eduKateSG · Why Science?
Use what is already known without pretending children are small adults
Follow FDA’s June 2016 final guidance through similarity, reference populations, device use, disease course, partial borrowing, pediatric data, uncertainty and labeling.
Children need devices sized, studied and labelled for them, yet pediatric trials can be difficult when conditions are rare or populations small. Extrapolation may use relevant existing clinical data to reduce unnecessary repetition while still addressing pediatric-specific uncertainty.
FDA’s June 2016 final guidance provides a roadmap for leveraging existing data in PMA and De Novo requests and HDE approvals for pediatric uses. It explains when extrapolation may be appropriate and statistical methods that can increase precision.
Did you know? Extrapolation is not copying an adult conclusion. It is a structured comparison of disease, device effects, use conditions and remaining uncertainty, followed by collection of pediatric evidence where needed.
For students, this is first-principles science: identify what is shared, what differs and how much information can travel. The reasoning supports future careers in pediatrics, biomedical engineering, statistics and regulation.
Chapter index
Chapters 1–9: pediatric question, reference data and similarity assessment
- 1. Pediatric extrapolation uses a reference population
- 2. The June 2016 device guidance is final
- 3. The pediatric need should be explicit
- 4. Disease similarity supports transport
- 5. Device effect similarity is a separate question
- 6. Use conditions may change in children
- 7. Growth can create unique long-term risks
- 8. Existing data must be relevant and reliable
- 9. A similarity assessment should be explicit
Chapters 10–17: statistical borrowing, residual data, uncertainty and communication
- 10. Full extrapolation still needs a justified model
- 11. Partial borrowing can moderate uncertainty
- 12. Prior-data conflict needs a response
- 13. Pediatric safety often needs direct evidence
- 14. Endpoint meaning should be pediatric-relevant
- 15. Sample size reflects borrowed and new information
- 16. Labels should state the evidence basis
- 17. A complete extrapolation receipt preserves the logic
CHAPTER 1 OF 17 · Define the pediatric need
1. Pediatric extrapolation uses a reference population
Existing data may come from adults or another pediatric subgroup and inform inference for a target pediatric population. The proposal should name both populations, the device use and the exact safety or effectiveness question.
For pediatric extrapolation uses a reference population, a useful result begins with an explicit population and a visible evidence route. Existing data may come from adults or another pediatric subgroup and inform inference for a target pediatric population. The proposal should name both populations, the device use and the exact safety or effectiveness question. Keep the boundary beside the conclusion: Data do not transfer merely because the device name is the same.
Try this
Build a source-first evidence card for pediatric extrapolation uses a reference population. Write reference population, target population, estimand and regulatory purpose. Add the denominator, date and one plausible source of bias. Keep this limit: Data do not transfer merely because the device name is the same.
A strong science note separates observation from interpretation. Write reference population, target population, estimand and regulatory 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
Write reference population, target population, estimand and regulatory purpose. Keep this boundary visible: Data do not transfer merely because the device name is the same.
CHAPTER 2 OF 17 · Define the pediatric need
2. The June 2016 device guidance is final
FDA’s final guidance is dated June 2016 under docket FDA-2015-D-1376. It addresses PMA, De Novo and HDE contexts for pediatric device uses and statistical leveraging.
The science in the june 2016 device guidance is final is strongest when design, data and interpretation stay connected. FDA’s final guidance is dated June 2016 under docket FDA-2015-D-1376. It addresses PMA, De Novo and HDE contexts for pediatric device uses and statistical leveraging. The result stops here: The framework is device-specific and should not be confused with drug extrapolation guidance.
Try this
Test the june 2016 device guidance is final with a counterexample. Create a source receipt with scope, pathways, date, status and docket. Ask what new evidence would change the conclusion. Preserve this boundary: The framework is device-specific and should not be confused with drug extrapolation guidance.
A strong science note separates observation from interpretation. Create a source receipt with scope, pathways, date, status and docket. 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 source receipt with scope, pathways, date, status and docket. Keep this boundary visible: The framework is device-specific and should not be confused with drug extrapolation guidance.
The submission should define the pediatric condition, age range, current options, unmet need and proposed device role. Age ranges may need subdivision when anatomy, development or care setting changes materially.
A careful reader can retrace the pediatric need should be explicit from question to method to uncertainty. The submission should define the pediatric condition, age range, current options, unmet need and proposed device role. Age ranges may need subdivision when anatomy, development or care setting changes materially. Preserve this guardrail: A broad label such as pediatric can hide important developmental stages.
Try this
Explain the pediatric need should be explicit to a classmate using claim, evidence, reasoning and limitation. Map age or developmental groups to the clinical decision and device use. Make sure the official source and version remain visible. Respect this limit: A broad label such as pediatric can hide important developmental stages.
A strong science note separates observation from interpretation. Map age or developmental groups to the clinical decision and device use. 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 age or developmental groups to the clinical decision and device use. Keep this boundary visible: A broad label such as pediatric can hide important developmental stages.
Extrapolation is more credible when disease course, relevant pathophysiology and outcome relationships are sufficiently similar between reference and target populations. Differences should be described rather than averaged away.
For disease similarity supports transport, a useful result begins with an explicit population and a visible evidence route. Extrapolation is more credible when disease course, relevant pathophysiology and outcome relationships are sufficiently similar between reference and target populations. Differences should be described rather than averaged away. Keep the boundary beside the conclusion: Similar diagnosis does not guarantee similar natural history or endpoint meaning.
Try this
Build a source-first evidence card for disease similarity supports transport. Compare disease mechanism, progression, prognosis and treatment context. Add the denominator, date and one plausible source of bias. Keep this limit: Similar diagnosis does not guarantee similar natural history or endpoint meaning.
A strong science note separates observation from interpretation. Compare disease mechanism, progression, prognosis and treatment context. 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 disease mechanism, progression, prognosis and treatment context. Keep this boundary visible: Similar diagnosis does not guarantee similar natural history or endpoint meaning.
A device may act through a physical or physiological mechanism whose effect can differ with body size, growth, tissue properties or behaviour. Engineering and clinical evidence should support the proposed similarity.
The science in device effect similarity is a separate question is strongest when design, data and interpretation stay connected. A device may act through a physical or physiological mechanism whose effect can differ with body size, growth, tissue properties or behaviour. Engineering and clinical evidence should support the proposed similarity. The result stops here: Mechanistic plausibility is not the same as demonstrated clinical performance.
Try this
Test device effect similarity is a separate question with a counterexample. Trace the device mechanism through pediatric anatomy and expected outcome. Ask what new evidence would change the conclusion. Preserve this boundary: Mechanistic plausibility is not the same as demonstrated clinical performance.
A strong science note separates observation from interpretation. Trace the device mechanism through pediatric anatomy and expected outcome. 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
Trace the device mechanism through pediatric anatomy and expected outcome. Keep this boundary visible: Mechanistic plausibility is not the same as demonstrated clinical performance.
Implant technique, caregiver involvement, adherence, activity, communication and clinical setting can affect exposure and outcome. Human-factors and training evidence may be needed even when core device mechanics are similar.
A careful reader can retrace use conditions may change in children from question to method to uncertainty. Implant technique, caregiver involvement, adherence, activity, communication and clinical setting can affect exposure and outcome. Human-factors and training evidence may be needed even when core device mechanics are similar. Preserve this guardrail: Shared mechanism does not guarantee shared use performance.
Try this
Explain use conditions may change in children to a classmate using claim, evidence, reasoning and limitation. Create a task analysis for child, caregiver and clinician roles. Make sure the official source and version remain visible. Respect this limit: Shared mechanism does not guarantee shared use performance.
A strong science note separates observation from interpretation. Create a task analysis for child, caregiver and clinician roles. 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 task analysis for child, caregiver and clinician roles. Keep this boundary visible: Shared mechanism does not guarantee shared use performance.
A device may need resizing, replacement or adjustment as the child grows. Durability, migration and developmental effects may require follow-up beyond the reference dataset.
For growth can create unique long-term risks, a useful result begins with an explicit population and a visible evidence route. A device may need resizing, replacement or adjustment as the child grows. Durability, migration and developmental effects may require follow-up beyond the reference dataset. Keep the boundary beside the conclusion: Short-term similarity cannot establish safety through growth.
Try this
Build a source-first evidence card for growth can create unique long-term risks. Map growth-related failure modes to monitoring intervals. Add the denominator, date and one plausible source of bias. Keep this limit: Short-term similarity cannot establish safety through growth.
A strong science note separates observation from interpretation. Map growth-related failure modes to monitoring intervals. 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 growth-related failure modes to monitoring intervals. Keep this boundary visible: Short-term similarity cannot establish safety through growth.
Reference evidence should be traceable, of suitable quality and aligned with endpoints, device version and use. Published summaries alone may not support detailed borrowing.
The science in existing data must be relevant and reliable is strongest when design, data and interpretation stay connected. Reference evidence should be traceable, of suitable quality and aligned with endpoints, device version and use. Published summaries alone may not support detailed borrowing. The result stops here: More data do not help when key variables or versions differ.
Try this
Test existing data must be relevant and reliable with a counterexample. Build a reference-data receipt with source, population, device, endpoint and quality. Ask what new evidence would change the conclusion. Preserve this boundary: More data do not help when key variables or versions differ.
A strong science note separates observation from interpretation. Build a reference-data receipt with source, population, device, endpoint and quality. 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 reference-data receipt with source, population, device, endpoint and quality. Keep this boundary visible: More data do not help when key variables or versions differ.
Evidence should identify which features are sufficiently similar, which are uncertain and which differ. The degree of extrapolation follows this assessment rather than an all-or-nothing label.
A careful reader can retrace a similarity assessment should be explicit from question to method to uncertainty. Evidence should identify which features are sufficiently similar, which are uncertain and which differ. The degree of extrapolation follows this assessment rather than an all-or-nothing label. Preserve this guardrail: Similarity in one outcome does not establish similarity in every safety endpoint.
Try this
Explain a similarity assessment should be explicit to a classmate using claim, evidence, reasoning and limitation. Create a matrix of evidence for each effectiveness and safety component. Make sure the official source and version remain visible. Respect this limit: Similarity in one outcome does not establish similarity in every safety endpoint.
A strong science note separates observation from interpretation. Create a matrix of evidence for each effectiveness and safety component. 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 matrix of evidence for each effectiveness and safety component. Keep this boundary visible: Similarity in one outcome does not establish similarity in every safety endpoint.
CHAPTER 10 OF 17 · Borrow with evidence
10. Full extrapolation still needs a justified model
When existing data strongly support the pediatric effect, fewer new pediatric observations may be needed for that component, but the statistical model and assumptions should be shown.
For full extrapolation still needs a justified model, a useful result begins with an explicit population and a visible evidence route. When existing data strongly support the pediatric effect, fewer new pediatric observations may be needed for that component, but the statistical model and assumptions should be shown. Keep the boundary beside the conclusion: Full borrowing is conditional on strong similarity assumptions.
Try this
Build a source-first evidence card for full extrapolation still needs a justified model. Simulate the pediatric inference under correct and incorrect similarity. Add the denominator, date and one plausible source of bias. Keep this limit: Full borrowing is conditional on strong similarity assumptions.
A strong science note separates observation from interpretation. Simulate the pediatric inference under correct and incorrect similarity. 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
Simulate the pediatric inference under correct and incorrect similarity. Keep this boundary visible: Full borrowing is conditional on strong similarity assumptions.
CHAPTER 11 OF 17 · Borrow with evidence
11. Partial borrowing can moderate uncertainty
Hierarchical, Bayesian or other methods may combine reference and pediatric data while limiting how much information transfers. Dynamic borrowing can respond to agreement, but its behaviour should be evaluated prospectively.
The science in partial borrowing can moderate uncertainty is strongest when design, data and interpretation stay connected. Hierarchical, Bayesian or other methods may combine reference and pediatric data while limiting how much information transfers. Dynamic borrowing can respond to agreement, but its behaviour should be evaluated prospectively. The result stops here: A complex borrowing method adds assumptions and cannot discover truth automatically.
Try this
Test partial borrowing can moderate uncertainty with a counterexample. Plot effective borrowed information across agreement and conflict scenarios. Ask what new evidence would change the conclusion. Preserve this boundary: A complex borrowing method adds assumptions and cannot discover truth automatically.
A strong science note separates observation from interpretation. Plot effective borrowed information across agreement and conflict 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
Plot effective borrowed information across agreement and conflict scenarios. Keep this boundary visible: A complex borrowing method adds assumptions and cannot discover truth automatically.
Pediatric results may disagree with the reference population because of real differences, chance or bias. Robust models, reduced borrowing and sensitivity analyses can show how conclusions change.
A careful reader can retrace prior-data conflict needs a response from question to method to uncertainty. Pediatric results may disagree with the reference population because of real differences, chance or bias. Robust models, reduced borrowing and sensitivity analyses can show how conclusions change. Preserve this guardrail: Forcing agreement can hide the very pediatric difference the study should detect.
Try this
Explain prior-data conflict needs a response to a classmate using claim, evidence, reasoning and limitation. Predefine a conflict metric and the action it triggers. Make sure the official source and version remain visible. Respect this limit: Forcing agreement can hide the very pediatric difference the study should detect.
A strong science note separates observation from interpretation. Predefine a conflict metric and the action it triggers. 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
Predefine a conflict metric and the action it triggers. Keep this boundary visible: Forcing agreement can hide the very pediatric difference the study should detect.
CHAPTER 13 OF 17 · Close remaining gaps
13. Pediatric safety often needs direct evidence
Rare events, growth, development and child-specific use may not be inferable from adult data. The plan should identify which risks need pediatric observation and why.
For pediatric safety often needs direct evidence, a useful result begins with an explicit population and a visible evidence route. Rare events, growth, development and child-specific use may not be inferable from adult data. The plan should identify which risks need pediatric observation and why. Keep the boundary beside the conclusion: Effectiveness similarity does not establish safety similarity.
Try this
Build a source-first evidence card for pediatric safety often needs direct evidence. Separate borrowed effectiveness components from directly observed safety components. Add the denominator, date and one plausible source of bias. Keep this limit: Effectiveness similarity does not establish safety similarity.
A strong science note separates observation from interpretation. Separate borrowed effectiveness components from directly observed safety components. 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
Separate borrowed effectiveness components from directly observed safety components. Keep this boundary visible: Effectiveness similarity does not establish safety similarity.
CHAPTER 14 OF 17 · Close remaining gaps
14. Endpoint meaning should be pediatric-relevant
Measurements should be valid for developmental stage, communication ability and clinical goals. A scale or threshold validated in adults may not carry the same meaning for children.
The science in endpoint meaning should be pediatric-relevant is strongest when design, data and interpretation stay connected. Measurements should be valid for developmental stage, communication ability and clinical goals. A scale or threshold validated in adults may not carry the same meaning for children. The result stops here: Numerically identical scores can represent different function across ages.
Try this
Test endpoint meaning should be pediatric-relevant with a counterexample. Document endpoint validity and meaningful change in the target group. Ask what new evidence would change the conclusion. Preserve this boundary: Numerically identical scores can represent different function across ages.
A strong science note separates observation from interpretation. Document endpoint validity and meaningful change in the target group. 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
Document endpoint validity and meaningful change in the target group. Keep this boundary visible: Numerically identical scores can represent different function across ages.
CHAPTER 15 OF 17 · Close remaining gaps
15. Sample size reflects borrowed and new information
The design should show how reference data affect precision and how much pediatric data remain necessary. Simulations can examine bias, coverage and decision error when similarity assumptions fail.
A careful reader can retrace sample size reflects borrowed and new information from question to method to uncertainty. The design should show how reference data affect precision and how much pediatric data remain necessary. Simulations can examine bias, coverage and decision error when similarity assumptions fail. Preserve this guardrail: A smaller sample is justified by information, not by convenience alone.
Try this
Explain sample size reflects borrowed and new information to a classmate using claim, evidence, reasoning and limitation. Compare operating characteristics under no, partial and full borrowing. Make sure the official source and version remain visible. Respect this limit: A smaller sample is justified by information, not by convenience alone.
A strong science note separates observation from interpretation. Compare operating characteristics under no, partial and full borrowing. 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 operating characteristics under no, partial and full borrowing. Keep this boundary visible: A smaller sample is justified by information, not by convenience alone.
Pediatric indications and instructions should align with ages, sizes, configurations and evidence actually evaluated. Limitations and required monitoring should be clear to clinicians and caregivers.
For labels should state the evidence basis, a useful result begins with an explicit population and a visible evidence route. Pediatric indications and instructions should align with ages, sizes, configurations and evidence actually evaluated. Limitations and required monitoring should be clear to clinicians and caregivers. Keep the boundary beside the conclusion: Extrapolated evidence should not imply direct observation where none occurred.
Try this
Build a source-first evidence card for labels should state the evidence basis. Draft a label statement distinguishing borrowed and pediatric data. Add the denominator, date and one plausible source of bias. Keep this limit: Extrapolated evidence should not imply direct observation where none occurred.
A strong science note separates observation from interpretation. Draft a label statement distinguishing borrowed and pediatric data. 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
Draft a label statement distinguishing borrowed and pediatric data. Keep this boundary visible: Extrapolated evidence should not imply direct observation where none occurred.
CHAPTER 17 OF 17 · Close remaining gaps
17. A complete extrapolation receipt preserves the logic
The receipt records pediatric need, target and reference populations, disease and device similarity, use conditions, device versions, endpoints, reference-data quality, borrowing model, pediatric sample, conflict checks, safety evidence, uncertainty and labeling.
The science in a complete extrapolation receipt preserves the logic is strongest when design, data and interpretation stay connected. The receipt records pediatric need, target and reference populations, disease and device similarity, use conditions, device versions, endpoints, reference-data quality, borrowing model, pediatric sample, conflict checks, safety evidence, uncertainty and labeling. The result stops here: The receipt clarifies assumptions; it does not remove pediatric evidence gaps.
Try this
Test a complete extrapolation receipt preserves the logic with a counterexample. Ask an independent reviewer to identify every assumption required for transfer. Ask what new evidence would change the conclusion. Preserve this boundary: The receipt clarifies assumptions; it does not remove pediatric evidence gaps.
A strong science note separates observation from interpretation. Ask an independent reviewer to identify every assumption required for transfer. 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 to identify every assumption required for transfer. Keep this boundary visible: The receipt clarifies assumptions; it does not remove pediatric evidence gaps.
Pediatric extrapolation evidence map
| Question | Evidence to capture | Boundary to retain |
|---|---|---|
| What is the target question? | Age range, condition, device use, outcome and pathway | Pediatric is not one homogeneous group |
| What can travel? | Disease, mechanism, use and endpoint similarity | Similarity is component-specific |
| How is information combined? | Reference quality, borrowing model and conflict checks | Borrowing adds assumptions |
| What must still be observed? | Pediatric safety, growth, use and labeling evidence | Adult effectiveness does not prove pediatric safety |
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.
Pediatric extrapolation is careful reuse: it avoids asking children to repeat questions already answered while insisting on new evidence for what is truly different. That balance is both scientifically efficient and humane.
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: Leveraging Existing Clinical Data for Extrapolation to Pediatric Uses of Medical Devices — Final guidance, June 2016.
- FDA: Full final guidance PDF — Official framework and statistical methods.
- FDA: Pediatric Medical Devices — Current pediatric-device resources, updated July 2026.