eduKateSG · Why Science?
Design the decisive device study around the claim it must support
Read FDA’s November 2013 final guidance through study objectives, controls, randomisation, blinding, endpoints, estimands, missing data and clinically meaningful interpretation.
A pivotal clinical investigation is designed to provide the principal evidence for a medical-device premarket decision. Its strength comes from alignment: the intended-use claim, participant population, comparator, endpoints and analysis should answer the same question.
FDA’s November 2013 final guidance describes design principles for pivotal medical-device investigations intended to meet premarket clinical-data requirements. It does not prescribe one universal trial design and is not a comprehensive statistics textbook; the suitable design depends on the device and regulatory question.
Did you know? Randomisation is powerful, yet FDA’s guidance discusses several possible control structures. The scientific task is to justify why the chosen comparison can separate device effect from natural history, placebo effects, concurrent care, operator learning and other explanations.
For students, pivotal-trial design is applied reasoning: define the question before choosing the test, protect measurements from bias and interpret statistical results through clinical meaning. Those skills connect science education to medicine, engineering and biostatistics careers.
Chapter index
Chapters 1–9: regulatory objective, population, controls and allocation
- 1. A pivotal study is built for a premarket decision
- 2. The November 2013 guidance is final
- 3. Objectives should separate safety and effectiveness questions
- 4. The target population should match intended use
- 5. Concurrent randomised controls often strengthen causal inference
- 6. Other control choices require a strong rationale
- 7. Sham controls can isolate procedural effects
- 8. Blinding protects assessments and behaviour
- 9. Operator learning can change outcomes over time
Chapters 10–17: endpoints, analysis, missing data and clinical meaning
- 10. Endpoints should measure a clinically meaningful concept
- 11. Safety endpoints need exposure and time context
- 12. Sample size follows effect, variability and decision rule
- 13. Multiplicity needs a prospectively controlled plan
- 14. Missing data can change the treatment comparison
- 15. Protocol deviations should be visible, not erased
- 16. Statistical significance needs clinical interpretation
- 17. A complete pivotal-study receipt reconstructs the claim
CHAPTER 1 OF 17 · Define the claim
1. A pivotal study is built for a premarket decision
FDA’s guidance concerns clinical investigations intended to support medical-device premarket submissions. The study should connect a regulatory objective and device claim to a design capable of producing valid scientific evidence. “Pivotal” describes evidentiary role, not simply size, cost or late timing.
For a pivotal study is built for a premarket decision, the design should make the final claim answerable. FDA’s guidance concerns clinical investigations intended to support medical-device premarket submissions. The study should connect a regulatory objective and device claim to a design capable of producing valid scientific evidence. “Pivotal” describes evidentiary role, not simply size, cost or late timing. Keep this limit beside the conclusion: A large study can still be poorly aligned with the intended claim.
Try this
Turn a pivotal study is built for a premarket decision into a design table. Write the exact decision and claim, then identify the one result most necessary to support them. Add the estimand, analysis population and missing-data rule. Preserve this limit: A large study can still be poorly aligned with the intended claim.
A strong science note separates observation from interpretation. Write the exact decision and claim, then identify the one result most necessary to support them. 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 the exact decision and claim, then identify the one result most necessary to support them. Keep this boundary visible: A large study can still be poorly aligned with the intended claim.
The final guidance is dated November 2013 and identified by docket FDA-2011-D-0567. It addresses industry, clinical investigators, IRBs and FDA staff. FDA explains that design should be selected for the particular device while evidentiary burden is tailored to the regulatory requirement.
A pivotal investigation is a chain from question to comparison to endpoint. In the november 2013 guidance is final, The final guidance is dated November 2013 and identified by docket FDA-2011-D-0567. It addresses industry, clinical investigators, IRBs and FDA staff. FDA explains that design should be selected for the particular device while evidentiary burden is tailored to the regulatory requirement. Respect this boundary: The document gives principles, not product-specific approval criteria.
Try this
Challenge the november 2013 guidance is final with a rival explanation. Create a source receipt with title, date, docket, audience and stated limitation. Identify the design feature that blocks or measures it. Keep this boundary: The document gives principles, not product-specific approval criteria.
A strong science note separates observation from interpretation. Create a source receipt with title, date, docket, audience and stated limitation. 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 title, date, docket, audience and stated limitation. Keep this boundary visible: The document gives principles, not product-specific approval criteria.
CHAPTER 3 OF 17 · Define the claim
3. Objectives should separate safety and effectiveness questions
A protocol may have several objectives, but primary objectives should be explicit and tied to endpoints and decision criteria. Safety and effectiveness may require different time windows, populations or analyses. Secondary and exploratory questions can add context without quietly replacing a failed primary objective.
The strongest feature of objectives should separate safety and effectiveness questions is clarity about what would count as convincing evidence. A protocol may have several objectives, but primary objectives should be explicit and tied to endpoints and decision criteria. Safety and effectiveness may require different time windows, populations or analyses. Secondary and exploratory questions can add context without quietly replacing a failed primary objective. The conclusion stops here: Success on an exploratory outcome does not rescue an unmet primary objective.
Try this
Trace objectives should separate safety and effectiveness questions from claim to endpoint. Make a hierarchy of primary, secondary and exploratory objectives and link each to an analysis. Ask whether a clinically important effect would be detected and interpreted. Respect this limit: Success on an exploratory outcome does not rescue an unmet primary objective.
A strong science note separates observation from interpretation. Make a hierarchy of primary, secondary and exploratory objectives and link each to an analysis. 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
Make a hierarchy of primary, secondary and exploratory objectives and link each to an analysis. Keep this boundary visible: Success on an exploratory outcome does not rescue an unmet primary objective.
CHAPTER 4 OF 17 · Define the claim
4. The target population should match intended use
Eligibility criteria translate intended use into a study population. Criteria should create an interpretable study while representing the people for whom the device claim is sought. Excessive exclusions can produce clean data that do not travel to practice; broad entry can add heterogeneity and risk.
For the target population should match intended use, the design should make the final claim answerable. Eligibility criteria translate intended use into a study population. Criteria should create an interpretable study while representing the people for whom the device claim is sought. Excessive exclusions can produce clean data that do not travel to practice; broad entry can add heterogeneity and risk. Keep this limit beside the conclusion: Study eligibility and final labelling population are related but not automatically identical.
Try this
Turn the target population should match intended use into a design table. Map each intended-use element to eligibility, stratification or a planned subgroup analysis. Add the estimand, analysis population and missing-data rule. Preserve this limit: Study eligibility and final labelling population are related but not automatically identical.
A strong science note separates observation from interpretation. Map each intended-use element to eligibility, stratification or a planned subgroup analysis. 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 intended-use element to eligibility, stratification or a planned subgroup analysis. Keep this boundary visible: Study eligibility and final labelling population are related but not automatically identical.
CHAPTER 5 OF 17 · Choose comparisons
5. Concurrent randomised controls often strengthen causal inference
Random allocation can balance known and unknown prognostic factors on average, while a concurrent control experiences the same calendar time and care environment. Device trials may face practical or ethical constraints, but those constraints should be described and alternative bias controls justified.
A pivotal investigation is a chain from question to comparison to endpoint. In concurrent randomised controls often strengthen causal inference, Random allocation can balance known and unknown prognostic factors on average, while a concurrent control experiences the same calendar time and care environment. Device trials may face practical or ethical constraints, but those constraints should be described and alternative bias controls justified. Respect this boundary: Randomisation does not prevent bias from missing outcomes, unblinding or protocol deviations.
Try this
Challenge concurrent randomised controls often strengthen causal inference with a rival explanation. List the major rival explanations and mark which randomisation addresses and which remain. Identify the design feature that blocks or measures it. Keep this boundary: Randomisation does not prevent bias from missing outcomes, unblinding or protocol deviations.
A strong science note separates observation from interpretation. List the major rival explanations and mark which randomisation addresses and which remain. 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
List the major rival explanations and mark which randomisation addresses and which remain. Keep this boundary visible: Randomisation does not prevent bias from missing outcomes, unblinding or protocol deviations.
CHAPTER 6 OF 17 · Choose comparisons
6. Other control choices require a strong rationale
Possible comparisons may include active treatment, sham, performance goal, objective performance criterion or historical information, depending on context. Each carries assumptions about comparability, constancy and measurement. The control should let the study estimate the effect relevant to the regulatory claim.
The strongest feature of other control choices require a strong rationale is clarity about what would count as convincing evidence. Possible comparisons may include active treatment, sham, performance goal, objective performance criterion or historical information, depending on context. Each carries assumptions about comparability, constancy and measurement. The control should let the study estimate the effect relevant to the regulatory claim. The conclusion stops here: Historical or performance benchmarks can fail when practice or populations change.
Try this
Trace other control choices require a strong rationale from claim to endpoint. Create a control-choice table with ethical feasibility, bias risks, data availability and interpretability. Ask whether a clinically important effect would be detected and interpreted. Respect this limit: Historical or performance benchmarks can fail when practice or populations change.
A strong science note separates observation from interpretation. Create a control-choice table with ethical feasibility, bias risks, data availability and interpretability. 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 control-choice table with ethical feasibility, bias risks, data availability and interpretability. Keep this boundary visible: Historical or performance benchmarks can fail when practice or populations change.
For some procedures, a sham control may separate device-specific effects from attention, expectations and the procedure itself. Its scientific value must be weighed against risks and burdens to control participants. Consent, rescue options and minimisation of unnecessary procedures are central ethical considerations.
For sham controls can isolate procedural effects, the design should make the final claim answerable. For some procedures, a sham control may separate device-specific effects from attention, expectations and the procedure itself. Its scientific value must be weighed against risks and burdens to control participants. Consent, rescue options and minimisation of unnecessary procedures are central ethical considerations. Keep this limit beside the conclusion: Scientific informativeness does not automatically justify a risky sham.
Try this
Turn sham controls can isolate procedural effects into a design table. Write the causal question a sham answers and the safeguards required to make it ethically supportable. Add the estimand, analysis population and missing-data rule. Preserve this limit: Scientific informativeness does not automatically justify a risky sham.
A strong science note separates observation from interpretation. Write the causal question a sham answers and the safeguards required to make it ethically supportable. 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 the causal question a sham answers and the safeguards required to make it ethically supportable. Keep this boundary visible: Scientific informativeness does not automatically justify a risky sham.
Masking participants, investigators or outcome assessors can reduce expectation and measurement bias, though complete blinding may be difficult for visible or procedural devices. When blinding is impossible, objective endpoints, independent adjudication, standardised care and scripted assessments can help. The report should evaluate whether masking was maintained.
A pivotal investigation is a chain from question to comparison to endpoint. In blinding protects assessments and behaviour, Masking participants, investigators or outcome assessors can reduce expectation and measurement bias, though complete blinding may be difficult for visible or procedural devices. When blinding is impossible, objective endpoints, independent adjudication, standardised care and scripted assessments can help. The report should evaluate whether masking was maintained. Respect this boundary: Calling an endpoint objective does not guarantee that care or ascertainment was unbiased.
Try this
Challenge blinding protects assessments and behaviour with a rival explanation. List who can be blinded and one compensating control for each role that cannot. Identify the design feature that blocks or measures it. Keep this boundary: Calling an endpoint objective does not guarantee that care or ascertainment was unbiased.
A strong science note separates observation from interpretation. List who can be blinded and one compensating control for each role that cannot. 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
List who can be blinded and one compensating control for each role that cannot. Keep this boundary visible: Calling an endpoint objective does not guarantee that care or ascertainment was unbiased.
CHAPTER 9 OF 17 · Protect measurement
9. Operator learning can change outcomes over time
Device performance may depend on training and experience. Early cases can have different outcomes from later cases, and centres may progress at different rates. The design can standardise training, require proficiency, track experience and analyse learning effects while preserving the intended-use setting.
The strongest feature of operator learning can change outcomes over time is clarity about what would count as convincing evidence. Device performance may depend on training and experience. Early cases can have different outcomes from later cases, and centres may progress at different rates. The design can standardise training, require proficiency, track experience and analyse learning effects while preserving the intended-use setting. The conclusion stops here: Excluding all early cases after seeing outcomes can bias the result.
Try this
Trace operator learning can change outcomes over time from claim to endpoint. Predefine an experience metric and how learning will be described or incorporated in analysis. Ask whether a clinically important effect would be detected and interpreted. Respect this limit: Excluding all early cases after seeing outcomes can bias the result.
A strong science note separates observation from interpretation. Predefine an experience metric and how learning will be described or incorporated in analysis. 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 an experience metric and how learning will be described or incorporated in analysis. Keep this boundary visible: Excluding all early cases after seeing outcomes can bias the result.
CHAPTER 10 OF 17 · Protect measurement
10. Endpoints should measure a clinically meaningful concept
An endpoint should correspond to how a person feels, functions or survives, or to another valid clinical or performance concept relevant to the device claim. Composite and surrogate endpoints need clear justification. Definitions, assessment windows and adjudication rules should be set before enrolment.
For endpoints should measure a clinically meaningful concept, the design should make the final claim answerable. An endpoint should correspond to how a person feels, functions or survives, or to another valid clinical or performance concept relevant to the device claim. Composite and surrogate endpoints need clear justification. Definitions, assessment windows and adjudication rules should be set before enrolment. Keep this limit beside the conclusion: A statistically convenient measure may not demonstrate meaningful patient benefit.
Try this
Turn endpoints should measure a clinically meaningful concept into a design table. Write the clinical concept, measurement method, time point and meaningful-difference rationale. Add the estimand, analysis population and missing-data rule. Preserve this limit: A statistically convenient measure may not demonstrate meaningful patient benefit.
A strong science note separates observation from interpretation. Write the clinical concept, measurement method, time point and meaningful-difference rationale. 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 the clinical concept, measurement method, time point and meaningful-difference rationale. Keep this boundary visible: A statistically convenient measure may not demonstrate meaningful patient benefit.
CHAPTER 11 OF 17 · Protect measurement
11. Safety endpoints need exposure and time context
Adverse events should be defined, collected and analysed with denominators, device exposure and follow-up. Procedure-related, device-related and disease-related events may overlap, so attribution methods should be transparent. Rare or late harms may require evidence beyond the pivotal study.
A pivotal investigation is a chain from question to comparison to endpoint. In safety endpoints need exposure and time context, Adverse events should be defined, collected and analysed with denominators, device exposure and follow-up. Procedure-related, device-related and disease-related events may overlap, so attribution methods should be transparent. Rare or late harms may require evidence beyond the pivotal study. Respect this boundary: Absence of a rare event in a finite trial does not prove zero risk.
Try this
Challenge safety endpoints need exposure and time context with a rival explanation. Create a safety table with event definition, risk window, adjudication and person-time or participant denominator. Identify the design feature that blocks or measures it. Keep this boundary: Absence of a rare event in a finite trial does not prove zero risk.
A strong science note separates observation from interpretation. Create a safety table with event definition, risk window, adjudication and person-time or participant denominator. 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 safety table with event definition, risk window, adjudication and person-time or participant denominator. Keep this boundary visible: Absence of a rare event in a finite trial does not prove zero risk.
CHAPTER 12 OF 17 · Protect measurement
12. Sample size follows effect, variability and decision rule
The sample-size calculation should identify the primary hypothesis or precision target, expected control performance, clinically important difference, variability, error rates, attrition and any multiplicity or interim looks. Assumptions should be supported and sensitivity-tested because optimistic inputs can underpower the study.
The strongest feature of sample size follows effect, variability and decision rule is clarity about what would count as convincing evidence. The sample-size calculation should identify the primary hypothesis or precision target, expected control performance, clinically important difference, variability, error rates, attrition and any multiplicity or interim looks. Assumptions should be supported and sensitivity-tested because optimistic inputs can underpower the study. The conclusion stops here: A precise calculation is only as credible as its assumptions.
Try this
Trace sample size follows effect, variability and decision rule from claim to endpoint. Recalculate sample size under a worse control rate, higher variability and greater attrition. Ask whether a clinically important effect would be detected and interpreted. Respect this limit: A precise calculation is only as credible as its assumptions.
A strong science note separates observation from interpretation. Recalculate sample size under a worse control rate, higher variability and greater attrition. 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
Recalculate sample size under a worse control rate, higher variability and greater attrition. Keep this boundary visible: A precise calculation is only as credible as its assumptions.
CHAPTER 13 OF 17 · Interpret the result
13. Multiplicity needs a prospectively controlled plan
Multiple primary endpoints, populations, time points or analyses create more chances for a favourable finding. A hierarchical testing strategy, adjustment or explicit success rule can control the overall error according to the study objective. Exploratory analyses should be labelled rather than blended with confirmatory claims.
For multiplicity needs a prospectively controlled plan, the design should make the final claim answerable. Multiple primary endpoints, populations, time points or analyses create more chances for a favourable finding. A hierarchical testing strategy, adjustment or explicit success rule can control the overall error according to the study objective. Exploratory analyses should be labelled rather than blended with confirmatory claims. Keep this limit beside the conclusion: Many plausible analyses do not create many independent confirmations.
Try this
Turn multiplicity needs a prospectively controlled plan into a design table. Draw the testing sequence and show what happens after each success or failure. Add the estimand, analysis population and missing-data rule. Preserve this limit: Many plausible analyses do not create many independent confirmations.
A strong science note separates observation from interpretation. Draw the testing sequence and show what happens after each success or failure. 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
Draw the testing sequence and show what happens after each success or failure. Keep this boundary visible: Many plausible analyses do not create many independent confirmations.
CHAPTER 14 OF 17 · Interpret the result
14. Missing data can change the treatment comparison
Withdrawal, missed visits, device removal or rescue therapy may relate to outcome and treatment. Prevention starts with feasible follow-up and participant support. The analysis should define the estimand, handling method and sensitivity analyses for departures from assumptions.
A pivotal investigation is a chain from question to comparison to endpoint. In missing data can change the treatment comparison, Withdrawal, missed visits, device removal or rescue therapy may relate to outcome and treatment. Prevention starts with feasible follow-up and participant support. The analysis should define the estimand, handling method and sensitivity analyses for departures from assumptions. Respect this boundary: A single imputation method cannot prove robustness to informative missingness.
Try this
Challenge missing data can change the treatment comparison with a rival explanation. List reasons for missingness by group and design two clinically plausible sensitivity scenarios. Identify the design feature that blocks or measures it. Keep this boundary: A single imputation method cannot prove robustness to informative missingness.
A strong science note separates observation from interpretation. List reasons for missingness by group and design two clinically plausible sensitivity 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
List reasons for missingness by group and design two clinically plausible sensitivity scenarios. Keep this boundary visible: A single imputation method cannot prove robustness to informative missingness.
CHAPTER 15 OF 17 · Interpret the result
15. Protocol deviations should be visible, not erased
Eligibility errors, treatment crossovers, procedural departures and assessment-window violations can affect interpretation. The protocol should define important deviations, and the report should describe frequency, timing, cause and analysis impact. Per-protocol analyses may inform sensitivity but do not automatically replace the randomised comparison.
The strongest feature of protocol deviations should be visible, not erased is clarity about what would count as convincing evidence. Eligibility errors, treatment crossovers, procedural departures and assessment-window violations can affect interpretation. The protocol should define important deviations, and the report should describe frequency, timing, cause and analysis impact. Per-protocol analyses may inform sensitivity but do not automatically replace the randomised comparison. The conclusion stops here: Excluding participants after allocation can reintroduce selection bias.
Try this
Trace protocol deviations should be visible, not erased from claim to endpoint. Build a deviation table linked to participant flow and both primary and sensitivity analyses. Ask whether a clinically important effect would be detected and interpreted. Respect this limit: Excluding participants after allocation can reintroduce selection bias.
A strong science note separates observation from interpretation. Build a deviation table linked to participant flow and both primary and sensitivity analyses. 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 deviation table linked to participant flow and both primary and sensitivity analyses. Keep this boundary visible: Excluding participants after allocation can reintroduce selection bias.
CHAPTER 16 OF 17 · Interpret the result
16. Statistical significance needs clinical interpretation
A p-value, confidence interval or posterior probability should be interpreted alongside effect size, uncertainty, endpoint meaning, safety and benefit-risk context. A tiny effect can be statistically persuasive in a large sample yet unimportant to patients; a promising estimate with wide uncertainty may need more evidence.
For statistical significance needs clinical interpretation, the design should make the final claim answerable. A p-value, confidence interval or posterior probability should be interpreted alongside effect size, uncertainty, endpoint meaning, safety and benefit-risk context. A tiny effect can be statistically persuasive in a large sample yet unimportant to patients; a promising estimate with wide uncertainty may need more evidence. Keep this limit beside the conclusion: Statistical success is not identical to clinical benefit or regulatory approval.
Try this
Turn statistical significance needs clinical interpretation into a design table. Write separate sentences for magnitude, precision, clinical meaning and remaining uncertainty. Add the estimand, analysis population and missing-data rule. Preserve this limit: Statistical success is not identical to clinical benefit or regulatory approval.
A strong science note separates observation from interpretation. Write separate sentences for magnitude, precision, clinical meaning and remaining 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
Write separate sentences for magnitude, precision, clinical meaning and remaining uncertainty. Keep this boundary visible: Statistical success is not identical to clinical benefit or regulatory approval.
CHAPTER 17 OF 17 · Interpret the result
17. A complete pivotal-study receipt reconstructs the claim
The receipt records claim, intended use, objectives, population, control, allocation, blinding, training, endpoints, sample-size assumptions, protocol versions, participant flow, missing data, deviations, analyses, effect estimates, uncertainty, safety and limitations. It lets a reviewer retrace why the study supports—or does not support—the proposed conclusion.
A pivotal investigation is a chain from question to comparison to endpoint. In a complete pivotal-study receipt reconstructs the claim, The receipt records claim, intended use, objectives, population, control, allocation, blinding, training, endpoints, sample-size assumptions, protocol versions, participant flow, missing data, deviations, analyses, effect estimates, uncertainty, safety and limitations. It lets a reviewer retrace why the study supports—or does not support—the proposed conclusion. Respect this boundary: Documentation clarifies the evidence; it cannot repair a design that asked the wrong question.
Try this
Challenge a complete pivotal-study receipt reconstructs the claim with a rival explanation. Give the receipt to a reviewer and ask them to trace the final claim back to prespecified evidence. Identify the design feature that blocks or measures it. Keep this boundary: Documentation clarifies the evidence; it cannot repair a design that asked the wrong question.
A strong science note separates observation from interpretation. Give the receipt to a reviewer and ask them to trace the final claim back to prespecified evidence. 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
Give the receipt to a reviewer and ask them to trace the final claim back to prespecified evidence. Keep this boundary visible: Documentation clarifies the evidence; it cannot repair a design that asked the wrong question.
Pivotal device-investigation design map
| Question | Evidence to capture | Boundary to retain |
|---|---|---|
| What claim must be supported? | Regulatory objective, intended use and primary estimand | Size alone does not make a study pivotal |
| What comparison identifies effect? | Randomisation, control rationale and bias protections | Every nonconcurrent comparison carries assumptions |
| What outcome matters? | Endpoint definition, time point and clinical meaning | Convenient measures may miss patient benefit |
| Can the conclusion withstand loss and variation? | Missing data, deviations, multiplicity and sensitivity | Statistical success is not the whole benefit-risk decision |
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.
Pivotal evidence is persuasive when every design choice points to the same practical question. That alignment is a happy discipline: readers can see what was compared, what changed, how sure we are and why the difference matters.
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: Design Considerations for Pivotal Clinical Investigations for Medical Devices — Final guidance, November 2013.
- FDA: Full guidance PDF — Official detailed design guidance.
- FDA: PMA clinical studies — Current PMA clinical-study context.