Human Factors
Human Factors: The Device That Is Safe on Paper and Dangerous in a Tired Person's Hands
IEC 62366-1:2015+AMD1:2020 is the international standard for usability engineering of medical devices, recognised by FDA as a consensus standard, and FDA's guidance 'Applying Human Factors and Usability Engineering to Medical Devices' (2016, revised August 2026) sets what reviewers expect. The process: specify users and use environment, identify use-related hazards, design the interface to mitigate them, validate with representative users. Formative testing uses early prototypes with 5–8 participants per user group; summative validation uses the finalised, production-equivalent device with at least 15 participants per distinct user group. On average 85% of utilisation problems are found with 5 representative users, 95% with 10. Between 2012 and 2015 there were 423 medical device recalls resulting from UI software errors. Acceptance is the correct execution of a task without use errors or difficulties, a percentage of users completing a task in a time is not a pass mark, because the users who cannot complete it are the signal, not the noise.
First, a simple picture
A door, and a tired person.
Imagine a door with a handle that looks like you should pull it, and the door only opens when you push. The door is not broken. The handle is not broken.
The person is not stupid. The door is safe on paper: the hinge works, the latch works, the lock works. And the door is dangerous in a tired person's hands, because the handle tells them the wrong thing, and they push when they should pull, and they pull when they should push, and the door stays shut.
Usability engineering, also called human factors engineering, is the job of making the door tell the truth: the handle that looks like a pull opens when you pull, the button that looks like a stop stops the machine, and the light that means "live" never also means "charging". It is the discipline of designing a device so that a tired, distracted, or untrained person can use it safely, because the person who uses it in the field will be tired, distracted, or untrained, and the device that only works in the hands of the engineer who built it does not work.
Two kinds of failure, and only one of them is the device's fault
There are two kinds of failure, and only one of them is the device's fault. A device failure is when the part broke: the battery died, the motor stalled, the radio dropped, the sensor drifted. The device did not do what it was built to do. A use error is when the person did the wrong thing, or did nothing, because the interface let them: the battery went in backwards, the alarm was misread, the button sequence was confused, the display data was misread, the lid was shut on an unseated seal. The device did exactly what it was built to do. The person did exactly what the interface told them to do. And the outcome was the one nobody wanted. In the medical field, problems with usability are referred to as "use errors" instead of "user errors" to defer blame away from the user, it is not the user's fault when the interface sets them up for failure, and the use error is defined as something the user does or fails to do that results in an outcome the user or manufacturer does not expect. Use errors can be reasonably predicted to occur, and their risk of happening can be minimised through contextual inquiry, risk analysis, and usability testing.
The numbers that make the case are the recalls. Between 2012 and 2015 alone, three years, September 2012 to August 2015, there were 423 medical device recalls resulting from UI software errors, about 140 a year, accounting for 5.44% of all recalls studied and 46.33% of recalls resulting from software errors (Zhang, Masci, Jones, Thimbleby, 2019, in the AAMI journal Biomedical Instrumentation & Technology, the analysis of the FDA's public and internal recall databases, PMID 31162965). A total of 499 UI software errors were identified as the root causes of those recalls, classified into 20 categories, and the identified UI software errors emerged from a lack of adherence with widely recognised human factors design principles or as consequences of simple design and implementation errors. The UI software governs the interaction between a device and its user, and it can increase the chance of use errors and in turn pose threats to patient safety if it does not appropriately manage design flaws and implementation mistakes. Mitigation of use errors is a top priority in the design of medical devices, and the same priority applies to any device a tired person will use, because the interface is where the field report comes from.
And the use errors are not exotic. A glucose meter was recalled because its users incorrectly interpreted the numbers on the display: when it showed "2.2", the square decimal point was so small and so close to the first "2" that users read the number as "22", so the users thought their blood glucose levels were ten times higher than they really were, and would mistakenly believe they needed to adjust their insulin injections to a much higher level than necessary, with consequences from severe hypoglycaemia to diabetic coma or death. In response, the FDA required the manufacturers to remove the capability of changing units on all glucose meters. An ultrasound system was recalled in 2008 because the graphics made users misunderstand the image orientations of the patient's left and right sides, the users assumed the patient's right and left were oriented in the same direction as the transducer, and the assumption was incorrect. A diabetes management mobile app was recalled because its UI software could erroneously reset the recommended insulin bolus dosage when the user changed the smartphone's orientation, which might cause the user to inadvertently command and receive unsafe insulin therapies. A wheelchair was recalled in 2008 because it had the potential to pinch a user's fingers between seat bars when the user was opening it, resulting in injuries as severe as fracture and severing. Usability testing reveals the assumptions that users make, and that can prevent designers from overlooking the characteristics that lead to a higher risk for users, and ultimately, a recall. And the share of the blame: when use errors occur, the user is solely responsible in just 1 out of 100 instances (Wiklund, via MedTech Intelligence), so the interface is the cause in the other 99.
How common is use error, honestly? The pipeline's figure, "a significant majority of adverse medical device incidents stem from use error", is not verified as stated, and the file says so. The checked figure is from Knisely et al. (2020), an analysis of FDA MAUDE adverse-event data from 2010–2018: use error is significantly represented in adverse event reporting, constituting 28.1% of reports labelled with device problem codes, a large share, not a majority. The UK's MHRA attributed 21% of incidents involving infusion devices to user error (MHRA 2013). The French health authority found a little over 25% of health-product serious adverse events were errors related to incorrect use of medical devices. The FDA reported that between 2005 and 2009, infusion pumps and related devices accounted for 35% of medical errors that resulted in significant patient harm, and a large number of the adverse events stemmed from programming errors attributed to poor device usability. And in one infusion-pump case, human error had caused 68% of the fatalities and serious injuries associated with the device. The honest statement is the one the checked sources support: use error is a leading cause of device incidents, roughly a quarter to a third of reported adverse events, and the majority of the harm in the worst cases, and the design is the cause in the vast majority of those cases. The pipeline's "significant majority" is replaced by those figures.
The standard, the guidance, and the process
IEC 62366-1:2015+AMD1:2020 is the international standard for usability engineering of medical devices, the standard that covers the application of usability engineering to medical devices, offering a standardised process for analysing, specifying, developing, and evaluating the usability of a device as it relates to safety. Correct use and use error together make up normal use, and both are inside the standard's scope. IEC 62366-1:2015 with Amendment 1, published June 17, 2020, remains the edition in force. IEC/TR 62366-2:2016 is guidance, a technical report that contains background information and guidance on implementing IEC 62366-1, and does not specify any requirements. The most recent versions of the two parts are IEC 62366:2015 + A1:2020 and IEC 62366-2:2016, and they are recognised consensus standards by FDA, which means FDA will accept a declaration of conformity from a medical device manufacturer to satisfy part of a premarket review requirement. IEC 62366-1 Edition 1.1 is a recognised consensus standard for US submissions under recognition number 5-129, and the US adoption comes from ANSI and AAMI as ANSI/AAMI/IEC 62366-1:2015+AMD1:2020. One thing IEC 62366-1 is not: a standalone certifiable standard, there is no "IEC 62366 certificate". Conformity is assessed within an ISO 13485 QMS audit, a notified-body review, or an FDA review, through the completeness of the Usability Engineering File. And the EU note, checked today: EN IEC 62366-1, including EN 62366-1:2015+A1:2020, is absent from the current EU MDR harmonised standards list, so conformity to it confers no presumption of conformity under Article 8(1), and the obligation itself comes from Annex I, which requires manufacturers to reduce ergonomic and use-environment risks as far as possible, and to account for users' training and physical condition, or on lay-use devices their skills and means.
FDA's guidance "Applying Human Factors and Usability Engineering to Medical Devices" sets what reviewers expect. The guidance was originally issued on February 3, 2016, and FDA revised it on August 3, 2026, the first revision since issue, updating the definitions and the documentation section for consistency with the new content guidance. FDA finalised a companion guidance, "Content of Human Factors Information in Medical Device Marketing Submissions", on May 29, 2026, which sorts filings into three HF Submission Categories, and the category a manufacturer claims decides how much of the file travels with the application. Neither change altered what IEC 62366-1 requires. The FDA's Quality Management System Regulation (QMSR, effective February 2, 2026) incorporates ISO 13485:2016 into Part 820 by reference, so usability work also reaches FDA through the design and development requirements it now points at. And the important nuance, checked today: FDA's 2016 guidance and IEC 62366-1 are not identical, where they differ, FDA's guidance reflects FDA's current regulatory expectations, and manufacturers should understand both. The FDA's guidance is the binding reference; IEC 62366-1 is the implementation framework most commonly used to meet it.
The process, in four steps: specify the users and the use environment (who will use it, where, in what light, with what training, with what else in their hands); identify the use-related hazards (the tasks where a use error could cause harm, the critical tasks, defined by FDA as a user task which, if performed incorrectly or not performed at all, would or could cause serious harm to the patient or user, where harm is defined to include compromised medical care, and FDA puts severity ahead of probability, because the probability of a use error is hard to estimate before a device is in use, so the severity of the potential harm is more meaningful for deciding what needs attention); design the UI to mitigate them (and the hierarchy matters, the FDA explicitly warns reviewers not to accept "revising the IFU" or "re-training users" as the primary mitigation when a physical or software interface redesign is feasible, because inherent design safety comes first, and labelling and training come last); and validate with representative users (on the finalised, production-equivalent device, in the intended use environment). The process is tightly linked to ISO 14971: usability engineering is part of the manufacturer's risk management, it begins with a use specification, identifies hazard-related use scenarios, uses formative evaluation to refine the interface iteratively during design, then summative evaluation to prove the device can be used without unacceptable residual risk. The evidence lives in a Usability Engineering File (UEF), and the ISO 14971 risk file and the hazard-related use scenario list are two records of one phenomenon, kept by two teams, and nothing in either process forces them to agree, which is why the file has to be checked for agreement, not assumed.
Formative and summative: the rounds, and the number fifteen
Formative testing is the rounds. It is conducted during the design process to identify usability problems and inform redesign, qualitative, focusing on discovering what goes wrong and why rather than measuring performance with statistical precision. Formative testing uses small samples (5–8 participants per user group), think-aloud protocols, and iterative cycles of testing and redesign. The participants can be company employees. The method is flexible: cognitive walk-throughs, early prototypes, a facilitator who gives instructions without steering, realistic tasks, and a realistic participant who usually thinks out loud, and then mostly keeping quiet.
Summative testing is the proof. It evaluates a finished or near-finished design against defined criteria, quantitative, measuring task completion rates, task times, error rates, and satisfaction scores, with larger samples and statistical analysis. The device must be finalised and production-equivalent. The participants must be representative intended users, generally at least 15 per distinct user group, not employees, and US residents if the device is for US users, and the FDA guidance names the quantitative sample-size expectation as at least 15 representative US participants per distinct user group for summative human factors validation testing. The number fifteen has an empirical basis: a 2003 study by Faulkner found that 15 users detected a minimum of 90% and an average of 97% of software problems, which is why it became the regulatory standard. And for high-risk medical devices where use errors could be fatal, some manufacturers routinely test 20+ participants per group.
The numbers that make the case: on average 85% of utilisation problems are found with 5 representative users, 95% with 10. The figure comes from Nielsen and Landauer (1993), and the mathematics behind it is the Nielsen–Landauer formula: N(n) = N × (1 − (1 − L)^n), where N is the total number of problems, n is the number of users, and L is the share of problems a single user uncovers, typically about 31%, meaning each user has roughly a one-in-three chance of discovering any given usability issue. The first test user reveals about 31% of usability problems. The second uncovers some of the same issues but also discovers new ones, bringing the total to roughly 55%. By the third user, you are at 70%. The fourth pushes you to 81%, and the fifth gets you to that magical 85% threshold. Ten users find about 98% by the model (the commonly quoted figure is 95%, and the file says so, the model gives 97.6% at L = 0.31, and Faulkner's empirical study found 15 users found a minimum of 90% and an average of 97% of software problems, so the 95% is the rounded figure the guides quote, not the model's exact output). Fifteen users find 99.6% by the model. But the 85% figure is an average, and it applies to problems with a discovery probability of about 31% per participant, less common problems require more participants. The honest arithmetic, verified before writing: a problem affecting 31% of users has a 0.4% chance of being missed by all 15 summative users; a problem affecting 10% of users has a 20.6% chance of being missed; a problem affecting 5% of users has a 46.3% chance of being missed, so fifteen per group is a floor, not a guarantee, and rare problems need the formative rounds (or more users) to be caught at all.
And the point of the five-user rule is not to test five people and stop. Nielsen's argument is that detecting 85% of problems is better than 100%, correct the design, and then perform the research again. Repeat the process. Instead of conducting one study with 15 respondents, it is better to perform three studies at three stages of the design with 5 respondents in each study, because three iterative tests with 5 users each outperform one test with 15 users: the first round finds most of the problems, you fix the design, the second round checks whether the fixes actually work, and the third round catches the issues the first round missed. Testing 15 people at once can mean watching the same problem ten times. The rule works when the audience is fairly similar; if the site serves very different groups, you need to test each group on its own, and multiple user groups require 3–4 users per segment, not 5 total. For quantitative measurement, such as comparing task times statistically, Nielsen Norman Group recommends about 20 users. Five users find problems, not numbers to quote.
Acceptance: the users who cannot complete it are the signal
Acceptance criteria must be the correct execution of a task without use errors or difficulties, and a percentage is not a pass mark. "X% of users must complete a task in Y minutes" is not an acceptable criterion, because the users who cannot complete it are the signal, not the noise. The checked position, from the FDA's August 2026 guidance as summarised by the practitioners who read it: there is no success rate, and that is by design, validation of a medical device interface is not a percentage of tasks passed. The results count distinct participants against each kind of use problem, by user population and scenario, and every issue goes to a root cause before the residual risk is reviewed. The purpose of HF validation testing is to show the final design can be used safely by intended users, with no use errors that could cause serious harm and could be designed out. The participant-by-task grid records whether the task was completed and, independently, what was seen on the way: a use error, a close call, a difficulty, help needed, or none of these, and every issue takes a root cause (interface, labelling, training, test artefact, or other), read as explained or undetermined, and a cause nobody could find is recorded as undetermined, not guessed. Participants, not percentages. And the hierarchy behind it: the risk controls that matter most are proven in validation, and the controls are ordered, inherent design safety first, then protective measures, then information for safety (labelling and training) last, because a label is not a fix, and training is not a fix, and the FDA explicitly warns reviewers not to accept revising the IFU or re-training users as the primary mitigation when a physical or software interface redesign is feasible.
The acceptance test is the critical task, done correctly, by a representative user, on the final device, in the intended use environment, without a use error, a close call, a difficulty, or help. If one user in fifteen cannot complete it, the design has a problem that affects one in fifteen users, and the file has to say so, with the root cause, and the residual risk reviewed, before the device ships. The users who cannot complete it are not the noise. They are the signal.
DEMO
Example Labs is a made-up company we use for worked examples.
The Example Labs field sensor hub, 100 × 60 × 40 mm, stationary on the fence post, the same hub from Articles #6 and #7, with the CR2032 battery (230 mAh, 2,484 J) from Article #6 and the enclosure from Article #7. The demo is the article's own worked example, and its arithmetic is verified before writing. The task: a tired field technician, at the end of a shift, in low light, with gloves on, swaps the hub's battery and reconnects the sensor cable, then confirms the hub is live.
Six steps: open the lid, remove the old battery, insert the new battery the right way round, close the lid, press the wake button, read the status LED. The person the demo tests is the person who will actually do it, tired, distracted, in low light, with gloves on, because the device that only works in the hands of the engineer who built it does not work.
Step 1, The three use errors. Each one is a real hazard, and each one is mapped to its mitigation, and the mitigations are ordered by the hierarchy, inherent safety first. Use error 1: the battery goes in backwards. There is no polarity marking visible in low light, and the CR2032 fits either way round, so the hub never wakes, and the technician assumes the hub is dead and drives back to the depot, or worse, leaves a dead hub on the fence post. The hazard is a dead hub in the field, and the mitigation is inherent safety: key the battery tray so the battery only fits one way round, plus a moulded "+" in the tray that can be felt with a glove on, not a label, and not a line in the manual. Use error 2: the lid is shut on an unseated seal. The seal is not seated, and the lid closes anyway, so the IP rating from Article #7 is gone, and the first rain kills the hub, weeks later, in the field, with no one watching. The hazard is water ingress and a dead hub, and the mitigation is inherent safety: detent the lid so it cannot close unless the seal is seated, a physical stop, not a warning, plus a visible seal indicator that shows green when the seal is seated. Use error 3: the status LED lies. Green means "live", and green also means "charging", so the technician walks away from a hub that is charging, not live, and the site goes dark. The hazard is a silent site, and the mitigation is inherent safety: two colours, not one, green means live, amber means charging, plus a distinct blink pattern for each state, so the state is readable in low light and from a distance.
Step 2, The formative rounds. Three rounds of five users, with a redesign between rounds, and the arithmetic is the Nielsen–Landauer model, verified before writing. The design starts with 20 problems. Round 1 (5 field technicians, early prototype, think aloud): finds about 85% of the problems, 17 found, 3 remain, and the team fixes all 17. Round 2 (5 new technicians, the fixed design): finds about 85% of the remaining 3, 2–3 found, about 0.5 remain, and the team fixes them. Round 3 (5 new technicians, the fixed design again): finds the last one, about 0.07 problems remain. Fifteen sessions, three rounds, and the residual problem count drops from 20 to effectively zero, because each round tests a better design. One round of fifteen would find 99.6% of the problems in that design, but it would test a design that never gets fixed between rounds, and it would mean watching the same problem ten times. The three rounds are the work; the five users are the round.
Step 3, The summative validation. Two distinct user groups, field technicians and site managers, because the site manager is the person who checks the hub is live, and the technician is the person who swaps the battery, and they are different users with different tasks. 15 participants per distinct user group, 30 participants in total, on the finalised, production-equivalent hub, in the intended use environment (low light, end of shift, gloves on). The tasks are the critical tasks: swap the battery, seat the seal, close the lid, confirm the hub is live, and the acceptance criterion is the correct execution of each task without use errors or difficulties. Every participant must complete the battery swap and confirm the hub is live without a use error, a close call, a difficulty, or help, and the one who cannot is the signal, not the noise. And the honest arithmetic, verified before writing: a problem affecting 31% of users has a 0.4% chance of being missed by all 15 summative users in a group; a problem affecting 10% has a 20.6% chance of being missed; a problem affecting 5% has a 46.3% chance of being missed, so fifteen per group is a floor, not a guarantee, and the rare problems are the ones the formative rounds have to catch, because the summative test will miss nearly half of them.
Step 4, The cost of not doing it. The recall numbers, checked today: 423 recalls in three years from UI software errors, about 140 a year, 5.44% of all recalls studied, 46.33% of recalls from software errors, 499 UI software errors as root causes, in 20 categories, and the identified errors emerged from a lack of adherence with widely recognised human factors design principles or as consequences of simple design and implementation errors. The glucose meter that showed "2.2" and was read as "22", a ten-times insulin error, from severe hypoglycaemia to diabetic coma or death, was a square decimal point, too small and too close to the first "2". The diabetes app that reset the insulin bolus when the phone was rotated was a UI software error. The hub's version of the same failure is the LED that means two things, the battery that fits both ways, and the lid that closes on an unseated seal, and the fix is three rounds of five users, thirty summative participants, and a design that cannot be used wrong, because the person who uses it will be tired, and the device that only works in the hands of the engineer who built it does not work.
Three use errors on the worked example's hub, and fixes that design the error out
| Use error | Hazard | Fix |
|---|---|---|
| Battery fitted backwards | No polarity mark visible in low light; the hub never wakes | Key the battery so it fits one way only |
| Lid shut on an unseated seal | The IP rating is gone; the first rain kills the hub | A lid that cannot close unless the seal is seated |
| Green LED for two states | Charging looks like live; the site goes dark | A distinct signal for each state |
The decision test
Four questions. They sort the devices that are safe in a tired person's hands from the devices that are safe on paper:
- Can you name the users, and the use environment they will actually be in? No, and the users are "the user", the environment is "the field", and the device was tested on a bench, in good light, by the engineer who built it. Yes, and the users are named (field technicians and site managers, with their training, their gloves, their light, and their shift), the environment is specified (low light, end of shift, rain, gloves on), and the test is run in that environment, on the finalised device.
- Can you name the critical tasks, and the use errors that could hurt someone on each one? No, and the tasks are "use the device", the errors are "the user will figure it out", and the first field report will be a use error, not a hardware failure. Yes, and the critical tasks are named (swap the battery, seat the seal, close the lid, confirm the hub is live), each one carries the use errors that could hurt someone, each use error carries its risk controls, and the controls that matter most are proven in validation.
- Did you test in rounds, and did the design change between rounds? No, and the test was one round of fifteen, on a design that never got fixed, and the team watched the same problem ten times. Yes, and the formative rounds were three rounds of five, with a redesign between rounds, and the residual problem count dropped from 20 to effectively zero, and the summative test was fifteen participants per distinct user group on the finalised, production-equivalent device.
- Is the acceptance the correct execution of the task without use errors or difficulties, and is the hierarchy inherent safety first? No, and the acceptance is "X% of users must complete a task in Y minutes", and the mitigation is a label or a line in the manual, and the users who cannot complete it are treated as noise. Yes, and the acceptance is the correct execution of the task without use errors, close calls, difficulties, or help, and the mitigations are inherent safety first (the battery only fits one way, the lid cannot close on an unseated seal, two colours not one), and the users who cannot complete it are the signal.
If the answers are yes, yes, yes, yes, the device is safe in a tired person's hands, and the review is a final check, not a rescue. If any answer is no, the job is a usability/human-factors review, and the honest thing is to do it before the device ships, not after it, when the first field report is a use error and the recall is already being written. And if the answer to question 4 is "we will train them", the problem is not the users.
The problem is the interface, and the review starts by testing it in the hands it will actually be in.
"But our users are trained", and two more fair doubts
"But our users are trained." Training is the last control in the hierarchy, not the first, and the FDA explicitly warns reviewers not to accept revising the IFU or re-training users as the primary mitigation when a physical or software interface redesign is feasible. The trained user is still tired at the end of a shift, still in low light, still with gloves on, and the use error that a trained user makes is the same use error an untrained user makes, because the interface is the same. The glucose meter that showed "2.2" and was read as "22" was used by trained patients. The ultrasound that confused left and right was used by trained clinicians. The diabetes app that reset the bolus when the phone was rotated was used by trained patients. Training does not fix a decimal point that is too small, a graphic that points the wrong way, or a battery that fits both ways. The design does.
"But we tested with fifteen people and found nothing." Fifteen people find 99.6% of the problems that affect a third of users, and 53.7% of the problems that affect one in twenty users. The problem that affects one in twenty users has a 46.3% chance of being missed by all fifteen, so "we tested with fifteen and found nothing" means "we tested with fifteen and found nothing that affects a third of users", and the rare problems are still there, waiting for the field. The formative rounds are the ones that catch them, because three rounds of five, with a redesign between rounds, test a better design each time, and the residual problem count drops from 20 to effectively zero. And the summative test is not a search for problems. It is a proof that the final design can be used safely, and the proof is the correct execution of the critical tasks without use errors or difficulties, by fifteen representative users per group, on the finalised device, in the intended use environment.
"But usability engineering is only for medical devices." IEC 62366-1 is the international standard for usability engineering of medical devices, and the process it specifies is the process for any device a tired person will use. The hub is not a medical device. The lid, the battery, and the cable are still the interface. The technician is still tired. The low light is still low. The gloves are still on. The use error is still the first field report. The standard is written for medical devices because the harm there is the most visible, but the method (specify the users and the use environment, identify the use-related hazards, design the interface to mitigate them, validate with representative users) is the method for any device whose first field report will be a use error, not a hardware failure. The gate in Article #10, the human approval, is the same job: design the approval so a tired person cannot approve the wrong thing, and test it in the hands it will actually be in.
Sources
- Jama Software, "What Is IEC 62366? Medical Device Usability Guide"
What we used
FDA finalised a companion guidance on marketing submission content in May 2026 and revised the 2016 human factors guidance itself in August 2026, and neither change altered what IEC 62366-1 requires; IEC 62366-1:2015 with Amendment 1, published June 17, 2020, remains the edition in force; IEC 62366-1 is a recognised consensus standard for US submissions and absent from the current EU harmonised standards list, so conformity to it doesn't mean the same thing in both markets; IEC 62366-1 Edition 1.1 is a recognised consensus standard for US submissions under recognition number 5-129; the US adoption comes from ANSI and AAMI as ANSI/AAMI/IEC 62366-1:2015+AMD1:2020; the QMSR has been in effect since February 2, 2026 and incorporates ISO 13485:2016 into Part 820 by reference; EN IEC 62366-1, including EN 62366-1:2015+A1:2020, is absent from the current EU MDR harmonised standards list, the standard therefore confers no presumption of conformity under Article 8(1); the obligation itself comes from Annex I, which requires manufacturers to reduce ergonomic and use-environment risks as far as possible, and to account for users' training and physical condition, or on lay-use devices their skills and means; the FDA finalised Content of Human Factors Information in Medical Device Marketing Submissions on May 29, 2026, then revised the 2016 guidance on August 3, 2026, deleting Appendix A with its eight-section report outline and redirecting Section 9 on documentation to the new guidance; the submission guidance sorts filings into three HF Submission Categories, and the category a manufacturer claims decides how much of the file travels with the application
- Greenlight Guru, "IEC 62366: What You Need To Know About Usability Engineering"
What we used
IEC 62366 is the international standard that covers the application of usability engineering to medical devices, offering a standardised process for analysing, specifying, developing, and evaluating the usability of a device; IEC 62366-2 is a technical report which contains background information and guidance on implementing IEC 62366-1 and does not specify any requirements; the most recent versions are IEC 62366:2015 + A1:2020 and IEC 62366-2:2016, and they are recognised consensus standards by FDA, FDA will accept a declaration of conformity from a medical device manufacturer to satisfy part of a premarket review requirement; FDA's guidance "Applying Human Factors and Usability Engineering to Medical Devices" explains what FDA reviewers typically expect to see for human factors/usability work, including formative studies and summative (validation) testing
- MedDeviceGuide, "Human Factors Testing for Medical Devices: FDA and IEC 62366 Guide"
What we used
Human factors engineering (HFE), also called usability engineering, exists to prevent use errors by designing medical devices that people can use safely, effectively, and intuitively; the FDA requires human factors validation testing data in premarket submissions for many devices; EU Notified Bodies expect a Usability Engineering File (UEF) under IEC 62366-1 as part of MDR conformity assessment; FDA's guidance "Applying Human Factors and Usability Engineering to Medical Devices" (2016) describes when human factors validation testing is expected, how to conduct formative evaluations during development, how to design and execute summative (validation) testing, and what to include in a human factors engineering report for FDA submissions; IEC 62366-1 is the international standard for usability engineering of medical devices, recognised by the FDA as a consensus standard; IEC 62366-2:2016 is a technical report that provides guidance on implementing IEC 62366-1
- BuzzBox Media, "Human Factors Engineering for Medical Devices: IEC 62366 and FDA Usability Requirements"
What we used
FDA's primary HFE guidance is "Applying Human Factors and Usability Engineering to Medical Devices", finalised February 2016; the international standard is IEC 62366-1:2015 + AMD1:2020; FDA recognises IEC 62366-1 in its recognised standards database; conformance with IEC 62366-1 is voluntary, but it provides a widely accepted framework for demonstrating that FDA's HFE guidance requirements have been met; FDA's 2016 guidance and IEC 62366-1 are not identical, where they differ, FDA's guidance reflects FDA's current regulatory expectations; IEC 62366-2:2016 provides supplementary (informative) guidance on applying IEC 62366-1 in practice
- Lakshy, "IEC 62366 Usability Consultants in India"
What we used
IEC 62366-1 lays out the process a manufacturer follows to study, define, build and test how usable a medical device is, wherever usability bears on safety; correct use and use error together make up normal use, both inside the standard's scope; the current text is IEC 62366-1:2015, consolidated with Amendment 1 (2020); the process is tightly linked to ISO 14971, it begins with a use specification, identifies hazard-related use scenarios, uses formative evaluation to refine the interface iteratively during design, then summative evaluation to prove the device can be used without unacceptable residual risk; the evidence lives in a Usability Engineering File (UEF); there is no standalone IEC 62366 certificate, conformity is assessed within an ISO 13485 QMS audit, a notified-body or FDA review, through the completeness of the UEF
- FDA, "Applying Human Factors and Usability Engineering to Medical Devices"
What we used
Guidance for industry, document issued on August 3, 2026, originally issued on February 3, 2016; the guidance represents the current thinking of the FDA on this topic and does not establish any rights for any person and is not binding on FDA or the public; human factors validation testing is sometimes referred to as "summative testing", but summative testing can be defined differently and some definitions omit essential components of human factors validation testing as described in this guidance; the results of the testing should facilitate analysis of the root causes of use errors or problems found during the testing; Appendix A: Considerations for Determining Sample Sizes for Human Factors Validation Testing; Appendix B: Analyzing Results of Human Factors Validation Testing
- Palo Alto Health Labs, "Human Factors & Usability for SaMD and Wearables"
What we used
FDA expects you to find the user tasks where a use error could cause serious harm (critical tasks), reduce those risks mainly through design, and then show with human factors validation testing that representative users can do those tasks safely on the final design; formative evaluations shape the design along the way; HF validation testing (often called summative testing) confirms it at the end; "Applying Human Factors and Usability Engineering to Medical Devices" is the "how to do it" guidance, first issued on February 3, 2016, and FDA revised it on August 3, 2026; "Content of Human Factors Information in Medical Device Marketing Submissions" is the "what to submit" guidance, final issued May 29, 2026; critical task: "A user task which, if performed incorrectly or not performed at all, would or could cause serious harm to the patient or user, where harm is defined to include compromised medical care"; FDA puts severity ahead of probability, because the probability of a use error is hard to estimate before a device is in use, "the severity of the potential harm is more meaningful" for deciding what needs attention; summative participants: representative intended users, generally at least 15 per distinct user group, not employees, US residents if the device is for US users
- CASRAI, "FDA Human Factors Guidance: URRA & 15-User Rule"
What we used
A quantitative sample-size expectation of at least 15 representative U.S. participants per distinct user group for summative human factors validation testing; a standardised 8-section HFE/UE Report structure stored in the Design History File and submitted to the agency; the FDA explicitly warns reviewers not to accept "revising the IFU" or "re-training users" as the primary mitigation when a physical or software interface redesign is feasible; the FDA identifies critical tasks based on the severity of potential harm alone, independent of the estimated probability or frequency of the use error occurring
- Ergosphere, "Medical Device HF"
What we used
From the use specification to the HFE/UE report, one thread from user need to validation, aligned with IEC 62366-1, ISO 14971 and the FDA's 2026 Human Factors guidance; "There is no success rate, and that is by design. Validation of a medical device interface is not a percentage of tasks passed. The results count distinct participants against each kind of use problem, by user population and scenario, and every issue goes to a root cause before the residual risk is reviewed"; "Participants, not percentages", distinct participants per kind of use problem, by population and scenario; every issue takes a root cause (interface, labelling, training, test artefact or other), read as explained or undetermined; a cause nobody could find is recorded as undetermined, not guessed
- Articos, "How Many Test Users in a Usability Study Are Needed?"
What we used
In 1993, researchers Jakob Nielsen and Thomas Landauer published what would become UX research's most cited recommendation: test with just 5 users, and five participants could uncover approximately 85% of usability problems in an interface; the Nielsen–Landauer formula N(n) = N × (1, (1, L)^n), with L averaging around 31%, each user has roughly a one-in-three chance of discovering any given usability issue; the first test user reveals about 31% of usability problems, the second brings the total to roughly 55%, the third to 70%, the fourth to 81%, and the fifth to that magical 85% threshold; three iterative tests with 5 users each outperform one test with 15 users; for validation (summative) testing, IEC 62366-2 standards recommend 15 participants per distinct user group as a baseline; a 2003 study by Faulkner found that 15 users detected a minimum of 90% and an average of 97% of software problems, which is why this became the regulatory standard; for high-risk medical devices where use errors could be fatal, some manufacturers routinely test 20+ participants per group
- Textbook of Usability, "Usability Testing"
What we used
Formative testing uses small samples (5–8 participants), think-aloud protocols, and iterative cycles of testing and redesign; five users find about 85% of usability problems, and beyond 15 users, diminishing returns dominate; summative testing uses larger samples (20+ participants) and statistical analysis; the 85% figure applies to problems with a discovery probability of ~31% per participant, less common problems require more participants; for formative testing, 5–8 participants per user group is a practical starting point
- Zhang, Masci, Jones & Thimbleby (2019), "User Interface Software Errors in Medical Devices: Study of U.S. Recall Data"
What we used
*Biomedical Instrumentation & Technology* 53(3):182–194, PMID 31162965, the analysis assessed U.S. medical device recalls during 2012–15 (September 2012 to August 2015) and identified 423 (~140/year) medical device recalls resulting from UI software errors, which accounted for nearly one-half of recalls caused by software errors during the same period, 5.44% of total recalls studied, 46.33% of recalls resulting from software errors; a total of 499 UI software errors were identified as the root causes, classified into 20 categories; mitigation of use errors is a top priority in the design of medical devices; UI software governs the interaction between a device and its user, and can increase the chance of use errors and in turn pose threats to patient safety if it does not appropriately manage design flaws and implementation mistakes; the identified UI software errors emerged from a lack of adherence with widely recognised human factors design principles or as consequences of simple design/implementation errors
- UXPA Magazine, "Total Recall: The Consequence of Ignoring Medical Device Usability"
What we used
In the medical field, problems with usability are referred to as "use errors" instead of "user errors" to defer blame away from the user, it is not the user's fault when the interface sets them up for failure; use errors are defined as "something that the user does or fails to do that results in an outcome that the user or manufacturer does not expect"; use errors can be reasonably predicted to occur, and their risk of happening can be minimised through contextual inquiry, risk analysis, and usability testing; one glucose meter was recalled because its users incorrectly interpreted the numbers on the display, when it showed "2.2", the square decimal point was so small and so close to the first "2" that users read the number as "22", so the users thought their blood glucose levels were ten times higher than they really were, with consequences from severe hypoglycaemia to diabetic coma or death; in response, the FDA required the manufacturers to remove the capability of changing units on all glucose meters; in 2008, the FDA recalled an ultrasound system because the graphics made users misunderstand the image orientations of the patient's left and right sides; in 2008, the FDA recalled a wheelchair that had the potential to pinch a user's fingers between seat bars when the user was opening it, resulting in injuries as severe as fracture and severing; usability testing reveals the assumptions that users make, which can prevent designers from overlooking medical device characteristics that may lead to a higher risk for users, and ultimately, a recall
- Knisely, Levine, Kharod & Vaughn-Cooke (2020), "An Analysis of FDA Adverse Event Reporting Data for Trends in Medical Device Use Error"
What we used
*Proceedings of the Human Factors and Ergonomics Society*, a comprehensive evaluation of use error adverse events in MAUDE (2010–2018): use error is significantly represented in adverse event reporting, constituting 28.1% of reports labelled with device problem codes; most use error reports were attributed to issues with device output, using the device in accordance with manufacturer expectations, and physically activating, positioning, or separating device components
- MedTech Intelligence, "Medical Device Use Error: Stop Blaming Users"
What we used
When use errors occur, the user is solely responsible in just one out of 100 instances (Wiklund, Wiklund R&D); the root cause analysis process inspects the device for user interface design flaws (hardware, software, or instructional materials associated with the device)
- Boston UX, "Focus on Usability Can Limit Medical Device Recalls"
What we used
The FDA reported that between 2005 and 2009, infusion pumps and related devices accounted for 35% of medical errors that resulted in significant patient harm, and a large number of the adverse events stemmed from programming errors attributed to poor device usability; in one infusion-pump case, human error had caused 68% of the fatalities and serious injuries associated with the device, and after the redesign the number of required programming steps decreased by 55% and user errors dropped by 56%
- Rölleke (2009), "Incident reports to BfArM support the importance of usability for patient safety"
What we used
Due to various international regulations and standards, medical devices are generally safe as far as mechanical or electrical characteristics are concerned, and use errors caused by inadequate medical device usability have become an increasing cause for concern; in most complex work systems 70 to 80% of all failures are caused by human errors, and several studies conclude that 50 to 80% of human errors are a result of a non adequate product design or complex working conditions