The evaluation in the design file.
+ The questionHow does a biological evaluation become a design input and a piece of evidence?
The evaluation starts with the first material
A biological evaluation begins when the first material for a device is chosen, long before any test is run, and it ends in a report. Chapter 21 ended on an estimate of risk, and estimating risk is one step of risk management, which runs through the whole life of a design. The 2025 edition of ISO 10993-1 places the evaluation "within the ISO 14971 risk management framework", in the heading of its clause 4.1; ISO 14971 is the risk management standard that Chapter 1 introduced. FDA said the same in its 2023 guidance on the standard, written before that edition: the evaluation "should be conducted within the framework of a risk management process".
A design moves around a loop. Requirements come first, and the design answers them. Verification checks the design against its requirements, validation checks the finished device against the needs of its users, production makes it, and use in the field returns complaints and reports that become new requirements. The biological evaluation runs on that loop. It takes in the materials being chosen, the contacts that Chapter 2 sorted and the useful life, and it gives out a plan during design, a report at verification, and specifications and instructions for production and the user. In the United States, design controls were required first under the Quality System Regulation and, since 2 February 2026, are required under the Quality Management System Regulation, which incorporates ISO 13485:2016 (Chapter 26).
From a released substance to harm
Chapter 1 gave ISO 14971's words for hazard, harm and risk. The current edition of the standard dates from 2019, and ISO reviewed and confirmed it in 2025. The standard links hazard to harm in a chain, which FDA's training slides on risk reproduce: a hazard, a sequence of events, a hazardous situation and harm. A hazardous situation is one of the "circumstances in which people, property or environment is/are exposed to one or more hazards", and the sequence of events carries a person from the hazard into that exposure. The chain carries two probabilities. P1 is the probability that the hazardous situation occurs and P2 the probability that it leads to harm, and the slides multiply the two into the probability of occurrence of harm. Risk combines that probability with the severity of the harm.
The slides illustrate the chain with one example, driving in heavy traffic, and none with a chemical. The reading that follows is therefore this book's own, set against the chain from material to the body's response that Chapter 1 set out. The hazard is what the material releases: on the running device, a volatile compound left in its foam from manufacture or a product of the foam's breakdown (Chapter 5). The sequence of events is the route, as the compound evaporates into the air at the temperature of use and the blower's flow carries it through the humidifier, the hose and the mask. The hazardous situation is the exposure, the user breathing that air for hours every night, for years, and harm is the response to a dose above what the body tolerates.
Each link is a question that one part of the evaluation answers. Chemistry names the hazard (Chapters 12 and 13), and the exposure estimate follows the sequence of events to the amount that reaches the patient (Chapters 15 and 19). Toxicology sets the tolerable intake (Chapter 14), and the dose that arrives, set against that intake, decides P2. Under the 2025 edition the evaluator estimates biological risk "based on the severity and probability of harm", and does so "either qualitatively or quantitatively", in the account of NAMSA, a contract laboratory that sells such testing. FDA, however, does not recognize the clause that asks for this estimate (Chapters 21 and 26).
- The hazardous situation. For a substance whose harm needs long exposure, the hazardous situation is years of breathing the machine's air. P1 is then the chance that a new user is still on the machine after those years, taking the release and its travel to the mask as certain on every night of use.
- The data. In the French cohort of Chapter 2, Pépin and colleagues' 2021 analysis of national insurance claims found that treatment had ended for 47.7 % of 480,000 new users after three years, counted as the end of reimbursement. ResMed funded the study with an unrestricted grant and paid for its medical writing, and three of its authors were ResMed employees.
- The arithmetic. 100 % less 47.7 % leaves 52.3 %, so P1 for three years of continued treatment is about 0.5.
- The result. The probability of occurrence of harm, P1 times P2, is then about half of P2. Rare exposure cannot keep the risk low, so the evaluation must show that P2 is small, which means a dose well below the tolerable intake.
- Assumptions. The release is taken as certain every night, which overstates P1. The end of reimbursement is not the same as no use, and continued reimbursement is not nightly use either, since French rules let it continue at 2 to 4 hours a night. The cohort is one country's in one period, and the reading of the chain is this book's.
The plan comes before the tests
A plan written first fixes what each result will be judged against before the result exists, so an acceptance criterion for the foam's emissions cannot be fitted afterward to a reading that came out high. The evaluation's first output is therefore the biological evaluation plan, written while the materials are still being chosen. Under the 2025 edition the plan names the hazards, the exposure and the acceptance criteria, in the account of TÜV NORD, an EU notified body. The evaluation must also "consider not only the intended use of a product, but also reasonably foreseeable misuse", TÜV NORD adds. NAMSA gives an example of such misuse: "the use for longer than the period intended by the manufacturer, resulting in a longer duration of exposure".
FDA expects a submission to start from the same place: its 2023 guidance asks for a risk assessment that identifies "the knowledge gaps that remain" and a plan "to address the knowledge gaps", placed "at the beginning of the biocompatibility section". The plan's questions also reach back into the design while the design can still change. They decide what a material's supplier must disclose (Chapter 3), which processing the test articles must go through (Chapters 6 and 7), and whether chemistry can answer alone or a biological test is needed (Chapter 18).
The last output is the biological evaluation report, which closes the plan. Under the 2025 edition the report must "document conformity with the evaluation plan", in NAMSA's account. The report is therefore verification evidence: it shows that the design meets the biological requirements set at the start. Tests gathered at the end with no plan behind them can show that each test passed, but not that they covered every question the device's contacts raise. Philips Respironics's own reports that the opening lists, dated from 2016 to 2021, are test results and risk assessments of the kind this book describes. A result of either kind protects patients only once the evaluation carries it through the chain of hazard, exposure and harm into a decision about the device's risk.
What the evaluation asks of the design
The evaluation also creates requirements of its own, which enter the design file as design inputs. A material specification names the grade and the supplier of each material in contact, so that a change can be recognized when it comes (Chapter 23). A cleaning instruction names the agents the user may apply, because ozone and UV cleaners can break down hoses and masks, FDA warns (Chapter 7). A useful life, such as one maker's 6 months for its nasal mask, fixes the duration the evaluation covered (Chapter 2). A contact limit states what the gas pathway may add to the air, against the limits of ISO 18562 (Chapter 19). Each input is a condition under which the report's conclusion holds.
These inputs are risk controls in ISO 14971's sense. The standard ranks such controls "in order of preference and effectiveness", as FDA's guidance on human factors reports: inherent safety by design first, then protective measures, then information for safety. The running device's foam is a polyether, this book's choice in Chapter 5. That choice is safety by design against hydrolysis, because the foam lacks the ester link that water cuts first in a warm, humid air path. The same foam has a weakness of its own, oxidation (Chapter 5), so the cleaning instruction that keeps ozone away protects the foam as well as the hoses and masks. An instruction to replace the mask at the end of its useful life is information for safety, last in the order. The risk left once the controls are in place is the residual risk, and a benefit–risk analysis justifies why the device's benefits outweigh it.
Traced from each contact to its conclusion
A reviewer reads the file from the other end, starting from a conclusion and asking what supports it. Traceability makes that reading possible. Each contact of the device is linked to its category and duration, to the biological effects listed for it, to the data that address each effect and to the conclusion drawn. A broken link appears as a contact with no effects listed, an effect with no data, or a conclusion that rests on data from another material. FDA's guidance asks for data "complete to the extent that an independent conclusion could be made", and an independent conclusion needs a path from contact to data that someone other than the author can follow.
FDA's guidance asks two things of those who test a device. It asks for "competent trained staff in the conduct of biocompatibility testing", and it says that the tests "should be conducted in accordance with recognized Good Laboratory Practice (GLP) regulations", the rules for conducting and recording a laboratory safety study. Both requests are written for the people who run the tests. The evaluator who writes the plan and the report has a different task: judging the chemistry, the toxicology and the device's use together. Naming the evaluator and their training in the file is this book's practice.
The running device's plan takes its rows from the contacts that Chapter 2 sorted and adds what that chapter left open: each row's questions, its data and its conclusion. The gas pathway's questions come from ISO 18562 (Chapter 19). In a test laboratory's account, the parts of that series test the air for particles and volatile compounds and the condensate for what it carries. The heated water of the water chamber is among the sources of what they test for. The housing's brief contact needs only a statement in the plan. The device's software releases nothing and has no row, so the software standard IEC 62304 plays no part.
The cushion's row shows the plan following the estimate. The cushion rests on intact skin every night, a long-term contact (Chapter 2). For intact skin the endpoints of concern are cytotoxicity, sensitization and irritation, in FDA's account of its 2023 guidance. The 2025 edition adds genotoxicity for contact of more than a day, two laboratories report. Chapters 12 to 17 named and weighed what the cushion's extract held, but one peak stayed unnamed (Chapter 13), and Chapter 16's thresholds leave sensitization and irritation open. FDA accepts information on listed materials in place of tests for devices on intact skin (Chapter 18), and the headgear's fabrics and foam could take that route if they are listed. This book's plan tests the cushion instead, because no statement of its material names the unknown peak or settles sensitization, and Chapter 18 sets out the tests.
Once the device is on the market, field data bring the evaluation back to the start of the loop. In the 2025 edition, biological safety is "continuously monitored and updated based on production and post-market data", in NAMSA's account. Complaints, returned devices and reports of injury are evidence about the hazardous situations that the plan assumed, and a pattern among them can reveal a sequence of events the plan never considered. In NAMSA's reading, the 2025 updates do not mandate "retesting of medical devices that are already marketed". A change to the device does reopen the evaluation, because the report's conclusion holds only for the materials and conditions it covered. Chapter 23 follows how a supplier's or a maker's change sends the evaluation around the loop again.
Write the biological evaluation plan when the first materials are chosen, giving each contact its category and duration, possible releases, assumed exposure, foreseeable misuse and acceptance criteria. Enter the plan's conditions in the design file as inputs: each material's grade and supplier, the permitted cleaning agents, the useful life and the contact limits. Trace every conclusion of the report to a contact and its data, and record who wrote the plan and the report, with their training.
- A biological evaluation runs inside ISO 14971's risk management, from a plan written at material selection to a report that serves as verification evidence, and field data bring it back to the start.
- ISO 14971 runs from a hazard through a sequence of events to a hazardous situation and harm, and for a released substance the dose that arrives, set against the tolerable intake, decides whether harm follows.
- About half of new users in one French claims cohort were still reimbursed for CPAP after three years, so for a machine kept in use the risk rests on a low dose, not on rare exposure.
- The evaluation's conditions enter the design as inputs, and traceability links each conclusion to its contact, its data and the estimate the plan followed.
Let's Talk