What only a living system can answer.
+ The questionWhen is a biological test still needed, where are animal tests being replaced, and where do they remain?
What the arithmetic leaves open
An unnamed compound in a cushion's extract can fall below every threshold of Chapter 16 and still cause an allergy, because those thresholds do not cover allergy. Chapters 12 to 17 turned each compound in an extract into a daily dose and weighed it against a tolerable intake or a threshold. That comparison answers harm that follows the dose through the body, and it leaves other effects open: ISO/TS 21726 sets no threshold for sensitization, irritation, harm to cells or effects on blood (Chapter 16).
The extract sets a second limit. A compound is weighed only if the extraction released it. The body also meets the protein coat on the material's surface, and an extract holds nothing of that coat (Chapter 8). An extract may also hold compounds that no one has named (Chapter 13). The doses of Chapter 15 can be added up only for compounds whose effects are known. Cells exposed to the extract respond to everything in it at once, named or not, so a test on cells reads the mixture as a whole.
Tests have limits of their own. Williams, a biomaterials scientist, argued in a 2019 review that most tests of the ISO 10993 series read the effect of a material's extracts on cells in culture. If a component is not extracted during the 72-hour extraction he cites, its effects "will not be detected" by those tests. He wrote that the response to debris from metal-on-metal hips "could not have been picked up by the standard biological safety tests". Wear makes that debris in use, so an extract of the new material holds none of it, and a test on the extract misses whatever the extract does not hold.
Cells read the whole extract
Cytotoxicity is harm to cells grown in a dish. ISO 10993-5, the device standard for it, lists tests on an extract, in direct contact and in indirect contact, and leaves the choice among them open. Its third edition, of 2009, is still the current one. Its methods read damage seen in the cells' shape, measured damage, the cells' growth or an aspect of their metabolism. The extract test is also called elution. Eurofins, a contract laboratory, runs it by extracting the material for 24 hours at 37 °C in culture medium with 10 % serum and placing the extract for 48 hours on a layer of L929 mouse connective-tissue cells. High-density polyethylene runs beside it as the negative control and two zinc compounds as the positive controls, so each run shows the cells ignoring an inert plastic and responding to a known toxic compound.
The test is read in two ways. A qualitative reading grades the cell layer by the share of cells rounded or destroyed, from 0, none, through slight, mild and moderate to 4, severe. Eurofins and Nelson Labs, another contract laboratory, pass a material graded no higher than 2. A quantitative reading measures the cells' metabolism, for example by a color reaction, and reports viability, the share of living cells against untreated ones. A material counts as cytotoxic when its extract lowers viability by more than 30 %, to below 70 % of the untreated value. Gruber and Nickel, of the Johner Institute, a regulatory consultancy, state that criterion in a 2023 paper, and Nelson Labs passes a material at 70 % or more.
The test responds to a wide range of substances, from monomers to metal ions. That sensitivity makes it "ideal as a screening test", in the words of the Johner Institute's blog, and Nelson Labs adds the test's low cost and short turnaround. In a 2021 article on failing cytotoxicity scores, Rääget and colleagues, all Nelson staff, reported that the test is still used routinely in initial screening.
A failure names no cause. The Johner Institute's blog states that the results do not show what caused the effect and that a failed test does not mean the device is not biocompatible. Rääget and colleagues recommend a toxicological risk assessment to judge whether a failure matters clinically, which sends the evaluation back to the chemistry of Chapters 12 to 15. A pass has limits too. The test stops responding when a compound's concentration is too low, even where that compound still matters toxicologically, so a passed test "is not a guarantee of the absence of critical residues", as the same blog warns.
The way a laboratory runs the test also changes the result. Gruber and Nickel's 2023 paper reports a comparison they ran among laboratories in 2020 and 2021, and the study had no external funding. The organizers sent every laboratory the same two tubes, to be tested by elution at 6 cm² of surface per mL of medium, 37 °C and 24 hours. Of 56 laboratories enrolled, 52 in 18 countries returned results.
- The tubes. Polyethylene tubing was expected to pass and PVC tubing to fail; before dispatch the organizers' own tests measured a fall in viability of about 89 % for the PVC and about 12 % for the polyethylene.
- The PVC. 32 of the 52 laboratories called it cytotoxic: 32 ÷ 52 = 0.62, or 62 %. The other 20, 38 %, passed it, more than one laboratory in three.
- The polyethylene. 48 of 52 passed it, 92 %; 4, or 8 %, failed it.
- Both. 30 of 52, 58 %, classified both tubes as expected, so 22, 42 %, misclassified at least one.
- Where they split. Readings of the PVC ran from 0 to 100 % reduction, with a mean of 43 %, near the 30 % line. Serum in the extraction medium and longer exposure of the cells raised the share that detected it.
- The result. The verdict on a tube that the organizers had found harmful to cells depended on the laboratory, and the authors conclude that the standard's specifications are "not explicit enough to obtain comparable results for an identical medical device".
- Conditions. Both tubes were sterilized with ethylene oxide and pooled before dispatch; two laboratories tested by direct contact instead.
Skin and DNA without animals
The irritation of Chapter 11 can now be read on reconstructed human epidermis, a tissue model of the skin's outer layer built from human cells. ISO 10993-23, the device standard for irritation, has included it among its test procedures since its first edition in 2021, and its 2025 amendment adds further models. In a 2018 round robin among laboratories, De Jong and colleagues exposed two commercial models to saline or sesame-oil extracts of four polymers, prepared as irritant or non-irritant, beside three non-irritant controls. The models received 100 µL of extract for 18 or 24 hours, and a fall in viability of more than 50 % marked an irritant. Both models identified "virtually all" irritant samples in both liquids, the authors report. The study's 37 authors included staff of both models' makers, of contract laboratories, of device makers and of FDA.
In a 2020 follow-up, De Jong and colleagues compared the models with rabbit tests on polymers, device components, chemicals and extracts. The authors included staff of contract laboratories, a device maker and China's device regulator. Their abstract calls the results essentially equivalent but gives no rate of agreement, and it concludes that the models are acceptable replacements for assessing device irritation.
Sensitization is harder to replace, because its two phases pass through several kinds of cell (Chapter 11). Methods in OECD Test Guidelines 442C, 442D and 442E each read one step of that pathway, and they are validated for chemicals within defined limits. GARDskin is one of them: it models the immune cells that carry the chemical with a human cell line, and it joined Guideline 442E in 2022. ISO/TS 11796, published in 2023, sets out how to show that such a method applies to devices, ending in an interlaboratory study.
In a 2024 bench study, Jenvert and colleagues applied GARDskin to extracts of 11 materials: silicone and thermoplastic polyurethane, each spiked with 10 % by weight of one of four sensitizers, unspiked controls and commercial tubes. Each material was extracted for 72 hours at 37 °C in saline, olive oil and sesame oil. GARDskin classified all 11 correctly, though one spiked silicone gave a weaker signal in olive oil. The extraction liquid therefore matters for a silicone such as the running device's cushion. SenzaGen, the method's developer, funded the study, and four of the six authors worked there. In the same paper, Jenvert and colleagues wrote that the methods without animals had not yet been sufficiently qualified for devices.
Genotoxicity is read by a battery, a set of tests run together because "no single test can detect all types of genotoxins", in the words of Measurlabs, a testing laboratory. Under ISO 10993-3:2026 the battery has at least two tests without animals, the bacterial reverse mutation test and one test in mammalian cells for damage to genes or chromosomes, as the laboratory's summary of 10 November 2026 describes it. The same summary reports that the edition lets chemical characterization and a toxicological risk assessment make testing unnecessary, drops animal tests from the follow-up of a positive result and keeps them only for highly specific cases, such as particles that accumulate in organs.
Where animals remain, and why fewer
The 3Rs are replacement of an animal test by a method without animals, reduction in the number of animals, and refinement of a test so that it causes less pain and distress. ISO 10993-11:2017, on systemic toxicity, states them in asking that "all reasonably and practically available replacement, reduction and refinement alternatives should be identified". Reconstructed skin in place of rabbits is a replacement. The local lymph node assay of Chapter 11, which spares the animals a challenge and an adjuvant, is a refinement, and its reduced version, with up to 40 % fewer animals, is a reduction. ISO 10993-2:2022, the third edition of the standard on animal welfare, aims at "maximizing the use of scientifically sound non-animal tests". One of its clauses is headed "Justification for animal tests".
Animal tests remain where no method without animals answers the question. In 2024 device sensitization was still tested mainly in guinea pigs, by the maximization or the Buehler test, Jenvert and colleagues wrote. Systemic toxicity that the chemistry of Chapters 14 and 15 cannot settle still needs a whole animal (Chapter 11). For acute studies ISO 10993-11 prefers the mouse or rat, with at least five rodents in a group. The implantation tests of Chapter 9 likewise place material in living tissue.
FDA's 2023 guidance on ISO 10993-1 aims "to replace, reduce, and/or refine animal use in testing, when feasible", as FDA's webinar on the guidance puts it. FDA's April 2025 plan to phase out animal testing covers drugs and biologics, and its roadmap does not mention devices. On 24 March 2026 the law firm Morgan Lewis wrote that FDA "has yet to issue a guidance" on non-animal methods for devices, pointing to a shortage of methods validated or qualified for device biocompatibility. For devices that touch only intact skin, Attachment G of the 2023 guidance already lets a maker supply information on listed materials, which include synthetic polymers and natural fabrics, "in lieu of biocompatibility testing or detailed rationales".
The running device's plan
The 2025 edition of ISO 10993-1 still sets out the endpoints of Chapter 2 in tables, calling them biological effects, and it lets the biological risk estimate decide which of them need new data (Chapter 21). Chapter 22 follows that estimate for the cushion, from its contact to what the chemistry left open, and the plan takes the tests the estimate calls for. Cytotoxicity stays as the screen of the whole extract. Sensitization still needs an animal test, for the reason Jenvert and colleagues gave in 2024. The plan takes the refinement, the local lymph node assay, over the guinea pig maximization and Buehler tests then mainly used. That choice is this book's own, and its justification must show that the assay suits the cushion's extracts under ISO 10993-10 (Chapter 11).
| Part and contact | Biological effect | How the plan answers it |
|---|---|---|
| Cushion, silicone; intact skin, long-term | cytotoxicity | elution test on the extract, as the screen |
| irritation | reconstructed human epidermis, saline and sesame-oil extracts | |
| sensitization | local lymph node assay, justified against ISO 10993-2; methods without animals not yet sufficiently qualified for devices in 2024 | |
| genotoxicity | chemistry first; the two-test battery only for a compound with a structural alert, or an unnamed peak, above the value Chapter 16 sets for compounds with an alert (Chapters 16 and 17) | |
| Headgear, foam and fabric; intact skin, long-term | cytotoxicity, irritation, sensitization | material information under Attachment G if its materials are listed, otherwise as the cushion |
| Air path and water chamber | particles, volatile compounds, condensate | ISO 18562 (Chapter 19) |
Write the biological evaluation plan after the chemistry. Ask a cytotoxicity laboratory to report the extraction medium and its serum, the ratio of surface to liquid, the cell line, the exposure time and the controls, and treat a failure as a question for the chemistry. Justify each animal test against ISO 10993-2 in the plan.
- Chemistry weighs compounds by dose. Tests answer the effects that the thresholds of toxicological concern do not cover, such as sensitization and irritation, and they read the extract as a whole mixture.
- In the cytotoxicity test a material is cytotoxic when the viability of cultured cells falls below 70 %, as a 2023 paper and a contract laboratory state the criterion. The test is a sensitive screen whose failure names no cause, and laboratories can disagree on the same tube.
- Reconstructed human epidermis now reads irritation, and the genotoxicity battery starts in cells, but in 2024 device sensitization was still tested mainly in guinea pigs.
- The 2025 edition of ISO 10993-1 keeps its tables of biological effects and lets the risk estimate decide which need data, so the running device's plan takes chemistry first, then methods without animals, then a justified animal test.
Let's Talk