved away so that the beads can be flushed out and the cell cleaned for the next test.

17 — Beads held on an electrode, read through it
WORKING ELECTRODE INLET OUTLET MAGNET PHOTOMULTIPLIER beads 2–3 µm, streptavidin-coated coreactant buffer (tripropylamine) unbound label washed out voltage on label on beads at the electrode gives the light 1 draw in; magnet holds beads 2 buffer washes unbound label 3 voltage on: read about 2 s 4 magnet away; flush and clean schematic, after a Roche/Hitachi patent; not to scale
AnalyteAntibodyLabel and signalTriggerFocal detail
Plate 17 — The measuring cell adds a second separation: a magnet holds the beads on the electrode while coreactant buffer flushes away free label, and only then is the voltage applied and the light counted for about two seconds.

Roche's troponin T assay follows this sequence. The FDA's 2017 decision summary describes the steps: a biotin-labeled antibody and a ruthenium-labeled antibody form a sandwich around troponin T, streptavidin-coated microparticles collect the sandwich, the microparticles are captured magnetically on the electrode, and a voltage stimulates the emission that a photomultiplier measures. The cobas e 801, an analyzer that now runs the assay, holds its measuring cells at 28 °C, separate from the 37 °C of its incubator, according to its maker's specifications.

Why the thin layer matters

The reactive fragments that excite the ruthenium last only a fraction of a millisecond. The 2020 study estimated their half-life at about 200 microseconds. It concluded that they are not expected to travel more than about 3 micrometers from the electrode before they decay. Light is made only in that thin layer.

Solved example: what the electrode can see

A bead 2.8 micrometers across that lies on the electrode has every part of its surface within 2.8 micrometers of the electrode, inside the 3-micrometer layer where light is made. Every label on the bead can therefore emit, though the labels nearest the electrode give most of the light. A label floating 30 micrometers away in the liquid is ten times farther than the layer reaches and stays dark. And the two-second reading lasts about 10,000 times the fragments' 200-microsecond half-life, so the electrode keeps making fresh fragments throughout the read.

18 — A label the electrode re-arms
ruthenium complex one electron added excited: a second fragment takes an electron away emits a photon near 620 nm by a tripropylamine fragment the electrode's voltage oxidizes tripropylamine into fragments returns unchanged: ready again acridinium: reacts once DISTANCE FROM THE ELECTRODE, µm ELECTRODE 0 1 2 3 30 30 µm away: dark 2.8 µm bead: every label inside the layer light made here: fragments live about 200 µs, reach about 3 µm a 2-second read ≈ 10,000 half-lives of the fragments the electrode sees only this layer
Label and signalTriggerFocal detail
Plate 18 — The ruthenium label is regenerated after each photon, so it shines for as long as the voltage is on, but only within about 3 µm of the electrode, where the short-lived coreactant fragments reach. A bead on the electrode lights; label in the bulk liquid does not.

Making the light at the electrode has three consequences. The method adds a second separation on top of the wash: label that is not on a bead held against the electrode contributes little light, whatever the wash left behind. There is no exciting light, so, as in chemiluminescence, there is no scatter and no autofluorescence, the sample's own glow, to subtract. And because each label can emit many photons during the read, the signal per label is larger than a single flash would give. A 1991 paper by IGEN's scientists reported that electrochemiluminescent labels could be measured over six factors of ten, with a detection limit of 200 femtomolar for the label. A femtomolar concentration is a millionth of a billionth of a mole per liter. Troponin T at 5 ng/L is about 140 femtomolar, so the label alone could be measured at about the concentration of troponin in Chapter 1.

The same 2020 study measured how much the light depends on geometry. Beads of 0.3 micrometer gave about eight times more light than beads of 2.8 micrometers, consistent with more of their labels sitting close to the electrode. A new coreactant raised the signal of commercial cobas e 801 assays by up to 47 percent. Both findings come from a research study, not from a product's specifications. Both show that the few micrometers next to the electrode are a design variable.

The electrode is part of the signal

In electrochemiluminescence, the light per label depends on the electrode surface, the coreactant reaching it and the voltage applied. A fouled electrode, a stale buffer or a drifting temperature in the cell changes the signal without any change in the sample. That is why the measuring cell is cleaned after every read, why its buffers are counted as consumables, and why its temperature is controlled separately from the incubator's.

Roche's troponin T assay, measured

The 2017 assay has a 9-minute protocol, a limit of blank of 2.5 ng/L on one analyzer and 3 ng/L on the other, and a measuring range that reaches 10,000 ng/L without dilution, more than three factors of ten above its limit of quantitation of 6 ng/L. Its limit of detection had fallen to 3 or 5 ng/L, depending on the analyzer, from the 10 ng/L of Roche's earlier, fourth-generation troponin T assay, but that fall came from changes to the assay itself, within the same detection method. Other companies use electrochemiluminescence differently. Meso Scale Discovery builds carbon electrodes into the bottoms of plastic plates, so that a research laboratory can read several spots, each with its own antibody, in one well. Only labels near the electrode surface are detected, according to the company.

Treat the measuring cell as a subsystem with its own requirements

An analyzer built on electrochemiluminescence carries a requirement set for the cell alone: electrode material and cleaning cycle, magnet travel and timing, voltage waveform, cell temperature, buffer consumption, and the photomultiplier's sensitivity to red light, since the light is near 620 nanometers. Each item needs its own verification before the assay is run on it.

+ What this chapter established
  • Electrochemiluminescence uses a ruthenium label that emits near 620 nanometers when it is excited by fragments of a coreactant, which a voltage at an electrode creates.
  • The label returns to its starting form after emitting, so it can give many photons while the voltage is on.
  • Beads are held on the electrode by a magnet, and light comes only from a layer about 3 micrometers thick, which adds a second separation to the wash.
  • Roche's troponin T assay forms its sandwich on streptavidin beads and reads it in this cell in 9 minutes.

+ Part III · The analyzer as a system

Holding the floor steady, test after test.

The cobas e 801 is the analyzer that now runs Roche's troponin T assay in many hospital laboratories, and its maker states that one module can report up to 300 immunoassay results an hour. Every one of those results comes from a reaction that must be timed, warmed, washed and read exactly like the calibrators it is compared with, which are samples of known concentration. The chemistry of Parts I and II sets how low a floor can be. This part is about the machine that has to reach that floor on every test: how a tube becomes a result, how liquids are moved without carrying one sample into the next, why minutes and tenths of a degree matter, how light becomes a number, and how the number is tied to a scale that another laboratory can share.

10 — Inside the analyzer

One test, many subsystems.

+ The questionWhat has to happen, and in what order, between a tube of blood and a reported number?

A tube's path through the machine

A blood tube arriving at an analyzer carries a barcode, and the barcode is the first input the machine reads. From the barcode, the laboratory's information system tells the analyzer which tests have been ordered on that sample. A sampling arm then fits a fresh disposable tip, finds the surface of the liquid, draws up a few microliters of plasma while sensors watch for clots, bubbles and foam, and dispenses it into a single-use reaction cup. On a cobas e 801 the sample volume per test can be set between 4 and 60 microliters, and every pipetting uses a new tip, according to the maker's specifications.

Reagents come next. Each test's antibodies, beads and buffers sit in packs in a refrigerated store on board. The e 801 holds up to 48 packs at 5 to 10 °C and identifies each by a radio-frequency tag. A reagent arm adds the right volumes to the cup, a mixer stirs it, and the cup moves into an incubator held at 37 °C, plus or minus 0.3 °C. At the end of the incubation the mixture goes to the measuring cell of Chapter 9, where the beads are captured, washed and read. The analyzer's software turns the light into a concentration with the calibration curve, checks the result against flags and limits, and sends it to the laboratory system.

19 — One tube's path through the analyzer
SAMPLE TUBE + BARCODE SAMPLING ARM new tip, 4–60 µL per test senses clots, bubbles, foam REACTION CUP INCUBATOR 37 °C ± 0.3 MEASURING CELL (Chapter 9) REAGENT STORE 5–10 °C, up to 48 packs, radio-frequency tag REAGENT ARM + MIXER SOFTWARE calibration curve, flags, limits LABORATORY INFORMATION SYSTEM RESULT which tests were ordered ±0.3 °C: the tightest tolerance in the path cobas e 801, maker's specifications; schematic layout
AnalyteReadoutFocal detail
Plate 19 — An analyzer runs one test as a sequence of subsystems: sampling, reagent handling, incubation, measurement and software. Each has its own specification, and each can change the number that reaches the laboratory system.

Each of those steps belongs to a subsystem that an engineer can name: sample handling, liquid handling, reagent storage, incubation, separation, detection, and the control software that schedules them. None of them is clever on its own. The design problem is that all of them must act on each test at exactly the right moment, the same way for every test and for every calibrator, while dozens of other tests are at other stages in the same machine.

Many tests in flight

The e 801 runs assays of fixed lengths, 9, 18 or 27 minutes. An 18-minute assay cannot be run one test at a time if the analyzer is to report hundreds of results an hour. The answer is a pipeline, in which many tests are in progress at once, each at a different step. On the e 801, a new cup starts every 12 seconds, so 300 tests can begin each hour. Each spends its incubation in one of 94 positions on the incubator, then leaves for the measuring cell as the next cup arrives.

Solved example: how many tests are in the machine at once

A new test every 12 seconds, each incubating for 18 minutes, means 18 × 60 ÷ 12 = 90 tests in the incubator at any moment, close to the e 801's 94 positions. A 9-minute protocol needs 45. The smaller cobas e 402 starts a test every 30 seconds, for 120 an hour, and holds 36 tests of 18 minutes in its 38 positions. The first result of a run arrives after the full assay time, and after that one result leaves every cycle. A 9-minute protocol shortens the wait for each result; it does not raise the number of results per hour, which the cycle time sets.

20 — A new test every 12 seconds
0 2 4 6 8 10 12 14 16 18 20 22 TIME FROM THE FIRST CUP, min CUP 1 CUP 2 CUP 10 … pipette incubate 18 min measure first result then one result every 12 s A NEW CUP EVERY 12 s about 90 tests in flight 18 × 60 ÷ 12 = 90 tests incubating at once (94 positions, e 801) 9-min protocol: 45 e 402: one every 30 s; 36 of 38 positions maker's figures; schematic; stagger drawn at twice scale
ReadoutFocal detail
Plate 20 — A pipeline turns an 18-minute assay into a result every 12 seconds: with a new cup every cycle, about 90 tests incubate at once. A shorter protocol shortens each wait but does not raise the number of results per hour, which the cycle time sets.

The same arithmetic shows where the limits lie. On the maker's own numbers, a run made only of 27-minute assays would need 135 positions at full speed, more than the incubator holds, so such a mix must slow down. For its smaller cobas e 402, the maker states that throughput "may differ based on the mix of test orders per sample". Consumables set another limit. The e 801 holds 15 magazines, each of 105 tips and 105 cups, enough for 1,575 tests. At 300 tests an hour that supply lasts 5 hours 15 minutes, the unattended running time that its specifications list.

A throughput rating is a best case for one test mix

A maker's "up to" figure assumes a favorable mix of assay times and few reruns, dilutions or calibrations. The real rate in a laboratory depends on its own test menu, and on how often samples need repeating. Comparing analyzers by their headline rate without the laboratory's own mix compares two assumptions.

The conversion chain

An analyzer never measures troponin. It measures something several steps removed and converts it back. Troponin molecules in the sample become sandwiches on beads, and the sandwiches put labels within a few micrometers of an electrode. The labels emit photons, a fraction of the photons release electrons in the detector, and the electrons become a count or a current. The count becomes a signal value for that test, the calibration curve turns the signal into a concentration, and software compares the concentration with limits and flags before it becomes a reported number.

21 — From troponin to a reported number
TROPONIN IN THE SAMPLE SANDWICHES ON BEADS LABELS AT THE ELECTRODE PHOTONS ELECTRONS: COUNTS OR CURRENT SIGNAL VALUE CALIBRATION CURVE → CONCENTRATION FLAGS AND LIMITS → REPORTED NUMBER capture share wash light collection counting calibration rules quantum efficiency the step that turns a signal back into troponin
AnalyteAntibodyLabel and signalReadoutFocal detail
Plate 21 — An analyzer never measures troponin directly. It converts molecules into sandwiches, sandwiches into light, light into counts and counts back into a concentration, and every link can lose or distort the signal.

Every conversion in that chain can change the result without any change in the troponin. A pipetting error changes how many molecules enter the cup; a warm incubator changes how many become sandwiches in nine minutes; a fouled electrode changes how many photons each label gives; a drifting detector changes the count; an out-of-date calibration changes the conversion. The rest of this part takes the conversions in turn. The quality control samples of Chapter 14 exist because no analyzer can check each conversion separately on every test.

Other makers, the same skeleton

Other analyzers make different choices within the same set of subsystems. Abbott rates its ARCHITECT i2000SR at up to 200 tests an hour. The analyzer draws samples with a fixed probe that is washed between samples, and the maker specifies the probe's carry-over, the fraction of one sample that passes into the next, at no more than 0.1 part per million (Chapter 11). Abbott's troponin assay on that analyzer has two incubations, each followed by a wash, and a flash label. The 2017 specification sheet for Siemens Healthineers' Atellica IM 1600 gives a rate of up to 440 tests an hour. That analyzer uses disposable tips, an acridinium label read by a photomultiplier, sample volumes of 10 to 100 microliters and assay times from 10 to 54 minutes, according to the same sheet. The subsystems are the same in each machine; what differs is where each maker spends the time, the space and the cost.

Draw the conversion chain before writing requirements

For an analyzer or a reader, list every conversion from molecule to reported number, then give each one an owner, a requirement and a check. The requirements of the subsystems follow from the chain: pipetting precision from the volume step, incubator tolerance from the binding step, detector noise from the photon step. The verification plan of Chapter 23 can then be traced to them.

+ What this chapter established
  • A tube's barcode starts a sequence of sampling, reagent addition, incubation, separation, detection and calculation, each a subsystem.
  • Analyzers pipeline tests: on the e 801 a new test starts every 12 seconds, and an 18-minute assay keeps about 90 tests in flight.
  • The analyzer converts troponin into sandwiches, photons, counts and finally a concentration, and every conversion can move the result.
  • Makers choose differently among tips and probes, flash and electrode labels, one or two incubations, within the same set of subsystems.
11 — Liquids and magnets

Moving microliters, washing beads.

+ The questionHow does a machine pipette a few microliters, wash particles a few micrometers across, and leave nothing behind for the next sample?

A drop that should not travel

On 9 August 2021 Beckman Coulter began a recall of more than 290,000 kits of its Access high-sensitivity troponin I reagent, according to the FDA's recall record. The firm attributed the problem to carry-over: after a sample with an extremely high troponin result, traces could pass "through the probe into the particle well" of the reagent pack and contaminate later tests. Its instructions to laboratories were concrete, and their trigger was any result above 270,000 picograms per milliliter, which is 270,000 ng/L, the unit used for troponin everywhere else in this part. After such a result, laboratories were to discard the open reagent packs, load a new one, run a low-level control sample on every pipettor, and repeat any positive results that might have been affected.

Carry-over is the transfer of material from one sample, or one reagent, into the next. A troponin assay is especially exposed to it because its range is so wide: Roche's troponin T assay, for example, was tested and found free of the high-dose hook up to 100,000 ng/L (Chapter 13). Troponin I assays meet samples above 270,000 ng/L, as the trigger value in the Beckman recall shows, while their makers' decision limits sit at about 9 to 58 ng/L (Chapter 14). A sample ten thousand times above the decision limit needs to leave behind less than one part in ten thousand of itself, and in practice far less, to keep the next result accurate.

Solved example: what a part per million does

Suppose a sample of 100,000 ng/L is followed by one containing almost no troponin. A carry-over of 1 part per million adds 0.1 ng/L to the second result, below the limit of blank. Abbott specifies the carry-over of the fixed probe on its ARCHITECT i analyzers at no more than 0.1 part per million. In the FDA's 2019 review of Abbott's high-sensitivity troponin I assay, a sample of at least 500,000 ng/L left about 0.3 ng/L in the next, at most 0.6 part per million. A 2014 letter reported an older troponin I assay, measured on another analyzer, that carried over 41 parts per million: after a sample of about 484,000 ng/L, a sample of about 60 ng/L read 80. At 41 parts per million, the 100,000 ng/L sample would add 4.1 ng/L to a blank, above the limit of blank of Roche's troponin T assay.

One common protocol measures carry-over with a fixed sequence: a high sample run twice, then a low sample three times. If the first low result is higher than the third, the difference, divided by the high concentration, estimates the fraction carried. The test has to be planned as part of the design, because routine quality control cannot reveal the problem. The Clinical and Laboratory Standards Institute's guideline on statistical quality control excludes random errors that affect individual samples, such as carry-over, from its scope, because control samples run between patients will not reveal them.

22 — What a part per million does
0.1 1 10 100 0.01 0.1 1 10 TROPONIN ADDED TO THE NEXT SAMPLE AFTER A 100,000 ng/L SAMPLE, ng/L CARRY-OVER, PARTS PER MILLION (LOG) Roche's limit of blank 2.5–3 ng/L fixed-probe specification (maker's figure) FDA 2019 review: at most 0.6 ppm 1 ppm: 0.1 ng/L 2014 letter: 41 ppm → 4.1 ng/L above the limit of blank CARRY-OVER TEST H H L1 L2 L3 run order: high twice, then low three times carry-over ≈ (L1 − L3) ÷ H
AnalyteInterferentReadoutFocal detail
Plate 22 — After a sample of 100,000 ng/L, a carry-over of 1 part per million adds 0.1 ng/L to the next result; 41 parts per million adds 4.1 ng/L, above the limit of blank of a high-sensitivity assay. Carry-over is tested with a fixed high-low sequence because routine controls cannot see it.
Statistical quality control misses carry-over

Control samples measured between patients test whether the whole system has shifted. A carry-over event affects one sample, the one after an extreme result, and leaves the next control untouched. Carry-over has to be tested by a dedicated sequence at verification and guarded against by design, by probe washing, tips, flags on extreme results and rules for repeating the following samples.

Tips, probes and pressure

The first defense is the pipetting hardware. A disposable tip used once and thrown away cannot carry sample on the probe to the next sample. Carry-over can still arise through reagent probes, wash stations or a shared measuring cell. Roche describes the cobas e 801 as using "one disposable AssayTip per pipetting". Siemens Healthineers and QuidelOrtho both say that their disposable tips eliminate sample carry-over, which is the makers' claim, not a measured value. A fixed probe is instead washed inside and out between samples. It saves the cost and waste of tips and can be cleaned to a specified carry-over, as Abbott's figure shows. If the firm's account of the cause is right, the Beckman recall shows what happens when sample can take a path that the cleaning does not reach.

The second defense is to watch the liquid as it moves. A probe that touches a clot, a bubble or foam draws the wrong volume, and the result is wrong with no visible sign. A 2000 patent from Bayer describes one way to catch a wrong draw, by recording the vacuum inside the probe while it aspirates, or draws in, the liquid. Bayer's diagnostics business later passed to Siemens. In the patent, a normal draw of about 100 microliters shows a smooth rise and decay within about half a second. A clot blocks the tip and drives the pressure higher and the draw longer, foam lets the pressure collapse early, and a second rise before the decay means that the tip lost contact with the liquid. Analyzers also sense the liquid surface before aspirating, and the e 801 adds a camera for foam.

23 — Listening to the probe as it draws
PROBE surface sensed before drawing foam camera (e 801) 0 0.2 0.4 0.6 0.8 TIME, s VACUUM IN THE PROBE schematic, after a 2000 patent (Bayer) normal draw, about 100 µL tip left the liquid: second rise foam: collapses early clot: higher, longer wrong volume, caught before it becomes a result
ReadoutFocal detail
Plate 23 — A probe that draws a clot, a bubble or foam takes the wrong volume with no visible sign. Watching the pressure inside the probe during the draw turns each fault into a recognizable trace that the analyzer can flag.

Small volumes make every error larger in relation to the volume drawn. A 4-microliter draw that loses a tenth of a microliter to a droplet on the tip is 2.5 percent short, and the troponin result is 2.5 percent low with it. Pipetting precision at such volumes does not appear in the makers' public specifications. It is included, however, in the total precision of the assay, which the studies of Chapter 23 measure.

Magnets and beads

Most analyzers separate bound from free label with magnetic microparticles. The reagent of Roche's troponin T assay contains streptavidin-coated particles at 0.72 milligram per milliliter, according to the FDA's decision summary. Patents describe the Elecsys beads as typically 2 to 3 micrometers across, and research products use beads of that size too, such as one supplier's 2.8-micrometer streptavidin beads, which bind about 500 to 750 picomoles of biotin per milligram. Beads that small stay suspended long enough to collect the sandwiches throughout the reaction volume, as Chapter 3 required, yet a permanent magnet can still pull them out of suspension.

The wash then depends on the magnet, which holds the beads against a wall or an electrode while the liquid around them is drawn off and fresh buffer is added. How much unbound label survives depends on the volume of liquid left clinging to the pellet of beads, the number of exchanges, and whether the beads are released and resuspended between them. Those figures do not appear in public sources, but each exchange works as a dilution: two exchanges that each leave 1 percent behind leave one part in ten thousand.

Magnets attract more than beads. In October 2019 Roche issued a field safety notice for one lot of its troponin T reagent on the cobas e 801, after laboratories had reported high results that could not be reproduced. A field safety notice is a maker's formal warning to the users of a product. Roche issued the final version of the notice in January 2020, and that version gave the cause as contamination with magnetic particles that were not beads, introduced during filling of the reagent for that analyzer only. Until the lot was replaced, laboratories were told to repeat any result of 14 ng/L or more, not to shake the reagent packs, and to discard each pack after its first 200 tests.

Test carry-over at the highest concentration the assay will meet

Run the carry-over sequence with the high sample at or above the hook-free limit, not at the top calibrator, and express the result in the units of the decision limit. Then write the instruction for laboratories as the Beckman recall had to: which result triggers which action, on which pipettors, for which following samples.

+ What this chapter established
  • Carry-over moves traces of one sample into the next; for troponin it matters because results span more than four factors of ten around a low decision limit.
  • Disposable tips and washed probes limit carry-over to a claimed or specified level, and pressure monitoring and level sensing catch wrong draws.
  • Magnetic microparticles a few micrometers across collect sandwiches throughout the sample and are held still while the liquid is exchanged.
  • Carry-over and stray particles affect single samples, and routine controls do not reveal them, so they must be tested by design.
12 — Time and temperature

Minutes at 37 °C.

+ The questionWhy do the analyzer's clock and thermostat shape the result, and how closely must they hold?

A tolerance of three tenths of a degree

Roche specifies the incubator of its cobas e 801 at 37 °C, plus or minus 0.3 °C, and the module's measuring cell at 28 °C, plus or minus 0.5 °C. Those numbers are requirements on the chemistry as much as on the heaters. Every rate in an immunoassay rises with temperature: molecules diffuse faster in warmer liquid, antibodies bind and release faster, and enzymes convert substrate faster. In a 9- or 18-minute protocol the reaction is usually stopped and read before it is complete, and a warmer reaction then gives more signal for the same amount of target.

How much more depends on the assay. An application note from Revvity, the maker of DELFIA reagents, gives one measured case. In a two-step assay for thyroid-stimulating hormone, a second incubation of two minutes reached 67 percent of its maximum signal at 35 °C, against 25 percent at 25 °C. From such data the note derives a rate that roughly triples for every 10 °C. That rise corresponds to an activation energy of about 84 kilojoules per mole, where the activation energy is the energy barrier the reaction must climb. The measurement comes from one assay on one platform, but the order of magnitude is common: chemistry slows sharply as it cools.

Solved example: what three tenths of a degree is worth

With an activation energy of 84 kilojoules per mole, the reaction rate near 37 °C changes by about 11 percent for each degree, from the standard relation between rate and temperature. A tolerance of plus or minus 0.3 °C therefore allows the rate to vary by about plus or minus 3 percent between one incubation and another. About 3 percent either way is the most the rate can vary within the tolerance, not the error in a result. Calibrators and patient samples run in the same incubator under the same protocol, so a steady offset largely cancels, and a reaction close to completion responds less to a change in rate than one far from it. A fluctuation between the calibration run and the patient's run does not cancel.

24 — What three tenths of a degree is worth
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 25 30 35 37 40 REACTION RATE, RELATIVE TO 37 °C TEMPERATURE, °C a steady offset cancels; a fluctuation does not 37 ± 0.3 °C → rate ± about 3 % 0.97 1.00 1.03 36.7 37.0 37.3 °C about 11 % per degree near 37 °C computed for an activation energy of 84 kJ/mol (from one maker's application note): rate roughly triples per 10 °C
AntibodyReadoutFocal detail
Plate 24 — For a reaction whose rate roughly triples every 10 °C, each degree near 37 °C changes the rate by about 11 percent, so a tolerance of ±0.3 °C allows about ±3 percent. Calibrators share a steady offset with patient samples; a fluctuation between runs they do not share.

The measuring cell has its own temperature for a related reason. Its light comes from reactions at the electrode surface, which are also faster when warmer. In one patent example the cell is read for about two seconds, so a difference in the cell's temperature from one test to the next changes the signal per label directly. Holding the cell at a fixed temperature, separate from the incubator's, keeps that signal the same from test to test.

Temperature tolerance is a requirement on the result

A heater that holds 37.0 °C at its sensor can still leave a reaction cup a few tenths of a degree away, through the cup's position, the reagent's temperature as it is added, or the room. What matters is the temperature of the liquid during the reaction, so the tolerance has to be verified in the cup, across positions and room conditions, not only at the controller.

The same minutes for every cup

Timing matters for the same reason as temperature. A reaction that has not reached equilibrium gives a signal that depends on how long it ran, so every cup must run exactly as long as the calibrators did. Analyzers therefore run on a fixed cycle, a new cup every 12 seconds on the e 801, and their protocols come in fixed steps of 9, 18 or 27 minutes. The scheduler can then guarantee that each cup spends its incubation in the same number of cycles, whatever else is happening in the machine.

The fixed protocol also makes a useful check possible. Chapter 3 showed that the fraction of target caught after a given time depends on how fast the target binds. A target that binds unusually slowly, because it is attached to something else in the blood, can give different results in a 9-minute and an 18-minute version of the same assay. Chapter 19 describes laboratories that used exactly that difference to find an interference, a distortion of the result caused by something in a patient's sample other than the target.

Shorter protocols, higher floors

When a protocol is shortened, binding has less time to approach equilibrium, so fewer sandwiches form for the same amount of troponin, and the signal from target shrinks while the background does not. The floor can rise. The International Federation of Clinical Chemistry publishes a table of troponin assay characteristics supplied by their makers. Its 2019 version gives two limits of detection for the high-sensitivity troponin T assay on one family of Roche analyzers: 2.05 ng/L for the 18-minute protocol and 2.85 ng/L for the 9-minute one. Both figures are below the 5 ng/L that a 2017 trade report gives as the limit on the assay's European label, a reminder that a limit depends on how it was measured. On the newer cobas e 801, both protocols were listed at 3 ng/L.

In the US clearance of 2017 the same trade-off was made a different way. On the cobas e 411 the 9-minute assay adds the beads in a second incubation, after the sandwich has formed, and its limit of detection was 5 ng/L. On the cobas e 601 the beads are present from the start, and its limit was 3 ng/L, according to the FDA's decision summary. The reagent was the same, but the two analyzers differ both in how they arrange the minutes and in their measuring hardware. The Federation's table also lists a higher limit for the e 411 than for Roche's other analyzers in the 18-minute protocol. The e 411's floor is therefore higher in both protocols, and the timing of the bead step explains at most part of the difference.

25 — The same reagent, arranged in time
ONE STEP cobas e 601 TWO STEPS cobas e 411 sample biotin antibody ruthenium antibody beads first incubation second incubation beads added 0 3 6 9 min schematic timelines 0 1 2 3 4 5 6 LIMIT OF DETECTION, ng/L one Roche family, 18-min protocol (IFCC 2019) 2.05 same, 9-min protocol 2.85 cobas e 801, both protocols 3 FDA 2017, cobas e 601, one step, 9 min 3 FDA 2017, cobas e 411, two steps, 9 min 5 same reagent, two analyzers, different floor limits of detection as listed by makers (IFCC table) and in the FDA decision summary
AnalyteAntibodyLabel and signalReadoutFocal detail
Plate 25 — A shorter protocol leaves binding less time and can raise the floor, from 2.05 to 2.85 ng/L in one maker-supplied table. The arrangement matters too: in the 2017 US clearance the same reagent reached 3 ng/L in one step and 5 ng/L in two.

Reagents on board and on the shelf

Time acts on the reagents too. Antibodies slowly lose activity, labels decay or degrade, and the beads can clump, so every reagent pack has a shelf life in the refrigerator and a shorter life once opened on the analyzer. The e 801 keeps its packs at 5 to 10 °C on board, and its maker gives on-board stabilities of up to four months, depending on the assay. Abbott gives 14 to 30 days on board for its ARCHITECT i reagents, and Siemens Healthineers gives 4 to 90 days for the Atellica IM.

A calibration loses its validity as its reagent ages. The FDA's decision summary for Roche's troponin T assay states that a calibration remains valid for 7 days while the same reagent pack stays on the analyzer, and for 12 weeks with packs from the same lot. After that the analyzer requires a new calibration, because the reagent's response may have drifted. In 2023 the Clinical and Laboratory Standards Institute published the second edition of its guideline EP25, which describes how makers establish reagent shelf-life and in-use stability, by real-time and accelerated studies. Accelerated studies store reagents at raised temperatures so that aging shows sooner.

Write time and temperature as one budget

Measure how the assay's signal changes with incubation temperature and with time around the chosen protocol, then set the tolerance on each so that their combined effect stays inside the precision the assay must meet. The same data show how far a protocol can be shortened before the floor moves, and they belong in the design file as the rationale for both numbers.

+ What this chapter established
  • Every rate in an immunoassay rises with temperature, and in one measured assay the rate roughly tripled per 10 °C.
  • A tolerance of plus or minus 0.3 °C at 37 °C allows the rate to vary by about plus or minus 3 percent, and that variation cancels only when calibrators and samples see the same conditions.
  • Fixed cycles and fixed protocol times make every cup run as long as the calibrators did.
  • Shorter protocols and different step arrangements can raise the floor, and reagents and calibrations age on board.
13 — From light to number

From a few photons to a few million.

+ The questionHow does one detector count a few hundred photons and also a few million, and what limits the count at each end?

A detector that multiplies electrons

A photomultiplier tube is a glass vacuum tube a few centimeters long with a thin metal-alkali film, the photocathode, on the inside of its window. A photon striking the film can knock out one electron. Inside the tube sits a chain of electrodes called dynodes, each at a higher voltage than the last. The voltage accelerates the electron into the first dynode, where it knocks out several more, and each dynode down the chain repeats the step, so that one electron at the cathode becomes a pulse of a million to ten million electrons at the end. Hamamatsu, which makes such tubes, gives that range in its handbook as the tube's gain, the number of electrons out for each electron in.

Not every photon starts a pulse. The fraction of photons that release an electron is called the quantum efficiency, and it depends on the cathode material and the wavelength. Hamamatsu quotes up to 43 percent at 350 nanometers for its best blue-sensitive cathodes. A further loss occurs because only 60 to 90 percent of released electrons reach the first dynode. For one photon-counting module, Hamamatsu gives 380,000 counts a second per picowatt of 400-nanometer light, which means that the module counts about 19 percent of the photons that arrive. At 600 nanometers the same module counts only about 3 percent. That loss matters for the orange-red light of electrochemiluminescence near 620 nanometers, and an analyzer reading that light is better served by a cathode chosen for red light.

26 — One photon, a million electrons
photon PHOTOCATHODE DYNODES, EACH AT HIGHER VOLTAGE ANODE quantum efficiency up to 43 % at 350 nm (best blue cathodes) 60–90 % of electrons reach the first dynode one pulse: 1 to 10 million electrons (maker's handbook) schematic SHARE OF ARRIVING PHOTONS COUNTED, ONE MODULE 0 % 10 % 20 % 300 400 500 600 700 WAVELENGTH, nm maker's figures 477 nm dioxetane glow 620 nm electrochemiluminescence 19 % about 3 % schematic between points orange-red light needs a red-sensitive cathode
Label and signalReadoutFocal detail
Plate 26 — A photomultiplier turns one photon into a pulse of millions of electrons, but only a fraction of photons start a pulse. One maker's module counts about 19 percent at 400 nm and about 3 percent at 600 nm, so the cathode must suit the label's color.

The tube also produces pulses when no light falls on it, from electrons released by heat in the cathode. The rate of these pulses is called the dark count. It is small, about 50 a second typical for that module, but it is never zero, and it is the detector's own floor. A light-tight chamber and a label that makes its own light let a blank reading approach that floor.

Counting statistics

Counting light is counting random events. Photons from a steady source arrive at random moments, so a count made over a fixed time scatters from one reading to the next even when nothing changes. The scatter follows a simple rule from probability: the standard deviation of the count is the square root of its average. A reading of 100 counts carries a scatter of about 10, or 10 percent; 10,000 counts carry 100, or 1 percent; and a million counts carry 1,000, or 0.1 percent. The more light a measurement collects, the more precisely it knows the light.

Solved example: the smallest signal a 2-second read can see

Read for 2 seconds, a detector with 50 dark counts a second collects on average 100 dark counts, with a scatter of the square root of 100, which is 10. A signal has to add about three times that scatter, roughly 30 counts, before it stands clearly above the dark reading. If 19 percent of photons are counted, 30 counts need about 160 photons to reach the tube in those 2 seconds. In a real assay the blank also contains light from wrongly bound label, which raises the count and its scatter, so the practical floor sits higher (Chapter 5).

27 — Counts scatter by their square root
A 2-SECOND READ HOW OFTEN 60 80 100 120 140 160 COUNTS IN 2 s dark: mean 100, scatter 10 plus 30 counts from label about three times the dark scatter about 160 photons at 19 % counted give the extra 30 counts SCATTER SHRINKS WITH THE COUNT 0.1 % 1 % 10 % 100 10,000 1,000,000 RELATIVE SCATTER (LOG) AVERAGE COUNT (LOG) 100 counts: 10 % 10,000: 1 % 1,000,000: 0.1 % scatter = 1 ÷ √N
InterferentReadoutFocal detail
Plate 27 — Light arrives as random photons, so a count of N scatters by about √N. A dark reading of 100 counts scatters by 10, and a signal must add about 30 counts to stand clearly above it; more counts mean proportionally less scatter.

Counting statistics apply at every step of the chain. A digital assay counts beads, each lit or dark according to whether it carries a labeled target molecule, and its bead counts obey the same square-root rule (Chapter 17). In any assay, the scatter of a low-level sample is set partly by how few labels, and how few photons, it produces. Collecting more of an analyzer's light, or reading for longer, lowers the scatter of the blank, and with it the limit of blank, when photon counting contributes much of that scatter. When label bound without target dominates the blank, as Chapter 5 described, those steps help little.

From dozens to millions

At the other end of the scale, the detector cannot count fast enough. Each pulse takes time to register, and if pulses arrive faster than the electronics can separate them, some are lost. The same Hamamatsu module keeps its count in proportion to the light up to about 5 million counts a second, at which point it loses about 10 percent. From 50 dark counts to 5 million is five factors of ten, a far wider span than the less than three factors of absorbance in Chapter 6.

Analyzers report the result of this counting as a signal value. Abbott's documents call it relative light units and state that there is a direct relationship between the troponin in a sample and the light units detected. Above the counting range, an analyzer can measure the tube's current instead of counting pulses, a mode called analog, or it can reduce the light reaching the tube, or dilute the sample. Roche's troponin T assay quantifies troponin from 6 to 10,000 ng/L without dilution. Its US clearance allows a tenfold dilution, which extends reporting to 100,000 ng/L. The cobas e 801 can dilute automatically by factors up to 27,000, according to its maker.

The hook, seen from the detector

A detector reports how much light it saw, not which side of a curve the light came from. In a one-step sandwich assay, a very high concentration of target occupies the capture antibodies and the labeled antibodies separately, so fewer complexes carry both antibodies and the light falls again after a peak (Chapter 4). A single reading cannot tell a moderate sample on the rising side of the curve from an enormous one past the peak.

A single reading cannot reveal a hook

Because the curve rises and then falls, one light value matches two concentrations, and the analyzer reports the lower one. The defense is in the design: a stated hook-free limit far above the measuring range, a two-step format where the target allows it, and dilution rules for samples that the clinical picture suggests are extreme.

28 — One light value, two concentrations
measuring range of Roche's assay 1 100 10,000 1,000,000 SIGNAL, RELATIVE TARGET CONCENTRATION, ng/L (LOG) no hook up to 100,000 ng/L (FDA 2017) two steps: levels off one step: rises, then falls moderate sample enormous sample reported as the lower one schematic: curve shapes, not the assay's data
ReadoutFocal detail
Plate 28 — In a one-step sandwich, enormous concentrations fill capture and labeled antibodies separately and the signal falls again, so one light value matches two concentrations. Makers state a hook-free limit far above the range; a two-step design levels off instead of falling.

Makers therefore test how far the signal keeps rising and state a limit. The FDA's decision summary for Roche's one-step troponin T assay reports no hook effect up to 100,000 ng/L, ten times the top of its measuring range. Abbott's high-sensitivity troponin I assay captures the target and washes before adding the labeled antibody, and no hook was reported for it up to 500,000 ng/L, a hundred times its top of range. A two-step design avoids the classic hook because excess free target is washed away before the label arrives. Its signal levels off at the capacity of the capture antibody instead of falling. Roche's method sheet for its insulin assay reports no hook up to 20,000 microunits per milliliter, twenty times the assay's range.

Specify the detector from both ends of the range

Write the detector's requirements from the blank and from the top of the range at once: dark count and collection efficiency set by the limit of blank, linear count rate or analog range set by the highest signal before dilution, and the cathode chosen for the label's wavelength. Then confirm the hook-free limit with real high samples, since the curve beyond the top calibrator is where the detector's numbers and the chemistry interact.

+ What this chapter established
  • A photomultiplier turns a photon into a pulse of millions of electrons, with about 19 percent of 400-nanometer photons counted in one module and far fewer in the red.
  • Photon counts scatter by the square root of the count, so a 2-second read with 50 dark counts a second needs about 30 extra counts to show a signal.
  • Counting spans about five factors of ten before saturating; analog modes and dilution extend the range.
  • A one-step sandwich can hook at very high concentrations, so makers state a hook-free limit well above the range.
14 — Calibration

Two numbers for one sample.

+ The questionTwo well-made troponin assays give different numbers for the same blood. Which one is right?

Ten times apart on the same samples

In 2015 a working group of the International Federation of Clinical Chemistry published a pilot study in which sixteen troponin I assays from six manufacturers measured the same 79 patient samples. When one assay's results were plotted against another's, the slope of the best-fit line ran from 0.33 to 2.87, so the lowest-reading and highest-reading assays differed by a factor of almost nine, which the authors describe as about tenfold. None of the assays was broken. Each did what its maker designed it to do, with its own antibodies, its own calibrators and its own definition of a nanogram per liter of troponin I.

That is the puzzle of this chapter. An immunoassay does not count the molecules in a sample. It reads its result off a curve that converts the analyzer's signal into a concentration, and the curve is only as true as the materials used to draw it. Chapter 2 showed that two antibody pairs can see different subsets of the troponin in blood. This chapter shows how each assay is tied to a scale, and why for troponin I there is still no single scale to tie them all to.

A curve made by the maker, adjusted in the laboratory

A calibration is the set of measurements that relates an assay's signal to concentration. Calibrators are samples with assigned concentrations. An immunoassay's curve through them is not a straight line. Plotted against the logarithm of concentration, it is flat at the bottom, rises through the middle and flattens again near the top. A standard S-shaped curve, the four-parameter logistic function, has been found to fit that shape well. A 1997 patent on immunoassay calibration describes the usual practice as five or six standards, each measured in duplicate.

Roche splits the work in two. For each reagent lot, Roche measures six master calibrators on many instruments and in many runs, and encodes the resulting master curve with the lot. In the laboratory, the analyzer measures two calibrators and uses the two results to adjust the master curve to that instrument and that reagent pack. For Roche's troponin T assay, the two calibrators sit at about 18 and about 4,200 ng/L. Abbott's earlier ARCHITECT troponin I assay worked differently. It was calibrated on the analyzer itself with six calibrators, a zero made from human serum and five containing recombinant cardiac troponin complex, as its FDA decision summary records.

Solved example: why the two-point adjustment matters

Suppose, for illustration, that a fresh electrode and a new reagent pack make one analyzer give 10 percent more light per label than the analyzers on which the master curve was made. Near the bottom of the range the signal is close to proportional to concentration, so without adjustment a patient sample whose true concentration is 18 ng/L would read about 10 percent high: about 19.8 ng/L, just above the US 99th-percentile limit of 19 ng/L. The low calibrator, assigned 18 ng/L, shows the same 10 percent excess, and the adjustment scales it away. A shift too small for anyone to notice in a single result moves a patient across the decision line.

29 — A master curve, adjusted on site
10 100 1,000 10,000 SIGNAL, RELATIVE TROPONIN T, ng/L (LOG) this analyzer: 10 % more light (illustration) master curve (per reagent lot, made by the maker) low calibrator about 18 high calibrator about 4,200 two-point adjustment scales the excess away a true 18 ng/L, read without adjustment US 99th percentile 19 ng/L true 18 19.8 unadjusted 17 18 19 20 21 an unnoticed 10 % crosses the decision line holds 12 weeks per lot, 7 days per pack on board (FDA decision summary) schematic curves
AnalyteReadoutFocal detail
Plate 29 — Roche makes a master curve for each reagent lot and each analyzer adjusts it with two calibrators. Without that adjustment, a 10 percent excess of light would read a true 18 ng/L as 19.8, across the US 99th-percentile limit.

Calibrations do not last. Chapter 12 gave how long a calibration of Roche's troponin T assay holds, as the FDA's decision summary states: 7 days while the same pack stays on the analyzer, and 12 weeks on the same reagent lot. Roche's method sheets also call for recalibration with each new lot and whenever quality control fails.

Lots, drift and quality control

Every new lot of reagent is a slightly different product: antibodies from a new production run, calibrators from a new batch, beads from a new coating. The maker assigns each lot its own curve, but the assignment can be wrong. In July 2020 Siemens Healthineers began a recall of three lots of its Dimension Vista high-sensitivity troponin I reagent because their results read lower than they should, a fault called negative bias. The remedy was a correction factor specific to each affected lot, according to the FDA's recall record. Laboratories check each new lot themselves with patient samples, following protocols such as the Clinical and Laboratory Standards Institute's guideline EP26.

Between lots, laboratories watch for drift with quality control samples of known concentration, run at intervals and plotted on charts. In 1981 James Westgard and colleagues published a set of rules for judging these charts, now known as the Westgard rules. A run is rejected, for example, when one control falls more than three standard deviations from its target, or when two in a row fall more than two standard deviations from it on the same side. Each rule trades catching real shifts against rejecting good runs. Westgard's own guidance gives the cost of a stricter rule: treating a single result beyond two standard deviations as a failure would reject about 9 percent of good runs that use two controls.

Who decides what a nanogram is

For a measurement to mean the same in every laboratory, each assay's calibrators must be linked to a reference that everyone accepts through an unbroken chain of comparisons, a property called traceability. The reference is ideally a reference measurement procedure and a certified reference material, both tied in turn to the International System of Units. In laboratory medicine the international standard for that chain is ISO 17511. Its 2020 edition is still current as of October 2026: ISO reviewed it in February 2026 and kept it without revision.

30 — An unbroken chain, or a missing top
THE IDEAL (ISO 17511) TROPONIN T (ONE MAKER) TROPONIN I (MANY MAKERS) SI unit reference measurement procedure certified reference material maker's master calibrator product calibrator patient result none none none Roche's earlier troponin T assay master calibrator product calibrator patient result agreement by design, within one generation no accepted reference procedure no common scale to tie them to SRM 2921a, NIST, 10 Mar 2026: for validating methods maker A calibrator result maker B calibrator result maker C calibrator result 2015 IFCC pilot: 16 assays, slopes 0.33–2.87
ReadoutFocal detail
Plate 30 — A result is comparable between laboratories only if its calibrators trace back to a common reference. Troponin T agrees because one company makes it; troponin I has no complete chain, so its assays read up to about tenfold apart on the same samples.

Troponin I has no such chain from end to end. In 2006 a committee of the American Association for Clinical Chemistry found that the scatter between 15 assays ran from 82 to 97 percent before harmonization. Adjusting them with shared serum pools cut that to 9 to 23 percent. But the two purified troponin materials tested behaved like patient samples in only 45 percent and 39 percent of the assays respectively, too few for either to serve as a common calibrator. The 2015 pilot that opened this chapter recalibrated its assays to common samples, which cut the scatter between them from 40 to 22 percent at low concentrations. Its authors concluded that troponin I results were not yet traceable to a higher-order reference. On 10 March 2026 the US National Institute of Standards and Technology issued a new certified troponin complex for validating methods, SRM 2921a, and its certification report notes that the assays still need to be standardized.

Troponin T is in a different position, because one company makes the assay. Its results agree between laboratories by design, and the FDA's decision summary states that the 2017 assay is traceable to Roche's own earlier troponin T assay. Even that agreement has limits across generations. By Roche's account, its sixth-generation troponin T assay carried the CE mark by September 2025. The CE mark allows a test to be sold in the European Union. Roche's own reference study gives the new assay 99th-percentile limits of 27 ng/L overall, 18 for women and 32 for men, against 19, 14 and 22 ng/L for the fifth-generation assay cleared in the US. Its results therefore cannot be read against the older limits.

Traceability to one material does not make assays agree

An assay can be traceable to a reference material, as the FDA's record says of Abbott's high-sensitivity troponin I assay, and still read differently from another assay traceable to something else, or to the same material if that material does not behave like patient blood in both. Agreement between assays is shown with patient samples, not inferred from a traceability statement.

The practical consequence runs through every later chapter. In the International Federation's 2019 table of high-sensitivity assays, the makers' 99th-percentile limits for troponin I ran from 8.67 to 58.2 ng/L. The European guideline's one-hour algorithm gives a separate set of cut-offs for each assay. A troponin result has meaning only against the limits of the assay that produced it.

31 — Each assay its own limits
IFCC 2019 table: troponin I 99th percentiles 8.67–58.2 ng/L about sevenfold between troponin I assays 5 10 20 50 100 TROPONIN, ng/L (LOG) Roche troponin T Gen 5 (US) 14 · 19 · 22 ◆ 52 Roche troponin T Gen 6 (CE, 2025) 18 · 27 · 32 Abbott ARCHITECT troponin I 17 · 28 · 35 ◆ 64 Abbott i-STAT troponin I 13 · 21 · 28 QuidelOrtho VITROS troponin I 9 · 11 · 12 ◆ 40 99th percentile (women · overall · men) ESC 2020 rule-in cut-off makers' figures and FDA decision summaries; Gen 6 not cleared in the US
AnalyteReadoutFocal detail
Plate 31 — Each troponin assay carries its own 99th-percentile limits and its own algorithm cut-offs; troponin I limits ran from 8.67 to 58.2 ng/L in the 2019 IFCC table. A number means something only against the limits of the assay that produced it.
Name the calibration chain in every specification

For a new assay, state what the calibrators are traceable to, how the master curve is made and transferred, which on-site calibration adjusts it and how often, and which patient-sample studies show agreement with the assays already in use. For a laboratory, the same list decides whether results from two analyzers can be pooled or must be reported against separate limits.

+ What this chapter established
  • An immunoassay's number comes from a curve through calibrators, and is only as true as the calibrators and the curve's adjustment.
  • Roche ships a master curve per lot and adjusts it on site with two calibrators; a 10 percent drift left uncorrected can push a result across a decision limit.
  • Lot changes and drift are caught with patient-sample checks and quality control rules, which balance missed shifts against false rejections.
  • Troponin I has no complete reference chain and its assays see different forms of the protein, so they differ severalfold, and even successive troponin T generations need separate limits.

+ Part IV · On a strip

The same chemistry without a machine.

The Science Museum Group in Britain keeps a Clearblue One Step pregnancy test made in Bedford, England, in 1988. It is a plastic stick with a pen-shaped sampler to hold in a stream of urine, and a window where, three minutes later, a thin blue line meant the hormone hCG was present. The brand's present owner describes it as the world's first lateral flow test, although the museum's record does not say so. Everything an analyzer does with pumps, magnets, heaters and a photomultiplier, that stick did with a strip of porous material, dried reagents and the eye. The two chapters of this part take the strip apart and then ask what changes when a machine, or a phone, reads the line instead of a person.

15 — Lateral flow

A test on a strip of paper.

+ The questionHow does a strip of porous material do, in fifteen minutes and with no power, what an analyzer does with pumps and magnets?

The parts of a strip

A lateral flow test is an immunoassay run inside a strip of porous material, in which the liquid sample moves by capillary action and carries the reagents with it. Capillary action is the same pull that draws water up a paper towel. A 2009 review of the format lists the parts. A sample pad receives the liquid. A conjugate pad holds the labeled antibody, dried. A membrane, usually nitrocellulose, carries a printed test line of capture antibody and, beyond it, a control line. An absorbent pad at the far end draws the liquid through, and a backing and a plastic housing hold it all together.

The Unilever patent behind the one-step pregnancy test describes the principle that makes the format work. Its first priority date, the date of the earliest filing on which it rests, is 27 April 1987. The labeled antibody is dry on its pad, but it becomes "freely mobile within the porous carrier when in the moist state": the arriving sample dissolves it and carries it along. If the sample contains hCG, the labeled antibody binds it on the way. At the test line a second antibody is "permanently immobilised", and it catches the complex as it passes. Labeled antibody that bound nothing flows on to the control line, which catches the labeled antibody itself and shows that the liquid has run the whole strip.

32 — A sandwich that runs itself
CAPILLARY FLOW TEST LINE CONTROL LINE binding time here is seconds SAMPLE PAD CONJUGATE PAD NITROCELLULOSE MEMBRANE ABSORBENT PAD unbound label flows on test line: catches target with label control line: catches the label itself THE ANALYZER SUBSYSTEM EACH PART REPLACES sample handling reagent addition solid phase incubation + transport detector: eye or reader wash after the Unilever patent, priority 27 April 1987; schematic
AnalyteAntibodyLabel and signalFocal detail
Plate 32 — A lateral flow strip runs a sandwich immunoassay by capillary flow: the sample dissolves a dried labeled antibody, the test line catches target with label, and the control line shows that the liquid ran. Each part does the job of one analyzer subsystem.

Each part maps onto a subsystem of the analyzer in Chapter 10. The sample pad is sample handling; the conjugate pad is reagent addition; the flow along the membrane is incubation and transport together; the test line is the solid phase; the stream that carries unbound label past the line and into the absorbent pad is the wash; and the eye, or a reader, is the detector. The strip has no clock and no thermostat. The material sets its timing.

A label that can be seen

The labels are particles, not molecules, so that a line holding enough of them becomes visible. The 2009 review gives particle sizes from about 15 to 800 nanometers. Colloidal gold, a suspension of tiny gold particles, is the most common, and colored latex, carbon and others are also in use. Gold particles of 20 to 40 nanometers absorb most strongly at 524 to 530 nanometers, in the green, so the light they leave looks red. The effect is called surface plasmon resonance, and it comes from the free electrons of the metal oscillating together.

In 1979 J. H. W. Leuvering filed a patent that described gold sols, another name for colloidal gold, as immunoassay labels. The patent was assigned to Akzona, the US arm of the Dutch group Akzo. One of its examples, a sandwich assay run in tubes, detected hCG at about 5 milli-international units per milliliter. The international unit of a hormone such as hCG is defined by an international reference preparation of the hormone rather than by mass.

Capillary flow sets the clock

Membrane makers describe their products by how long water takes to climb a set distance. Sartorius gives 65 to 115 seconds per 40 millimeters for one of its faster nitrocellulose grades and 95 to 155 for a slower one. Cytiva names its grades by the same measure, from about 80 to about 170 seconds per 4 centimeters. Flow is fastest at the start and slows as the wetted length grows, because the pull of the advancing front has more and more liquid behind it to drag. Lucas and Washburn described the ideal case early in the twentieth century: the distance traveled grows with the square root of time.

Solved example: seconds over a line

Take a membrane in which water climbs 4 centimeters in 125 seconds, near the middle of the makers' ranges, and assume ideal capillary flow, so that the square of the distance grows in proportion to time. The square of the distance reaches 40 squared, or 1,600 square millimeters, in 125 seconds, which is 12.8 square millimeters per second. The front reaches a test line 20 millimeters from the start when the square of the distance is 400 square millimeters, after 400 divided by 12.8, about 31 seconds, and passes 21 millimeters at 441 divided by 12.8, about 34.5 seconds. It therefore crosses a line 1 millimeter wide in about 3 seconds. At the line the front moves at about 0.32 millimeter per second, which is 12.8 divided by twice the 20 millimeters traveled. Each part of the sample spends only seconds over the capture antibody, against the minutes an analyzer gives the same binding.

33 — Seconds over the line
0 10 20 30 40 0 25 50 75 100 125 DISTANCE OF THE LIQUID FRONT, mm TIME, S TEST LINE AT 20 mm LIQUID FRONT 31 s front speed 0.32 mm/s: crosses a 1 mm line in about 3 s seconds of binding, against an analyzer's minutes membrane: water climbs 4 cm in 125 s (middle of makers' ranges) halve the flow time, lose about 4× in sensitivity (Millipore rule of thumb) ideal flow (Lucas–Washburn): distance grows with the square root of time
AnalyteReadoutFocal detail
Plate 33 — Capillary flow slows as the wetted length grows, but at a test line 20 mm from the start the front still moves about 0.32 mm per second. Each part of the sample spends only seconds over the capture antibody, the strip's central limit.

The short contact time is the strip's central limit. Millipore's guide for membrane users states that the capillary flow rate sets the time reactants spend close enough to bind. It gives a rule of thumb: the effective concentration of analyte, the substance the test measures, falls with the square of the flow rate. By that rule, halving the flow time, from 180 to 90 seconds per 4 centimeters, makes the strip about four times less sensitive to the analyte. Designers lengthen the binding time with slower membranes, with test lines placed farther from the start, where the flow is slower, and with larger sample volumes. Those choices cost minutes and liquid.

How little a line can show

The limits that result appear in regulators' records. In January 2022 the FDA cleared one pregnancy test cassette for professional use through a 510(k), and its decision summary gives the cut-off as 10 milli-international units of hCG per milliliter in serum and 20 in urine. At 8 milli-international units per milliliter, 20 percent below the serum cut-off, 78 percent of serum samples still read positive. A strip's cut-off is not a sharp edge: near it, the line darkens gradually with concentration, so samples just below the cut-off read positive part of the time. The makers tested for a hook up to 2,000,000 milli-units per milliliter, a hundred thousand times the urine cut-off, and found none. Such testing is needed because a strip that runs sample and labeled antibody together can still saturate at extreme concentrations, so its makers have to find out how high a concentration it tolerates.

Troponin shows the limit of what strips can measure. Roche's cobas h 232 is a point-of-care reader, one used near the patient rather than in a central laboratory, for Roche's whole-blood cardiac strips. According to its maker, it measures troponin T from 100 to 2,000 ng/L in 12 minutes. The low end of that range is about seven times 14 ng/L, the 99th-percentile limit of Roche's high-sensitivity laboratory troponin T assay on its European label and for women in the US. The cardiac strip has not reached the floor of the laboratory's high-sensitivity assays. The instrument-read cartridges of Chapter 16 go lower.

A line read after the window answers a different test

Liquid keeps moving after the stated read time, carrying more label past the lines and letting faint background build. In June 2020 the Indian Council of Medical Research told users of the first rapid antigen test it approved for COVID-19, a strip that detects a protein of the virus, to read at 15 minutes and to discard the strip after 30 minutes. A line that appears later is not the result the test was validated to give, whatever its color.

Choose the membrane for binding time, not speed

A fast membrane gives a quick result and a weak line. Pick the membrane, the line position and the sample volume from the binding time the antibody pair needs, measured on strips. Then fix the read window from the same experiments, so that the instructions for use, the cut-off and the hook-free limit all describe one timed process.

+ What this chapter established
  • A lateral flow strip runs a sandwich immunoassay by capillary flow, with dried labeled antibody dissolved by the sample and captured at a printed line.
  • Its pads and membrane perform the analyzer's sampling, reagent addition, incubation, capture and washing, timed by the material.
  • Flow slows with distance, and each part of the sample spends only seconds over the test line, which limits how little the strip can detect.
  • One cleared pregnancy cassette has cut-offs of 10 milli-international units of hCG per milliliter in serum and 20 in urine, while one troponin strip reads only from 100 ng/L.
16 — Readers

When a line becomes a number.

+ The questionWhat does a reader add to a test a person can read by eye, and what changes when the person does the reading?

The same strips, read twice

From November 2020 the city of Liverpool, home to about 498,000 people, offered its residents rapid antigen tests for the virus that causes COVID-19, tests that detect a protein of the virus. The city used the Innova lateral flow test, in a program run as a pilot. As part of the pilot's evaluation, people without symptoms at 48 testing sites took two swabs minutes apart under supervision. One swab went to the strip and one to a laboratory PCR test, which detects the virus's genetic material after copying it many times.

On 23 December 2020 the University of Liverpool published an interim evaluation that reported 5,869 paired results. After void results were set aside, the strips read on site had found 28 of the 70 people whose PCR test was positive. That is a sensitivity of 40 percent, where sensitivity in this clinical sense is the share of infected people a test finds. Among the people with higher viral loads, meaning more virus in their samples, the strips found about two thirds.

The evaluators then had reviewers who did not know the PCR results re-read photographs of the same strips. On re-reading, the sensitivity rose to 53.4 percent overall, and among the higher viral loads from 66.7 to 83.3 percent. The specificity, the share of people without the infection whose strip read negative, was unchanged at 99.9 percent. Most of the discordant readings had come in the first week of the pilot. The report linked that pattern to the testing staff's inexperience, and it cautioned that the numbers were small. The strips had not changed. What changed was who looked at them, and how.

A separate national evaluation was carried out at the government laboratory at Porton Down and the University of Oxford, and it found that the strip's result also depended on who performed the test. A briefing for journalists on 11 November 2020 quoted its preliminary report. The report gave the strip's sensitivity as about 79 percent when laboratory scientists performed the test, 73 percent with trained health workers and 58 percent with members of the public who had trained themselves from the instructions and swabbed and tested themselves. The groups were not formed at random, so the figures are not a like-for-like comparison.

Solved example: what the Liverpool strip's results meant

Leaving aside void results, the paired study had 31 positive strips, and PCR confirmed 28 of them, so a positive strip was right about 90 percent of the time. Of 5,434 people whose PCR test was negative, 5,431 had a negative strip: a specificity of 99.9 percent. But 42 of the 70 PCR-positive people had a negative strip. In this population, where 70 of the 5,504 people with valid results, about 1.3 percent, were infected, a negative strip was right 5,431 times in 5,473, or 99.2 percent, because almost everyone was negative, yet the strip missed three in five of the infections it was meant to catch. The share of positive results that are right is called the positive predictive value, and the share of negative results that are right the negative predictive value; both depend on prevalence, the share of the people tested who have the condition. The PCR guide in this library works through the same arithmetic for test performance and prevalence.

34 — The same strips, read twice
0 25 50 75 100 SENSITIVITY, % 40 % read on site 53.4 % photos re-read blind ALL PCR-POSITIVE (70) 66.7 % on site 83.3 % re-read HIGHER VIRAL LOADS same strips, different reading specificity 99.9 % both ways; small numbers Liverpool pilot interim report, 23 Dec 2020 WHO PERFORMED THE TEST laboratory scientists 79 % trained health workers 73 % self-trained public 58 % 0 25 50 75 100 SENSITIVITY, % Porton Down/Oxford, as quoted 11 Nov 2020; groups not randomized
ReadoutFocal detail
Plate 34 — Re-reading photographs of the same Liverpool strips raised sensitivity from 40 to 53.4 percent, and a national evaluation found it fell from 79 to 58 percent between laboratory scientists and self-trained members of the public. The reader belongs to the test.

What an instrument reader measures

An instrument reader replaces the eye with a light source and a detector. The FDA cleared the BD Veritor system for influenza on 23 March 2012. Its battery-powered bench reader measures the light reflected from each line of the strip. The decision summary records that the reader subtracts the signal at a negative-control line from the signal at each test line and scores the difference against a cut-off. The negative-control line is not the control line of Chapter 15, which catches labeled antibody to show that the strip has run. Its signal measures the strip's nonspecific color, the color that appears whether or not the target is present. The decision summary also states that the strip "cannot be interpreted visually" and that the subtraction is one that "the human eye is unable to accurately perform".

The subtraction is the reader's real contribution. The eye judges a faint line against the color of the membrane around it, under whatever light is in the room, and different people set their own threshold for "a line". A reader measures both the line and the background in fixed geometry, with its own light, and applies the same threshold every time. It can also save the reading, flag a strip that did not run, and send the result to a record.

35 — What an instrument reader subtracts
the eye: faint line against the membrane, room light, own threshold NEGATIVE-CONTROL LINE TEST LINE CONTROL LINE REFLECTANCE POSITION ALONG THE STRIP → test − negative control cut-off a difference the eye cannot make reliably after BD Veritor's FDA decision summary (2012): 'cannot be interpreted visually'; schematic
Label and signalReadoutFocal detail
Plate 35 — An instrument reader measures each line and the background in fixed geometry with its own light, subtracts the strip's nonspecific color and applies one threshold every time. That subtraction is what the eye cannot do reliably.

Readers also allow labels that the eye cannot see. The FDA cleared Quidel's Sofia analyzer on 24 October 2011. According to its decision summary, the analyzer scans the whole strip and reads beads dyed with a europium chelate, excited at 365 nanometers and glowing at 618. A 2012 study compared europium nanoparticles, read by time-resolved fluorescence, with colloidal gold in two model lateral flow assays. The fluorescent label detected concentrations about 7 times lower than gold could in one assay and about 300 times lower in the other. In a 2017 study, academic developers of a thermal reader reported that it detected concentrations about ten times lower than reading by eye could. Their reader detects the heat that gold particles release when lit by a laser.

Phones as readers

A phone carries a camera, a light and a computer, and has become a reader in its own right. A 2023 study tested one smartphone reader, built around a 3D-printed holder and an app that works in real time. At a low antigen concentration, volunteers judged only 20 percent of test strips positive, while the app classified all of them correctly. The study used a single phone model. In August 2021 BD and Scanwell Health received emergency use authorization in the US for a home COVID-19 test read by a phone's camera, which BD described as the first of its kind. Through an emergency use authorization, the FDA allows a test to be used during a declared public health emergency without clearance or approval.

India used the phone differently. On 19 May 2021 the Indian Council of Medical Research issued an advisory on home testing with rapid antigen tests, naming one approved kit. It required users to photograph the strip with the phone they had used to register in a mandatory app, and the app recorded the result and sent it to the national testing portal. The same council had validated 95 rapid antigen kits by April 2021. Its threshold for use at the point of care was a sensitivity of at least 50 percent and a specificity of at least 95 percent.

Troponin at the bedside

The readers that now bring high-sensitivity troponin to the bedside are not strips read by eye but cartridges read by instruments. The FDA cleared the PATHFAST high-sensitivity troponin I test on 20 March 2024, and its distributor called it the first such test cleared for use at the point of care in the US. Abbott's handheld i-STAT high-sensitivity troponin I cartridge followed on 3 January 2025. According to its FDA decision summary, it is an assay with electrochemical detection that uses about 22 microliters of whole blood, gives a result in about 15 minutes and has 99th-percentile limits of 13 ng/L for women and 28 for men. Siemens Healthineers states that its Atellica VTLi gives a result from a fingerstick in eight minutes. The company announced the analyzer's CE mark in April 2021, and the analyzer was not found among US clearances as of October 2026.

A reader reads the strip's chemistry, faults included

A reader removes the variation of human judgment, but it cannot recover target that the strip failed to capture, and it can apply its threshold with great consistency to a strip that ran badly. Its cut-off is a design decision that has to be validated with the strips, samples and users it will meet, as Chapter 23 describes.

Validate the strip, the reader and the user as one system

For a reader-based test, run the clinical study with the reader, software version and intended users the product will have, and include strips that are faint, late or partly run. For a visually read test, test reading by the intended users under realistic light, since the Liverpool figures show that the reader, human or electronic, is part of the test, and the Porton Down figures show that the operator is too.

+ What this chapter established
  • Re-reading photographs of the same Liverpool strips raised sensitivity from 40 to 53.4 percent, so the reader changed the result.
  • Instrument readers measure line and background in fixed geometry, subtract nonspecific color and apply one threshold every time.
  • Fluorescent labels, and thermal reading of gold labels, lowered detection limits severalfold to a few hundredfold in published comparisons.
  • High-sensitivity troponin at the bedside comes from instrument-read cartridges, two of them cleared in the US by January 2025.

+ Part V · One molecule at a time

The floor, counted.

The disc at the heart of one digital immunoassay analyzer holds 216,000 wells. Each has a volume of about 40 femtoliters, a twenty-five-millionth of a microliter, and is just large enough to hold one bead. In wells that small, a single enzyme molecule soon fills its well with enough fluorescent product to see, and the instrument can count the wells that light instead of measuring a glow. The two chapters of this part follow that idea from its first immunoassay paper to its limits, and they complete the answer to the central question: what sets the floor once single molecules can be seen.

17 — Counting single molecules

Counting instead of measuring.

+ The questionIf a few hundred labels are too few to see as a glow above an assay's background, can each one be seen on its own?

One enzyme in fifty femtoliters

On 23 May 2010 the journal Nature Biotechnology published a paper describing an immunoassay that counted single protein molecules in blood. Fourteen of its fifteen authors worked at Quanterix, a company in Cambridge, Massachusetts. The fifteenth, David Walt of Tufts University, was an adviser to the company, and all held shares or options in it. The paper called the method a single-molecule enzyme-linked immunosorbent assay, or digital ELISA.

The chemistry is a sandwich on beads, as in Chapter 9. Capture antibodies on beads 2.7 micrometers across catch the target; a detection antibody carrying biotin binds it; and streptavidin attached to an enzyme, beta-galactosidase, binds the biotin. When target is scarce, most beads catch no molecule at all, and the few others catch one each. The beads are then loaded into an array of tiny wells, about 50 femtoliters each in the 2010 paper, so that each well holds at most one bead. A substrate is added, the wells are sealed, and any well holding an enzyme fills with a fluorescent product.

Concentration is what makes a single molecule visible. One molecule of product in a tenth of a milliliter of liquid, the volume of an ordinary assay, is undetectable. The same molecule in 50 femtoliters is at a concentration of about 33 picomolar, and the enzyme goes on making more. Each well is either bright or dark. The instrument photographs the array and counts the bright ones. According to the paper's abstract, the method detected as few as 10 to 20 enzyme-labeled complexes in 100 microliters of sample, and it measured prostate-specific antigen at 14 femtograms per milliliter, or 0.4 femtomolar, in the serum of men who had had their prostates removed. The prefix femto- means a thousandth of pico-.

The ratio of molecules to beads is what makes the count digital. A company poster reporting the same work describes one experiment with about 400,000 beads and a disposable array of more than 50,000 wells. Suppose that many beads share the paper's lowest level, 10 to 20 labeled complexes in 100 microliters. There is then one complex for every 20,000 to 40,000 beads, so a bead that lights almost certainly carries exactly one molecule, and the number of lit beads is the number of molecules seen. The same arithmetic shows the cost: a few lit wells have to be found among tens of thousands of dark ones, so the array and the camera that reads it matter as much as the label.

36 — One enzyme lights a 40-femtoliter well
target biotin streptavidin–enzyme (β-galactosidase) BEAD 2.7 µm one bead per well SEALED ARRAY, TOP VIEW count wells, not brightness most beads caught nothing: dark one enzyme: the well lights 1 product molecule in 50 fL ≈ 33 pM HD-1 disc: 216,000 wells of 40 fL; 25,000–40,000 beads imaged per test
AnalyteAntibodyLabel and signalFocal detail
Plate 36 — In digital ELISA most beads catch no target. Sealed one to a well of about 40 to 50 femtoliters, a single enzyme raises its product to tens of picomolar and the well lights, so the instrument counts lit wells instead of measuring a faint glow.

Counting the lit beads

A count of bright wells becomes a concentration through a rule from probability. When molecules are spread at random over many beads, the number on any one bead follows what statisticians call a Poisson distribution, which describes counts of independent events that occur at a fixed average rate. Under that distribution, the average number per bead fixes the share of beads carrying none. The company's papers call that average the AEB, for average enzymes per bead, and use it as the common unit: the instrument counts the share of beads that are lit and converts it into AEB.

Solved example: counting at one enzyme per ten beads

At an average of 0.1 enzyme per bead, the Poisson rule gives about 90.5 percent of beads with no enzyme, 9.0 percent with exactly one and 0.5 percent with two or more. So about 9.5 percent of beads light, and 95 percent of the lit beads carry a single enzyme. If the instrument reads 25,000 beads, about 2,380 are lit, and by the square-root rule of Chapter 13 the count has a scatter of about 49, or 2 percent. A tenfold lower concentration, 0.01 enzyme per bead, lights about 250 beads, with a scatter of about 6 percent.

As concentration rises, more beads carry two or more enzymes, and the count of lit beads no longer follows the concentration in a simple way. Once more than 70 percent of the beads are lit, the analyzer described in 2016 switches to measuring their average brightness, an analog mode. A 2011 paper by the company states the same limit in another form, an average of about 1.2 enzymes per bead. The digital range spans about two and a half factors of ten, and with the analog mode added the total exceeds four.

37 — Counting lit beads
DIGITAL: COUNT LIT BEADS ANALOG: AVERAGE BRIGHTNESS 0 25 50 75 100 0.001 0.01 0.1 1 10 SHARE OF BEADS LIT, % AVERAGE ENZYMES PER BEAD (AEB), LOG SCALE switch at 70 % lit (AEB 1.2) AEB 0.1: 9.5 % lit above here, counting stops tracking BEADS AT AEB 0.1 none 90.5 % one enzyme 9.0 % two or more 0.5 % at AEB 0.1, 95 % of lit beads carry one enzyme 25,000 beads read: about 2,380 lit, scatter about 2 %
ReadoutFocal detail
Plate 37 — Random loading makes the share of lit beads a fixed function of the average enzymes per bead, so a count converts to a concentration. Above about 70 percent lit, beads carry two or more enzymes and the analyzer switches to measuring brightness.

From a paper to an analyzer

The method reached laboratories as an automated analyzer, the Simoa HD-1. Authors from Quanterix and from Stratec, the company that built the instrument, described it in 2016. Its disc holds 216,000 wells of 40 femtoliters. A test for a single analyte loads 400,000 to 500,000 beads, and the instrument images about 25,000 to 40,000 of them. The analyzer runs 66 samples an hour, with the first result after about an hour. The paper defines the limit of detection as 2.5 standard deviations above the background. It reports limits such as 0.020 picogram per milliliter for prostate-specific antigen and 0.0020 picogram per milliliter for interleukin 6, and an average analytical sensitivity more than 1,200 times that of ELISA across 16 proteins.

A fold gain depends on the comparator chosen

In a poster from 2010 reporting the same work, the company compared the method with one commercial prostate-specific antigen test and found it about 15,000 times more sensitive. The 2016 paper reported an average of more than 1,200 times against ELISA. Both comparisons are the maker's, and each depends on which conventional assay was picked. A claim of greater analytical sensitivity is only meaningful against a named assay measured on the same samples.

The HD-1's limits are for other proteins, and they sit a hundred to a thousand times below the floor of the troponin assays in Part III. The current analyzer, the HD-X, is labeled by its maker for research use only, which excludes its use in diagnosing patients. Quanterix offers blood tests for two proteins linked to damage and disease in the brain, neurofilament light and phosphorylated tau 217 (pTau217), the tau fragment of Chapter 1. It offers them as laboratory-developed tests, which are designed and run within a single laboratory. In early 2026 the company submitted a 510(k) for an algorithmic Alzheimer's test combining several markers, and no FDA clearance of it had been found as of October 2026. Chapter 18 asks why, with single molecules countable, none of the FDA clearances of Alzheimer's blood tests went to a test on a digital analyzer.

Count the molecules in the sample before choosing digital

Before specifying a digital assay, work out how many target molecules the sample volume holds at the lowest concentration that matters, how many the capture step can catch, and how many of those the reader will actually see. If that final number is in the tens, counting statistics set the floor whatever the reader's sensitivity, and a larger sample or more efficient capture will help more than a better detector.

+ What this chapter established
  • Digital ELISA loads beads into wells of tens of femtoliters, where one enzyme makes enough product to light its well.
  • Counting lit wells, with the Poisson rule, gives the average number of enzymes per bead and so the concentration.
  • The 2010 paper detected 10 to 20 labeled complexes in 100 microliters; the 2016 analyzer reported limits near a hundredth of a picogram per milliliter.
  • The method's analyzers are labeled for research use, and no FDA clearance of a digital ELISA test had been found as of October 2026.
18 — Beyond one label

What sets the floor now.

+ The questionOnce single labels can be counted, what limits a test?

Four cleared tests, none of them digital

By October 2026 the FDA had cleared at least four blood tests that aid in assessing the amyloid plaques of Alzheimer's disease, and none of them ran on a digital analyzer. The count comes from this guide's search of the FDA's records, and the search may have missed a clearance. Amyloid plaques are deposits of the protein fragment beta-amyloid in the brain. Three of the tests were immunoassays on conventional analyzers of the kinds described in Part III, and two of those measured tau protein that carries a phosphate group at amino acid 217, called pTau217. The fourth used mass spectrometry, which identifies and measures molecules by their mass. Quanterix, whose method Chapter 17 described, had submitted its own test for clearance, and no decision on it had been found.

The puzzle is that pTau217 circulates at very low concentrations. In a small 2024 study on one analyzer, people with amyloid plaques had on average about 0.66 picogram per milliliter in their plasma and people without them about 0.11. A picogram per milliliter is the same as a nanogram per liter, so these concentrations sit well below the troponin limits of Part III. Yet the conventional assays reach them. Fujirebio's pTau217 assay runs on its Lumipulse analyzer as a chemiluminescent enzyme immunoassay of the kind in Chapter 7. According to the FDA's decision summaries, that assay has a limit of detection of 0.041 pg/mL, and Roche's electrochemiluminescent pTau217 assay has a limit of 0.056 pg/mL. Roche's cut-offs sit at 0.223 and 0.378 pg/mL, four to seven times above its limit of detection.

The makers of the digital analyzer of Chapter 17, Quanterix and Stratec, reported limits of about 0.002 to 0.02 pg/mL for other proteins. Those limits are lower again, although limits for different proteins and antibodies are not directly comparable. Whatever the commercial reasons for the outcome, the arithmetic shows that the conventional assays had already reached the concentrations at which patients differ. That finding changes the question this part set out to answer. Counting single labels removed the detector as the limit. What remains is the label that lights without a target, the share of the sample the assay actually looks at, how strongly the antibody holds its target, and the number of molecules in the drop.

The blank, one bead at a time

In a digital assay the blank is not a faint glow but a count. The 2016 paper on the HD-1 analyzer reported that backgrounds were generally between 0.005 and 0.01 enzymes per bead. At those levels, 0.5 to 1 percent of the beads light in a sample with no target, about 125 to 250 of 25,000 beads read. The paper does not say which molecules light them. By definition they are beads carrying label that no target holds, the background of Chapter 5 seen one molecule at a time.

Solved example: where a digital assay's floor sits

Take a blank of 0.005 enzymes per bead and 25,000 beads read, so that about 125 beads light with no target. By the square-root rule of Chapter 13, repeated blanks scatter by at least about 11 beads. Real blanks scatter more, since the 2016 paper notes that counting noise dominates only below about 50 background beads. The paper set its limit of detection at 2.5 standard deviations above the background, so a sample must light about 28 more beads than the blank to be detected, an extra 0.0011 enzyme per bead, and more when the blank scatters more than this minimum. The analyzer read 25,000 of the roughly 500,000 beads in the test, so those 28 beads stand for about 560 labeled molecules or more, spread over all the beads. Where counting noise dominates, halving the blank lowers the threshold only by the square root of two, to about 20 beads.

The example shows why a more sensitive camera would not help: every lit bead in the blank is already seen. What sets the floor is how many beads light without a target and how much that number scatters from one blank to the next. A 2012 theoretical paper by the company's scientists notes that lower concentrations of label reduce the background. But binding is a balance that shifts with concentration, as Chapter 3 showed, so less label also binds fewer of the captured molecules. The trade between signal and background that Chapter 5 described therefore returns at the scale of single molecules.

Molecules in the drop

Counting also exposes a limit that a glow hides: the number of molecules there are to count. On the HD-1, only about 5 percent of the beads in a test are sealed into wells and read. Researchers in the laboratory of David Walt, a co-author of the 2010 method, gave that figure in a 2020 paper. At 10 attomolar, a hundredth of a femtomolar, a 100-microliter sample holds about 600 molecules of the target. Even if every one were caught and labeled, reading one bead in twenty would see about 30 of them, with a counting scatter near 18 percent.

The same group sealed beads into droplets instead of wells and read up to 60 percent of 100,000 beads. For two signaling proteins of the immune system, limits of detection fell to 30 attomolar for interferon gamma, against 350 attomolar on the commercial analyzer, and to 20 attomolar for interleukin 2, against 550. At the lowest limit, that is about 1,200 molecules in the whole sample. The method was a research technique, not a product.

The droplet study's authors also noted that the antibodies' dissociation constants limit how low a concentration the assay can detect. As Chapter 3 showed, the share of target an antibody holds depends on how the antibody's concentration compares with that constant, and a complex that comes apart during the washes takes its label with it. In their 2012 theoretical paper, the company's scientists put the efficient range for digital ELISA at antibodies with dissociation constants from about 10 picomolar to 1 nanomolar.

38 — Where the molecules go
IN THE SAMPLE 10 aM in 100 µL CAUGHT AND LABELED BEADS READ about 5 % (HD-1) READOUT about 600 molecules not caught or lost in washes a share set by the antibodies (Chapter 3) about 30 counted if all were caught the result rests on a few dozen molecules 30 ± about 5 scatter near 18 % droplet method (research, 2020): up to 60 % of beads read; limits 20–30 aM blank: about 125 of 25,000 beads lit (AEB 0.005); +28 needed to detect arithmetic from the 2016 and 2020 papers; schematic proportions
AnalyteAntibodyLabel and signalInterferentReadoutFocal detail
Plate 38 — At 10 attomolar, a 100 µL sample holds about 600 molecules, and an analyzer that reads one bead in twenty sees at most about 30 of them. Below the detector's limit, the floor is set by molecules lost in capture and sampling and by beads that light without target.

Sample volume is the last factor that can be raised. A larger sample holds proportionally more molecules, and the tests in the 2016 paper used from 25 microliters of sample for most proteins up to 144 microliters for one viral protein. The limit on volume is practical: more sample means more time to capture from it, more beads to read, and more of everything else in blood that can stick.

Two antibodies that must meet

Another way to lower the background is to demand more before a signal counts. In a sandwich, a detection antibody stuck to the surface reads as if a target held it in place. In 2002 a group at Uppsala University in Sweden described a proximity ligation assay. In this assay, two binders, each carrying a short strand of DNA, must attach to the same molecule before their strands lie close enough to be joined into a new piece of DNA. The joined piece is then copied and measured, and a binder on its own leaves nothing to copy. The group used short DNA molecules called aptamers as binders and detected a growth factor at 40 zeptomoles, about 24,000 molecules, "without washes or separations". A zeptomole is a thousandth of an attomole.

In 2014, authors mostly from the company Olink described a version of the proximity extension assay that they called 96-plex. It used pairs of antibodies carrying matched DNA strands, which an enzyme extends into a new piece of DNA when both antibodies bind the same molecule, and it ran 92 protein assays together in 1 microliter of sample. The omics guide in this library explains how the resulting DNA barcodes are read. Running several assays in one sample is called multiplexing, and the Olink authors argued that conventional multiplex sandwich assays stop at about ten targets, because the chances of a wrong antibody pairing multiply with every target added. In their format a signal needs both antibodies on the same molecule and DNA strands that match each other.

Each added target adds ways to be wrong

With ten capture antibodies and ten detection antibodies in one well, there are ten intended pairs and ninety unintended ones, each a possible bridge if one antibody cross-reacts. The authors of the HD-1 paper expected cross-reactivity between antibodies to limit the analytical sensitivity of multiplex tests. Their multiplex tests read about 10,000 beads per target, against 25,000 to 40,000 in a single test, so the counting scatter grows as well. A floor measured for one target alone does not carry over to a panel.

39 — Two binders must meet
PROXIMITY LIGATION (2002) joined into new DNA binders: DNA aptamers in 2002 copied and measured alone: nothing to copy MULTIPLEXING (A) CONVENTIONAL MULTIPLEX SANDWICH A B C wrong pairing reads as signal each added target adds ways to be wrong (B) PROXIMITY EXTENSION A B tails do not match: no signal 10 capture + 10 detection antibodies: 10 intended pairs, 90 unintended conventional multiplexing stops at about ten targets (authors of the 2014 paper); omics guide, Chapter 9
AnalyteAntibodyLabel and signalInterferentReadoutFocal detail
Plate 39 — Proximity methods require two binders on the same molecule before their DNA strands can be joined or extended, so a binder stuck alone gives nothing. That also removes most wrong pairings, which multiply with every target added to a conventional multiplex.

What sets the floor

Chapter 4 traced the troponin T floor from 100 ng/L in 1992 to a quantitation limit of 6 ng/L in the 2017 assay. Abbott's high-sensitivity troponin I assay detected 1.7 ng/L in its 2019 US clearance. In 2007 scientists at the company Singulex described a single-molecule counting assay that detected 0.2 ng/L. Across those years, what set the limit changed. The early floors in immunoassays were set by the label and the way it was read: a radioactive count, a color in a photometer. Labels that give light from a chemical reaction or at an electrode, read in the dark, cut the reader's own noise until the detector stopped being the limit. The analyzer's washing, timing and calibration kept the floor steady from run to run.

With labels counted one by one, the floor is set by what the antibodies do and by what the sample offers: the labels left stuck without a target, the share of target caught and kept through the washes, and the number of molecules in the volume tested. The answer to the central question therefore comes in sequence. The label and its reader set the floor first, the instrument held it, and today the antibodies and the drop set it.

40 — Who set the floor
healthy people: about 1–50 ng/L 0.1 1 10 100 1990 1995 2000 2005 2010 2015 2020 2025 TROPONIN DETECTION LIMIT (OR BOTTOM OF RANGE), ng/L, LOG LABEL AND READER SET THE FLOOR THE INSTRUMENT HOLDS IT T, first assay (Boehringer Mannheim): range from 100 T, 2nd generation: 12 T, high-sensitivity (Europe): 5 I, Abbott ARCHITECT: 1.7 I, VITROS hs (maker's claim): 0.59 T, US clearance: LoD 3–5, reported from 6 I, single-molecule counting (Singulex staff): 0.2 ANTIBODIES AND THE DROP below the detector, the blank and the drop decide T = troponin T, I = troponin I
AnalyteReadoutFocal detail
Plate 40 — The troponin floor fell from about 100 ng/L in 1992 to a few ng/L and, in single-molecule counting, to 0.2 ng/L. The limit passed from the label and its reader to the analyzer that held it steady, and then to the antibodies and the number of molecules in the drop.

A lower floor is worth having only where the biology calls for it. An international task force, whose criteria Chapter 20 sets out, tabulated results from a 2012 study that measured several assays in one healthy population. At the limit of detection of 5 ng/L then on its label, Roche's high-sensitivity troponin T assay measured troponin in 43 percent of the men and 7 percent of the women. Abbott's ARCHITECT troponin I assay measured troponin in 80 percent of the men and 67 percent of the women, at the limit of detection of 1.9 ng/L that the task force's table gave it, against 1.7 ng/L in its US clearance. Chapter 20 shows why measuring healthy people matters for a decision made within an hour. For pTau217 the conventional analyzers already reach the concentrations that separate patients. Below that point a lower floor adds little to the decision the test supports, unless it also improves precision at the cut-offs, where about a fifth of patients in two of the cleared tests, which use two cut-offs, get an intermediate result that is neither positive nor negative.

Find the binding limit before paying for sensitivity

Before specifying a more sensitive assay or analyzer, work out which limit sets the floor at the decision concentration: the scatter of the blank, the share of molecules caught, the share of the sample read, or the number of molecules in the volume. A better detector helps only with the first, and only when the blank's scatter comes from the detector rather than from label stuck where no target holds it.

+ What this chapter established
  • On a digital analyzer the blank is a count of beads lit without target, and its scatter, not the camera, sets the floor.
  • Reading a twentieth of the beads, and catching only part of the target, can leave a result resting on a few dozen molecules.
  • Proximity methods require two binders on one molecule before a signal counts, which cuts background and lets many assays share a sample.
  • The floor passed from the label and its reader to the antibodies and the drop, and a test needs only the floor its biology sets.

+ Part VI · Reading the number

When a result is right, and when it is not.

On 28 November 2017 the FDA issued a safety communication warning that biotin can make laboratory tests read falsely high or falsely low. Biotin is a B vitamin, sold in high doses in supplements for hair, skin and nails. The agency reported that one patient taking high levels of biotin had died after falsely low troponin test results. Biotin harms neither a reagent nor an analyzer; it works through the chemistry many tests are built on. The three chapters of this part read results: what makes a well-run assay wrong for one patient, how the numbers from Roche's troponin T assay become a decision within an hour, and why the first blood test cleared to aid the diagnosis of Alzheimer's disease reports a ratio, with a zone that gives no answer.

19 — When the number misleads

Hook, heterophiles and biotin.

+ The questionWhat can make a well-run assay give one patient a wrong number?

Seven samples in 21,000

A 2004 review of interference in immunoassays cites a study of hospital samples run on immunoassays that contained blocking reagents. In that study, a patient's own antibodies interfered in seven samples in 21,000. The same review reports that human antibodies against mouse proteins, which can bind the mouse antibodies many reagents use, have been found in 30 to 40 percent of patient samples. Both numbers can be true at once, because the blocking reagents of that study and the other defenses that makers build into the reagent, described below, account for much of the gap between them.

Human antibodies that bind the antibodies of other animals take two forms. Heterophile antibodies have no clearly defined origin and bind the antibodies of two or more species. Human anti-animal antibodies are made against one animal's proteins, for example after treatment with mouse antibodies, and bind them strongly and specifically; the most common are human anti-mouse antibodies, or HAMA. In practice the two names are often used for each other. A 1999 review notes that they can arise from treatment with animal antibodies and from other causes, can reach high levels and can persist for years. Many immunoassay antibodies are made in mice, as Chapter 2 described, so a heterophile antibody can bind the capture antibody with one arm and the labeled antibody with the other. The analyzer then reads a sandwich with no target in it, and the result is falsely high. An antibody that binds only one reagent antibody can instead stop it catching the target, and the 2004 review offers that blocking as the likely route to a falsely low result. Rheumatoid factor, an antibody that binds the stem of other antibodies, can bridge in the same way, and the same review notes that it can cause false positives in troponin assays.

41 — Four ways a sandwich can mislead
BRIDGE: FALSELY HIGH heterophile antibody or HAMA a sandwich with nothing in it BLOCK: FALSELY LOW antibody against the reagent antibody FREE BIOTIN: FALSELY LOW IN A SANDWICH washed away biotin occupies the streptavidin MACROCOMPLEX: HIGH ON SOME ASSAYS target bound to the patient's own antibody schematic; mechanisms as described in reviews (1999, 2004) and FDA communications
AnalyteAntibodyLabel and signalInterferentFocal detail
Plate 41 — A patient's own antibodies can bridge or block the reagent antibodies, free biotin can occupy the streptavidin that holds the sandwich, and a target bound to an antibody can read high on one assay and normal on another. Each gives a wrong number from a well-run test.

Makers build their defenses against all of these into the reagent. They add non-immune animal serum or mouse antibodies that bind the patient's antibodies before they reach the sandwich; they use antibody fragments that lack the stem; and they use antibodies built partly from human sequence, such as the labeled antibody of Roche's troponin T assay. These defenses are not complete. The 2004 review summarizes a study in which 66 laboratories in seven countries measured 74 analytes in the blood of ten donors, chosen because their blood was known to interfere. About 6 percent of the results were false positives, and the kits' blockers failed to prevent about two thirds of the interferences. Because the donors were selected for interference, the 6 percent is a measure of how hard the problem is, not of how often it occurs.

Biotin, the reagent's own glue

Biotin interferes for a different reason: many assays use it themselves. Chapter 4 described Roche's troponin T assay, in which the capture antibody carries biotin and the magnetic beads carry streptavidin, a protein that binds biotin very tightly. Free biotin in the patient's blood occupies the streptavidin on the beads, so the capture antibody has nowhere to attach and is washed away with the troponin and the label it holds. In a sandwich assay the result reads falsely low. In a competitive assay, where less label means more target, the same loss reads falsely high.

A 2017 study in JAMA measured the effect directly. Six healthy adults took 10 milligrams of biotin a day for a week. Their blood was drawn about two hours after the last dose and run on 37 immunoassays on four systems. Interference appeared in 9 of the 23 assays that used biotin and in none of the 14 that did not. Five of eight competitive assays read falsely high and four of fifteen sandwich assays falsely low; the result of one assay for thyroid-stimulating hormone fell by about 94 percent.

The FDA's warnings of 2017 and 2019 described supplements of up to 20 milligrams, then pills of up to 100 milligrams. They also described doses of up to 300 milligrams a day, recommended by physicians for some conditions, and put the resulting blood levels at up to 1,200 ng/mL. In its 2020 guidance the FDA asked makers to test biotin up to 3,500 ng/mL, about three times the highest concentration expected in patients. Roche's troponin T assay had been cleared in 2017 after testing biotin only up to 20 ng/mL. Its instructions said not to draw blood from patients taking more than 5 milligrams a day until at least 8 hours after the last dose.

Solved example: a vitamin dose against a reagent's tolerance

A 2021 study measured plasma biotin peaks of about 1,130 to 1,160 ng/mL one to one and a half hours after a 300-milligram dose; its authors included the maker's staff. Against the original tolerance of 20 ng/mL, 1,160 ng/mL is 58 times too much. In molecules the contrast is starker. Biotin weighs 244 grams per mole, so 1,160 ng/mL is about 4.7 micromolar. Troponin T at 14 ng/L, the US 99th-percentile limit for women, is about 0.39 picomolar, so the blood holds about 12 million biotin molecules for every troponin molecule. Each of those biotin molecules can take a place on the streptavidin.

Roche changed the reagent rather than the instructions. It added an antibody that captures free biotin and lengthened the linker that holds biotin on the capture antibody. The FDA cleared the reformulated assay on 21 September 2021, and the decision summary records no interference up to 1,500 ng/mL. Abbott states that its Alinity assays are free from biotin interference, a maker's claim. In June 2022 the FDA's list of troponin assays whose biotin risk had not been addressed still named the 2017 version of Roche's troponin T assay, because the list goes by clearance number.

42 — A vitamin dose against a reagent's tolerance
0.1 1 10 100 1,000 10,000 BIOTIN IN BLOOD, ng/mL, LOG SCALE usual level about 0.77 ng/mL (774 pg/mL) 10 mg a day for a week: above 3.6 ng/mL (JAMA 2017) peak after 300 mg: 1,130–1,160 (2021 study, maker's staff among authors) up to 1,200 (FDA 2017) 20: Roche's assay, 2017 1,500: reformulated assay, FDA 2021 3,500: FDA 2020 test level 58 times the original tolerance REAGENT, 2021 longer linker scavenger antibody
AntibodyInterferentReadoutFocal detail
Plate 42 — High-dose biotin can raise blood levels above 1,100 ng/mL, against the 20 ng/mL Roche's troponin T assay tolerated in 2017. Roche added a biotin-capturing antibody and a longer linker, and the reformulated assay was cleared in 2021 with no interference up to 1,500 ng/mL.

Too much target, or target in disguise

The high-dose hook of Chapters 4 and 13 turns into a clinical error when an enormous concentration reads as a modest one. The 2004 review cites a patient whose prolactin, a pituitary hormone, read 1,500 milliunits per liter undiluted and 950,000 on dilution, about 630 times higher. A 2017 case report describes a woman with an advanced molar pregnancy, an abnormal growth of placental tissue, who had two negative urine pregnancy tests. Measured after proper dilution, her serum hCG was above 900,000 milli-units per milliliter. The defense is dilution: a laboratory that suspects an extreme value reruns the sample diluted, and an analyzer can do so automatically.

Some targets circulate bound to the patient's own antibodies. Macroprolactin, prolactin bound mostly to immunoglobulin G, is generally considered biologically inactive, yet assays detect it. The 2004 review reports it in 24 percent of sera with raised prolactin, and differences of 2.3- to 7.8-fold between assays. Troponin can do the same. In a 2022 case report, a patient's troponin I read 872 ng/L on one analyzer, against that assay's upper reference limit of 54 ng/L for women, and peaked at 3,165. Roche's high-sensitivity troponin T assay gave 12 ng/L, against its limit of 14, and a second troponin I assay 11 ng/L, against its own limit of 9. Polyethylene glycol is a polymer that makes antibodies and other large proteins fall out of solution. Precipitating the antibodies with it left only 31 percent of the troponin I, below the 40 percent threshold the authors cited for suspecting a complex of troponin and antibody. The patient had already been treated in error for inflammation of the heart, and the troponin I was still 130 ng/L a year later.

Small molecules bring a different problem. The antibody of a competitive assay for a drug or hormone can bind related molecules of similar shape, so the assay measures them too. The 2004 review cites immunoassays for the transplant drug cyclosporin that read up to 174 percent higher than a chromatography method, because they also counted the drug's breakdown products, which chromatography separates from the drug before measuring it.

A plausible number is the dangerous one

An interference that produces an impossible value is caught at once. One that moves a result into the normal range looks like any other normal result: a falsely low troponin in a patient taking biotin, or a modest prolactin from an enormous tumor. Interference checks therefore have to be triggered by disagreement between the number and the patient's clinical picture, not only by the number itself.

What the analyzer and the laboratory can do

The reagent carries most of the defense: blockers, fragments, partly human antibodies, an antibody that captures free biotin, a two-step format that washes away excess target, and a hook-free limit far above the measuring range. The analyzer adds rules: it flags results above the measuring range and reruns them diluted.

The laboratory runs the last checks. Comparing results that should agree can expose a complex. In a 2021 case series, four patients with troponin T bound to antibodies were identified when the 9-minute and 18-minute versions of the same assay disagreed, with a ratio between the two results below 0.80. The 2004 review lists the other checks: diluting the sample to see whether the result falls in proportion, adding a blocking reagent, precipitating the antibodies, or measuring with a method that uses antibodies from another species. For biotin, the matching check is a method that uses no biotin. A result that changes by more than the test's own scatter under one of those checks was not a measurement of the target alone.

Ship the follow-up tests with the assay

When specifying a new assay, specify its interference checks too: the biotin, heterophile and rheumatoid factor levels it tolerates, its hook-free limit, the dilution a laboratory should use to test linearity, and a second method with different antibodies for confirmation. Write the conditions that should prompt a check into the instructions, so that a laboratory can act on a result that disagrees with the patient.

+ What this chapter established
  • Patients' own antibodies can bridge or block reagent antibodies; blockers and partly human antibodies make that rare but not impossible.
  • Free biotin occupies the streptavidin that biotin-based assays use, reading falsely low in sandwiches and falsely high in competitive assays.
  • Extreme concentrations can hook, and targets bound to antibodies can read high on one assay and normal on another.
  • Reagents carry most defenses; analyzers add flags and dilution; laboratories confirm with dilution, blocking and a second method.
20 — Troponin

Ruling out in an hour.

+ The questionHow does a number measured at a few nanograms per liter let an emergency department send a patient home within an hour?

An hour in a guideline

At the end of August 2015 the European Society of Cardiology released new guidelines for patients with a suspected heart attack whose electrocardiogram does not show one particular pattern, called ST-segment elevation. An electrocardiogram is the tracing of the heart's electrical activity. The guidelines recommended a decision from two blood samples taken an hour apart, as an alternative to waiting three hours, wherever a high-sensitivity troponin assay with a validated algorithm was available. High-sensitivity describes an assay's analytical performance at low concentrations, not the share of heart attacks it finds; the last section of this chapter gives the definition. The guidelines added that the cut-off levels in that one-hour algorithm are specific to each assay. In the 2020 edition the one-hour algorithm became the best option and two hours the second best, and the 2023 guideline on acute coronary syndromes kept both. Acute coronary syndromes are the conditions in which blood flow to the heart muscle is suddenly reduced, heart attacks among them.

An hour is not long for a protein to accumulate. Troponin leaks from dying heart muscle over hours, and a patient who arrives soon after the pain began may not yet have much of it in the blood. The opening scene showed the problem: in the Basel study, the standard assay separated heart attacks from other causes much less well in patients who had arrived within three hours. The guideline's answer uses two properties of the sensitive assays together: a limit set in healthy people and a change measured between two blood draws.

The 99th percentile

In its fourth version, from 2018, the international definition of a heart attack calls troponin above the 99th-percentile upper reference limit a sign of injury to heart muscle. That limit is the value below which 99 percent of a healthy reference population falls. An acute heart attack requires a rise or fall in troponin, with at least one value above that limit, together with clinical evidence that the heart muscle was starved of blood. The definition also requires the high-sensitivity assays to measure with a coefficient of variation of 10 percent or less at the 99th percentile, and it recommends separate limits for women and men. The coefficient of variation is the standard deviation of repeated measurements divided by their mean, expressed as a percentage.

For Roche's troponin T assay, the FDA's decision summary gives limits of 19 ng/L overall from 1,301 healthy people, 14 ng/L from 656 women and 22 ng/L from 645 men. A percentile at the extreme of a distribution rests on few people: 1 percent of 656 is between six and seven, so the women's limit is set by the handful of highest values among them.

For Abbott's ARCHITECT high-sensitivity troponin I assay, cleared in 2019, the FDA's decision summary gives an overall limit of 28 ng/L from 1,531 people, with a 90 percent confidence interval of 22 to 33. A confidence interval is the range within which the true value plausibly lies, at the stated level of confidence, and this one shows how loosely even a large reference study fixes the limit. The European label for Roche's troponin T assay, the same chemistry as the US version above, gave 14 ng/L overall, from a different reference population. Chapter 14 showed that the next generation of the assay has different limits again.

43 — A limit set by the top 1 percent
below the limit of detection (3–5 ng/L): not measured 0 5 10 15 20 25 30 TROPONIN T, ng/L women (656) men (645) WOMEN 14 OVERALL 19 MEN 22 14 19 24 top 1 % of women, height × 21 the 6 or 7 highest of 656 women set this limit a percentile at the edge rests on a handful schematic distributions; limits from the 2017 FDA decision summary Abbott troponin I, 1,531 people: 28 ng/L, 90 % interval 22–33
AnalyteReadoutFocal detail
Plate 43 — The 99th-percentile limit comes from the highest values in a healthy reference group, so it rests on a handful of people: six or seven of 656 women for Roche's troponin T assay. Separate limits for women and men follow the different distributions.

The FDA's review of Roche's troponin T assay shows why one number is not enough early on. Using the 718 Basel patients of the opening, the FDA found that a single sample on arrival, judged against 19 ng/L, identified 93.5 percent of the heart attacks, and a sample at three hours 98.3 percent. Judged against the 99th percentile alone, a single sample taken on arrival misses one heart attack in fifteen and cannot be used to send patients home.

From one number to a change

The one-hour algorithm adds the change between two draws. With Roche's troponin T assay, a heart attack is ruled out under the 2020 guideline's table when the first result is below 12 ng/L and the second, an hour later, differs from it by less than 3 ng/L. A single very low first result, below 5 ng/L, can also rule a heart attack out under conditions the guideline sets. A heart attack is ruled in when the first result is 52 ng/L or more, or when the change in one hour is 5 ng/L or more. Everyone else is observed and tested again. The same table gives a separate row of numbers for each assay. For Abbott's troponin I assay, for instance, a heart attack is ruled out below 5 ng/L with a change under 2, or below 4 ng/L on the first result alone, and ruled in at 64 or more, or at a change of 6 or more.

The Basel researchers of the opening led a prospective study, published in 2015, at six centers in Switzerland, Spain and Italy. Such a study enrolls patients first and then follows them, rather than looking back at records. It tested the algorithm with Roche's troponin T assay on 1,320 patients, 229 of whom had a heart attack. The study was funded in part by several assay makers, and Roche donated the assay. The algorithm ruled out 786 patients, 59.5 percent, with a clinical sensitivity of 99.6 percent and a negative predictive value of 99.9 percent. It ruled in 216, of whom 78.2 percent had a heart attack, and left 318 for observation.

Solved example: what a change of 3 ng/L asks of the assay

According to its decision summary, Roche's troponin T assay has a coefficient of variation of 10 percent at about 11 ng/L, which is a standard deviation of about 1.1 ng/L. Independent scatters add as squares. Each measurement contributes 1.1 squared, about 1.2, so the difference between two measurements has a squared scatter of about 2.4, and its standard deviation is the square root of that, about 1.6 ng/L, or the square root of two times 1.1. A change of 3 ng/L is therefore about two standard deviations of the difference. In a patient whose troponin did not change at all, measurement noise alone would produce an apparent change of 3 ng/L or more, up or down, in roughly one pair of samples in twenty. The algorithm's smallest step sits only just above the assay's own scatter.

44 — Ruling out in one hour
0 h and 1 h samples, high-sensitivity troponin T (ESC 2020) OUTCOME IN THE 2015 STUDY RULE OUT 0 h below 12 ng/L and 1 h change below 3 (or 0 h very low, below 5, under the guideline's conditions) 786 (59.5 %) sensitivity 99.6 %, NPV 99.9 % 30-day deaths 0.0 % OBSERVE everyone else 318 (24.1 %) 18.6 % had a heart attack 30-day deaths 1.6 % RULE IN 0 h 52 ng/L or more, or 1 h change 5 or more 216 (16.4 %) 78.2 % had a heart attack 30-day deaths 1.9 % the middle zone carries real risk 1,320 PATIENTS rule out 786 observe 318 rule in 216 a change of 3 ng/L is about two standard deviations of the difference between two results near 11 ng/L outcomes from a 2015 study: 1,320 patients, 229 heart attacks; Basel group, makers among funders
ReadoutFocal detail
Plate 44 — The one-hour algorithm combines a low first value with a small change to rule out, and a high value or a large change to rule in. In a 2015 study it ruled out 59.5 percent of patients with a negative predictive value of 99.9 percent and left a quarter for observation.

A negative predictive value of 99.9 percent depends on how many of the patients tested have the disease, here 17 percent. The PCR guide in this library works through how predictive values depend on prevalence, and the same arithmetic applies here unchanged. The same algorithm, used in a population with more heart attacks, would leave more of them among the patients ruled out.

The middle zone carries real risk

The observation zone is not a near-negative. In the 2015 study, 18.6 percent of the 318 patients left for observation had had a heart attack. The observation group's 30-day mortality was 1.6 percent, close to the 1.9 percent of those ruled in, against none among those ruled out. An algorithm's middle zone is a decision to keep testing, and a test that sorts many patients into it has moved work, not removed it.

What "high-sensitivity" means

The word has a definition. In a 2017 paper, a task force of the International Federation of Clinical Chemistry restated two criteria it had first proposed in 2012. The criteria are a coefficient of variation of 10 percent or less at the 99th percentile, and results above the limit of detection in at least half of healthy people, reported in whole nanograms per liter. The authors of the task force's paper disclosed research funding or consulting for several troponin makers. In the healthy population the task force used, Roche's high-sensitivity troponin T assay measured troponin in 7 percent of women and 43 percent of men at its limit of detection of 5 ng/L. The assay met the precision criterion but not the second one, and the task force noted that its rates were lower than recommended.

As the opening noted, a trade report states that when the FDA cleared the assay in January 2017, it asked the maker to call it "next generation" rather than high-sensitivity. Assays that measure most healthy people do more than lower a limit. A value that can be measured in a healthy person can be compared with the same person's value an hour later, and the change the algorithm uses needs that comparison at the bottom of the range.

45 — Measurable in healthy people
IFCC criterion: at least half Roche troponin T (limit of detection 5 ng/L) men 43 % women 7 % Abbott ARCHITECT troponin I (1.9 ng/L) men 80 % women 67 % Beckman Access troponin I men 87.5 % women 74 % precise at the limit, but rarely measurable in healthy women 0 % 25 % 50 % 75 % 100 % HEALTHY PEOPLE ABOVE THE LIMIT OF DETECTION IFCC task force, 2017, common healthy population; second criterion: coefficient of variation ≤ 10 % at the 99th percentile
ReadoutFocal detail
Plate 45 — The IFCC defines a high-sensitivity assay as precise at the 99th percentile and able to measure troponin in at least half of healthy people. In the task force's population Roche's troponin T assay met the first criterion but measured only 7 percent of women and 43 percent of men.
Validate the algorithm on the analyzer that will run it

The cut-offs in a one-hour algorithm apply to one assay, and the smallest change depends on that assay's precision near 5 to 12 ng/L. A laboratory adopting the algorithm should confirm its own analyzers' scatter at those concentrations, report in whole nanograms per liter, and agree with the emergency department what happens to patients in the observation zone. A maker introducing an assay needs its own validated algorithm, not another assay's numbers.

+ What this chapter established
  • The 2015 European guideline let a heart attack be ruled out from two samples an hour apart, with cut-offs specific to each assay.
  • The 99th-percentile limit comes from healthy people, separately for women and men, and its extreme position makes it uncertain.
  • The algorithm combines a low first value with a small change; the smallest change sits near the assay's own scatter.
  • High-sensitivity means precise at the 99th percentile and measurable in at least half of healthy people, which Roche's troponin T assay only partly met.
21 — A ratio in blood

pTau217 and amyloid.

+ The questionWhy does the first cleared Alzheimer's blood test report a ratio of two proteins, with a middle zone that gives no answer?

Two numbers and a remainder

Of the positive results, 91.7 percent were confirmed by a brain scan or a spinal fluid test, and of the negative results, 97.3 percent. The FDA gave those figures on 16 May 2025, when it cleared what it called the first in vitro diagnostic device that tests blood to aid in diagnosing Alzheimer's disease. Fewer than 20 percent of the 499 patients in the study received neither a positive nor a negative result. The device was Fujirebio's Lumipulse G pTau217/β-Amyloid 1-42 Plasma Ratio. The FDA cleared it through a 510(k) about eight months after the application arrived. Its predicate was the company's spinal fluid test of 2022, the earlier device to which the FDA found it substantially equivalent (Chapter 26).

The test does not diagnose the disease. It looks for amyloid pathology: deposits of a protein fragment called beta-amyloid in the brain. In clinical practice, amyloid pathology is assessed by positron emission tomography, or PET, a brain scan that shows the deposits, or by a test of the fluid around the spinal cord. The FDA cleared the test only for adults aged 50 and older who are seen in a specialized care setting with signs and symptoms of cognitive decline. The cleared indications state that it is not intended as a screening or stand-alone diagnostic test. Its results must be interpreted with other clinical information about the patient.

Two assays, one number

The decision summary describes the test as two immunoassays run on one analyzer, Fujirebio's LUMIPULSE G1200, using 40 microliters of plasma. Each is a two-step sandwich with an alkaline phosphatase label and a dioxetane substrate that glows at 477 nanometers, the chemistry of Chapter 7. One assay measures pTau217: tau, a protein of nerve cells, carrying a phosphate group at amino acid 217. The pTau217 assay measures from 0.047 to 10 picograms per milliliter, with a limit of detection of 0.041 pg/mL. The other assay measures beta-amyloid 1-42, the 42-amino-acid form of the fragment, from 0.8 to 500 pg/mL. A picogram per milliliter equals a nanogram per liter, so the pTau217 range starts far below even the troponin limits of detection in Chapter 1.

The analyzer reports the two concentrations, and the ratio is then calculated by hand: pTau217 divided by beta-amyloid 1-42, rounded to five decimal places. The FDA's decision summary states that the ratio "must be manually calculated because the instrument does not report the final result". The maker adds that both values must come from the same sample on the same analyzer, and that results from other makers' assays must not be mixed in.

Neither plasma assay has an international reference of its own. The summary states that pTau217 is traceable only to an in-house standard, one the maker prepared itself. It also states that no international reference material or method exists for plasma beta-amyloid 1-42, and that the calibrators for that assay are tied to the maker's spinal fluid assay. Troponin I lacks a common reference in the same way, as Chapter 14 showed.

The decision summary does not say why the test divides one protein by the other. The predicate is itself a ratio of two amyloid fragments in spinal fluid, and the FDA's review of it did not explain that choice either. Arithmetic still shows what a ratio gains and what it costs. Any factor that scales both concentrations in one tube by the same proportion cancels when one is divided by the other. But each measurement brings its own scatter, and the ratio carries both.

The summary reports that scatter as coefficients of variation, each a standard deviation expressed as a percentage of the mean. Measured within one laboratory over many runs and days, the total coefficients of variation were 2.2 to 10.3 percent for pTau217, 2.5 to 3.8 percent for beta-amyloid 1-42 and 3.1 to 10.2 percent for the ratio, so at low concentrations the scarcer protein, pTau217, dominates the ratio's scatter. Between reagent lots, the ratio's coefficient of variation was 7.3 to 16.3 percent. The highest figure came from a low panel whose ratio lay close to the lower cut-off described below, so a result near that cut-off can cross it when the reagent lot changes, and the decision summary asks for caution with such results.

46 — Two assays, one number
40 µL plasma LUMIPULSE G1200 pTau217 assay two-step sandwich, alkaline phosphatase + dioxetane measures 0.047–10 pg/mL, limit of detection 0.041 pTau217, pg/mL beta-amyloid 1-42 assay 0.8–500 pg/mL Aβ1-42, pg/mL ratio = pTau217 ÷ Aβ1-42 calculated by hand, 5 decimals positive ≥ 0.00738 indeterminate negative ≤ 0.00370 the analyzer does not report the ratio within-lab CV: pTau217 2.2–10.3 %, Aβ1-42 2.5–3.8 %, ratio 3.1–10.2 % pTau217 traceable to an in-house standard; no international reference for plasma Aβ1-42
AnalyteAntibodyLabel and signalReadoutFocal detail
Plate 46 — The first cleared Alzheimer's blood test is two immunoassays on one analyzer and one tube, combined by hand into a ratio. The ratio carries the scatter of both measurements, and neither assay is tied to an international reference.

Two cut-offs and a middle zone

The test sorts each ratio into three results. A ratio of 0.00738 or more is positive, one of 0.00370 or less is negative, and anything between is indeterminate. A positive result therefore needs at least about 7.4 picograms of pTau217 for every 1,000 picograms of beta-amyloid 1-42, and a negative one no more than about 3.7. The cut-offs were set on 208 stored plasma samples from three research cohorts. The decision summary states that an indeterminate result "is consistent with patients who are uncertain to have amyloid pathology", and that those patients "should be considered for further testing", such as a PET scan or a spinal fluid test.

The clinical study tested 499 people aged 52 to 93 against a reference, the established test whose answer is taken as correct. For 152 of them the reference was a PET scan read by eye, and for the other 347 it was the spinal fluid test. Just over half, 51.1 percent, had amyloid pathology. The decision summary gives the results in a table.

Solved example: what the middle zone buys

Of 219 positive results, 201 were amyloid-positive by the reference, 91.8 percent by this arithmetic; the FDA's announcement gave 91.7 percent. Of 182 negative results, 177 were amyloid-negative, 97.3 percent. The 98 indeterminate results split exactly in half, 49 and 49, so for those patients the chance of amyloid pathology after the test, one in two, was close to the 51.1 percent of the whole study, and the blood test added almost no information. Leaving them aside, the test found 201 of 206 amyloid-positive patients, a clinical sensitivity of 97.6 percent, and 177 of 195 amyloid-negative ones, a clinical specificity of 90.8 percent. Counting each indeterminate result as a failure in either direction, sensitivity falls to 201 of 255, 78.8 percent, and specificity to 177 of 244, 72.5 percent. Counted simply as negative, the indeterminates would leave specificity at 226 of 244, 92.6 percent. The middle zone holds the results among which the test's errors would otherwise have fallen.

47 — What the middle zone holds
0 0.00370 0.00738 pTau217 ÷ Aβ1-42 NEGATIVE 177 5 negative results right 97.3 % INDETERMINATE 49 49 indeterminate: 19.6 % of results POSITIVE 18 201 positive results right 91.8 % 49 and 49: no information violet: amyloid-positive by PET or spinal fluid test gray: amyloid-negative sensitivity 97.6 % with indeterminates set aside; 78.8 % counting them as not positive FDA decision summary K242706, 499 patients
AnalyteReadoutFocal detail
Plate 47 — Two cut-offs leave a fifth of patients indeterminate, and those split exactly in half by the reference test. Setting them aside makes the remaining answers right more than nine times in ten; counting them as negative lowers the sensitivity from 97.6 to 78.8 percent.

Two cut-offs answer fewer patients, and answer them more reliably. A single cut-off would have given every patient a yes or no. If the ratio separates the patients inside the middle zone no better than the zone does as a whole, roughly half of them would have been sorted wrongly. With two cut-offs, a fifth of patients go on to the scan or spinal fluid test they would have needed without the blood test, and the other four fifths get an answer that matches the reference most of the time. The algorithm of Chapter 20 makes the same trade, with an observation zone in place of an indeterminate one. The predictive values also depend on how common amyloid pathology is among those tested, which in this specialist population was about half.

A test's accuracy depends on how its middle zone is counted

The same 499 results give a clinical sensitivity of 97.6 percent or 78.8 percent, depending on whether indeterminate results are set aside or counted as negative. In its 2025 clinical practice guideline, the Alzheimer's Association asks blood tests used for triage to reach at least 90 percent sensitivity and 75 percent specificity, and tests used to confirm amyloid pathology to reach at least 90 percent of both. A triage test decides who goes on to a confirmatory test. Whether a two-cut-off test meets those numbers depends on how its indeterminate results are counted. With them set aside, the Fujirebio ratio's 97.6 and 90.8 percent meet the thresholds for confirmation; with each counted as a failure, its 78.8 and 72.5 percent meet neither set. A performance claim should state the share of indeterminate results and how they were counted.

What the clearance established, and what came after

The clearance established that the ratio, run as specified on the named analyzer, agrees with amyloid PET or the spinal fluid test closely enough for the FDA to find the test substantially equivalent to its spinal fluid predicate. That finding covers symptomatic adults in specialist care. The clearance did not establish use in people without symptoms, for screening, or in primary care, and it did not make a positive result a diagnosis. The handling of samples matters too. The summary states that specimens "degrade rapidly at room temperature", so a result depends on how the tube was kept before it reached the analyzer, as well as on the analyzer's limit of detection.

Three more blood tests for amyloid pathology were cleared within fifteen months. The FDA cleared Roche's Elecsys pTau181 assay on 8 October 2025. It uses a single cut-off to rule out amyloid pathology in primary care. In 312 people, of whom 13.1 percent were amyloid-positive, a negative result was right 97.9 percent of the time and a positive one only 22.4 percent of the time.

On 19 August 2026 the FDA cleared two more. C2N's PrecivityAD2 measures by mass spectrometry (Chapter 18). Roche's Elecsys pTau217 assay uses a single protein and two cut-offs. According to the decision summary for that assay, 19.9 percent of results in Roche's study of 958 people fell between its two cut-offs. The share is about the same as for the Fujirebio ratio. In the EU, Fujirebio announced a CE mark under the IVDR for its single-protein pTau217 plasma assay on 11 May 2026. The IVDR is the EU's regulation for in vitro diagnostics, explained in Chapter 26. No approval of such a test by India's regulator was found as of October 2026.

48 — Blood tests for amyloid, 2022 to 2026
2022 2023 2024 2025 2026 4 May 2022 spinal fluid ratio, De Novo (Fujirebio) the predicate for the first blood test 19 Aug 2026 Elecsys pTau217 (Roche) single protein, two cut-offs, 55+ 19 Aug 2026 PrecivityAD2 (C2N): mass spectrometry 8 Oct 2025 Elecsys pTau181 (Roche) one cut-off, rule-out in primary care, 55+ 16 May 2025 Lumipulse pTau217/Aβ1-42 plasma ratio (Fujirebio) two cut-offs, specialist care, 50+ first blood test cleared early 2026 Quanterix submits a 510(k), multi-marker test not cleared by Oct 2026 11 May 2026 CE mark under the IVDR, EU Lumipulse pTau217 plasma (single protein) OCTOBER 2026 above the line: cleared in the US; below: EU mark and a pending US submission
ReadoutFocal detail
Plate 48 — Within fifteen months of the first clearance in May 2025, the FDA cleared three more blood tests for amyloid pathology, three of them immunoassays on conventional analyzers. Designs differ: one cut-off or two, ratio or single protein, primary or specialist care.
Plan for the middle zone before launch

For a test with two cut-offs, state three things in the specification and the instructions: the share of results expected between the cut-offs in the intended population, the confirmatory test those patients go to, and the performance both with indeterminate results set aside and with them counted. Then confirm the share in the clinical study, since a larger middle zone than planned moves cost and waiting time onto the confirmatory test.

+ What this chapter established
  • The first cleared blood test for amyloid pathology divides a pTau217 result by a beta-amyloid 1-42 result, both from one analyzer and one tube.
  • The FDA's review does not explain why the test uses a ratio, which by arithmetic cancels factors shared by both measurements but carries the scatter of both assays.
  • Two cut-offs leave about a fifth of patients indeterminate, where the test carries no information, and make the other answers more reliable.
  • The clearance covers symptomatic adults in specialist care, later tests use other designs, and the test's clinical sensitivity depends on how the middle zone is counted.

+ Part VII · The field in 2026

Who builds the analyzers.

A cobas pro integrated solutions line joins clinical chemistry modules, which measure substances such as glucose, with immunoassay modules in one machine. In figures Roche supplied to the trade magazine CAP TODAY in July 2026, the largest configuration is rated at up to 6,400 tests an hour. Roche first sold one in the US in 2019 and reported more than 1,000 installed there and more than 5,000 elsewhere. Those lines belong to a business that Roche calls Core Lab and defines as immunoassays, clinical chemistry and its CustomBiotech unit. Core Lab sold 7.6 billion Swiss francs' worth in 2025, 55 percent of the company's diagnostics sales. This part is a single chapter, dated October 2026. It sorts the makers of such machines, and of the other analyzers and strips in use, by the way each reads its label.

22 — Who builds them

Analyzers and strips, by technique and maker.

+ The questionWho makes the analyzers and strips in use in 2026, and which way of reading a label does each rely on?

Four ways of reading, many makers

Two hundred tests an hour is a common rating for one immunoassay module in 2026, and the makers reach it with the different labels of Part II. The figures below come from the makers' own product pages unless a trade survey is named. Each is a best case for one mix of tests, as Chapter 10 warned.

Among the central-laboratory makers in this chapter, electrochemiluminescence belongs to Roche: light switched on at an electrode, the method of Chapter 9. Roche's cobas e 801 module is rated at up to 300 immunoassay tests an hour and its smaller e 402 at up to 120, both with reaction times of 9, 18 or 27 minutes.

Abbott's Alinity i, rated at up to 200 tests an hour per module, reads a flash from an acridinium label, the first family of Chapter 7. The FDA's review of one Alinity assay describes it as a chemiluminescent microparticle immunoassay with acridinium-labeled antibodies. Abbott states that its Alinity assays are free from biotin interference. Siemens Healthineers uses acridinium esters on its Atellica IM 1300 and 1600, rated at 220 and 440 tests an hour. It also uses them on the ADVIA Centaur XPT, rated at 240 tests an hour with a first result in 18 minutes.

Glow from an enzyme, the second family of Chapter 7, is the widest group. Beckman Coulter's Access analyzers use alkaline phosphatase with a light-producing substrate. For its DxI 9000, announced in 2023, the company moved to a faster substrate, which its 2019 poster described as producing its light within a minute instead of about six. Fujirebio rates its LUMIPULSE G1200 at 120 tests an hour. According to the FDA's decision summary for Fujirebio's Alzheimer's ratio test, the two assays of that test use alkaline phosphatase and a dioxetane substrate. Mindray states the same principle for its CL-2000i, which is rated at 240 tests an hour. QuidelOrtho's VITROS systems measure a peroxidase label with luminol, a method the maker calls enhanced chemiluminescence.

Snibe's MAGLUMI X8 is rated at up to 600 tests an hour per module and 2,400 with four modules. Snibe's product page describes chemiluminescence without naming the label.

Fluorescence, the method of Chapter 8, is still used in smaller systems. bioMérieux's VIDAS KUBE runs single-dose tests read by fluorescence at up to 36 an hour per module, and up to six modules can be stacked. Thermo Fisher Scientific's KRYPTOR analyzers use the energy-transfer method TRACE, which needs no washing. Quanterix sells single-molecule counting, the subject of Chapter 17, on analyzers labeled for research use only. In July 2025 the company completed its purchase of Akoya Biosciences.

49 — Who reads which label
ELECTROCHEMILUMINESCENCE ACRIDINIUM FLASH ENZYME GLOW CHEMILUMINESCENCE, LABEL NOT STATED FLUORESCENCE SINGLE-MOLECULE COUNTING Roche cobas e 801 300 Roche cobas e 402 120 Abbott Alinity i 200 per module, 800 per system, 1,550 with chemistry Siemens Healthineers Atellica IM 1600 440 Atellica IM 1300 220 ADVIA Centaur XPT 240 Beckman Coulter DxI 9000 450 maker gives 215 per m² Beckman Access 2 100 Fujirebio LUMIPULSE G1200 120 Mindray CL-2000i 240 QuidelOrtho VITROS 3600 189 Snibe MAGLUMI X8 600 per module bioMérieux VIDAS KUBE 36 per module Thermo Fisher KRYPTOR no hourly figure given Quanterix HD-X research use only makers' own figures unless marked; each is a best case for one test mix 0 100 200 300 400 500 600 TESTS PER HOUR, MAKER'S RATING vendor data, CAP TODAY Jul 2026
ReadoutFocal detail
Plate 49 — Central-laboratory makers split by the way they read a label: electrochemiluminescence at Roche, acridinium flash at Abbott and Siemens Healthineers, enzyme glow at Beckman Coulter, Fujirebio, Mindray and QuidelOrtho. Throughput ratings count different things and are best cases.
Solved example: what a throughput figure counts

Beckman Coulter gives the DxI 9000 a throughput of 215 tests an hour per square meter of floor. In a 2026 trade survey the vendor gave it up to 450 tests an hour, which at 215 per square meter implies about 2.1 square meters. Roche's cobas e 402 is rated at 120 tests an hour on 1.2 square meters, which works out at 100 per square meter. Abbott gives 200 tests an hour for one Alinity i module and 1,550 for an integrated system that includes chemistry, and the trade survey gives up to 800 for a four-module system. Three honest numbers for the same platform count three different things, so a comparison has to fix the unit first: per module, per system, per square meter, immunoassay alone or with chemistry.

At the bedside

Chapter 16 described the high-sensitivity troponin tests that reached the bedside: Abbott's handheld i-STAT, cleared in January 2025, the PATHFAST system of PHC Corporation of Tokyo, cleared in March 2024, and Siemens Healthineers' Atellica VTLi, CE-marked in 2021. Other point-of-care analyzers carry broader menus of tests. Radiometer's AQT90 FLEX runs up to 30 tests an hour from whole blood or plasma, with results in 11 to 21 minutes. Its menu includes troponin I and T, natriuretic peptide, D-dimer, procalcitonin, C-reactive protein and hCG, among others. In December 2023 Roche agreed to buy the parts of LumiraDx tied to that company's point-of-care platform, which runs immunoassays and clinical chemistry on one instrument. The agreed price was 295 million US dollars, and the sale was to be made out of administration, a UK insolvency procedure.

India

Judged from their own pages, India's makers sell mostly rapid tests, ELISA kits and chemiluminescence technology used under license. In June 2020 India's medical research council recommended a rapid antigen test for point-of-care use, SD Biosensor's Standard Q, which was made at a unit in Manesar, Gurugram. Agappe Diagnostics of Kochi signed a supply and license agreement with Fujirebio in March 2023. Under it, Fujirebio supplies the Mispa i60 and i121 analyzers and reagent raw materials, and Agappe makes the cartridge reagents in India and sells them under its own name. Agappe's 2024 brochure also lists acridinium-ester reagents and two chemiluminescence analyzers rated at 200 and 180 tests an hour, without saying where the analyzers are made.

J. Mitra & Co. of New Delhi sells rapid tests, ELISA kits, two bench-top chemiluminescence analyzers and an analyzer developed with a technology center at the Indian Institute of Technology Madras. Again, no place of manufacture is stated for the two chemiluminescence analyzers. Transasia Bio-Medicals sells ELISA kits and an automated ELISA processor that reads absorbance. Mindray sells its chemiluminescence analyzers in India from an office in Gurugram. No maker's page opened for this guide described a central-laboratory immunoassay analyzer designed in India.

Every "first" in a press release carries a qualifier

Snibe calls the MAGLUMI 2000 the first chemiluminescence analyzer from China to be cleared by the FDA. Siemens Healthineers called the VTLi the first high-sensitivity troponin I test from a fingerstick. Fujirebio and Agappe called Agappe the first local Indian company with a full chemiluminescence solution, and Roche called its pTau217 assay the first and only FDA-cleared single-biomarker blood test for both ruling in and ruling out amyloid pathology across primary and specialty care. Each claim is the maker's own, none was checked against a regulator's records for this guide, and each can hold only within its qualifier.

A field in motion

The field changes from one quarter to the next. Reporting their own counts to the same 2026 trade survey, the makers put about 3,625 of Abbott's Alinity ci-series systems in the US and about 16,600 elsewhere, and more than 950 of Diasorin's LIAISON XL analyzers in the US and more than 6,500 elsewhere. These counts are self-reported and unaudited. Siemens Healthineers said in 2026 that it was preparing to separate its diagnostics business, after a quarter in which that business's reported revenue fell 12 percent, according to a trade report. At constant currency, with the effect of exchange rates removed, the same report put the fall at 7 percent. Fujirebio has developed reagents for Sysmex's HISCL analyzers since 2020, and the two companies announced a basic agreement on wider collaboration around the start of 2024.

The products change too. By the company's account, Roche's sixth-generation troponin T assay was CE-marked by September 2025, with new reference limits (Chapter 14). The FDA cleared QuidelOrtho's VITROS high-sensitivity troponin I assay on 29 October 2025, with a claimed limit of detection of 0.59 ng/L. Four blood tests for amyloid pathology were cleared between May 2025 and August 2026, three of them immunoassays on conventional analyzers (Chapter 21).

50 — A field in motion
2023 2024 2025 2026 Mar 2023 Agappe–Fujirebio supply and license agreement (India) around start of 2024 Fujirebio–Sysmex basic agreement Jul 2025 Quanterix completes Akoya purchase 29 Oct 2025 VITROS hs troponin I cleared, LoD 0.59 ng/L claimed 19 Aug 2026 two more amyloid blood tests cleared Dec 2023 Roche agrees to buy LumiraDx point-of-care platform May 2025 first US blood test for amyloid (Ch 21) by Sep 2025 Roche troponin T Gen 6 CE mark 2026 Siemens Healthineers prepares diagnostics carve-out (trade report) THIS CHAPTER: OCTOBER 2026 every statement here is dated hollow: approximate date
ReadoutFocal detail
Plate 50 — Between 2023 and 2026 makers bought, merged and licensed platforms, and regulators cleared new troponin and Alzheimer's tests. The chapter's picture of who builds what is dated to October 2026.
Choose a platform by its menu and approvals before its speed

For a laboratory, the deciding facts are which cleared or CE-marked assays the platform runs in its own jurisdiction, how its assays are protected against biotin and heterophile antibodies, its throughput measured on the laboratory's own test mix, and the supply of reagents and service where the laboratory is. For a maker entering a market, the same list defines the competition: a faster analyzer without the assays a laboratory orders every day does not displace an installed one.

+ What this chapter established
  • Central-laboratory makers split by label: electrochemiluminescence at Roche, acridinium flash at Abbott and Siemens Healthineers, enzyme glow at Beckman Coulter, Fujirebio, Mindray and QuidelOrtho.
  • Throughput figures count different things, per module, per system or per square meter, and need a common unit before comparison.
  • India's makers sell mostly rapid tests, ELISA kits and chemiluminescence technology used under license, and no central-laboratory analyzer designed in India was found.
  • Companies and clearances changed through 2025 and 2026, so every statement in this chapter is dated to October 2026.

+ Part VIII · Quality and regulation, as of October 2026

What a test system must show.

About 8 percent of the in vitro diagnostics on the EU market needed a notified body's assessment under the 1998 directive. Under the regulation that replaced it, around 80 percent do. The European Commission gave those figures in October 2021, when only six notified bodies had been designated to do the work; by February 2026 there were 19. In vitro diagnostics, or IVDs, are the reagents, instruments, software and other devices used to examine samples taken from the human body. A notified body is an independent organization that an EU country designates to check a maker's evidence before a device is sold. The four chapters of this part set out what a reagent and its analyzer must show before they are sold, as of October 2026. Each chapter explains why a requirement exists and what evidence it calls for, and gives the clauses with their status as of October 2026, because the clauses will change sooner than the reasons.

23 — Analytical performance

Proving the floor and the scatter.

+ The questionHow many measurements does it take to show a test's floor and precision, and why that many?

A definition that is also an experiment

The limit of blank is the highest result expected, 95 times in 100, from a sample that contains none of the target. Chapter 5 worked the arithmetic once. This chapter asks what that definition commits a maker to measure, because each of its terms leaves a choice open: which samples count as blank, on how many analyzers and reagent lots, over how many days, and how many results are enough to locate the 95th percentile of a spread of results.

The Clinical and Laboratory Standards Institute, a US standards body for laboratory medicine, answers in a series of documents whose names begin with EP. In its second edition of 2012, EP17 covers the limits of blank, detection and quantitation. It requires that they fall in that order: the limit of detection above the limit of blank, and the limit of quantitation at or above the limit of detection.

Roche's high-sensitivity troponin T assay followed EP17. According to the FDA's decision summary, the assay's limit of blank rests on 60 results from blank samples for each reagent lot on each type of analyzer, and its limit of detection on another 60 per lot from low-level samples. Its limit of quantitation rests on 84 results per sample per lot. That limit is the lowest concentration measured with a coefficient of variation of no more than 20 percent. A laboratory that only wants to check a maker's claim needs far fewer results. The implementation guide to EP17 specifies at least 20 blank results in all, with at least two blank samples, each measured at least twice a day, on at least three days. Two samples measured twice on each of five days give 20 results; five samples measured twice on each of three days give 30.

Why precision takes eighty results

A measurement's scatter has more than one source. Results within one run differ because pipetting, mixing and counting are never identical, and this part of the scatter is called repeatability. Runs on the same day differ a little more, because each run starts afresh with its own controls and handling. Days differ again, with new reagent packs, operators or calibrations. The within-laboratory precision is the total scatter a laboratory sees over weeks, and it combines all three. A standard deviation from one run alone understates it.

51 — Where the scatter comes from
ONE SAMPLE 20 DAYS 2 RUNS 2 REPLICATES day 1 day 2 day 3 day 20 20 × 2 × 2 = 80 results (CLSI EP05-A3) Roche's assay: 21 × 2 × 2 = 84 per lot WITHIN-LABORATORY VARIANCE = repeatability + between-run + between-day SD = √(sum of the three variances) schematic proportions 0 % 10 % 20 % 30 % 40 % 5 10 20 50 100 RELATIVE UNCERTAINTY OF THE SD NUMBER OF RESULTS (LOG) 5: 35 % 25: 14 % 80: 8 % twenty days: about 8 %, and the day-to-day part is visible approximate, treating results as independent
ReadoutFocal detail
Plate 51 — Precision has three layers, within run, between runs and between days, and only a design that repeats each layer can separate them. Eighty results over twenty days pin a standard deviation to about 8 percent; five results leave it uncertain by about a third.

To separate the sources, the design has to repeat each one. In the third edition of CLSI's EP05, of 2014, each sample is measured in two replicates, in two runs a day, on 20 days. That gives 80 results per sample, and a statistical method called analysis of variance splits their scatter into its within-run, between-run and between-day parts. Roche's troponin T assay was tested to the previous edition's design: two replicates in each of two runs a day on 21 days, or 84 results per lot. For the Alzheimer's ratio of Chapter 21, five panels of samples were measured in two replicates, in two runs a day, on 20 days. Two further studies followed, both with two runs of three replicates: one on three reagent lots over five days each, and one at three sites over five days each.

Solved example: what twenty days buy

A standard deviation estimated from a handful of results is itself uncertain. If the results are treated as independent, the relative uncertainty is about one divided by the square root of twice the number one less than the count of results. A nested design, in which replicates sit within runs and runs within days, only approximates that independence. From 5 results, the count less one is 4, twice that is 8, and the uncertainty is one divided by the square root of 8, about 35 percent, so a true scatter of 5 percent could easily be reported as 3 or 7. From 20 results it is about 16 percent, and from 80 about 8 percent. With 80 results, a coefficient of variation of 5 percent can at best be placed between roughly 4.6 and 5.4 percent, one standard error either side. When day-to-day scatter is large, the effective count is closer to the number of days, so 20 days leave an uncertainty of about 16 percent, the same as 20 results. Spreading those results over 20 days, rather than running them in one afternoon, is what makes the between-day part visible at all.

A laboratory, again, checks a claim rather than establishing one. Under CLSI's EP15, also of 2014, a laboratory verifies a maker's precision claim and estimates bias, a consistent offset of results from their true value, with at least 25 measurements over at least five days. CLSI's product page for EP15 states that such verification cannot establish precision. The difference matters in practice: the maker's 80 results support a claim, and the laboratory's 25 test whether its own analyzer meets that claim.

Range, interference and carry-over

Linearity is the claim that results stay proportional to concentration across the measuring range. The second edition of EP06, published in 2020, dropped the curve-fitting of its predecessor. It recommends a weighted straight-line fit, with the deviation at each level judged against what would matter clinically. Roche's troponin T assay was cleared with a measuring range of 6 to 10,000 ng/L and no hook effect up to 100,000 ng/L, the hook-free limit that Chapter 13 explained.

Interference testing measures how much a result moves when a possible interferent is added to a real sample. The third edition of EP07, of 2018, describes the method, and a companion document, EP37, lists the concentrations at which common substances should be tested. Chapter 19 showed the cost of choosing too low a level: the 2017 version of Roche's troponin T assay had been tested for biotin only up to 20 ng/mL, and in its 2020 guidance the FDA asks for 3,500 ng/mL.

Carry-over is the part of one sample that reaches the next, the subject of Chapter 11. It is tested by running high and low samples in alternation and is reported as a fraction. In 2021 Beckman Coulter recalled its high-sensitivity troponin I assay, and the recall showed what a small fraction can cost. By the firm's account, the reason was that a sample above 270,000 ng/L could contaminate the reagent pack, and so the results that followed. As an illustration, take 14 ng/L, the 99th-percentile limit for women of Roche's troponin T assay. A carry-over of about 52 parts per million from such a sample would by itself turn a blank into a result at that limit.

Five days confirm a claim that took twenty to make

A laboratory's verification is sized to catch a gross failure of the maker's claim, not to measure precision anew. Twenty-five results over five days leave a standard deviation uncertain by at least roughly 14 percent, and by much more when day-to-day scatter dominates. A laboratory that finds a coefficient of variation a little better than the maker's has therefore learned only that its analyzer is not far worse. Claims at a decision limit, such as precision at a 99th percentile, rest on the maker's full design.

Comparison, reference limits and the documents regulators recognize

A new assay is compared with an existing one on patient samples. The method is that of EP09, in its third edition of 2018, which estimates the bias between the two assays across the range. Reference intervals, the range of results expected in healthy people, follow EP28. EP28's minimum of 120 healthy people per group gives a nonparametric interval between the 2.5th and 97.5th percentiles, read directly from the ranked results without assuming the shape of their distribution. A 99th percentile needs many more people: with 120 it would be the single highest value. That is why the 99th-percentile limits of Roche's troponin T assay drew on 1,301 healthy people.

Regulators adopt these documents instead of writing their own. As of October 2026, the FDA's database of recognized consensus standards listed, among others, the third editions of EP05, EP07, EP15 and EP28, the second editions of EP06, EP17 and EP37, and the third edition of EP09 in part. The FDA recognized the second edition of EP37 in December 2025. The database also listed an older CLSI document on interference by the patient's own antibodies in immunoassays. Once the FDA recognizes a standard, a maker can cite it in a submission instead of justifying its method from first principles.

In the EU, the IVDR sets out the evidence for an IVD in three parts: scientific validity, analytical performance and clinical performance. Scientific validity is the link between the substance measured and a clinical condition. In 2022 the Medical Device Coordination Group, made up of experts appointed by the EU countries, issued guidance with a list of analytical characteristics. Beside the limits, linearity and precision of this chapter, the list names carry-over and comparison between instruments. Clinical performance studies follow ISO 20916 of 2019, which was confirmed in 2025. The PCR guide in this library explains how often a positive or negative result is right.

52 — Each claim and its document
CLAIM RECOGNIZED DOCUMENT (FDA NUMBER, ENTRY DATE) ROCHE'S ASSAY OR A NOTE limits of blank, detection, quantitation CLSI EP17-A2 (7-233, 2013) 60 blanks per lot and analyzer type; LoQ from 84 per sample precision EP05-A3 (7-251, 2015); user check EP15-A3 (7-253) 21 × 2 × 2 per lot linearity EP06, 2nd ed. (7-306, 2021) 6–10,000 ng/L interference EP07, 3rd ed. (7-275, 2018); EP37, 2nd ed. (7-348, Dec 2025) biotin tested to 20 ng/mL in 2017; FDA 2020 asks 3,500 method comparison EP09, 3rd ed. (7-296, in part) reference interval EP28-A3c (7-224, 2014) 99th percentile from 1,301 people antibody interference in immunoassays I/LA30-A (7-176, 2009) clinical performance studies ISO 20916:2019 (not in the FDA database; harmonized in the EU) a test level chosen too low EU: MDCG 2022-2 groups the evidence as scientific validity, analytical performance, clinical performance
ReadoutFocal detail
Plate 52 — Each analytical claim has a recognized method: EP17 for the detection limits, EP05 for precision, EP06 for linearity, EP07 and EP37 for interference, EP09 for comparison and EP28 for reference intervals. Following a recognized document lets a maker cite it in a submission.
Write the claims first, then size the studies

List every analytical claim the labeling will make, with its concentration: limits of blank, detection and quantitation, precision at each decision level, the linear range, the interferents and their test levels, the hook-free limit and carry-over. Assign each claim its recognized design, and decide the lots, analyzers, sites and days before the first run, since a study that did not repeat a source of scatter cannot later report it.

+ What this chapter established
  • The limits of blank, detection and quantitation each rest on a stated design, which for Roche's troponin T assay meant 60 results per reagent lot and analyzer type for the limit of blank, 60 per lot for the limit of detection and 84 per sample per lot for the limit of quantitation.
  • Precision is built from repeatability, between-run and between-day scatter, and only a design that repeats each can separate them, as CLSI's EP05 does with 80 results over 20 days.
  • Linearity, interference, hook and carry-over each have their own recognized methods, and testing an interferent at too low a level can let wrong results reach patients later.
  • The FDA recognizes the CLSI documents used here, and the EU groups the same evidence under scientific validity, analytical performance and clinical performance.
24 — The system is the product

Reagent, calibrator, analyzer, software.

+ The questionWhy is an assay cleared together with the analyzer it runs on, and what happens when either changes?

A recall with nothing wrong in the bottle

On 2 January 2018 Siemens Healthineers began a recall of its troponin I assay for the Dimension Vista analyzer. The reagent itself had not changed. The instructions for use did not state the concentration of biotin below which the assay was free of interference, and the FDA's record gives the root cause as an error in labeling. The recall followed an earlier one, in March 2017, over biotin values printed in the wrong units in the same instructions.

A product's labeling is the instructions and other text supplied with it, and it is part of the product, because a result means only what the labeling says it means. The same is true of the calibrators that set the curve, the controls that check it, the analyzer that runs the protocol and the software that turns light into a number. Earlier chapters showed each of them shaping the result: calibration in Chapter 14, timing and temperature in Chapter 12, the detector in Chapter 13. Regulators therefore judge the performance of the combination. This chapter follows what that means when one part changes.

Cleared as a system

The FDA states the principle in one sentence in its Replacement Reagent and Instrument Family Policy: clearance of a test system "is based on assay performance characteristics demonstrated with an instrument (or instruments) specified by the assay manufacturer". The same guidance notes that most automated clinical instruments by themselves are in class I, the lowest of the FDA's three risk classes, and exempt from premarket review. The risk lies in the performance claim, and that claim is made for the assay run on named analyzers.

Roche's troponin T assay shows the arrangement. Its 2017 clearance covered the reagent, its calibrator set, its control material and a calibration-check material, and the decision summary names the two analyzers, the cobas e 411 and the cobas e 601. Troponin assays are in class II, the FDA's middle class. Read literally, the regulation that classifies them describes tests for creatine kinase, the older cardiac marker, and troponin falls under it through its own product code, the three-letter code the FDA gives each type of device. The exemption for analyzers has limits. A general section of the same part of the FDA's regulations withdraws it from devices used at the point of care and from devices that assess the risk of cardiovascular disease, so an analyzer built for near-patient testing cannot rely on it.

Solved example: one reagent, two analyzers

Chapter 12 showed the same reagent running in two steps on the cobas e 411 and in one step on the cobas e 601, with limits of detection of 5 and 3 ng/L. The two analyzers shared a limit of quantitation, 6 ng/L, and under the US clearance results are reported from there. In the European guideline, the "very low" value for this assay is below 5 ng/L. That value therefore sits under the US reporting floor of the same reagent: in the US, every such patient's result falls below the measuring range, which starts at 6 ng/L. A cut-off published for one labeled version of an assay cannot be applied to the reports of another, even when the chemistry in the bottle is identical, because the system includes its protocol and its labeling.

The EU sorts the parts as the US does, and India's list places the analyzer in its lowest class. Under the EU's guidance on classifying IVDs, revised in September 2026, a troponin test for acute patients is class C and needs a notified body. Instruments intended specifically for diagnostic procedures are class A, which a maker can declare itself. India's regulator updated its list of IVD classes in October 2023, and the list puts a chemiluminescent immunoassay analyzer in class A and a clinical chemistry analyzer for near-patient testing in class C. In the US and the EU the reagent carries the higher class and the analyzer the lower. India's class for a troponin reagent was not found for this guide. In each place the performance claim ties the two together.

53 — What was cleared in 2017
THE CLEARED SYSTEM, K162895 (18 JAN 2017) REAGENT biotin capture antibody ruthenium detection antibody streptavidin beads calibrator set control material calibration check ANALYZERS NAMED cobas e 411 two steps cobas e 601 one step calibration curve, flags SOFTWARE INSTRUCTIONS FOR USE limits, biotin warning ONE PERFORMANCE CLAIM 6–10,000 ng/L; 99th percentile 19 ng/L; limits of detection 3–5 ng/L the label is part of the product (recall, 2018) US: reagent class II; analyzers alone mostly class I, exempt EU: troponin test class C; instrument class A India: chemiluminescent immunoassay analyzer class A
AntibodyLabel and signalReadoutFocal detail
Plate 53 — A clearance covers performance shown with named analyzers, so the reagent, calibrators, controls, analyzers, software and instructions form one product. Regulators in the US, EU and India put the reagent in a higher class than the analyzer while the claim binds them.

Open systems, families and migration

An assay that runs only on its maker's analyzers is a closed system. An assay whose maker validates it on other instruments is open, at least to that maker. The FDA's Replacement Reagent and Instrument Family Policy has been final since 17 August 2022. Under it, a maker may move a cleared assay to another cleared instrument in the same family without a new 510(k), provided it performs a risk assessment, verifies and validates against the protocols and acceptance criteria of the original clearance, and documents the work. The policy applies only to open systems. It does not cover class III devices, the FDA's highest-risk class, or blood-bank tests, home tests or tests waived under CLIA, the US law on clinical laboratories (Chapter 26). The instruments in a family share architecture, design, tolerances, regulation and product code.

Moving an assay to a new kind of system is a bigger step. Under the FDA's 2013 guidance on assay migration, a sponsor, the company that applies to the FDA, can bring a cleared or approved assay to a new combination of instrument and software with focused analytical studies instead of new clinical trials. The studies cover detection limits compared with the old system, precision over days, runs and calibrations, reproducibility at three or more sites, linearity, and comparison on panels of samples. The intended use must stay the same: the purpose of the test and the patients for whom it is meant. A migration is a way of showing that nothing important changed, and it cannot add a claim.

When a part changes

Every change raises the same question: could it move a result that a patient's care depends on? The FDA's 2017 guidance on deciding when to submit a new 510(k) for a changed device has a section and a flowchart for IVDs, and a companion document for software. Roche's response to biotin interference, described in Chapter 19, went through that process. Adding a scavenger antibody and lengthening a linker changed the reagent's formulation, and Roche obtained a new clearance for the reformulated assay in September 2021 rather than updating the old one.

Reagent lots change routinely, each with its own assigned curve, as Chapter 14 described. When a lot's results show a bias, the remedy is a correction, such as the lot-specific correction factors in the Siemens recall of that chapter. The laboratory's own check is a comparison of the old and new lots on patient samples, by the method of CLSI's EP26.

A software update is a change to the test

An analyzer's software schedules incubations, applies the calibration curve, flags hooks and converts counts into results. A change to any of those can move results without any change to the reagent, which is why regulators treat software versions as part of the cleared or certified system and ask makers to assess each one.

For devices still in transition from the old directive, EU guidance of May 2022 describes which changes are significant and which are not. Changing an essential ingredient such as a capture antibody, introducing a new algorithm, making a major change of operating system, or moving from visual to automated reading is significant. Replacing an ELISA instrument, adjusting incubation time or temperature, fixing bugs or adding a manufacturing site is not. "Significant" here decides whether a legacy device keeps its transitional status under Article 110(3) of the IVDR; a change outside the list can still move results, and its effect on the performance claims still has to be assessed. The guidance states that a maker in doubt should ask its notified body. In December 2025 the European Commission proposed to simplify the rules. Its proposal would sort changes to certified devices into those needing no notification, prior notification or a notified body's approval, and would allow plans that agree certain changes in advance. As of October 2026 it had not been adopted.

In India, a major change after licensing needs prior approval, which is deemed granted if the licensing authority does not respond within 60 days, and a minor change must be notified within 30 days. In an addendum of 13 March 2026 to its questions and answers on IVDs, the regulator states that each new version of approved software needs a post-approval change application.

54 — What a change triggers
US EU INDIA REAGENT FORMULATION (e.g. biotin fix) new 510(k): cleared Sep 2021 capture-antibody change is significant (MDCG 2022-6, legacy devices) major change: prior approval, deemed after 60 days ASSAY ONTO ANOTHER INSTRUMENT family policy (2022): documented, no new 510(k), open systems only; new system: migration studies (2013) replacing an ELISA instrument: not significant post-approval change SOFTWARE VERSION assess under the 2017 software change guidance new algorithm or major OS change: significant every new version: post-approval change (13 Mar 2026) each new version needs an application REAGENT LOT quality system; lab checks by EP26 quality system quality system EU proposal of Dec 2025 (not adopted Oct 2026) would add no-notification, prior-notification and approval tiers
Focal detail
Plate 54 — The same change triggers different steps in each jurisdiction: a reformulated reagent needed a new 510(k), an assay moved within an instrument family needs documented verification, and in India every new software version needs a post-approval change application.
Keep one change record that answers every regulator

For each change to a reagent, calibrator, analyzer, firmware or labeling, record which performance claims it could move, the verification that was run, and the regulatory decision for each market: a documented assessment or a new 510(k) in the US, notification or approval through the notified body in the EU, and a post-approval change application in India. The same assessment then serves all three, and a recall like the one that opened this chapter shows up as a missing line in the record before it shows up in the field.

+ What this chapter established
  • The FDA clears assay performance demonstrated on named analyzers, and most analyzers alone are class I and exempt from premarket review.
  • The EU also puts the troponin reagent in a higher class than the analyzer, and India puts the analyzer in class A, while the claim binds reagent and analyzer together.
  • The same reagent can carry different limits of detection on two analyzers, and different reporting floors under its US and European labeling, so the system, not the bottle, is the product.
  • Changes to reagents, software or labeling are assessed against the claims they could move, under rules that differ by jurisdiction.
25 — The analyzer's own standards

Safety, EMC, risk and software.

+ The questionAn analyzer never touches a patient. What can it harm, and which standards cover that?

A medical device with laboratory hazards

An immunoassay analyzer is a medical device, yet the standard that governs its electrical safety is written for laboratory equipment. That standard is IEC 61010-1, for laboratory equipment in general, and IEC 61010-2-101 of 2018 is its particular part for in vitro diagnostic equipment, including equipment for self-testing. A particular part adds requirements for one kind of equipment and is used together with the general standard. IEC 60601-1, the standard for medical electrical equipment, belongs to a different family. The choice makes sense once the question is asked the other way round: what could this machine harm?

An analyzer can harm the people who run it. It holds mains electricity, moving pipetting arms, heated incubators, magnets, blood that may carry infection, and chemicals such as the acidic peroxide and sodium hydroxide that trigger a flash label (Chapter 7). Those are the hazards of laboratory equipment. An analyzer can also disturb other equipment, or be disturbed by it, through electromagnetic interference. And it can harm a patient it never meets, through a wrong number. In its overview of IVD regulation, the FDA links the safety of an IVD to its false-negative and false-positive results, and that third route to harm is the one most particular to a diagnostic analyzer.

The wrong result as the hazard

Risk management for medical devices follows ISO 14971, whose 2019 edition was confirmed in 2025. Applied to an analyzer, risk management traces each fault forward to a wrong result, then to a situation in which the result is acted on, then to harm. The PCR guide in this library works through that chain for a thermal cycler, the heating and cooling instrument of PCR. For an immunoassay analyzer the faults are often the ones earlier chapters described, and real cases show the chain from end to end.

Carry-over is one such fault. Chapters 11 and 23 described the 2021 Beckman Coulter recall; by the firm's account in the FDA's recall record, the reason was that a sample above 270,000 ng/L could carry over through the probe into the particle well of the reagent pack, and the record gives the root cause as undetermined. The fault as described is mechanical, the wrong result a falsely raised troponin in a later patient, and the harm a needless cardiac workup or a missed alternative diagnosis. The remedy was a procedure: after any such result, the open packs are discarded, a new one is loaded, low-level quality control is run on every pipettor and the affected positive results are repeated.

Biotin is another. In assays such as the original version of Roche's troponin T assay, the fault was a design tolerance for biotin as low as 20 ng/mL, the wrong result a falsely low troponin, and the harm of the kind the FDA reported in 2017, a death. The biotin risk was controlled first by a warning in the instructions and later by a change to the reagent, the route Chapter 19 described.

Solved example: a hazard counted in results

In 2021 authors who included Roche staff published an evaluation of the reformulated troponin T assay. They estimated that the original assay's susceptibility to biotin would misclassify about 0.026 percent of samples, about 1 in 3,800, in the population they modeled. A hospital that measures 100 troponins a day runs 36,500 a year, so at that rate it would expect about ten misclassified results a year: rare for any one patient, routine for the laboratory. Risk management works with such counts, set against the severity of a missed heart attack, rather than with the impression that a rare interference is negligible.

55 — From a fault to a harm
FAULT WRONG RESULT RESULT ACTED ON HARM CARRY-OVER (2021 RECALL) probe carries a sample above 270,000 ng/L into the reagent pack falsely raised troponin in a later patient acted on needless cardiac workup CONTROL procedure: discard open packs, new pack, low-level QC on every pipettor, repeat positives BIOTIN (2017 WARNING) reagent tolerance 20 ng/mL falsely low troponin acted on missed heart attack; a death reported CONTROLS first: warning in the instructions then: reagent changed (scavenger antibody, longer linker; 1,500 ng/mL) a design control does not rely on a reader COUNTED IN RESULTS 1 in 3,800 samples misclassified (2021 estimate, maker's staff among authors) × 36,500 tests a year ≈ 10 results a year risk management per ISO 14971:2019
InterferentReadoutFocal detail
Plate 55 — An analyzer harms a patient through a wrong number, so risk management traces each fault forward to a result and a harm. The biotin risk was first controlled by a warning and later by changing the reagent; at 1 in 3,800 samples, a busy laboratory would still meet about ten such results a year.

Safety and compatibility, as of October 2026

The general laboratory safety standard, IEC 61010-1, stands at its third edition of 2010 with an amendment of 2016. A second amendment would align it with the standard for audio, video and information technology equipment. According to a test house, a company that tests equipment against standards, that amendment was at the committee draft vote in April 2025. Its publication could not be confirmed as of October 2026. The FDA recognizes IEC 61010-1 with US national differences, in its consolidated version, which merges the standard with its amendment. The FDA's database did not list the IVD part, IEC 61010-2-101, which a maker can still apply as the state of the art, the generally accepted current practice. The part for automatic laboratory equipment in general, IEC 61010-2-081, excludes IVD equipment and refers it to Part 2-101.

Electromagnetic compatibility, or EMC, is the ability of equipment to work without disturbing other equipment or being disturbed by it. Its standards changed more recently. IEC 61326-2-6, the particular part for IVD equipment, reached its fourth edition in June 2025. The FDA recognized that edition in full on 28 July 2025, having recognized the 2020 edition only in part. A maker designing an analyzer in 2026 therefore tests against an edition little more than a year old.

56 — Hazard, standard and status
HAZARD STANDARD (EDITION) FDA RECOGNIZED EU HARMONIZED (LIST TO OCT 2024) electric shock, heat, moving parts, chemicals, infection (operator) IEC 61010-1 Ed 3.1; IEC 61010-2-101:2018 (IVD) 61010-1 yes (US differences); 2-101 not listed no electromagnetic disturbance IEC 61326-2-6 Ed 4.0 (Jun 2025) yes , in full (28 Jul 2025) no wrong result: risk management ISO 14971:2019 (confirmed 2025) yes yes software IEC 62304 Ed 1.1 (2015); Ed 2 still a draft yes no use errors IEC 62366-1 + A1:2020 yes no information and symbols ISO 18113 parts 1–5 (2022); ISO 15223-1 + A1:2025 18113 not listed ; 15223-1 yes 15223-1 yes ; labeling and symbols decisions 2026 quality system ISO 13485:2016 (confirmed Oct 2025) incorporated in the QMSR, 2 Feb 2026 yes met as state of the art, without presumption status as of October 2026; the EU list is the consolidated list of October 2024
ReadoutFocal detail
Plate 56 — An analyzer's operator hazards fall under IEC 61010 with its IVD part, its electromagnetic compatibility under IEC 61326-2-6, its software under IEC 62304. The FDA recognizes most of them; the EU's harmonized list leaves out electrical safety, EMC, software and usability.

Software, usability and labels

The software that schedules an analyzer's steps and computes its results is developed under IEC 62304, which in October 2026 was still its 2006 edition as amended in 2015, the version the FDA recognizes. A second edition, widened to health software in general, remained a draft. In July 2026 the German standards association VDE described it as a working draft, expected to replace the three software safety classes with two levels of process rigor. The safety classes set how rigorous development must be according to the harm a software failure could cause. One software firm reported that the new edition was published in August 2026, but the IEC's own catalog listed no newer version, so this guide treats it as unpublished.

Usability engineering, which addresses the errors users make, follows IEC 62366-1 as amended in 2020, and the FDA recognizes that standard. The five parts of ISO 18113 set out the information a maker must supply with its products. All five are in second editions of October 2022, with Part 3 for instruments used by professionals and Part 5 for self-testing instruments, and the FDA database does not list them. The symbols on labels follow ISO 15223-1 of 2021, whose 2025 amendment the FDA recognized in December 2025.

Under all of these sits the quality management system, the procedures by which a maker controls design, production and change. Its standard is ISO 13485 of 2016, confirmed again in October 2025. The FDA incorporated it by reference into its Quality Management System Regulation, which has been in force since 2 February 2026.

The EU's harmonized list leaves out the analyzer's electrical standards

In the EU, a standard listed as harmonized under the IVDR gives a presumption of conformity: a device that meets the standard is presumed to meet the requirements it covers. The list as consolidated to October 2024 held 15 entries, among them the standards for quality management, risk management, symbols, traceability and clinical performance studies, and none for electrical safety, electromagnetic compatibility, software or usability. Judging by their titles, decisions in 2026 added the labeling standard and the amended symbols standard. For the other requirements, a maker applies the IEC standards as the state of the art, but without the presumption, and has to argue its case in its technical documentation.

The regulators thus agree on what an analyzer must show, and differ on which documents carry formal weight. The FDA recognizes most of the IEC set. The EU's harmonized list, as consolidated to October 2024 and as far as the titles of the 2026 decisions show, includes none of it; a 2025 amendment was not read. Even so, regulators on both sides treat the set as the established way to demonstrate safety, compatibility and sound software.

Trace every standard to a hazard

Start the standards list from the hazard analysis, not from a template: for each hazard to the operator, to other equipment and to the patient through the result, name the standard and the edition that addresses it, and record its status with the FDA, in the EU's harmonized list and in India. Then record the date each edition was checked, since several documents in this chapter changed status between 2025 and 2026, among them the EMC part, the symbols standard, the EU's harmonized list and the FDA's quality regulation.

+ What this chapter established
  • An analyzer's own hazards are those of laboratory equipment, so its safety standards are IEC 61010-1 and the IVD part, IEC 61010-2-101.
  • Its most particular hazard is a wrong result, traced through risk management from fault to harm, as in the carry-over and biotin cases.
  • The EMC part for IVD equipment reached a fourth edition in 2025, and IEC 62304's second edition was still a draft in October 2026.
  • The FDA recognizes most of these standards, and the EU's harmonized list leaves out electrical safety, EMC, software and usability.
26 — Four routes to market

US, EU, India, and WHO prequalification.

+ The questionWhat does each regulator, and the program that UN buyers of rapid tests rely on, ask of a reagent and its analyzer, as of October 2026?

The European Union, from 26 May 2022

On 26 May 2022 the EU's In Vitro Diagnostic Medical Devices Regulation, the IVDR, began to apply, five years after its publication. Its central mechanism is the one the three regulators in this chapter share: the risk a wrong result poses decides who must check the evidence before a test is sold. The IVDR sorts devices into four classes, A to D, by rules set out in an annex. For class B and above, and for sterile class A devices, a notified body designated under the regulation must take part in assessing the maker's quality system and technical documentation, more fully for the higher classes.

Under the EU's guidance on classification, as revised in September 2026, the troponin tests, strips and analyzers of this guide fall into the following classes. The parts of the guidance read for this guide give no example for the Alzheimer's blood tests, and their EU class was not found. A troponin test for patients who arrive with acute symptoms is class C, as a cardiac marker. Tests for self-testing are class C, except for a short list, including pregnancy tests, that is class B. Instruments intended specifically for in vitro diagnostic procedures are class A, for which the maker declares conformity itself, and class B is the default for anything the rules do not place elsewhere. The analyzer and its troponin reagent therefore reach the market by different routes of assessment, as Chapter 24 showed.

The number of notified bodies has lagged behind the need for them. At the end of February 2026 a Commission survey counted 19 notified bodies designated under the IVDR, which together had received 3,418 applications and issued 2,318 certificates. A regulation of 2024 extended the transition period for devices that need a notified body for the first time. During the extension such devices may stay on the market under the old directive, on conditions and with deadlines that depend on the class.

Solved example: the calendar for a class C reagent

Take a troponin reagent sold under the old directive that needs a notified body for the first time under the IVDR. Under the 2024 extension it may stay on the market until 31 December 2028, provided its maker had its quality management system in place by 26 May 2025, lodged its application with a notified body by 26 May 2026, and signed a written agreement with that body by 26 September 2026. By October 2026 that last deadline had passed. For a class D device the application and agreement deadlines and the end date each fall a year earlier: application by 26 May 2025, agreement by 26 September 2025, and the market until the end of 2027. For class B and sterile class A devices they fall a year later: application by 26 May 2027, agreement by 26 September 2027, and the market until the end of 2029. The quality management system was due by 26 May 2025 in every class.

57 — The IVDR transition, by class
2022 2023 2024 2025 2026 2027 2028 2029 2030 class D → 31 Dec 2027 class C → 31 Dec 2028 class B and sterile A → 31 Dec 2029 quality system in place, 26 May 2025 application, 26 May 2025 signed agreement, 26 Sep 2025 26 MAY 2022: IVDR APPLIES 8 OCT 2026 THIS GUIDE passed twelve days before this guide's date Regulation (EU) 2024/1860
ReadoutFocal detail
Plate 57 — Under the 2024 extension, legacy class D devices may stay on the EU market to the end of 2027, class C to the end of 2028 and class B to the end of 2029, if their makers met deadlines for a quality system, an application and a signed agreement with a notified body.

The rules may change again. In December 2025 the European Commission proposed to amend the IVDR and its companion regulation for other medical devices. The proposal would replace the five-year expiry of certificates with periodic reviews based on risk, reduce notified-body involvement for lower-risk devices, and allow changes to be agreed in advance. As of October 2026 it had not been adopted. The Council of the EU, where the member states' governments sit, had issued a progress report in May. In July the European Parliament's committee rapporteur, the member who leads its work on the proposal, had published a draft report.

The United States

The FDA also sorts devices into classes, I to III, and the class decides the route. A device similar to one already on the market reaches it through a 510(k), a demonstration of substantial equivalence to that earlier device, which is called its predicate. A low- or moderate-risk device with no predicate goes through a De Novo request, in which the FDA sets up a new classification for that type of device. A class III device needs premarket approval, in which the FDA reviews evidence of the device's own safety and effectiveness. Roche's troponin T assay is in class II and was cleared through a 510(k). The first spinal fluid test for amyloid had no predicate and was granted through De Novo in May 2022. It then served as the predicate for the blood test of Chapter 21.

A second decision follows clearance. Under the Clinical Laboratory Improvement Amendments, or CLIA, the law that governs US laboratories, the FDA assigns each test a complexity: waived, moderate or high. Seven criteria, each scored from 1 to 3, separate moderate from high complexity. A test whose scores sum above 12 is high complexity, and one at or below 12 is moderate, while waived status is decided separately. Until it is categorized, a test is treated as high complexity. The category decides which laboratories may run the test, so a maker designs for it as much as for the clearance.

Tests developed and run within a single laboratory have a separate history. In May 2024 the FDA issued a rule that would have made explicit that such laboratory-developed tests are devices under the law. A federal court in Texas set the rule aside, or vacated it, on 31 March 2025, and in a rule published that September the FDA removed from its regulation the words the 2024 rule had added, "including when the manufacturer of these products is a laboratory". As of October 2026 the Alzheimer's blood tests that Quanterix sells as laboratory-developed tests remained outside FDA clearance. Quanterix submitted a 510(k) for one of them in early 2026, and no decision on it had been found.

India

India regulates IVDs under the Medical Devices Rules, 2017, which took effect at the start of 2018 and use the same four classes, A to D. India's national regulator for drugs and devices, the Central Drugs Standard Control Organisation or CDSCO, houses the Central Licensing Authority. That authority licenses all imports and the manufacture of class C and D devices, approves new IVDs and their clinical performance evaluation, and issues test licenses, which allow a device to be made or imported in small quantities for testing, evaluation, clinical investigation or training. State licensing authorities license the manufacture of class A and B devices and the sale of devices. An importer applies on Form MD-14 and receives a license on Form MD-15. Analyzers, instruments and software came under licensing in two phases, a year apart: the lower classes on 1 October 2022, and classes C and D on 1 October 2023. The regulator's 2023 list puts a chemiluminescent immunoassay analyzer in class A.

India's distinctive requirement is local evidence. The rules call an IVD new until the Central Licensing Authority has approved it for manufacture or import in India. A new IVD in class B, C or D needs a clinical performance evaluation in India, a study of its performance on patient samples, whatever its approvals in the EU, the US or elsewhere. For imported kits, the evaluation uses three batches made from three lots of key raw materials and takes place at central or registered testing laboratories. The regulator updated its lists of those laboratories three times between June 2025 and June 2026. In September 2025 the regulator and the medical research council published standard evaluation protocols and a portal for kit validation. The class of a troponin reagent in the regulator's lists was not found for this guide.

WHO prequalification

The fourth route is not a regulator. The World Health Organization says of itself that it "is not a regulatory authority". It assesses products for prequalification, a benchmark that UN and other international procurement agencies use when they buy, and that national regulators can rely on. Its program for IVDs became operational in 2010, growing out of a scheme for evaluating test kits begun in 1988. An assessment has three parts: a review of the dossier, an inspection of the manufacturing site and a review of the labeling. The dossier is the maker's file of evidence, and an abridged one is allowed where a recognized regulator has already assessed the product stringently. WHO has also made the laboratory evaluation of performance a separate procedure that comes first, done by designated evaluation laboratories.

Prequalification serves buyers; registration stays national

A prequalified rapid test still needs registration in each country that uses it. Prequalification shortens national review where a regulator chooses to rely on it.

Eligibility is limited to priority diseases, mainly infections such as HIV, hepatitis B and C, malaria, syphilis and tuberculosis, and a few point-of-care tests. There is no category for cardiac markers or for Alzheimer's disease, so neither of this guide's two cases has a route through prequalification. In October 2026 the list held 125 prequalified IVDs. The entries from makers whose names indicate Indian companies were all rapid tests on strips, for HIV, hepatitis C, syphilis and malaria. Through prequalification, the strip technology of Chapter 15 reaches international procurement.

58 — Four routes, one evidence base
US · FDA classes I–III → 510(k), De Novo or premarket approval then CLIA category: waived, moderate, high (7 criteria) LDT rule of 2024 vacated, 2025 EU · IVDR classes A–D; notified body for B–D and sterile A troponin: class C; instrument: class A 19 notified bodies (Feb 2026) INDIA · CDSCO classes A–D (Medical Devices Rules, 2017) import: MD-14 → MD-15 new class B–D IVD: clinical performance evaluation in India, 3 batches from 3 lots the route that asks for local data WHO PREQUALIFICATION not a regulator: a benchmark for UN buyers dossier, site inspection, labeling priority infections only; no cardiac or Alzheimer's route 125 IVDs listed; Indian entries all rapid tests one evidence base: analytical studies (Ch 23) + a well-designed clinical study
ReadoutFocal detail
Plate 58 — The US, the EU and India all sort tests by risk class and decide who must check the evidence; WHO prequalification is a procurement benchmark for priority infections. One well-planned evidence base serves all four, with India's local clinical evaluation the distinct requirement.
Plan the evidence once and file it four ways

The analytical studies of Chapter 23 and a well-designed clinical study serve every route. Plan them once, with the strictest requirement of each in view: the US classification and CLIA category, the EU class, notified body and transition dates, an Indian clinical performance evaluation on three batches from three raw-material lots, and prequalification only where the disease is eligible. Start the Indian evaluation and the notified-body agreement early, since both run on calendars a maker does not control.

Back to Basel

The Basel-coordinated study drew blood from 718 patients between 2006 and 2008 and sought nothing from a regulator. The blood was frozen and measured blind in batches, and the results were judged against a diagnosis made from the records, which is the design of a clinical performance study. The paper appeared in August 2009, and Roche's sensitive troponin T assay reached laboratories in Britain and Ireland that year. The European guideline based a one-hour decision on such assays in 2015, and in January 2017 the FDA cleared the assay for the US, using the same 718 patients as part of its evidence. Nearly seven and a half years separated the paper from the clearance.

Most chapters of this guide describe part of what had to hold in that interval. The antibodies and the label set how few molecules the test could see; the analyzer held that floor from run to run; the calibration tied it to a scale; the studies showed its blank, its scatter and its interferents; and the regulators decided, each in its own way, what evidence was enough. A few molecules in a drop became a number a physician could act on in an hour only when all of those held at once.

+ What this chapter established
  • The IVDR has applied since 26 May 2022, and under it troponin reagents are class C and need a notified body while analyzers are class A, with transition deadlines running to 2029.
  • The FDA routes devices by class through a 510(k), De Novo or premarket approval and then assigns a CLIA complexity, and a court vacated its 2024 rule on laboratory-developed tests in 2025.
  • India licenses IVDs under the Medical Devices Rules, 2017, and requires local clinical performance evaluation for new class B to D tests.
  • WHO prequalification is not regulation but a procurement benchmark for priority infections, and it covers neither troponin nor Alzheimer's tests.
§ Lessons

Lessons.

The chapters reduce to thirteen working rules for people who design, verify, buy, regulate or write about immunoassays and the analyzers that run them. Each rule points to the chapters that explain why it holds.

  1. Write the target as a count of molecules before choosing a method. A concentration of a few nanograms per liter is a few million molecules in a drop, and the count that reaches the detector is smaller still after capture, washing and sampling (Chapters 1, 17 and 18).
  2. Name the epitopes and the circulating forms, not only the protein. Two antibody pairs against the same troponin can count different subsets of it, and assays with different epitopes and calibrators read the same blood differently (Chapters 2 and 14).
  3. Budget the minutes of binding against antibody concentration and distance. An excess of antibody speeds binding, beads spread through the sample shorten the distance each molecule must diffuse, and a strip gives each part of the sample only seconds over its line (Chapters 3, 12 and 15).
  4. Choose the format from the size of the molecule, and state the hook-free limit. Large proteins go between two antibodies and small molecules compete for one; a one-step sandwich can read falsely low at extreme concentrations, the hook effect, so the hook-free limit belongs in the specification (Chapters 4, 13 and 19).
  5. Treat the blank as the floor, and report every limit with the design that produced it. The signal from label that remains when no target is present sets the limit of blank, and limits of blank, detection and quantitation mean little without the replicates, lots, analyzers and days behind them (Chapters 5, 18 and 23).
  6. Judge a label by the background it leaves, not by its brightness alone. Color reads a small dip in a bright beam; chemiluminescence, delayed fluorescence and electrochemiluminescence each succeeded by measuring against less unwanted light, and an amplifying label also amplifies the signal from wrongly bound label (Chapters 6 to 9).
  7. Specify the analyzer as a chain of subsystems, and test the joins. Sampling, reagent handling, incubation, washing, detection and software each move the result, and carry-over and temperature drift can escape routine controls (Chapters 10 to 12).
  8. Specify the detector from both ends of the range. Dark count and light collection set the detector's share of the blank, counting statistics set the scatter of small signals, and saturation and dilution set the top (Chapter 13).
  9. Tie every result to its calibration chain and to the limits of its own assay. A master curve adjusted on site, a lot change or a new generation can move results, and troponin cut-offs cannot be carried from one assay to another (Chapters 14 and 20).
  10. Validate the reader with the test, whether it is a person or a machine. The same strips gave different clinical sensitivities when read by different people, and an instrument reader applies its threshold just as consistently to a strip that ran badly (Chapters 15 and 16).
  11. Ship the interference checks with the assay, and run them when a result disagrees with the patient. Heterophile antibodies, biotin, macrocomplexes and the hook effect produce wrong results that look plausible, and a mismatch with the patient's condition, or between methods, exposes them (Chapter 19).
  12. Count the middle zone and the prevalence in every performance claim. An observation zone or an indeterminate result moves the work of reaching an answer to a later test instead of removing it, and predictive values change with the share of patients who have the condition (Chapters 20 and 21).
  13. Plan the evidence once for the whole system, then file it by each route. Regulators judge the reagent, calibrators, analyzer, software and labeling together, sort them by risk in different ways, and change their rules on calendars a maker does not control (Chapters 23 to 26).
§ Glossary

Glossary.

Terms are defined as they are used in this guide.

510(k)
A US premarket submission showing that a device is substantially equivalent to a similar device already on the market, its predicate; the guide's troponin T assay and the first Alzheimer's blood test were cleared this way.
99th-percentile upper reference limit
The troponin value below which 99 percent of a healthy reference population falls; above it, troponin is a sign of injury to heart muscle. For the guide's troponin T assay in the US it is 19 ng/L overall, 14 ng/L for women and 22 ng/L for men.
Absorbance
A logarithmic measure of the light a colored solution removes from a beam, read by a photometer or plate reader: 1 means a tenth of the light got through, 2 a hundredth and 3 a thousandth. Readers stay linear only to about 2 or 3.
Acridinium ester
A chemiluminescent label that reacts once with hydrogen peroxide in alkali and gives its light in a flash lasting a few seconds; Abbott and Siemens Healthineers analyzers use it.
AEB
Average enzymes per bead: the unit in which a digital ELISA reports its result, worked out from the share of beads that light.
Amyloid pathology
Deposits in the brain of the protein fragment beta-amyloid, assessed in clinical practice by a PET scan or a spinal fluid test. The cleared Alzheimer's blood tests aid in assessing it; they do not diagnose the disease.
Analyte
The substance an assay measures; this guide mostly calls it the target.
Antibody
A protein the immune system makes to bind one molecular shape. The kind used in immunoassays, immunoglobulin G, is a Y-shaped protein of about 146,000 grams per mole with an identical binding site at the tip of each arm.
Assay
A procedure that measures how much of a particular substance a sample contains.
Assay migration
Moving a cleared or approved assay to a new combination of instrument and software, supported by focused analytical studies instead of new clinical trials; the intended use must stay the same, and no claim can be added.
Association and dissociation rate constants
The rate at which two molecules bind, typically about a million per molar per second for proteins, and the rate at which a bound pair comes apart, between about 1 in 1,000 and 1 in 100,000 per second for good antibodies. Their ratio is the dissociation constant.
Beta-amyloid
A protein fragment that forms deposits in the brain; its 42-amino-acid form, beta-amyloid 1-42, is measured in plasma beside pTau217 in the first cleared Alzheimer's blood test.
Biotin
A small B vitamin attached to an antibody so that streptavidin will grab it. Free biotin from high-dose supplements occupies the streptavidin, so assays built on it read falsely low in a sandwich and falsely high in a competitive format.
Blocker
An inert protein, such as bovine serum albumin or milk protein, that coats any surface the capture antibody left bare; added animal serum or mouse antibody likewise soaks up a patient's interfering antibodies.
Calibration curve
The curve, drawn through calibrators, that converts an analyzer's signal into a concentration. For an immunoassay it is S-shaped, flat near zero and near the top, and in a competitive assay it slopes downward.
Calibrator
A sample with an assigned concentration, measured to draw or adjust the calibration curve; calibration is repeated with each new reagent lot, at set intervals and whenever quality control fails.
Capillary flow
The pull that draws liquid through a porous material, as water climbs a paper towel. On a lateral flow strip it moves the sample and sets the timing, slowing as the wetted length grows.
Carry-over
The transfer of material from one sample, or one reagent, into the next, reported as a fraction of the high sample, such as parts per million; routine quality control does not reveal it.
CDSCO
The Central Drugs Standard Control Organisation, India's regulator; its Central Licensing Authority licenses all imported IVDs, the manufacture of class C and D devices and the approval of new IVDs.
CE mark
The marking under which a device is sold in the EU; under the IVDR most IVDs need a notified body's assessment first, while the maker of a class A instrument declares conformity itself.
Chemiluminescence
Light from a chemical reaction whose product forms in an excited state. Measured against darkness, with no lamp, it reaches lower concentrations and a wider range than color.
CLIA categorization
The FDA's assignment of a test to waived, moderate or high complexity under the Clinical Laboratory Improvement Amendments, which decides which US laboratories may run it. Seven criteria are each scored 1 to 3, and a sum above 12 means high complexity.
Clinical performance study
A study that judges a test's results against a reference diagnosis of each patient, as the Basel troponin study did; in the EU it follows ISO 20916, and India requires a local clinical performance evaluation for new class B to D IVDs.
Closed and open system
An assay that runs only on its maker's analyzers is a closed system; one whose maker validates it on other instruments is open, at least to that maker. The FDA's 2022 instrument family policy applies only to open systems.
CLSI
The Clinical and Laboratory Standards Institute, a US standards body for laboratory medicine; its EP documents, such as EP05 for precision and EP17 for detection limits, set study designs that the FDA recognizes.
Coefficient of variation
The standard deviation of repeated results divided by their mean, as a percentage; a high-sensitivity troponin assay must reach 10 percent or less at the 99th percentile.
Colloidal gold
Gold particles a few tens of nanometers across, the most common label on lateral flow strips; particles of 20 to 40 nanometers absorb most strongly in the green, at 524 to 530 nanometers, so a line of them looks red.
Competitive assay
An immunoassay in which a labeled copy of the target competes with the target in the sample for a scarce antibody, so the signal falls as the concentration rises. Small molecules, too small to hold two antibodies, are measured this way, as was insulin in the first radioimmunoassay.
Coreactant
The chemical, tripropylamine in the guide's troponin assay, that sits in large excess beside an electrochemiluminescent label; oxidized at the electrode, it breaks into short-lived fragments that excite the label.
Cross-reactivity
Binding of an antibody to a molecule other than its intended target, such as skeletal troponin T in a cardiac assay or a drug's breakdown products in a competitive assay.
Cut-off
A value that divides results into answers, such as positive and negative, or rule-out, observation and rule-in; troponin cut-offs belong to one assay and cannot be carried to another.
Dark count
Pulses a detector gives with no light falling on it, from electrons released by heat in the cathode; about 50 a second in one photon-counting module, it is the detector's own floor.
De Novo
The US route for a new low- or moderate-risk device with no predicate; the first spinal fluid test for amyloid was granted this way in May 2022.
Decision summary
The FDA's published summary of the review behind a clearance, the source of most of this guide's figures for cleared assays.
Digital ELISA
An immunoassay that seals beads into wells of tens of femtoliters, so that a single enzyme lights its well, and counts the lit wells instead of measuring a glow. Its analyzers are labeled for research use only.
Dioxetane
A light-producing substrate for alkaline phosphatase: the enzyme removes a phosphate group, and the unstable remainder falls apart and emits light, glowing for minutes.
Dissociation constant
The concentration of target at which half of an antibody's binding sites are occupied at equilibrium; smaller is tighter. With antibody at twenty times this constant, more than 90 percent of a scarce target is bound.
Electrochemiluminescence
Light from a label, a ruthenium complex in the guide's troponin assay, driven by a voltage at an electrode together with a coreactant. Each label returns to its starting form and can emit many photons, and only labels within about 3 micrometers of the electrode light.
ELISA
Enzyme-linked immunosorbent assay: an immunoassay read by the color an enzyme label makes from its substrate, first published in 1971 and now most often a sandwich in the wells of a plastic plate.
Energy transfer (TRACE)
A method in which a europium label on one antibody passes its energy to an acceptor on a second antibody only when both bind the same target molecule, so bound and free label are told apart without washing; Thermo Fisher Scientific's KRYPTOR analyzers use it under the name TRACE.
Enzyme label
An enzyme, usually horseradish peroxidase or alkaline phosphatase, attached to an antibody so that each bound antibody turns a substrate into many molecules of colored, fluorescent or light-giving product.
Epitope
The patch of a target's surface that an antibody touches, typically about 20 amino acids; a protein the size of troponin has room for several antibodies, each at its own epitope.
HAMA
Human anti-mouse antibodies: antibodies made against mouse proteins, for example after treatment with mouse antibodies, which bind them strongly and specifically; they matter because many reagent antibodies are made in mice. In practice the name is often used interchangeably with heterophile antibody.
Heterophile antibody
A patient's antibody, of no clearly defined origin, that binds the antibodies of two or more other species. Bridging the capture and labeled antibodies, it gives a falsely high result; binding only one of them, it can give a falsely low one.
High-dose hook
The fall in signal at very high target concentrations in a one-step sandwich, when free target fills the capture and labeled antibodies separately, so an enormous concentration can read as a modest one. Makers state a hook-free limit, 100,000 ng/L for the guide's troponin T assay.
High-sensitivity troponin assay
A troponin assay with a coefficient of variation of 10 percent or less at the 99th percentile and results above its limit of detection in at least half of healthy people, as defined by a task force of the International Federation of Clinical Chemistry in 2017.
IEC 61010
The safety standard for laboratory equipment, which analyzers follow instead of the medical electrical standard IEC 60601-1: Part 1 covers laboratory equipment in general, and Part 2-101, of 2018, in vitro diagnostic equipment.
IEC 61326-2-6
The electromagnetic compatibility standard for IVD equipment, covering how an instrument disturbs other equipment or is disturbed by it; its fourth edition, of June 2025, was recognized in full by the FDA on 28 July 2025.
IEC 62304
The standard for developing medical device software, such as the software that schedules an analyzer's steps and computes its results; in October 2026 its 2006 edition, amended in 2015, was still current.
Immunoassay
An assay in which the part that recognizes the substance being measured is an antibody.
Indeterminate result
A result between two cut-offs, which gives neither a positive nor a negative answer and sends the patient to a confirmatory test; about a fifth of results in the first cleared Alzheimer's blood test.
Interference
A change in a result caused by something in the sample other than the target, such as biotin, heterophile antibodies, rheumatoid factor or molecules related to the target; the substance responsible is an interferent.
International Federation of Clinical Chemistry (IFCC)
The international body whose working groups and task forces compare troponin assays, publish a table of their characteristics supplied by the makers, and in 2017 defined a high-sensitivity troponin assay.
ISO 13485
The international standard for the quality management systems of medical device makers, in its 2016 edition; the FDA's Quality Management System Regulation (QMSR), in force since 2 February 2026, incorporates it by reference.
ISO 14971
The standard for risk management of medical devices, in a 2019 edition; applied to an analyzer, it traces each fault to a wrong result, to a situation in which that result is acted on, and to harm.
IVD
In vitro diagnostic device: a reagent, calibrator, analyzer or other product used to test samples, such as blood, outside the body.
IVDR
The EU's In Vitro Diagnostic Medical Devices Regulation, applied since 26 May 2022; it sorts IVDs into classes A to D and requires a notified body for around 80 percent of them, against about 8 percent under the 1998 directive it replaced.
Label
The part of an immunoassay that makes a bound antibody visible: an enzyme that makes color, a molecule that gives off light, or a particle that shows as a colored line.
Laboratory-developed test
A test developed and run within a single laboratory. A 2024 FDA rule that would have made such tests explicitly devices was vacated by a federal court in March 2025.
Lateral flow test
An immunoassay run inside a strip of porous material: the sample moves by capillary flow, dissolves a dried labeled antibody and carries it past a printed test line of capture antibody and then a control line.
Limit of blank
The highest result expected from samples that contain none of the target, set so that only 5 percent of blank results exceed it; in its simplest form, the mean of the blanks plus 1.645 standard deviations.
Limit of detection
The lowest concentration whose results exceed the limit of blank 95 percent of the time; 3 to 5 ng/L, depending on the analyzer, for the guide's troponin T assay in the US.
Limit of quantitation
The lowest concentration measured with an agreed precision, often a coefficient of variation of no more than 20 percent; 6 ng/L for the guide's troponin T assay in the US.
Macrocomplex
A target bound to the patient's own antibodies, such as macroprolactin, which is prolactin bound mostly to immunoglobulin G, or troponin bound to antibodies; it can read high on one assay and normal on another.
Magnetic microparticles
Beads a few micrometers across, typically coated with streptavidin, that collect sandwiches throughout the reaction volume and are held by a magnet while the liquid around them is exchanged.
Master curve
A calibration curve the maker measures for each reagent lot, from master calibrators on many instruments, and encodes with the lot; the analyzer adjusts it with a two-level calibrator set.
Measuring range
The span of concentrations an assay reports without dilution, across which results are claimed to stay proportional to concentration; 6 to 10,000 ng/L for the guide's troponin T assay in the US.
Medical Device Coordination Group (MDCG)
The EU group that issues guidance under the IVDR; its 2022 guidance divides the evidence for an IVD into scientific validity, analytical performance and clinical performance.
Medical Devices Rules, 2017
The rules under which India regulates IVDs, in effect since the start of 2018, with four classes, A to D, and licensing divided between the CDSCO and state authorities.
Mole
An amount of substance holding about 6.02 × 10²³ molecules; an attomole, a billionth of a billionth of a mole, is about 602,000 molecules. One mole per liter is one molar, so picomolar means trillionths of a mole per liter.
Monoclonal antibody
An antibody made by one line of cells, a hybridoma, that binds one epitope. Unlike a polyclonal antibody, the varying mixture in an immunized animal's serum, it can be supplied unchanged for decades, which made reagent lots consistent.
Multiplex assay
An assay that measures several targets in one sample. Each added target adds unintended antibody pairs that can cross-react, so a floor measured for one target alone does not carry over to a panel.
Notified body
An independent organization, designated by an EU member state, that assesses a maker's quality system and technical documentation before an IVD from class B upward can be sold; 19 were designated under the IVDR at the end of February 2026.
One-hour algorithm
The European Society of Cardiology's rule for ruling a heart attack in or out from two troponin samples an hour apart, using the first value and the change between them, with cut-offs specific to each assay; patients in neither group are observed.
PET
Positron emission tomography: a brain scan that, with a spinal fluid test, is how amyloid pathology is assessed in clinical practice.
Photomultiplier tube
A vacuum-tube detector in which a photon striking the photocathode releases an electron, and a chain of dynodes multiplies it into a pulse of a million to ten million electrons.
Poisson distribution
The pattern of counts of random events, which fixes, for example, the share of beads carrying no enzyme from the average number per bead. A count that follows it scatters by its own square root: 100 counts by 10 percent, 10,000 by 1 percent.
Positive and negative predictive value
The share of positive results that are right and the share of negative results that are right; both depend on prevalence among the people tested.
Premarket approval
The US route to market for a class III device, the highest of the FDA's three classes.
Prevalence
The share of the people tested who have the condition; 17 percent had a heart attack in the 2015 study of the one-hour algorithm.
Proximity ligation and extension assays
Methods in which two binders carrying DNA strands must attach to the same molecule before their strands yield a piece of DNA that is copied and measured; a binder on its own leaves nothing to copy, so background falls and many assays can share one sample.
pTau217
Tau protein carrying a phosphate group at amino acid 217, in forms that circulate as fragments, at fractions of a picogram per milliliter of plasma; measured to aid in assessing amyloid pathology.
Quality control
Samples of known concentration run at intervals and plotted on charts to catch drift; a run is rejected under rules such as one control more than three standard deviations from its target, or two in a row more than two on the same side.
Quantum efficiency
The fraction of photons striking a detector's cathode that release an electron; up to 43 percent at 350 nanometers for the best blue-sensitive cathodes, and much lower in the red.
Reader
An instrument that reads an assay's signal. A plate reader measures absorbance in many wells at once; a strip reader measures each line against the strip's own background with its own light and applies the same threshold every time.
Recall
A maker's correction or removal of a product already on the market, recorded by the FDA; the remedies in this guide include new instructions for laboratories and lot-specific correction factors.
Reference material
A material whose value is known well enough to anchor a calibration chain; it serves only if it behaves like patient samples in the assays that use it.
Rheumatoid factor
An antibody that binds the stem of other antibodies; it can bridge reagent antibodies like a heterophile antibody and cause false positives in troponin assays.
Sandwich assay
An immunoassay in which a capture antibody and a labeled antibody bind two different epitopes of the same target, so the signal rises with concentration from near zero; also called a two-site assay.
Sensitivity
Clinically, the share of people with the condition whose test is positive, such as 40 percent for the Liverpool strips read on site. Analytically, as in a more sensitive assay, how low a concentration a test can detect.
Solid phase
The surface that holds the capture antibody, such as the inside of a tube or well or the surface of beads, so that bound label stays while free label is washed away.
Specificity
Clinically, the share of people without the condition whose test is negative, 99.9 percent for the Liverpool strips. For an antibody, its selective binding to one shape, which weakens quickly as a molecule departs from that shape.
Standard deviation
The usual measure of the scatter of repeated results around their mean.
Streptavidin
A bacterial protein that grips biotin almost irreversibly; coated on beads, it collects the biotin-carrying capture antibody together with the sandwich it holds.
Time-resolved fluorescence
Reading a long-lived label such as europium, which glows for tens of microseconds to about a millisecond, after a delay long enough for scattered light and the sample's own fluorescence, which fade in about 10 nanoseconds, to disappear.
Traceability
The link, through an unbroken chain of comparisons, from an assay's calibrators to an accepted reference, ideally a reference measurement procedure and a certified reference material; troponin I still lacks such a chain end to end.
Troponin
A complex of three proteins, T, I and C, that controls muscle contraction; the cardiac forms of troponin T and I leak into the blood when heart cells die and are measured to rule a heart attack in or out.
Wash
Rinsing away everything that did not bind while the surface keeps what did, so that only label held by the target is left to measure; it works by repeated dilution, not as a filter.
WHO prequalification
The World Health Organization's assessment of an IVD as a benchmark for UN and other international buyers, limited to priority diseases such as HIV, hepatitis B and C, malaria, syphilis and tuberculosis; WHO is not a regulator, so registration stays national.
Within-laboratory precision
The total scatter of results a laboratory sees over weeks, combining repeatability within a run with scatter between runs and between days; CLSI's EP05 estimates it from 80 results per sample over 20 days.
§ Sources

Sources.

Each source is listed under the chapter whose text first relies on it, with the opening scene and the front matter first. Company documents and papers by a company's own staff are marked, as they are in the text. Regulatory status, standards editions and product facts are as of October 2026.

  1. Opening — Reichlin T., Hochholzer W., Bassetti S., Steuer S., Stelzig C., Hartwiger S., et al. Early diagnosis of myocardial infarction with sensitive cardiac troponin assays. N Engl J Med 361, 858–867 (2009). doi:10.1056/NEJMoa0900428 (abstract read, as reproduced at fis.uke.de and edoc.unibas.ch)
  2. Opening — US Food and Drug Administration. 510(k) K162895, Elecsys Troponin T Gen 5 STAT Assay (Roche Diagnostics): database record, decision summary, 510(k) summary and clearance letter; decision 18 Jan 2017. accessdata.fda.gov/cdrh_docs/reviews/K162895.pdf; accessdata.fda.gov/cdrh_docs/pdf16/K162895.pdf
  3. Opening — US Food and Drug Administration. FDA clears first blood test used in diagnosing Alzheimer’s disease. Press announcement, 16 May 2025. fda.gov/news-events/press-announcements/fda-clears-first-blood-test-used-diagnosing-alzheimers-disease
  4. Opening — Washington University School of Medicine, Emergency Medicine Journal Club. Critical appraisals of Reichlin et al. 2009 and Keller et al. 2009 (March 2011). emergencymedicine.wustl.edu/app/uploads/2018/10/March2011AK4.pdf and March2011AK3.pdf (secondary)
  5. Opening — Keller T., Zeller T., Peetz D., Tzikas S., Roth A., Czyz E., et al. Sensitive troponin I assay in early diagnosis of acute myocardial infarction. N Engl J Med 361, 868–877 (2009). doi:10.1056/NEJMoa0903515 (abstract read, as reproduced at fis.uke.de)
  6. Opening — LabMedica (27 Oct 2009) and Pathology in Practice (4 Dec 2009). Trade-press reports of the launch of Roche’s high-sensitive troponin T assay on cobas and MODULAR Analytics serum work areas in the UK and Ireland (secondary, reporting the manufacturer)
  7. Opening — Roffi M., Patrono C., Collet J.-P., et al. 2015 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J 37, 267–315 (2016; released online Aug 2015). doi:10.1093/eurheartj/ehv320 (abstract and partial text read)
  8. Opening — ADLM. First high-sensitivity troponin test cleared for use in the U.S. Clinical Laboratory News, Regulatory Roundup (1 Mar 2017). myadlm.org/cln/articles/2017/march/first-high-sensitivity-troponin-test-cleared-for-use-in-the-us (secondary)
  9. Opening — UniProt Consortium. UniProtKB P45379 (TNNT2_HUMAN), human cardiac troponin T. rest.uniprot.org/uniprotkb/P45379 (accessed 8 Oct 2026)
  10. Opening — Moman R.N., Gupta N., Singh C., Varacallo M.A. Physiology, Albumin. StatPearls (NCBI Bookshelf NBK459198), updated 31 Jan 2026 (secondary)
  11. Opening — Katus H.A., Looser S., Hallermayer K., et al. Development and in vitro characterization of a new immunoassay of cardiac troponin T. Clin Chem 38, 386–393 (1992). doi:10.1093/clinchem/38.3.386 (abstract read) [authors include company staff]
  12. Ch. 1 — Katrukha I.A., Katrukha A.G. Myocardial injury and the release of troponins I and T in the blood of patients. Clin Chem 67, 124–130 (2021). doi:10.1093/clinchem/hvaa281 [authors include company staff]
  13. Ch. 1 — Apple F.S., Sandoval Y., Jaffe A.S., Ordonez-Llanos J., for the IFCC Task Force on Clinical Applications of Cardiac Bio-Markers. Cardiac troponin assays: guide to understanding analytical characteristics and their impact on clinical care. Clin Chem 63, 73–81 (2017). doi:10.1373/clinchem.2016.255109 (as reproduced at cms.ifcc.org/media/478955/clin-chem-troponin-review-2017.pdf) [authors declare company ties]
  14. Ch. 1 — Vidarsson G., Dekkers G., Rispens T. IgG subclasses and allotypes: from structure to effector functions. Front Immunol 5, 520 (2014). doi:10.3389/fimmu.2014.00520
  15. Ch. 1 — Musso G., Gabelli C., Cagnin A., et al. Diagnostic performances and cut-off verification of blood pTau 217 on the Lumipulse platform for amyloid deposition in Alzheimer’s disease. Clin Chem Lab Med (2024). doi:10.1515/cclm-2024-1091
  16. Ch. 1 — UniProt Consortium. UniProtKB P10636-8 (MAPT isoform Tau-F, 2N4R, 441 amino acids). rest.uniprot.org/uniprotkb/P10636-8 (accessed 8 Oct 2026)
  17. Ch. 1 — US Food and Drug Administration. 510(k) K242706, Lumipulse G pTau217/β-Amyloid 1-42 Plasma Ratio (Fujirebio Diagnostics): database record, 510(k) summary and decision summary; decision 16 May 2025. accessdata.fda.gov/cdrh_docs/reviews/K242706.pdf; accessdata.fda.gov/cdrh_docs/pdf24/K242706.pdf
  18. Ch. 1 — Collet J.-P., Thiele H., Barbato E., et al. 2020 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J 42, 1289–1367 (2021; online 29 Aug 2020). doi:10.1093/eurheartj/ehaa575
  19. Ch. 1 — Giannitsis E., Kurz K., Hallermayer K., Jarausch J., Jaffe A.S., Katus H.A. Analytical validation of a high-sensitivity cardiac troponin T assay. Clin Chem 56, 254–261 (2010). doi:10.1373/clinchem.2009.132654 (abstract read) [authors include company staff]
  20. Ch. 2 — BioNumbers. BNID 104045, Stokes radius of IgG, citing Boyer P.M., Hsu J.T. AIChE J 38, 259–272 (1992). bionumbers.hms.harvard.edu (secondary)
  21. Ch. 2 — Goodsell D. Molecule of the Month: Antibodies. RCSB PDB-101 (September 2001). pdb101.rcsb.org/motm/21 (secondary)
  22. Ch. 2 — Rubinstein N.D., Mayrose I., Halperin D., Yekutieli D., Gershoni J.M., Pupko T. Computational characterization of B-cell epitopes. Mol Immunol 45, 3477–3489 (2008). doi:10.1016/j.molimm.2007.10.016
  23. Ch. 2 — Sela-Culang I., Kunik V., Ofran Y. The structural basis of antibody-antigen recognition. Front Immunol 4, 302 (2013). doi:10.3389/fimmu.2013.00302
  24. Ch. 2 — David G.S., Greene H.E. Immunometric assays using monoclonal antibodies. US Patent 4,376,110 (Hybritech Inc.), filed 4 Aug 1980, issued 8 Mar 1983. patents.google.com/patent/US4376110A [manufacturer]
  25. Ch. 2 — Ekins R., Kelso D. Single-molecule ELISA. Clin Chem (Papers in Press, 13 Oct 2010). doi:10.1373/clinchem.2010.152850 (accepted manuscript as reproduced at quanterix.com)
  26. Ch. 2 — Lipman N.S., Jackson L.R., Trudel L.J., Weis-Garcia F. Monoclonal versus polyclonal antibodies: distinguishing characteristics, applications, and information resources. ILAR J 46, 258–268 (2005). doi:10.1093/ilar.46.3.258
  27. Ch. 2 — Köhler G., Milstein C. Continuous cultures of fused cells secreting antibody of predefined specificity. Nature 256, 495–497 (1975). doi:10.1038/256495a0 (abstract read)
  28. Ch. 2 — Nobel Prize Outreach. The Nobel Prize in Physiology or Medicine 1984, summary. nobelprize.org/prizes/medicine/1984/summary/
  29. Ch. 2 — Jaffe A.S., Ordonez-Llanos J. High sensitivity troponin in chest pain and acute coronary syndromes. A step forward? Rev Esp Cardiol 63, 763–769 (2010)
  30. Ch. 2 — UniProt Consortium. UniProtKB P19429 (TNNI3_HUMAN), human cardiac troponin I. rest.uniprot.org/uniprotkb/P19429 (accessed 8 Oct 2026)
  31. Ch. 2 — Müller-Bardorff M., Hallermayer K., Schröder A., et al. Improved troponin T ELISA specific for cardiac troponin T isoform: assay development and analytical and clinical validation. Clin Chem 43, 458–466 (1997). doi:10.1093/clinchem/43.3.458 (abstract read) [three of nine authors at the Boehringer Research Center, Tutzing]
  32. Ch. 3 — BioNumbers. BNIDs 112534 and 112538 (protein–protein and antibody–antigen association rate constants), citing Northrup S.H., Erickson H.P. PNAS 89, 3338–3342 (1992) and Noble R.W., Reichlin M., Gibson Q.H. J Biol Chem 244, 2403–2411 (1969). bionumbers.hms.harvard.edu (secondary)
  33. Ch. 3 — Jarmoskaite I., AlSadhan I., Vaidyanathan P.P., Herschlag D. How to measure and evaluate binding affinities. eLife 9, e57264 (2020). doi:10.7554/eLife.57264
  34. Ch. 3 — BioNumbers. BNID 100612, diffusion coefficient of human serum albumin in water, citing Vogel S. Life’s Devices (Princeton University Press, 1988). bionumbers.hms.harvard.edu (secondary)
  35. Ch. 3 — Milo R., Phillips R. What are the time scales for diffusion in cells? In Cell Biology by the Numbers (Garland Science, 2015), online edition. book.bionumbers.org (secondary)
  36. Ch. 3 — Corning. Representative microplate dimensions, 96-, 384- and 1536-well plates (drawing, undated). corning.com/catalog/cls/documents/drawings/MicroplateDimensions96-384-1536.pdf [manufacturer]
  37. Ch. 3 — Zanut A., Fiorani A., Canola S., Saito T., Ziebart N., Rapino S., et al. Insights into the mechanism of coreactant electrochemiluminescence facilitating enhanced bioanalytical performance. Nat Commun 11, 2668 (2020). doi:10.1038/s41467-020-16476-2 [authors include company staff]
  38. Ch. 3 — Squires T.M., Messinger R.J., Manalis S.R. Making it stick: convection, reaction and diffusion in surface-based biosensors. Nat Biotechnol 26, 417–426 (2008). doi:10.1038/nbt1388 (abstract read)
  39. Ch. 3 — Roche Diagnostics. cobas e 801 analytical unit, product page and specifications (undated; accessed 8 Oct 2026). diagnostics.roche.com/global/en/products/instruments/cobas-e-801-ins-2202.html [manufacturer]
  40. Ch. 3 — Department of Health Abu Dhabi. Health Technology Review: Elecsys Troponin T hs / Elecsys Troponin T hs STAT, V1.0 (HTA23005; undated). doh.gov.ae (secondary)
  41. Ch. 3 — Collinson P.O., Saenger A.K., Apple F.S., on behalf of the IFCC Committee on Clinical Application of Cardiac Bio-Markers. High sensitivity, contemporary and point-of-care cardiac troponin assays: educational aids developed by the IFCC C-CB. Clin Chem Lab Med 57, 623–632 (2019). doi:10.1515/cclm-2018-1211 (as reproduced at cms.ifcc.org; table values as provided by the manufacturers) (secondary)
  42. Ch. 4 — Berson S.A., Yalow R.S., Bauman A., Rothschild M.A., Newerly K. Insulin-I131 metabolism in human subjects: demonstration of insulin binding globulin in the circulation of insulin treated subjects. J Clin Invest 35, 170–190 (1956). doi:10.1172/JCI103262
  43. Ch. 4 — Yalow R.S. Radioimmunoassay: a probe for fine structure of biologic systems. Nobel Lecture, 8 December 1977, pp. 447–468. nobelprize.org/uploads/2018/06/yalow-lecture.pdf
  44. Ch. 4 — Kahn C.R., Roth J. Berson, Yalow, and the JCI: the agony and the ecstasy. J Clin Invest 114, 1051–1054 (2004). doi:10.1172/JCI23316 (secondary)
  45. Ch. 4 — Nobel Prize Outreach. The Nobel Prize in Physiology or Medicine 1977, summary. nobelprize.org/prizes/medicine/1977/summary/
  46. Ch. 4 — Yalow R.S., Berson S.A. Assay of plasma insulin in human subjects by immunological methods. Nature 184, 1648–1649 (1959). doi:10.1038/1841648b0 (abstract read)
  47. Ch. 4 — Yalow R.S., Berson S.A. Immunoassay of endogenous plasma insulin in man. J Clin Invest 39, 1157–1175 (1960). doi:10.1172/JCI104130
  48. Ch. 4 — Miles L.E.M., Hales C.N. Labelled antibodies and immunological assay systems. Nature 219, 186–189 (1968). doi:10.1038/219186a0 (abstract read)
  49. Ch. 4 — Kobayashi N., Kato Y., Oyama H., Goto J. Antibody engineering-based approach for hapten immunometric assays with high sensitivity. Yakugaku Zasshi 127, 55–69 (2007)
  50. Ch. 4 — Piran U., Livshin L.A., Martinelli R.A., Riordan W.J., Unger J.T. Reduction of background in noncompetitive binding assays. US Patent 5,705,338 (Chiron Diagnostics), filed 4 May 1995, issued 6 Jan 1998. patents.google.com/patent/US5705338 [manufacturer]
  51. Ch. 4 — Technopolis Group for the European Patent Office. Presenting the (economic) value of patents nominated for the European Inventor Award 2012: Hugo A. Katus (30 Mar 2012). link.epo.org/web/katus_en.pdf (secondary)
  52. Ch. 4 — Ross G.M.S., Filippini D., Nielen M.W.F., Salentijn G.IJ. Unraveling the hook effect: a comprehensive study of high antigen concentration effects in sandwich lateral flow immunoassays. Anal Chem (2020). doi:10.1021/acs.analchem.0c03740 (as reproduced in the authors’ PDF)
  53. Ch. 4 — Zhang Y.V. The hook effect. ADLM Clinical Chemistry Trainee Council, Pearls of Laboratory Medicine (7 Oct 2014) (secondary)
  54. Ch. 4 — DrugBank. Insulin human (DB00030): average molecular weight 5808.0 Da (database entry, accessed 8 Oct 2026). go.drugbank.com/drugs/DB00030 (secondary)
  55. Ch. 5 — Institute of Isotopes Ltd. hPSA [I-125] IRMA kit, REF RK-10CT, instructions for use (updated March 2020) (as reproduced by the distributor at clinisciences.com) [manufacturer]
  56. Ch. 5 — Tate J., Ward G. Interferences in immunoassay. Clin Biochem Rev 25, 105–120 (2004). PMID 18458713; PMC1904417 (opened in the review round of 14 Oct 2026)
  57. Ch. 5 — Thermo Fisher Scientific. Blocking buffers for Western blot and ELISA (Pierce protein methods; web page, undated). thermofisher.com [manufacturer]
  58. Ch. 5 — CLSI. EP17-A2, Evaluation of Detection Capability for Clinical Laboratory Measurement Procedures, 2nd ed. (18 Jun 2012; reaffirmed 2017). Catalogue page, clsi.org/shop/standards/ep17/
  59. Ch. 5 — Analyse-it Software Ltd. LoB, LoD, LoQ explained (web page, undated). analyse-it.com/learn/lob-lod-loq-explained (secondary)
  60. Ch. 6 — Lequin R.M. Enzyme immunoassay (EIA)/enzyme-linked immunosorbent assay (ELISA). Clin Chem 51, 2415–2418 (2005). doi:10.1373/clinchem.2005.051532 (abstract read) (secondary)
  61. Ch. 6 — Engvall E., Perlmann P. Enzyme-linked immunosorbent assay (ELISA). Quantitative assay of immunoglobulin G. Immunochemistry 8, 871–874 (1971). doi:10.1016/0019-2791(71)90454-X (not opened; citation as given in Engvall 2010)
  62. Ch. 6 — van Weemen B.K., Schuurs A.H.W.M. Immunoassay using antigen-enzyme conjugates. FEBS Lett 15, 232–236 (1971). doi:10.1016/0014-5793(71)80319-8 (not opened; title and issue as cited in secondary sources)
  63. Ch. 6 — Aumiller W.M. Jr, Davis B.W., Hatzakis E., Keating C.D. Interactions of macromolecular crowding agents and cosolutes with small-molecule substrates: effect on horseradish peroxidase activity with two different substrates. J Phys Chem B 118, 10624–10632 (2014). doi:10.1021/jp506594f
  64. Ch. 6 — Gao L., Zhuang J., Nie L., et al. Intrinsic peroxidase-like activity of ferromagnetic nanoparticles. Nat Nanotechnol 2, 577–583 (2007). doi:10.1038/nnano.2007.260 (abstract read; peroxidase kinetic values as reproduced in the supplementary information of doi:10.1039/c8tb01132j)
  65. Ch. 6 — Engvall E. The ELISA, enzyme-linked immunosorbent assay. Clin Chem 56, 319–320 (2010). doi:10.1373/clinchem.2009.127803 (abstract read)
  66. Ch. 6 — Bowers G.N. Jr, McComb R.B., Christensen R.G., Schaffer R. High-purity 4-nitrophenol: purification, characterization, and specifications for use as a spectrophotometric reference material. Clin Chem 26, 724–729 (1980). doi:10.1093/clinchem/26.6.724 (abstract read)
  67. Ch. 6 — BMG LABTECH. SPECTROstar Nano, product page and specifications (accessed 8 Oct 2026). bmglabtech.com/en/spectrostar-nano/ [manufacturer]
  68. Ch. 6 — Opentrons. Flex Absorbance Plate Reader, documentation (accessed 8 Oct 2026). docs.opentrons.com/flex/modules/absorbance-plate-reader/ [manufacturer]
  69. Ch. 6 — R&D Systems (Bio-Techne). Human IL-6 Quantikine ELISA Kit D6050, product page (accessed 8 Oct 2026). rndsystems.com/products/human-il-6-quantikine-elisa-kit_d6050 [manufacturer]
  70. Ch. 7 — Hamamatsu Photonics. Photon counting heads H10682 series, datasheet TPMO1075E04 (Nov 2024). hamamatsu.com/content/dam/hamamatsu-photonics/sites/documents/99_SALES_LIBRARY/etd/H10682_TPMO1075E.pdf [manufacturer]
  71. Ch. 7 — Weeks I., Beheshti I., McCapra F., Campbell A.K., Woodhead J.S. Acridinium esters as high-specific-activity labels in immunoassay. Clin Chem 29, 1474–1479 (1983). doi:10.1093/clinchem/29.8.1474 (abstract read)
  72. Ch. 7 — Natrajan A., Wen D. Effect of branching in remote substituents on light emission and stability of chemiluminescent acridinium esters. RSC Adv (2014). doi:10.1039/c4ra02516d (author version) [authors include company staff]
  73. Ch. 7 — US Food and Drug Administration. 510(k) K191595, ARCHITECT STAT High Sensitivity Troponin-I (Abbott Laboratories): database record and decision summary; decision 13 Sep 2019. accessdata.fda.gov/cdrh_docs/reviews/K191595.pdf
  74. Ch. 7 — US Food and Drug Administration. 510(k) K223602, Traumatic brain injury (TBI) test on the Alinity i system (Abbott Laboratories): database record and decision summary; decision 2 Mar 2023. accessdata.fda.gov/cdrh_docs/reviews/K223602.pdf
  75. Ch. 7 — Siemens Healthineers. Atellica Solution IM 1300 and IM 1600 Analyzer specification sheet (08-2017). marketing.webassets.siemens-healthineers.com [manufacturer]
  76. Ch. 7 — Siemens Healthineers. ADVIA Centaur XPT immunoassay system, product page (accessed 8 Oct 2026). siemens-healthineers.com/en-us/immunoassay/systems/advia-centaur-xpt [manufacturer]
  77. Ch. 7 — Whitehead T.P., Thorpe G.H.G., Carter T.J.N., Groucutt C., Kricka L.J. Enhanced luminescence procedure for sensitive determination of peroxidase-labelled conjugates in immunoassay. Nature 305, 158–159 (1983). doi:10.1038/305158a0 (abstract read)
  78. Ch. 7 — Thorpe G.H.G., Kricka L.J., Moseley S.B., Whitehead T.P. Phenols as enhancers of the chemiluminescent horseradish peroxidase-luminol-hydrogen peroxide reaction: application in luminescence-monitored enzyme immunoassays. Clin Chem 31, 1335–1341 (1985). doi:10.1093/clinchem/31.8.1335 (abstract read)
  79. Ch. 7 — Bronstein I., Edwards B., Voyta J.C. 1,2-Dioxetanes: novel chemiluminescent enzyme substrates. Applications to immunoassays. J Biolumin Chemilumin 4, 99–111 (1989) (not opened; title from a catalogue listing, volume and pages unverified)
  80. Ch. 7 — Kricka L.J. Chemiluminescent and bioluminescent techniques. Clin Chem 37, 1472–1481 (1991). doi:10.1093/clinchem/37.9.1472 (abstract read) (secondary)
  81. Ch. 7 — US Food and Drug Administration. 510(k) K252393, VITROS hs Troponin I Reagent Pack (Ortho-Clinical Diagnostics): database record and decision summary; decision 29 Oct 2025. accessdata.fda.gov/cdrh_docs/reviews/K252393.pdf
  82. Ch. 7 — Sandison M., De Silva R., Melkamu G., Kilmartin P., Koti J. A new and improved chemiluminescent substrate. Poster (Lumigen Inc. and Beckman Coulter, 2019). media.beckmancoulter.com [manufacturer]
  83. Ch. 7 — Beckman Coulter. A better chemiluminescent substrate (clinical studies web page, accessed 8 Oct 2026). beckmancoulter.com/clinical-studies-and-trends/clinical-studies/a-better-chemiluminescent-substrate [manufacturer]
  84. Ch. 8 — Revvity. Time-resolved fluorometry with the DELFIA method, user guide 010765_01 (2023); DELFIA instrument options and settings; DELFIA time-resolved fluorescence assays (web pages, accessed 8 Oct 2026). resources.revvity.com/pdfs/gde-reagents-delfia-user-guide.pdf; revvity.com/ask [manufacturer]
  85. Ch. 8 — Soini E., Kojola H. Time-resolved fluorometer for lanthanide chelates: a new generation of nonisotopic immunoassays. Clin Chem 29, 65–68 (1983). doi:10.1093/clinchem/29.1.65 (abstract read)
  86. Ch. 8 — Mathis G. Rare earth cryptates and homogeneous fluoroimmunoassays with human sera. Clin Chem 39, 1953–1959 (1993). doi:10.1093/clinchem/39.9.1953 (abstract read) [authors include company staff]
  87. Ch. 8 — Thermo Fisher Scientific. Nobel Prize-winning TRACE Technology on B·R·A·H·M·S KRYPTOR Analyzers, white paper 901203.2 (2022). documents.thermofisher.com [manufacturer]
  88. Ch. 8 — US Food and Drug Administration. 510(k) K112161, AQT90 FLEX analyzer with myoglobin test kit (Radiometer Medical ApS): database record; substantially equivalent 20 Jan 2012; and decision summary K120326, AQT90 FLEX CK-MB test kit. accessdata.fda.gov (pmn.cfm?ID=k112161; cdrh_docs/reviews/K120326.pdf) (opened in the review round of 14 Oct 2026)
  89. Ch. 8 — US Food and Drug Administration. 510(k) K163462, AQT90 FLEX Myo and CKMB Test Kits (Radiometer Medical ApS): database record and decision summary; decision 21 Sep 2017. accessdata.fda.gov/cdrh_docs/reviews/K163462.pdf
  90. Ch. 8 — Radiometer. AQT90 FLEX immunoassay analyzer, product page (accessed 8 Oct 2026). radiometer.com/en/products/immunoassay-testing/aqt90-flex-immunoassay-analyzer [manufacturer]
  91. Ch. 8 — bioMérieux. VIDAS High sensitive Troponin I, product page (accessed 8 Oct 2026). biomerieux.com/corp/en/our-offer/clinical-products/vidas-high-sensitive-troponin-i.html [manufacturer]
  92. Ch. 8 — Fulton R.J., McDade R.L., Smith P.L., Kienker L.J., Kettman J.R. Jr. Advanced multiplexed analysis with the FlowMetrix system. Clin Chem 43, 1749–1756 (1997). doi:10.1093/clinchem/43.9.1749 (abstract read) [authors include company staff]
  93. Ch. 9 — US Food and Drug Administration. 510(k) K961487, Elecsys HCG Assay (Boehringer Mannheim Corp.): database record; decision 12 Jun 1996. accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K961487
  94. Ch. 9 — Blackburn G.F., Shah H.P., Kenten J.H., Leland J., Kamin R.A., Link J., et al. Electrochemiluminescence detection for development of immunoassays and DNA probe assays for clinical diagnostics. Clin Chem 37, 1534–1539 (1991). doi:10.1093/clinchem/37.9.1534 (abstract read) [authors include company staff]
  95. Ch. 9 — BioVeris Corp. Form S-4/A registration statement. US Securities and Exchange Commission (2003), accession 0000950123-03-014266. sec.gov/Archives/edgar/data/0001264899/000095012303014266/y89943a3sv4za.txt
  96. Ch. 9 — Wikipedia. Tris(bipyridine)ruthenium(II) chloride (web page, accessed 8 Oct 2026). en.wikipedia.org (secondary)
  97. Ch. 9 — Roche Diagnostics GmbH, Hitachi High-Technologies Corp. Electrochemiluminescence method of detecting an analyte in a liquid sample and analysis system. European Patent EP 3 011 319 B1 (filed 16 May 2014; grant published 25 Mar 2020). data.epo.org [manufacturer]
  98. Ch. 9 — Mayo Clinic Laboratories. Test ID TRPS, Troponin T 5th Generation, Plasma: method description citing the Roche package insert (web page, undated). mayocliniclabs.com/test-catalog/Performance/65832 (secondary)
  99. Ch. 9 — Meso Scale Discovery. Electrochemiluminescence technology (web page, accessed 8 Oct 2026). mesoscale.com/en/technical_resources/our_technology/ecl [manufacturer]
  100. Ch. 10 — Roche Diagnostics. cobas e 402 analytical unit, product page and specifications (undated; accessed 8 Oct 2026). diagnostics.roche.com/global/en/products/instruments/cobas-e-402-ins-6350.html (also at /cobas-e-402.html) [manufacturer]
  101. Ch. 10 — Abbott. ARCHITECT i2000SR immunoassay analyzer, product page and specification table (accessed 8 Oct 2026). corelaboratory.abbott/us/en/offerings/brands/architect/architect-i2000SR [manufacturer]
  102. Ch. 11 — US Food and Drug Administration. Medical Device Recall Z-0032-2022, Access hsTnI (Beckman Coulter; 510(k) K172783, K172787), Class 2; initiated by the firm 9 Aug 2021. accessdata.fda.gov/scripts/cdrh/cfdocs/cfRes/res.cfm?id=188991
  103. Ch. 11 — Analytical assessment of the Beckman Coulter Unicel DxI AccuTnI+3 immunoassay (letter; authors not shown on the page read). De Gruyter (2014). doi:10.1515/dx-2013-0039
  104. Ch. 11 — CLSI. C24, Statistical Quality Control for Quantitative Measurement Procedures: Principles and Definitions, 4th ed. (29 Sep 2016). Catalogue page, clsi.org/standards/products/clinical-chemistry-and-toxicology/documents/c24/
  105. Ch. 11 — QuidelOrtho. VITROS XT 7600 Integrated System, product page (accessed 8 Oct 2026). quidelortho.com/global/en/products/vitros-systems/vitros-xt-7600-integrated-system [manufacturer]
  106. Ch. 11 — Dorenkott J.S., Panek C.F. Method of verifying aspirated volume in automatic diagnostic system. US Patent 6,121,049 (Bayer Corp.), filed 10 Jun 1999, granted 19 Sep 2000. patents.google.com/patent/US6121049 [manufacturer]
  107. Ch. 11 — Quint S., Bauer-Espindola K. Reliability indicating method for an electrochemiluminescence method. US Patent 11,796,477 B2 (Roche Diagnostics Operations), filed 25 Jul 2018, granted 24 Oct 2023. patents.google.com/patent/US11796477B2 [manufacturer]
  108. Ch. 11 — Thermo Fisher Scientific. Dynabeads M-280 Streptavidin, cat. 11205D, product page (accessed 8 Oct 2026). thermofisher.com/order/catalog/product/11205D [manufacturer]
  109. Ch. 11 — Roche Diagnostics. Urgent Field Safety Notice FSN-CPS-2019-022, Elecsys Troponin T hs on cobas e 801: V1 (Oct 2019) and V3 (Jan 2020), published by Swissmedic (FSCA Vk_20191031_01). fsca.swissmedic.ch/mep/api/publications/Vk_20191031_01/documents/15 and /documents/17 [manufacturer]
  110. Ch. 12 — Revvity. How to optimize rapid and simple immunoassays, application note (DELFIA and VICTOR, 2023). resources.revvity.com/pdfs/app-delfia-victor-immunoassay-optimization.pdf [manufacturer]
  111. Ch. 12 — Identification of macrotroponin T: findings from a case report and non-reproducible troponin T results (authors not displayed on the page read). Clin Chem Lab Med (2021). doi:10.1515/cclm-2021-0626 (abstract read)
  112. Ch. 12 — CLSI. EP25, Evaluation of Stability of In Vitro Medical Laboratory Test Reagents, 2nd ed. (26 Apr 2023). Catalogue page, clsi.org/shop/standards/ep25/
  113. Ch. 13 — Hamamatsu Photonics. Photomultiplier Tubes: Basics and Applications, 4th ed. (2017). hamamatsu.com/content/dam/hamamatsu-photonics/sites/documents/99_SALES_LIBRARY/etd/PMT_handbook_v4E.pdf [manufacturer]
  114. Ch. 13 — US Food and Drug Administration. 510(k) K041192, ARCHITECT STAT Troponin-I (applicant Fisher Diagnostics, for Abbott): decision summary (c. 2004). accessdata.fda.gov/cdrh_docs/reviews/K041192.pdf
  115. Ch. 13 — Roche Diagnostics. Elecsys Insulin, method sheet 07027559501 V2.0 (2022-10). elabdoc-prod.roche.com [manufacturer]
  116. Ch. 14 — Tate J.R., Bunk D.M., Christenson R.H., Barth J.H., Katrukha A., Noble J.E., et al., for the IFCC Working Group on Standardization of Cardiac Troponin I. Evaluation of standardization capability of current cardiac troponin I assays by a correlation study: results of an IFCC pilot project. Clin Chem Lab Med 53, 677–690 (2015). doi:10.1515/cclm-2014-1197 (as reproduced at sites.unimi.it/cirme) [authors include company staff]
  117. Ch. 14 — Brown E.N., Skates S.J. Method and system for calibration of immunoassay systems through application of Bayesian analysis. US Patent 5,616,504 (General Hospital Corp.), granted 1 Apr 1997. patents.google.com/patent/US5616504A
  118. Ch. 14 — US Food and Drug Administration. 510(k) K201441, Elecsys Troponin T Gen 5 (Roche Diagnostics): database record and decision summary; decision 21 Sep 2021. accessdata.fda.gov/cdrh_docs/reviews/K201441.pdf
  119. Ch. 14 — US Food and Drug Administration. Medical Device Recall Z-2927-2020, Dimension Vista High Sensitivity Troponin I (TNIH) (Siemens Healthcare Diagnostics; 510(k) K182225), Class 2; initiated 24 Jul 2020. accessdata.fda.gov/scripts/cdrh/cfdocs/cfres/res.cfm?id=182848
  120. Ch. 14 — CLSI. EP26, User Evaluation of Acceptability of a Reagent Lot Change, 2nd ed. (27 Apr 2022). Catalogue page, clsi.org/shop/standards/ep26/
  121. Ch. 14 — Westgard J.O., Barry P.L., Hunt M.R., Groth T. A multi-rule Shewhart chart for quality control in clinical chemistry. Clin Chem 27, 493–501 (1981). doi:10.1093/clinchem/27.3.493 (not opened; citation only)
  122. Ch. 14 — Westgard QC. Westgard Rules (web page, undated). westgard.com/mltirule.htm (secondary)
  123. Ch. 14 — ISO. ISO 17511:2020, 2nd ed. (published 24 Apr 2020; reviewed and confirmed 2026), catalogue page. iso.org/standard/69984.html
  124. Ch. 14 — Christenson R.H., Duh S.H., Apple F.S., Bodor G.S., Bunk D.M., Panteghini M., et al. Toward standardization of cardiac troponin I measurements part II: assessing commutability of candidate reference materials and harmonization of cardiac troponin I assays. Clin Chem 52, 1685–1692 (2006). doi:10.1373/clinchem.2006.068437 (abstract read)
  125. Ch. 14 — National Institute of Standards and Technology. Certificate of Analysis, Standard Reference Material 2921a, Human Cardiac Troponin Complex (issued 10 Mar 2026). tsapps.nist.gov/srmext/certificates/2921a.pdf; with Bunk D.M., Newton D. Certification of Standard Reference Material 2921a Human Cardiac Troponin Complex. NIST Special Publication 260-263 (Jan 2026). nvlpubs.nist.gov/nistpubs/SpecialPublications/NIST.SP.260-263.pdf
  126. Ch. 14 — Roche. Data show Roche’s sixth-generation Troponin T test offers a new level of accuracy critical for diagnosing heart attacks. Media release, Basel, 30 Sep 2025. roche.com/media/releases/med-cor-2025-09-30 [manufacturer]
  127. Ch. 15 — Science Museum Group Collection. Pregnancy test kit, England, 1988 (Clearblue One Step, Unipath Ltd), object 1988-990. collection.sciencemuseumgroup.org.uk/objects/co96440/
  128. Ch. 15 — Clearblue (SPD Swiss Precision Diagnostics). Pregnancy test history 101 (web page, undated; accessed 8 Oct 2026). clearblue.com/pregnancy-tests/pregnancy-test-history-101 [manufacturer]
  129. Ch. 15 — Posthuma-Trumpie G.A., Korf J., van Amerongen A. Lateral flow (immuno)assay: its strengths, weaknesses, opportunities and threats. A literature survey. Anal Bioanal Chem 393, 569–582 (2009). doi:10.1007/s00216-008-2287-2 (secondary)
  130. Ch. 15 — May K., Prior M.E., Richards I. Immunoassays and devices therefor. European patent application EP0291194A1 (Unilever NV), priority 27 Apr 1987, filed 26 Apr 1988. patents.google.com/patent/EP0291194A1/en [manufacturer]
  131. Ch. 15 — Sigma-Aldrich. Gold nanoparticles: properties and applications (technical article, undated; accessed 8 Oct 2026). sigmaaldrich.com [manufacturer]
  132. Ch. 15 — Leuvering J.H.W. Metal sol particle immunoassay. US Patent 4,313,734 (Akzona Inc.), filed 13 Jul 1979, granted 2 Feb 1982. patents.google.com/patent/US4313734A/en [manufacturer]
  133. Ch. 15 — Sartorius. Unisart nitrocellulose membranes, brochure (status 09/2020). sartorius.hr/media/slho0cqh/unisart-nitro-brochure-en-l-sartorius.pdf [manufacturer]
  134. Ch. 15 — Cytiva. FFHP Plus nitrocellulose membranes, product page (accessed 8 Oct 2026). cytivalifesciences.com/en/us/products/items/ffhp-plus-nitrocellulose-membranes-p-10020 [manufacturer]
  135. Ch. 15 — Washburn E.W. Phys Rev 17, 273 (1921) (not opened; cited for the Lucas–Washburn relation)
  136. Ch. 15 — EMD Millipore. Rapid lateral flow test strips: considerations for product development (Hi-Flow Plus membrane guide, 2013). sigmaaldrich.com, document tb500en00em [manufacturer]
  137. Ch. 15 — US Food and Drug Administration. 510(k) K203272, Alltest Pregnancy Rapid Combo Test Cassette (Hangzhou AllTest Biotech): decision summary (the database record gives received 6 Nov 2020, decision 31 Jan 2022, substantially equivalent; opened 14 Oct 2026). accessdata.fda.gov/cdrh_docs/reviews/K203272.pdf
  138. Ch. 15 — Roche Diagnostics. cobas h 232 system, product pages (global and Canada; accessed 8 Oct 2026). diagnostics.roche.com/global/en/products/instruments/cobas-h-232.html; rochecanada.com [manufacturer]
  139. Ch. 15 — Indian Council of Medical Research. Advisory on use of Rapid Antigen Detection Test for COVID-19 (14 Jun 2020) (as reproduced by the Delhi Health Department at health.delhi.gov.in)
  140. Ch. 16 — Department of Health and Social Care and University of Liverpool. Liverpool COVID-19 community testing pilot: interim evaluation report summary (GOV.UK, published 14 Jan 2021). gov.uk/government/publications/liverpool-covid-19-community-testing-pilot-interim-evaluation-report-summary
  141. Ch. 16 — University of Liverpool. Liverpool Covid-19 Community Testing Pilot: Interim Evaluation Report (23 Dec 2020). liverpool.ac.uk/media/livacuk/research/Liverpool,Community,Testing,Pilot,Interim,Evaluation.pdf
  142. Ch. 16 — Science Media Centre. Expert reaction to preliminary report evaluating lateral flow viral antigen detection devices (LFDs) (11 Nov 2020), quoting the Porton Down and University of Oxford preliminary report. sciencemediacentre.org/?p=42045 (secondary)
  143. Ch. 16 — US Food and Drug Administration. 510(k) K120049, BD Veritor System for Rapid Detection of Flu A+B (Becton Dickinson): database record and decision summary; decision 23 Mar 2012. accessdata.fda.gov/cdrh_docs/reviews/K120049.pdf
  144. Ch. 16 — US Food and Drug Administration. 510(k) K112177, Sofia Analyzer and Influenza A+B FIA (Quidel): database record and decision summary; decision 24 Oct 2011. accessdata.fda.gov/cdrh_docs/reviews/K112177.pdf
  145. Ch. 16 — Juntunen E., Myyryläinen T., Salminen T., Soukka T., Pettersson K. Performance of fluorescent europium(III) nanoparticles and colloidal gold reporters in lateral flow bioaffinity assay. Anal Biochem 428, 31–38 (2012). doi:10.1016/j.ab.2012.06.005 (abstract read, as reproduced in the University of Turku repository)
  146. Ch. 16 — Zhan L., Guo S., Song F., Gong Y., Xu F., Boulware D.R., et al. The role of nanoparticle design in determining analytical performance of lateral flow immunoassays. Nano Lett 17, 7207–7212 (2017). doi:10.1021/acs.nanolett.7b02302 (as reproduced in the authors’ PDF)
  147. Ch. 16 — Colombo M., Bezinge L., Rocha Tapia A., Shih C.-J., deMello A.J., Richards D.A. Real-time, smartphone-based processing of lateral flow assays for early failure detection and rapid testing workflows. Sens Diagn (2023; first published 1 Dec 2022). doi:10.1039/d2sd00197g
  148. Ch. 16 — BD. BD receives Emergency Use Authorization for first at-home COVID-19 test to use smartphone to interpret, deliver results. Press release, 25 Aug 2021. news.bd.com [manufacturer]
  149. Ch. 16 — Indian Council of Medical Research. Advisory for COVID-19 home testing using Rapid Antigen Tests (RATs) (19 May 2021) (as reproduced by PRS India at prsindia.org)
  150. Ch. 16 — Indian Council of Medical Research. Rapid Antigen Test Kits for COVID-19, list dated 26 Apr 2021 (as reproduced by the Odisha Health Department at health.odisha.gov.in)
  151. Ch. 16 — US Food and Drug Administration. 510(k) K231974, PATHFAST hs-cTnI-II (PHC Corporation): database record; decision 20 Mar 2024. accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K231974
  152. Ch. 16 — SelectScience. News item on the US point-of-care clearance of PATHFAST hs-cTnI-II (3 Apr 2024) (secondary, reporting the distributor’s claim)
  153. Ch. 16 — US Food and Drug Administration. 510(k) K240984, i-STAT hs-TnI cartridge with the i-STAT 1 System (Abbott Point of Care): database record and decision summary; decision 3 Jan 2025. accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K240984
  154. Ch. 16 — Siemens Healthineers. Expediting the diagnosis of heart attack with Atellica VTLi patient-side analyzer. Press release, 12 Apr 2021. siemens-healthineers.com/press/releases/atellica-vtli [manufacturer]
  155. Ch. 17 — Wilson D.H., Rissin D.M., Kan C.W., et al. The Simoa HD-1 Analyzer: a novel fully automated digital immunoassay analyzer with single-molecule sensitivity and multiplexing. J Lab Autom 21(4) (2016; online 15 Jun 2015). doi:10.1177/2211068215589580 [authors include company staff]
  156. Ch. 17 — Rissin D.M., Kan C.W., Campbell T.G., et al. Single-molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations. Nat Biotechnol 28, 595–599 (2010). doi:10.1038/nbt.1641 (abstract read) [authors include company staff]
  157. Ch. 17 — Rissin D.M., Kan C.W., et al. Attomolar detection of proteins in serum using single molecule enzyme-linked immunosorbent assays. Poster (Quanterix, 2010). quanterix.com/wp-content/uploads/2020/12/oakridge2010.pdf [manufacturer]
  158. Ch. 17 — Rissin D.M., et al. Simultaneous detection of single molecules and singulated ensembles of molecules enables immunoassays with broad dynamic range. Anal Chem (2011). doi:10.1021/ac103161b (abstract read, as reproduced on quanterix.com) [authors include company staff]
  159. Ch. 17 — Quanterix. HD-X Simoa bead-based immunoassay platform, brochure (2023), and product listings on quanterix.com (accessed 8 Oct 2026). quanterix.com/wp-content/uploads/2024/08/2023-HD-X-Brochure-download-version.pdf [manufacturer]
  160. Ch. 17 — Quanterix. Press release on its FDA 510(k) submission for a multi-analyte algorithmic Alzheimer’s blood test (Business Wire, 3 Feb 2026). businesswire.com/news/home/20260203562016/en/ [manufacturer]
  161. Ch. 18 — US Food and Drug Administration. 510(k) K252163, Elecsys Phospho-Tau (181P) Plasma (Roche Diagnostics): database record and decision summary; decision 8 Oct 2025. accessdata.fda.gov/cdrh_docs/reviews/K252163.pdf
  162. Ch. 18 — US Food and Drug Administration. 510(k) K261686, Elecsys Phospho-Tau (217P) Plasma (Roche Diagnostics): database record and decision summary; decision 19 Aug 2026. accessdata.fda.gov/cdrh_docs/reviews/K261686.pdf
  163. Ch. 18 — US Food and Drug Administration. 510(k) K253240, PrecivityAD2 Test (C2N Diagnostics): database record; decision 19 Aug 2026. accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=K253240
  164. Ch. 18 — Healio. FDA clears PrecivityAD2 Alzheimer’s disease blood test as early as age 40 years (21 Aug 2026). healio.com/news/neurology/20260821/fda-clears-precivityad2-alzheimers-disease-blood-test-as-early-as-age-40-years (secondary)
  165. Ch. 18 — Alzheimer’s Association. Statements on the FDA clearances of PrecivityAD2 (20 Aug 2026) and Elecsys pTau217 (24 Aug 2026). alz.org/news/2026/fda-clearance-precivityAD2-blood-test-alzheimers (secondary)
  166. Ch. 18 — Chang L., Rissin D.M., Fournier D.R., et al. Single molecule enzyme-linked immunosorbent assays: theoretical considerations. J Immunol Methods (2012). doi:10.1016/j.jim.2012.02.011 (abstract read, as reproduced on quanterix.com) [authors include company staff]
  167. Ch. 18 — Cohen L., Cui N., Cai Y., Garden P.M., Li X., Weitz D.A., Walt D.R. Single molecule protein detection with attomolar sensitivity using droplet digital enzyme-linked immunosorbent assay. ACS Nano (2020). doi:10.1021/acsnano.0c02378 (author manuscript, as reproduced at par.nsf.gov)
  168. Ch. 18 — Fredriksson S., Gullberg M., Jarvius J., Olsson C., Pietras K., Gustafsdottir S.M., et al. Protein detection using proximity-dependent DNA ligation assays. Nat Biotechnol 20, 473–477 (2002). doi:10.1038/nbt0502-473 (abstract read)
  169. Ch. 18 — Assarsson E., Lundberg M., Holmquist G., et al. Homogenous 96-plex PEA immunoassay exhibiting high sensitivity, specificity, and excellent scalability. PLoS One 9, e95192 (2014). doi:10.1371/journal.pone.0095192 [authors include company staff]
  170. Ch. 18 — Todd J., Freese B., Lu A., Held D., Morey J., Livingston R., et al. Ultrasensitive flow-based immunoassays using single-molecule counting. Clin Chem 53, 1990–1995 (2007). doi:10.1373/clinchem.2007.091181 (abstract read) [authors include company staff]
  171. Ch. 19 — US Food and Drug Administration. The FDA warns that biotin may interfere with lab tests: FDA Safety Communication (28 Nov 2017; updated 5 Nov 2019) (as reproduced at drugs.com/fda-alerts/2926-0.html; the FDA pages return 404)
  172. Ch. 19 — Kricka L.J. Human anti-animal antibody interferences in immunological assays. Clin Chem 45, 942–956 (1999). doi:10.1093/clinchem/45.7.942 (abstract read) (secondary)
  173. Ch. 19 — Li D., Radulescu A., Shrestha R.T., Root M., Karger A.B., Killeen A.A., et al. Association of biotin ingestion with performance of hormone and nonhormone assays in healthy adults. JAMA 318, 1150–1160 (2017). doi:10.1001/jama.2017.13705 (as reproduced in PMC, PMC5818818)
  174. Ch. 19 — US Food and Drug Administration. Testing for Biotin Interference in In Vitro Diagnostic Devices. Guidance for industry (Oct 2020). fda.gov/media/127915/download
  175. Ch. 19 — von Meyer A., et al. Evaluating the performance of an updated high-sensitivity troponin T assay with increased tolerance to biotin. Clin Chem Lab Med (2021). doi:10.1515/cclm-2020-0104 (abstract read) [authors include company staff]
  176. Ch. 19 — Abbott. Alinity ci-series, product page (accessed 8 Oct 2026). corelaboratory.abbott/int/en/offerings/brands/alinity/alinity-ci-series.html [manufacturer]
  177. Ch. 19 — US Food and Drug Administration. Biotin interference with troponin lab tests: assays subject to biotin interference (web page, content current as of 21 Jun 2022). fda.gov/medical-devices/in-vitro-diagnostics/biotin-interference-troponin-lab-tests-assays-subject-biotin-interference
  178. Ch. 19 — Winder A.D., Suarez Mora A., Berry E., Lurain J.R. The “hook effect” causing a negative pregnancy test in a patient with an advanced molar pregnancy. Gynecol Oncol Rep 21, 34–36 (2017). doi:10.1016/j.gore.2017.06.008
  179. Ch. 19 — Laguë M., Turgeon P.Y., Thériault S., Steinberg C. A false-positive troponin assay leading to the misdiagnosis of myopericarditis. CMAJ 194, E456–E459 (2022). doi:10.1503/cmaj.211842
  180. Ch. 20 — Byrne R.A., Rossello X., Coughlan J.J., et al. 2023 ESC Guidelines for the management of acute coronary syndromes. Eur Heart J 44, 3720–3826 (2023). doi:10.1093/eurheartj/ehad191
  181. Ch. 20 — Thygesen K., Alpert J.S., Jaffe A.S., et al. Fourth Universal Definition of Myocardial Infarction (2018). Circulation 138, e618–e651 (2018). doi:10.1161/CIR.0000000000000617
  182. Ch. 20 — Reichlin T., Twerenbold R., Wildi K., et al. Prospective validation of a 1-hour algorithm to rule-out and rule-in acute myocardial infarction using a high-sensitivity cardiac troponin T assay. CMAJ 187, E243–E252 (2015). doi:10.1503/cmaj.141349 (abstract and declarations read) [study supported in part by Abbott, Beckman Coulter, BRAHMS, Roche and Siemens; authors declare company payments]
  183. Ch. 20 — Clerico A., Lippi G. The state-of-the-art of "high-sensitivity" immunoassay for measuring cardiac troponin I and T. J Lab Precis Med 3, 53 (2018) (editorial; summarises the IFCC task force criteria first proposed by Apple F.S., Collinson P.O., Clin Chem 58, 54–61 (2012), which was not opened) (secondary)
  184. Ch. 21 — US Food and Drug Administration. De Novo DEN200072, Lumipulse G β-Amyloid Ratio (1-42/1-40) (Fujirebio Diagnostics): database record and decision summary; granted 4 May 2022. accessdata.fda.gov/cdrh_docs/reviews/DEN200072.pdf
  185. Ch. 21 — Fujirebio. Lumipulse G pTau 217/β-Amyloid 1-42 Plasma Ratio, product page (accessed 8 Oct 2026). fujirebio.com [manufacturer]
  186. Ch. 21 — Palmqvist S., …, Tampi M.P. Clinical practice guideline for blood-based biomarkers in the diagnostic workup of Alzheimer’s disease within specialized care settings. Alzheimers Dement 21(7), e70535 (2025). doi:10.1002/alz.70535 (not opened; thresholds as reported in the Alzheimer’s Association AAIC release of 29 Jul 2025, its guideline page and a Barcelonaβeta summary) (secondary)
  187. Ch. 21 — Fujirebio. Fujirebio announces CE marking of the fully automated Lumipulse G pTau 217 Plasma assay. Press release, Ghent and Tokyo, 11 May 2026. fujirebio.com/en/news-events/fujirebio-announces-ce-marking-of-the-fully-automated-lumipulser-g-ptau-217-plasma [manufacturer]
  188. Ch. 22 — CAP TODAY. Chemistry and immunoassay analyzers, mid- and high-volume: product guide (July 2026; vendor-supplied data). captodayonline.com/2026/ProductGuides/CAPTODAY_ChemImmuno_MidHigh_ProductGuide_0726.pdf (secondary)
  189. Ch. 22 — Roche Holding AG. Roche reports strong 2025 results with 7% sales growth. Media release (ad hoc announcement), 29 Jan 2026. roche.com/media/releases/med-cor-2026-01-29 [manufacturer]
  190. Ch. 22 — US Food and Drug Administration. 510(k) K242870, Access hsTnI on the DxC 500i (Beckman Coulter): decision summary. accessdata.fda.gov/cdrh_docs/reviews/K242870.pdf
  191. Ch. 22 — Beckman Coulter. Beckman Coulter unveils next generation immunoassay analyzer enabling elite laboratory performance with no daily maintenance. Press release, 15 May 2023. news.beckmancoulter.com [manufacturer]
  192. Ch. 22 — Fujirebio. LUMIPULSE G1200, product page (accessed 8 Oct 2026). fujirebio.com/en/products-solutions/lumipulse-g1200 [manufacturer]
  193. Ch. 22 — Mindray. CL-2000i chemiluminescence immunoassay analyzer, India product page (accessed 8 Oct 2026). mindray.com/in/products/laboratory-diagnostics/chemiluminescence-immunoassay/medium-test-volume/cl-2000i/ [manufacturer]
  194. Ch. 22 — QuidelOrtho. VITROS ECiQ Immunodiagnostic System, product page (accessed 8 Oct 2026). quidelortho.com/global/en/products/vitros-systems/vitros-eciq-immunodiagnostic-system [manufacturer]
  195. Ch. 22 — Snibe. MAGLUMI X8 and MAGLUMI M Series, product pages (accessed 8 Oct 2026). snibe.com/en/product/CLIA_analyzer/40.html; snibe.com/en/product/CLIA_analyzer/635.html [manufacturer]
  196. Ch. 22 — bioMérieux. VIDAS KUBE, product page and brochure (accessed 8 Oct 2026). biomerieux.com/corp/en/our-offer/clinical-products/vidas-kube.html [manufacturer]
  197. Ch. 22 — Quanterix. Quanterix completes acquisition of Akoya Biosciences. Press release (Business Wire, 7 Jul 2025). businesswire.com/news/home/20250707828058/en/ [manufacturer]
  198. Ch. 22 — Quanterix. Quanterix releases financial results for the fourth quarter and full year 2025. Press release (via BioSpace, Mar 2026). biospace.com [manufacturer]
  199. Ch. 22 — Roche. Media release, 29 Dec 2023, on the agreement to acquire the parts of LumiraDx tied to its point-of-care platform. roche.com/media/releases/med-cor-2023-12-29 [manufacturer]
  200. Ch. 22 — Fujirebio. Fujirebio and Agappe enter into an agreement on business collaboration in the field of CLIA based immunoassay. News release, 19 Jan 2024. fujirebio.com/en/news-events [manufacturer]
  201. Ch. 22 — Agappe Diagnostics. Corporate brochure 2024. agappe.com/media/Brochure/default/Corporate_Brochure_2024_latest.pdf [manufacturer]
  202. Ch. 22 — J. Mitra & Co. Home page and CLIA analyzer page (accessed 8 Oct 2026). jmitra.co.in; jmitra.co.in/products/clia-analyzer/ [manufacturer]
  203. Ch. 22 — Transasia Bio-Medicals. Lisa XL automated ELISA processor, product page (accessed 8 Oct 2026). transasia.co.in/instrument/immunology/lisa-xl/95 [manufacturer]
  204. Ch. 22 — Roche. Investor update, 24 Aug 2026, on the FDA clearance of the Elecsys pTau217 plasma test. roche.com/investors/updates/inv-update-2026-08-24 [manufacturer]
  205. Ch. 22 — 360Dx. Siemens Healthineers outlines diagnostics carve-out; business weighs down fiscal Q2 (2026). 360dx.com/business-news/siemens-healthineers-outlines-diagnostics-carve-out-business-weighs-down-fiscal-q2-0 (secondary)
  206. Ch. 22 — Fujirebio. Fujirebio and Sysmex enter into basic agreement on business collaboration in the field of immunoassay. News release (undated, early 2024). fujirebio.com/en/news-events [manufacturer]
  207. Ch. 23 — European Commission. Proposal for a Regulation amending Regulation (EU) 2017/746 as regards transitional provisions for certain in vitro diagnostic medical devices and the deferred application of requirements for in-house devices. COM(2021) 627 final, 14 Oct 2021, explanatory memorandum. eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021PC0627
  208. Ch. 23 — European Commission. Notified Bodies survey on certifications and applications (MDR/IVDR), 20th survey, data to 28 Feb 2026 (presentation dated 2 Jul 2026). health.ec.europa.eu
  209. Ch. 23 — CLSI. Evaluation of Detection Capability Implementation Guide (EP17-Ed2-IG), sample pages (undated). shop.clsi.org/media/fjedgxkz/ep17ed2ige_sample.pdf
  210. Ch. 23 — CLSI. EP05-A3, Evaluation of Precision of Quantitative Measurement Procedures, 3rd ed. (1 Oct 2014; reaffirmed 2019). Catalogue page, clsi.org/shop/standards/ep05/
  211. Ch. 23 — CLSI. EP15-A3, User Verification of Precision and Estimation of Bias, 3rd ed. (11 Sep 2014; reaffirmed 2019). Catalogue page, clsi.org/shop/standards/ep15/
  212. Ch. 23 — CLSI. EP06, Evaluation of Linearity of Quantitative Measurement Procedures, 2nd ed. (24 Nov 2020). Catalogue page, clsi.org/shop/standards/ep06/
  213. Ch. 23 — CLSI. EP07, Interference Testing in Clinical Chemistry, 3rd ed. (30 Apr 2018; reaffirmed 2022), and EP37, Supplemental Tables for Interference Testing in Clinical Chemistry (1st ed. 2018; updated 3 Jul 2025). Catalogue pages, clsi.org/shop/standards/ep07/; clsi.org/shop/subscription-products/ep37/
  214. Ch. 23 — CLSI. EP09, Measurement Procedure Comparison and Bias Estimation Using Patient Samples, 3rd ed. (20 Jun 2018). Catalogue page, clsi.org/shop/standards/ep09/
  215. Ch. 23 — CLSI. EP28-A3c, Defining, Establishing, and Verifying Reference Intervals in the Clinical Laboratory (19 Oct 2010; reaffirmed 2020). Catalogue page, clsi.org/shop/standards/ep28/
  216. Ch. 23 — Analyse-it Software Ltd. Choosing a reference interval method (web page, undated). analyse-it.com/learn/choosing-a-reference-interval-method (secondary)
  217. Ch. 23 — LaValley C., Graff J.P. Reference intervals: the process to verify outside sources. UC Davis Health, Lab Best Practice blog (Apr 2022). health.ucdavis.edu/blog/lab-best-practice (secondary)
  218. Ch. 23 — US Food and Drug Administration. Recognized Consensus Standards database: CLSI EP and I/LA documents (accessed 8 Oct 2026). accessdata.fda.gov/scripts/cdrh/cfdocs/cfStandards/search.cfm
  219. Ch. 23 — Medical Device Coordination Group. MDCG 2022-2, Guidance on general principles of clinical evidence for in vitro diagnostic medical devices (Jan 2022). health.ec.europa.eu/system/files/2022-01/mdcg_2022-2_en.pdf
  220. Ch. 23 — ISO. ISO 20916:2019, In vitro diagnostic medical devices: clinical performance studies using specimens from human subjects: good study practice (published 4 Jun 2019; confirmed 2025), catalogue page. iso.org/standard/69455.html
  221. Ch. 24 — US Food and Drug Administration. Medical Device Recall Z-1671-2018, Dimension Vista CTNI Flex reagent cartridge (Siemens Healthcare Diagnostics; 510(k) K063756), Class 2; initiated 2 Jan 2018. accessdata.fda.gov/scripts/cdrh/cfdocs/cfRes/res.cfm?id=162500
  222. Ch. 24 — US Food and Drug Administration. Replacement Reagent and Instrument Family Policy for In Vitro Diagnostic Devices. Guidance (17 Aug 2022). fda.gov/media/111186/download
  223. Ch. 24 — US Code of Federal Regulations (eCFR). 21 CFR 862.1215 (creatine phosphokinase/creatine kinase or isoenzymes test system) and 21 CFR 862.9 (limitations of exemptions) (accessed 8 Oct 2026). ecfr.gov
  224. Ch. 24 — Medical Device Coordination Group. MDCG 2020-16 rev.5, Guidance on classification rules for in vitro diagnostic medical devices under Regulation (EU) 2017/746 (9 Sep 2026). health.ec.europa.eu
  225. Ch. 24 — Central Drugs Standard Control Organisation, India. Classification list of IVD medical devices (updated 25 Oct 2023; file IVD/Misc/196/2020). cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/medical-device/IVD_classifiction25oc23.pdf
  226. Ch. 24 — US Food and Drug Administration. Assay Migration Studies for In Vitro Diagnostic Devices. Guidance (25 Apr 2013). fda.gov/media/73669/download
  227. Ch. 24 — US Food and Drug Administration. Deciding When to Submit a 510(k) for a Change to an Existing Device. Guidance (25 Oct 2017), with its companion guidance on software changes. fda.gov/media/99812/download
  228. Ch. 24 — Medical Device Coordination Group. MDCG 2022-6, Guidance on significant changes regarding the transitional provision under Article 110(3) of the IVDR (May 2022). health.ec.europa.eu/system/files/2022-05/mdcg_2022-6.pdf
  229. Ch. 24 — European Commission. Proposal to amend Regulations (EU) 2017/745 and 2017/746 (simplification), COM(2025) 1023 final, 2025/0404(COD), 16 Dec 2025, as transmitted in Council document ST 16919/25 (18 Dec 2025). data.consilium.europa.eu/doc/document/ST-16919-2025-INIT/en/pdf
  230. Ch. 24 — Austrian Parliament, EU database. Council progress report on the MDR/IVDR simplification proposal, RAT 9114/26 (27 May 2026). parlament.gv.at/gegenstand/XXVIII/EU/74474
  231. Ch. 24 — Arnold & Porter. How will the EU Parliament shape the MDR/IVDR revision? (via Mondaq, 7 Aug 2026). mondaq.com (secondary)
  232. Ch. 24 — Central Drugs Standard Control Organisation, India. IVD Medical Devices FAQ, Doc No. CDSCO/IVD/FAQ/04/2022, with addendum of 28 Mar 2025. cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/FAaddendum.pdf
  233. Ch. 24 — Central Drugs Standard Control Organisation, India. Addendum No. 02 to Doc No. CDSCO/IVD/FAQ/04/2022 (13 Mar 2026). cdsco.gov.in
  234. Ch. 25 — IEC. IEC 61010-2-101:2018, ed. 3.0 (5 Oct 2018), particular safety requirements for in vitro diagnostic medical equipment. IEC Webstore page, webstore.iec.ch/en/publication/60197
  235. Ch. 25 — IEC. IEC 60601-1:2005+AMD1:2012+AMD2:2020 CSV, consolidated ed. 3.2 (20 Aug 2020). IEC Webstore page, webstore.iec.ch/en/publication/67497
  236. Ch. 25 — US Food and Drug Administration. Overview of IVD regulation (web page, accessed 8 Oct 2026). fda.gov/medical-devices/ivd-regulatory-assistance/overview-ivd-regulation
  237. Ch. 25 — ISO. ISO 14971:2019, application of risk management to medical devices, 3rd ed. (10 Dec 2019; confirmed 2025), catalogue page. iso.org/standard/72704.html
  238. Ch. 25 — IEC. IEC 61010-1:2010/AMD1:2016 (published 16 Dec 2016). IEC Webstore page, webstore.iec.ch/en/publication/33314
  239. Ch. 25 — SIQ Ljubljana. IEC 61010-1 Amendment 2: upcoming changes to safety requirements (3 Apr 2025). siq.si/en/news (secondary)
  240. Ch. 25 — IEC. IEC 61010-2-081:2015, ed. 2.0 (23 Jan 2015; superseded by a 2019 edition). IEC Webstore page, webstore.iec.ch/en/publication/22287
  241. Ch. 25 — IEC. IEC 62304:2006+AMD1:2015 CSV, ed. 1.1 (26 Jun 2015). IEC Webstore page, webstore.iec.ch/en/publication/22794
  242. Ch. 25 — VDE. IEC 62304 Edition 2: Änderungen für Hersteller (22 Jul 2026). vde.com/iec-62304-edition-2-aenderungen (secondary)
  243. Ch. 25 — Critical Software. IEC 62304 Edition 2 changes (newsroom item, Aug 2026). asd.criticalsoftware.com/en/newsroom/iec-62304-edition-2-changes-august-2026 (secondary)
  244. Ch. 25 — ISO. ISO 18113-1, ISO 18113-3 and ISO 18113-5:2022, information supplied by the manufacturer for IVD medical devices, 2nd eds. (6 Oct 2022), catalogue pages. iso.org/standard/79866.html; iso.org/standard/79868.html; iso.org/standard/79870.html
  245. Ch. 25 — ISO. ISO 15223-1:2021, symbols to be used with information to be supplied by the manufacturer, 4th ed. (6 Jul 2021), with Amendment 1:2025, catalogue page. iso.org/standard/77326.html
  246. Ch. 25 — ISO. ISO 13485:2016, medical devices: quality management systems, 3rd ed. (25 Feb 2016; confirmed 31 Oct 2025), catalogue page. iso.org/standard/59752.html
  247. Ch. 25 — US Code of Federal Regulations (eCFR). 21 CFR Part 820, Quality Management System Regulation (final rule 89 FR 7523, 2 Feb 2024; effective 2 Feb 2026); and US Food and Drug Administration, Quality Management System Regulation (QMSR) web page. ecfr.gov; fda.gov
  248. Ch. 25 — Commission Implementing Decision (EU) 2021/1195 on harmonised standards for in vitro diagnostic medical devices, consolidated text to 9 Oct 2024. eur-lex.europa.eu/eli/dec_impl/2021/1195/2024-10-09/eng/pdf
  249. Ch. 25 — European Commission. Harmonised standards: in vitro diagnostic medical devices (web page), with Implementing Decisions (EU) 2026/197 and 2026/1313 (landing pages and titles read). single-market-economy.ec.europa.eu
  250. Ch. 26 — Regulation (EU) 2017/746 of the European Parliament and of the Council of 5 April 2017 on in vitro diagnostic medical devices. OJ L 117, 176–332 (5 May 2017). eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32017R0746 (read to Article 24)
  251. Ch. 26 — Regulation (EU) 2024/1860 of 13 June 2024, amending the transitional provisions of Regulation (EU) 2017/746. OJ L (9 Jul 2024). eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32024R1860
  252. Ch. 26 — US Food and Drug Administration. CLIA categorizations (web page, accessed 8 Oct 2026). fda.gov/medical-devices/ivd-regulatory-assistance/clia-categorizations
  253. Ch. 26 — US Food and Drug Administration. Medical Devices; Laboratory Developed Tests; Implementation of Vacatur. Final rule. Federal Register 90, 45134 (19 Sep 2025), FR Doc. 2025-18217. govinfo.gov/content/pkg/FR-2025-09-19/pdf/2025-18217.pdf
  254. Ch. 26 — Ministry of Health and Family Welfare, India. Medical Devices Rules, 2017, G.S.R. 78(E) (31 Jan 2017). main.mohfw.gov.in/sites/default/files/Medical%20Device%20Rules%2C%202017.pdf
  255. Ch. 26 — Central Drugs Standard Control Organisation, India. IVD FAQ, Doc No. CDSCO/IVD/FAQ/03/2022. cdsco.gov.in/opencms/export/sites/CDSCO_WEB/Pdf-documents/IVD/FAQs/CDSCO-IVD-FAQ-03-2022-.pdf
  256. Ch. 26 — Central Drugs Standard Control Organisation, India. In Vitro Diagnostics web page and notices, 2025–2026 (accessed 8 Oct 2026). cdsco.gov.in/opencms/opencms/en/Medical-Device-Diagnostics/InVitro-Diagnostics/
  257. Ch. 26 — World Health Organization. Prequalification of health products, fact sheet (11 Jun 2026). who.int/news-room/fact-sheets/detail/prequalification-of-medicines-by-who
  258. Ch. 26 — World Health Organization. IVD prequalification web pages: IVD prequalification assessment; IVD performance evaluation; IVDs eligible for WHO prequalification (tables updated 27 Sep 2026). extranet.who.int/prequal/vitro-diagnostics
  259. Ch. 26 — World Health Organization. List of prequalified in vitro diagnostic products (accessed 8 Oct 2026). extranet.who.int/prequal/vitro-diagnostics/prequalified/in-vitro-diagnostics
§ PDF edition

Take the PDF with you.

Reading online needs no sign-up. For the PDF edition, laid out for print and offline reading, tell us where to send it and we will email you.

We use your details to send the guide, and to tell you about new guides only if you tick the box. See the privacy policy.

Created with Claude (Anthropic).
Educational reference material, not engineering, regulatory or clinical advice. Product and company names referenced are the property of their respective owners.