Reading a biological evaluation.
+ The questionHow does one read an extractables table and a test report?
The conditions and the line come before the compounds
An extractables report runs to dozens of pages, and what decides it is often one row of a table. Every number in that table was made by the conditions set out on the pages before it: the solvents, heat and time of Chapter 12, the instruments of Chapter 13 and the reporting line of Chapter 17. A reader therefore reads a report in the order it was produced, the conditions first, then the line, then the compounds, and follows each compound to the patient. For the running device's cushion, the deciding row is the one peak that no library could name (Chapter 13).
The conditions decide whether the table is an upper estimate of what use releases. An extract's amounts are such an estimate for the substances present before use (Chapter 4), but only if the extraction went past use in the way the contact requires. The reader checks four conditions. The solvents should have the polarities that FDA's 2024 draft guidance on chemical analysis asks for, and the intensity should suit the contact. The temperature and cycle length should be stated, and an exhaustive series should end on a cycle that yielded less than 10 % of what the first cycle yielded (Chapter 12). The sample is checked on the same page. FDA's draft calls it the test article and defines it as the device in its final, finished form, with its sterilization and packaging where they apply. Chapter 12 adds aging where the material degrades in use. A report on any other sample has to say how that sample stands in for the device.
The reporting line comes next, because it decides which peaks the table lists at all. The line is the analytical evaluation threshold, or AET, and it is built from the six inputs that Chapter 17 sets out, from the daily threshold to the uncertainty factor. Each input belongs beside the result. The uncertainty factor belongs to the analytical method, so a report with one extract analyzed by GC-MS and another by LC-MS needs a line for each method. A line printed without its inputs cannot be checked.
Each row of the compound table answers four questions. The first is what was found, with its name and registry number where the identification allows them. The second is in which extract and by which technique it was found. The third is how far the identification goes, on the scale of FDA's draft from confirmed to unknown (Chapter 13). The fourth is how much came out per device, and whether that amount was counted against the compound's own standard or against an internal standard, which may miscount it many times over. A peak above the line stays in the table whether or not it has a name, and Chapter 16's thresholds decide what its amount is held to.
One row, followed to the patient
An evaluation carries each row of the table into the toxicology, and a reader can follow it there link by link. The D4 row of the cushion's illustrative report has four links. The first, the amount, breaks the rule set out above, because the report states no amount of D4 per cushion. The evaluation borrows one by analogy: about 269 µg per gram, the top of the range that Zhang and colleagues at FDA reported in 2012. Their measurements covered three cyclic siloxanes, D4 among them, in silicone nipples made for food contact. The tolerable intake, about 5,970 µg per kilogram per day, is this book's, derived in Chapter 15 from a draft EPA value for covered skin. The exposure assumes that the D4 in one gram of silicone leaves in one night and reaches the lightest user, taken at 30 kg (Chapter 15).
Reading the row means checking each link for the assumption it carries. The amount is a measurement in another product. The tolerable intake rests on an EPA draft, which EPA's science advisory committee reviewed in December 2025 and EPA has not made final. The body mass is the lightest user's, and the release period is the shortest, a single night. The margin is about 670 for each gram of silicone, and it is wide: the cushion would have to release the D4 of about 670 g of silicone in one night to bring it to 1. A reader who sees only the number, though, cannot tell whether the margin is wide because the compound does little harm at that dose or because a link was generous. The evaluation therefore states each link with its source and its status in the row that uses it, as Chapter 15 asks of the exposure table.
The cushion's other two rows have no tolerable intake to follow, so they are read against thresholds. The tentative compound takes the value for Cramer class III, 90 µg a day at 60 kg or 45 µg at 30 kg. It takes that value only if every candidate structure falls in that class, is free of structural alerts and lies outside the groups the threshold excludes (Chapter 16). The unknown peak has no structure that could clear it. Nelson Labs, a contract testing laboratory, treats such a peak as a possible carcinogen or mutagen until a structure clears it. For long-term contact that practice holds the peak to 1.5 µg a day, the dose that FDA's draft takes as conservatively protective (Chapter 16). FDA's draft itself sends the peak back to the analysis for a revised or added method. The peak's amount comes from the internal standard, so the row with the least information carries the least certain number, and that row decides what the cushion's chemistry can conclude.
A test report is read from its controls
A biological test reports a response, and a response means something only if the run could show one and could show its absence. The controls supply both. The cytotoxicity test of Chapter 18 runs untreated cells, a negative control and positive controls beside the extract. The reader checks first that the negative control left the cells unharmed and that the positive controls damaged them. Only then is the test article's grade or viability read against Chapter 18's criteria. Nelson Labs states that it usually tests several dilutions of the extract, so each grade is read with the dilution that gave it. A pass says that the extract as a whole did not harm the cells under the test's conditions. A failure names no cause (Chapter 18).
In practice a device's sensitization report is often a guinea pig maximization test report. Jenvert and colleagues wrote in 2024 that device sensitization was still tested mainly in guinea pigs (Chapter 18). Their paper was a bench study of a method without animals. The method's developer, SenzaGen, funded it, and four of the six authors worked for SenzaGen. Chapter 11 set out the test and its two phases under OECD Test Guideline 406. The report adds the data a reader checks: a score for every treated and control animal at each of the test's two readings after the challenge, on the Magnusson and Kligman scale. Its grades run from 0, no visible change, through 1, discrete or patchy redness, and 2, moderate and confluent redness, to 3, intense redness and swelling.
The laboratory's reliability check shows whether the test could have found a sensitizer. The guideline asks the laboratory to assess its technique with a substance known to be a mild-to-moderate sensitizer, such as hexyl cinnamic aldehyde, mercaptobenzothiazole or benzocaine, and to run that check no more than six months before the test. The maximization test is an adjuvant test, and in a properly conducted adjuvant test at least 30 % of the animals given the known sensitizer should respond. In a test without adjuvant, at least 15 % should. A report that shows no such check within those six months gives no evidence that the animals and the laboratory would have detected a sensitizer. The guideline sets no percentage that defines a sensitizer. The conclusion therefore rests on how the treated animals' scores compare with the controls' under ISO 10993-10, the device standard for sensitization, whose criterion for that comparison lies in the standard's own text.
The annotated report below belongs to a generic device, not to the cushion, which took the local lymph node assay (Chapter 18).
What FDA finds missing, and what to ask
At a town hall on 23 September 2026, FDA staff listed the most common deficiencies they see in extraction, in slides that are a presentation rather than guidance. A reader of a report can find three of them: inadequate or unclear extraction conditions; manipulation of extracts, such as pooling, concentration or dilution; and inadequate or unclear values and calculations, the AET and the uncertainty factor among them. Each takes away a reader's means of recomputing a number. Unclear conditions hide whether the table is an upper estimate of use, and pooling changes how many devices stand behind one extract. Dilution lowers the line (Chapter 17), so an unrecorded dilution leaves the printed line too high by the dilution factor.
The table below sets these three beside the four flags raised above, from a sample never aged where it degrades in use to a failed control. In the recall of Chapter 20 the polyester foam's degradation was found in use first.
Each flag becomes a question for one of the two parties to a report, the laboratory or the evaluator. The laboratory answers for the method: the conditions and the sample, the evidence that each exhaustive series stopped, each method's line with its inputs, and the controls' results. It also answers for the evidence that Chapter 13 asks for behind each name and each amount. The evaluator answers for the judgment: which tolerable intake or threshold each row takes and its status, the exposure assumptions, why each unknown is acceptable, and how chemistry and tests together address each biological effect the plan named. A question with no answer in the file is a broken link of the kind that Chapter 22's traceability exposes.
The running device's report, read
The excerpt below is this book's construction for the running device's cushion, labeled illustrative: no maker's record supports it, and every value comes from Chapters 12 to 18 or from the sources they cite. Its sensitization entry reports the local lymph node assay that this book's plan chose (Chapter 18). Besides the D4 row's missing amount per cushion, set out above, the excerpt carries one fault for the reader to find: it prints Chapter 17's GC-MS line for every method, although Chapter 17 also computed a line for LC-MS.
| Entry in the report | What it states | How it reads |
|---|---|---|
| Extraction | water, isopropanol and hexane; exhaustive, 72-hour cycles at 50 °C; three cushions per solvent, as made and as aged and cleaned | the plan of Chapter 12 for long-term contact |
| Reporting line | 0.0075 µg/mL, from 1.5 µg a day, one cushion in 50 mL, one cushion a night and a factor of 4 | correct for GC-MS; the LC-MS extract needs its own line (solved example) |
| D4, CAS 556-67-2; water extract, headspace GC-MS | confirmed; counted against a D4 standard; amount per cushion not reported | up to about 269 µg per gram, borrowed by analogy from food-contact silicone; margin about 670 per gram, on a draft's values (Chapters 14 and 15) |
| Peak; hexane extract, GC-MS | tentative, two candidate structures; counted by the internal standard | 90 µg a day, or 45 at 30 kg, only if both candidates meet Chapter 16's conditions |
| Peak; isopropanol extract, LC-MS | unknown; counted by the internal standard | the deciding row: 1.5 µg a day as a possible mutagen, by Nelson Labs' practice, or back to the analysis by FDA's draft |
| Cytotoxicity, elution | not above grade 2; controls as expected | passes as a screen of the whole extract, naming no cause (Chapter 18) |
| Sensitization, local lymph node assay | stimulation index below 3 at each concentration; vehicle and positive controls as expected | negative under a criterion written for chemicals; reads induction only (Chapter 11) |
The evaluator can carry three entries of the excerpt to a conclusion: cytotoxicity, sensitization and D4, the last only on a draft's values and a borrowed amount. The local lymph node assay reports a stimulation index, the multiplication of lymphocytes in treated mice divided by that in mice given the vehicle alone. The cushion's index stays below 3, the value at which EU test method B.42 counts a chemical as a sensitizer. That criterion was written for chemicals, not for devices, so the cushion's justification under ISO 10993-10 has to show that the assay and its criterion suit the cushion's extracts (Chapter 18). Three questions stay open. The tentative peak stays open until both candidates are shown to meet Chapter 16's conditions. In the isopropanol extract the LC-MS line is missing, and the unknown peak needs a name from the laboratory or a justified threshold from the evaluator.
- The inputs. The report states 1.5 µg a day for long-term contact, one cushion extracted in 50 mL, one cushion a night and no dilution (Chapter 17).
- The GC-MS line. 1.5 divided by 50 and by 1 is 0.03 µg/mL. Divided by 4, the lowest default factor that FDA's draft allows for GC methods without further justification, it gives 0.0075 µg/mL, the line the report prints.
- The LC-MS line. FDA's draft sets no default factor for LC-MS (Chapter 17). Nelson Labs gave a factor of 10 for its LC-MS list in its own slides at FDA meetings in late 2024, and 0.03 divided by 10 is 0.003 µg/mL.
- What the LC-MS line allows. A compound sitting on it delivers 0.003 µg/mL times 50 mL, or 0.15 µg a day, a tenth of the threshold, so a compound the method under-reports by up to ten times still stays within it.
- The result. The printed line holds for the GC-MS extracts. For the isopropanol extract, where the deciding peak appeared, it sits 0.0075 divided by 0.003, or 2.5 times, too high, and a peak between the two lines goes unlisted.
- Assumptions. The report is this book's illustration. The factor of 4 is a draft's default and the factor of 10 is Nelson Labs' figure for its own database; neither was measured on the cushion.
Before accepting a report as verification evidence (Chapter 22), check its conditions and sample against the extraction plan (Chapter 12), recompute each method's line from its printed inputs, and take each row above the line to its tolerable intake or threshold, with the status of every value. Read each test's controls before its result. Send the laboratory and the evaluator a list of every number that cannot be recomputed from the file.
- A chemical characterization report is read in the order it was produced: conditions and sample, then each method's reporting line with its inputs, then each compound's identification level and how its amount was counted.
- Each compound row is followed through its amount, its tolerable intake or threshold, the exposure and the margin, and every link carries an assumption and a status that the evaluation must state.
- A test report is read from its controls first. A device's sensitization report is often a guinea pig maximization test, and there the laboratory's check of its technique with a known sensitizer, run no more than six months before the test, shows whether the test could have found a sensitizer.
- FDA staff listed unclear extraction conditions, manipulated extracts and unclear AET and uncertainty-factor calculations among the most common deficiencies at a town hall in September 2026. Each flag becomes a question, to the laboratory on the method or to the evaluator on the judgment. In the running device's illustrative report, built with one reporting line for two methods, the tentative and unknown peaks remain open.
Let's Talk