eduKateSG · Why Science?
Decide whether learning from an investigational device justifies the risks of the study
Use FDA’s January 2017 final guidance through direct benefits, risks, uncertainty, alternatives, safeguards, study value, participant perspective and staged learning.
An investigational device study asks people to accept uncertainty so medicine can learn. FDA’s IDE benefit-risk framework examines whether the potential benefits and knowledge justify risks while participant protections remain appropriate.
The January 2017 final guidance explains principal factors FDA considers for original IDE applications, amendments and supplements for human clinical studies. It complements, rather than replaces, applicable IDE and human-subject rules.
Did you know? The “benefit” in an early study may include potential direct benefit to participants and the value of information, but those ideas must stay separate. A scientifically useful study cannot treat participants merely as data sources.
For students, benefit-risk reasoning combines science, ethics and decision-making. It builds research skills for healthcare, engineering, policy and regulation.
Chapter index
Chapters 1–9: anticipated benefit, clinical need, alternatives and risk characterisation
- 1. IDE benefit-risk asks whether the study should proceed
- 2. The January 2017 guidance is final
- 3. Potential direct benefit should be plausible
- 4. Knowledge value is not the same as direct benefit
- 5. Clinical need and alternatives shape the balance
- 6. Risk magnitude and probability both matter
- 7. Uncertainty is itself part of risk
- 8. Nonclinical evidence should reduce avoidable risk
- 9. Risk controls should form a layered system
Chapters 10–17: uncertainty, mitigations, study value and participant protections
- 10. Participant selection affects fairness and safety
- 11. Informed consent communicates uncertainty
- 12. Monitoring turns emerging evidence into action
- 13. Stopping rules should match serious concerns
- 14. Study design can reduce exposure and improve value
- 15. Patient perspective can inform acceptable trade-offs
- 16. Amendments and supplements need updated reasoning
- 17. A complete IDE benefit-risk receipt preserves the decision
CHAPTER 1 OF 17 · Name benefits
1. IDE benefit-risk asks whether the study should proceed
FDA assesses the potential benefits and risks of the proposed investigation, the uncertainty around them, and protections in the protocol. The decision concerns a clinical study, not marketing authorisation of the device.
For ide benefit-risk asks whether the study should proceed, a useful result begins with an explicit population and a visible evidence route. FDA assesses the potential benefits and risks of the proposed investigation, the uncertainty around them, and protections in the protocol. The decision concerns a clinical study, not marketing authorisation of the device. Keep the boundary beside the conclusion: Permission to investigate is not a conclusion that the device is safe and effective.
Try this
Build a source-first evidence card for ide benefit-risk asks whether the study should proceed. Write separate statements for the study decision and any future marketing decision. Add the denominator, date and one plausible source of bias. Keep this limit: Permission to investigate is not a conclusion that the device is safe and effective.
A strong science note separates observation from interpretation. Write separate statements for the study decision and any future marketing decision. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Write separate statements for the study decision and any future marketing decision. Keep this boundary visible: Permission to investigate is not a conclusion that the device is safe and effective.
The final guidance is dated January 2017 under docket FDA-2015-D-1777 and addresses IDE sponsors, sponsor-investigators and FDA staff.
The science in the january 2017 guidance is final is strongest when design, data and interpretation stay connected. The final guidance is dated January 2017 under docket FDA-2015-D-1777 and addresses IDE sponsors, sponsor-investigators and FDA staff. The result stops here: Guidance states current thinking and does not displace applicable regulations.
Try this
Test the january 2017 guidance is final with a counterexample. Record title, date, final status, docket, audience and scope. Ask what new evidence would change the conclusion. Preserve this boundary: Guidance states current thinking and does not displace applicable regulations.
A strong science note separates observation from interpretation. Record title, date, final status, docket, audience and scope. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Record title, date, final status, docket, audience and scope. Keep this boundary visible: Guidance states current thinking and does not displace applicable regulations.
Direct benefit may arise when the device could improve diagnosis, treatment or function for participants. Magnitude, probability, duration and evidence should be described without promotional language.
A careful reader can retrace potential direct benefit should be plausible from question to method to uncertainty. Direct benefit may arise when the device could improve diagnosis, treatment or function for participants. Magnitude, probability, duration and evidence should be described without promotional language. Preserve this guardrail: A theoretical mechanism does not establish that a participant will benefit.
Try this
Explain potential direct benefit should be plausible to a classmate using claim, evidence, reasoning and limitation. Build a benefit table with outcome, probability, duration, evidence and uncertainty. Make sure the official source and version remain visible. Respect this limit: A theoretical mechanism does not establish that a participant will benefit.
A strong science note separates observation from interpretation. Build a benefit table with outcome, probability, duration, evidence and uncertainty. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Build a benefit table with outcome, probability, duration, evidence and uncertainty. Keep this boundary visible: A theoretical mechanism does not establish that a participant will benefit.
A study can generate important information for future patients even when direct benefit is limited. Consent and benefit-risk reasoning should distinguish social or scientific value from likely personal benefit.
For knowledge value is not the same as direct benefit, a useful result begins with an explicit population and a visible evidence route. A study can generate important information for future patients even when direct benefit is limited. Consent and benefit-risk reasoning should distinguish social or scientific value from likely personal benefit. Keep the boundary beside the conclusion: Scientific importance cannot excuse disproportionate risk.
Try this
Build a source-first evidence card for knowledge value is not the same as direct benefit. Write plain-language sentences separating personal and societal benefit. Add the denominator, date and one plausible source of bias. Keep this limit: Scientific importance cannot excuse disproportionate risk.
A strong science note separates observation from interpretation. Write plain-language sentences separating personal and societal benefit. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Write plain-language sentences separating personal and societal benefit. Keep this boundary visible: Scientific importance cannot excuse disproportionate risk.
CHAPTER 5 OF 17 · Characterise risks
5. Clinical need and alternatives shape the balance
Severity, unmet need and available treatments affect how risks are viewed. Alternatives should include standard care, other trials and no intervention where relevant, with their benefits and burdens.
The science in clinical need and alternatives shape the balance is strongest when design, data and interpretation stay connected. Severity, unmet need and available treatments affect how risks are viewed. Alternatives should include standard care, other trials and no intervention where relevant, with their benefits and burdens. The result stops here: Few alternatives do not make every investigational risk reasonable.
Try this
Test clinical need and alternatives shape the balance with a counterexample. Compare the proposed study with each realistic alternative. Ask what new evidence would change the conclusion. Preserve this boundary: Few alternatives do not make every investigational risk reasonable.
A strong science note separates observation from interpretation. Compare the proposed study with each realistic alternative. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Compare the proposed study with each realistic alternative. Keep this boundary visible: Few alternatives do not make every investigational risk reasonable.
Risk characterisation should cover severity, likelihood, reversibility, duration and detectability. Device, procedure, anaesthesia, follow-up and data-collection risks may all contribute.
A careful reader can retrace risk magnitude and probability both matter from question to method to uncertainty. Risk characterisation should cover severity, likelihood, reversibility, duration and detectability. Device, procedure, anaesthesia, follow-up and data-collection risks may all contribute. Preserve this guardrail: A low-frequency catastrophic risk should not be hidden by an average score.
Try this
Explain risk magnitude and probability both matter to a classmate using claim, evidence, reasoning and limitation. Place each risk on separate severity and probability scales. Make sure the official source and version remain visible. Respect this limit: A low-frequency catastrophic risk should not be hidden by an average score.
A strong science note separates observation from interpretation. Place each risk on separate severity and probability scales. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Place each risk on separate severity and probability scales. Keep this boundary visible: A low-frequency catastrophic risk should not be hidden by an average score.
Early devices may have limited human use, incomplete testing or uncertain failure modes. The application should state what is unknown and why remaining uncertainty is acceptable for the study stage.
For uncertainty is itself part of risk, a useful result begins with an explicit population and a visible evidence route. Early devices may have limited human use, incomplete testing or uncertain failure modes. The application should state what is unknown and why remaining uncertainty is acceptable for the study stage. Keep the boundary beside the conclusion: Unknown risk is not equivalent to zero risk.
Try this
Build a source-first evidence card for uncertainty is itself part of risk. Create an uncertainty register with evidence gap, consequence and control. Add the denominator, date and one plausible source of bias. Keep this limit: Unknown risk is not equivalent to zero risk.
A strong science note separates observation from interpretation. Create an uncertainty register with evidence gap, consequence and control. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Create an uncertainty register with evidence gap, consequence and control. Keep this boundary visible: Unknown risk is not equivalent to zero risk.
CHAPTER 8 OF 17 · Characterise risks
8. Nonclinical evidence should reduce avoidable risk
Bench, biological, software, electrical, animal or modelling evidence should address foreseeable hazards before human exposure where applicable. Remaining gaps should be justified by why clinical experience is necessary.
The science in nonclinical evidence should reduce avoidable risk is strongest when design, data and interpretation stay connected. Bench, biological, software, electrical, animal or modelling evidence should address foreseeable hazards before human exposure where applicable. Remaining gaps should be justified by why clinical experience is necessary. The result stops here: Human participants should not be used to answer questions suitable nonclinical methods can resolve.
Try this
Test nonclinical evidence should reduce avoidable risk with a counterexample. Map each hazard to completed testing and residual uncertainty. Ask what new evidence would change the conclusion. Preserve this boundary: Human participants should not be used to answer questions suitable nonclinical methods can resolve.
A strong science note separates observation from interpretation. Map each hazard to completed testing and residual uncertainty. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Map each hazard to completed testing and residual uncertainty. Keep this boundary visible: Human participants should not be used to answer questions suitable nonclinical methods can resolve.
Design controls, eligibility, training, monitoring, rescue, stopping rules and follow-up can work together. The rationale should show which risk each control addresses and how effectiveness is verified.
A careful reader can retrace risk controls should form a layered system from question to method to uncertainty. Design controls, eligibility, training, monitoring, rescue, stopping rules and follow-up can work together. The rationale should show which risk each control addresses and how effectiveness is verified. Preserve this guardrail: A long control list is weak if ownership and triggers are unclear.
Try this
Explain risk controls should form a layered system to a classmate using claim, evidence, reasoning and limitation. Build a risk-control-owner-trigger matrix. Make sure the official source and version remain visible. Respect this limit: A long control list is weak if ownership and triggers are unclear.
A strong science note separates observation from interpretation. Build a risk-control-owner-trigger matrix. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Build a risk-control-owner-trigger matrix. Keep this boundary visible: A long control list is weak if ownership and triggers are unclear.
CHAPTER 10 OF 17 · Reduce uncertainty
10. Participant selection affects fairness and safety
Eligibility should identify people for whom the balance is supportable and the study informative. Vulnerability, disease stage and access to alternatives require attention.
For participant selection affects fairness and safety, a useful result begins with an explicit population and a visible evidence route. Eligibility should identify people for whom the balance is supportable and the study informative. Vulnerability, disease stage and access to alternatives require attention. Keep the boundary beside the conclusion: Narrow enrollment may protect participants but limit generalisability.
Try this
Build a source-first evidence card for participant selection affects fairness and safety. Link every inclusion and exclusion rule to safety or scientific purpose. Add the denominator, date and one plausible source of bias. Keep this limit: Narrow enrollment may protect participants but limit generalisability.
A strong science note separates observation from interpretation. Link every inclusion and exclusion rule to safety or scientific purpose. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Link every inclusion and exclusion rule to safety or scientific purpose. Keep this boundary visible: Narrow enrollment may protect participants but limit generalisability.
Consent should explain investigational status, alternatives, potential benefits, risks, unknowns, procedures and withdrawal. Language should avoid therapeutic misconception and support comprehension.
The science in informed consent communicates uncertainty is strongest when design, data and interpretation stay connected. Consent should explain investigational status, alternatives, potential benefits, risks, unknowns, procedures and withdrawal. Language should avoid therapeutic misconception and support comprehension. The result stops here: A signature does not prove understanding or voluntariness.
Try this
Test informed consent communicates uncertainty with a counterexample. Design a teach-back question for the largest uncertainty. Ask what new evidence would change the conclusion. Preserve this boundary: A signature does not prove understanding or voluntariness.
A strong science note separates observation from interpretation. Design a teach-back question for the largest uncertainty. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Design a teach-back question for the largest uncertainty. Keep this boundary visible: A signature does not prove understanding or voluntariness.
CHAPTER 12 OF 17 · Reduce uncertainty
12. Monitoring turns emerging evidence into action
Safety reviews, device checks and event reporting should occur quickly enough to protect current and future participants. Data flow and authority should be clear.
A careful reader can retrace monitoring turns emerging evidence into action from question to method to uncertainty. Safety reviews, device checks and event reporting should occur quickly enough to protect current and future participants. Data flow and authority should be clear. Preserve this guardrail: Collecting safety data without a decision pathway is not active protection.
Try this
Explain monitoring turns emerging evidence into action to a classmate using claim, evidence, reasoning and limitation. Specify one monitoring trigger, reviewer and response deadline. Make sure the official source and version remain visible. Respect this limit: Collecting safety data without a decision pathway is not active protection.
A strong science note separates observation from interpretation. Specify one monitoring trigger, reviewer and response deadline. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Specify one monitoring trigger, reviewer and response deadline. Keep this boundary visible: Collecting safety data without a decision pathway is not active protection.
CHAPTER 13 OF 17 · Protect the study
13. Stopping rules should match serious concerns
Prespecified pause or stop criteria can address serious adverse events, repeated deficiencies or an unfavourable trend. Restart criteria and required evidence should also be defined.
For stopping rules should match serious concerns, a useful result begins with an explicit population and a visible evidence route. Prespecified pause or stop criteria can address serious adverse events, repeated deficiencies or an unfavourable trend. Restart criteria and required evidence should also be defined. Keep the boundary beside the conclusion: Small samples make both absence and clustering of events uncertain.
Try this
Build a source-first evidence card for stopping rules should match serious concerns. Test the rule with three event scenarios. Add the denominator, date and one plausible source of bias. Keep this limit: Small samples make both absence and clustering of events uncertain.
A strong science note separates observation from interpretation. Test the rule with three event scenarios. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Test the rule with three event scenarios. Keep this boundary visible: Small samples make both absence and clustering of events uncertain.
CHAPTER 14 OF 17 · Protect the study
14. Study design can reduce exposure and improve value
Staged enrollment, sentinel participants, adaptive review or smaller early cohorts may limit exposure while learning. Design efficiency matters because an uninformative study imposes risk without sufficient knowledge gain.
The science in study design can reduce exposure and improve value is strongest when design, data and interpretation stay connected. Staged enrollment, sentinel participants, adaptive review or smaller early cohorts may limit exposure while learning. Design efficiency matters because an uninformative study imposes risk without sufficient knowledge gain. The result stops here: A smaller study is not safer if it cannot answer its question.
Try this
Test study design can reduce exposure and improve value with a counterexample. Show how each design feature improves protection or information. Ask what new evidence would change the conclusion. Preserve this boundary: A smaller study is not safer if it cannot answer its question.
A strong science note separates observation from interpretation. Show how each design feature improves protection or information. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Show how each design feature improves protection or information. Keep this boundary visible: A smaller study is not safer if it cannot answer its question.
CHAPTER 15 OF 17 · Protect the study
15. Patient perspective can inform acceptable trade-offs
Patients may value benefits, burdens and uncertainty differently from investigators. Relevant patient input can clarify outcomes and tolerance, but it should be collected and interpreted systematically.
A careful reader can retrace patient perspective can inform acceptable trade-offs from question to method to uncertainty. Patients may value benefits, burdens and uncertainty differently from investigators. Relevant patient input can clarify outcomes and tolerance, but it should be collected and interpreted systematically. Preserve this guardrail: One advocate or small sample does not represent all patients.
Try this
Explain patient perspective can inform acceptable trade-offs to a classmate using claim, evidence, reasoning and limitation. Record whose perspective was obtained, by what method and with what limits. Make sure the official source and version remain visible. Respect this limit: One advocate or small sample does not represent all patients.
A strong science note separates observation from interpretation. Record whose perspective was obtained, by what method and with what limits. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Record whose perspective was obtained, by what method and with what limits. Keep this boundary visible: One advocate or small sample does not represent all patients.
CHAPTER 16 OF 17 · Protect the study
16. Amendments and supplements need updated reasoning
New device versions, events, procedures or cohorts can change the benefit-risk balance. The update should compare old and new evidence and controls rather than reusing the original conclusion unchanged.
For amendments and supplements need updated reasoning, a useful result begins with an explicit population and a visible evidence route. New device versions, events, procedures or cohorts can change the benefit-risk balance. The update should compare old and new evidence and controls rather than reusing the original conclusion unchanged. Keep the boundary beside the conclusion: Past IDE approval does not automatically cover materially changed conditions.
Try this
Build a source-first evidence card for amendments and supplements need updated reasoning. Create a change-impact table for benefit, risk, uncertainty and protection. Add the denominator, date and one plausible source of bias. Keep this limit: Past IDE approval does not automatically cover materially changed conditions.
A strong science note separates observation from interpretation. Create a change-impact table for benefit, risk, uncertainty and protection. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Create a change-impact table for benefit, risk, uncertainty and protection. Keep this boundary visible: Past IDE approval does not automatically cover materially changed conditions.
CHAPTER 17 OF 17 · Protect the study
17. A complete IDE benefit-risk receipt preserves the decision
The receipt records study purpose, stage, population, benefits, knowledge value, alternatives, risks, uncertainties, nonclinical evidence, controls, consent, monitoring, stopping, patient input, changes and residual questions.
The science in a complete ide benefit-risk receipt preserves the decision is strongest when design, data and interpretation stay connected. The receipt records study purpose, stage, population, benefits, knowledge value, alternatives, risks, uncertainties, nonclinical evidence, controls, consent, monitoring, stopping, patient input, changes and residual questions. The result stops here: The receipt explains the decision; it does not guarantee favourable outcomes.
Try this
Test a complete ide benefit-risk receipt preserves the decision with a counterexample. Ask an independent reviewer whether each residual risk has a proportionate safeguard. Ask what new evidence would change the conclusion. Preserve this boundary: The receipt explains the decision; it does not guarantee favourable outcomes.
A strong science note separates observation from interpretation. Ask an independent reviewer whether each residual risk has a proportionate safeguard. Record the source, version or date, population or device context, method, result and one uncertainty. If the official record does not show a field, write “not shown” instead of filling the gap from memory.
Evidence checkpoint
Ask an independent reviewer whether each residual risk has a proportionate safeguard. Keep this boundary visible: The receipt explains the decision; it does not guarantee favourable outcomes.
IDE benefit-risk evidence map
| Question | Evidence to capture | Boundary to retain |
|---|---|---|
| What may be gained? | Direct benefit, probability, duration and knowledge value | Study value is not personal benefit |
| What may go wrong? | Severity, probability, reversibility and uncertainty | Unknown is not zero |
| How are risks reduced? | Nonclinical evidence, design controls, monitoring and stopping | Controls need owners and triggers |
| Why proceed now? | Need, alternatives, study value and participant perspective | IDE approval is not marketing authorisation |
Science learning becomes useful when a student can show where a claim came from, why it applies and where it stops. Continue with the eduKate Science Learning Hub, compare choices in the Education Hub, and connect these habits to career pathways built from subject capabilities.
A responsible IDE decision is optimistic without being careless. It allows learning when the question matters, avoidable risks have been reduced and every participant is treated as a person whose welfare comes before the dataset.
Primary sources and verification date
Source status checked 25 October 2026. FDA guidance describes the agency’s current thinking and does not replace applicable statutes, regulations or product-specific communication with FDA.
- FDA: Factors to Consider When Making Benefit-Risk Determinations for Medical Device IDEs — Final guidance, January 2017.
- FDA: Final guidance PDF — Official detailed framework.
- FDA: Investigational Device Exemption overview — Current IDE purpose and pathway context.