The rule set
29 checks read from one recorded episode
Each check states what it looks for, why it exists and what resolves it. The source is the FDA Watch episode on real-world evidence in accelerated approvals, from 19:57 to 39:36, with the timestamp on every rule. High-severity checks are the ones a review division is most likely to raise first.
Read from one recorded source, not invented. Every check carries its timestamp in the episode, so you can hear the sentence behind the finding. The legal basis is read from 29 held controls across 7 frameworks.
Prospective planning
The protocol and the statistical analysis plan are written, dated and shared before the study starts. A plan assembled after the data arrives reads as a plan fitted to the answer.
The protocol is pre-specified and shared with FDA before the study starts
Does the outline say the protocol was shared with the review division before the study was conducted?
highPlanned before the dataA2The statistical analysis plan is pre-specified, dated and locked before analysis
Is there a dated statistical analysis plan, fixed before the analysis began?
highPlanned before the dataA3The estimand, primary endpoint and time zero are defined in advance
Does the outline define the estimand, the primary endpoint and the start of follow-up before any data is seen?
highPlanned before the dataA4The analysis populations and eligibility rules are fixed in advance
Are the analysis population and the inclusion and exclusion rules defined before the cohort is assembled?
mediumPlanned before the dataA5Every protocol and SAP version is retained and handed over
Does the outline say the protocol and SAP versions are kept and available to the reviewer?
mediumPlanned before the dataA6The confirmatory study is underway, with milestones, not only on paper
Does the outline show the confirmatory study started, with enrolment and a completion date, rather than a protocol that exists?
highPlanned before the dataFitness for use
Relevance and reliability are the two halves of fitness for use. The data has to answer this question, and it has to be accurate, complete and traceable back to a source that can be verified.
The data is relevant to the question, the population and current care
Does the outline connect the data source to this question, this population and the care patients receive now?
highFit for the questionB2Data source fit is assessed, with the limits written down
Is there a feasibility or fit-for-purpose assessment that names the limitations of this source?
highFit for the questionB3Outcome and exposure definitions are validated, not just coded
Does the outline say how each outcome and exposure is identified, and how that identification was validated?
highFit for the questionB4Missing data is characterised and handled by a stated method
Does the outline describe how much is missing, why, and what the analysis does about it?
highFit for the questionB5Every value is traceable from the source to the analysis dataset
Does the outline describe an audit trail from the source record to the analysis dataset?
highFit for the questionB6Source data verification is planned, and the records can be retrieved
Does the outline say who will verify source data, on what sample, and that the records can be produced?
highFit for the questionB7The data is converted to the standards FDA can read
Does the outline say the analysis dataset will be converted to CDISC or another accepted standard?
mediumFit for the questionRobustness
Sensitivity analyses, subgroup analyses and a stated view of what would change the conclusion are fixed in advance, so the result does not rest on one assumption or one analytical choice.
Sensitivity analyses are pre-specified, each with the assumption it tests
Does the outline name the sensitivity analyses and the assumption each one varies?
highDoes not hinge on one choiceC2Subgroup analyses are pre-specified, with a multiplicity position
Are the subgroups named in advance, and does the outline say how many comparisons it is making?
mediumDoes not hinge on one choiceC3The result is shown to hold across methods, not to depend on one
Does the outline show the result will be examined under more than one method or data cut?
mediumDoes not hinge on one choiceC4The plan says in advance what would change the conclusion
Does the outline state, before the result, what evidence would change the conclusion?
mediumDoes not hinge on one choiceExternal control
An external control arm is judged on whether it is the same patient, the same endpoint and the same standard of care, and whether the analysis addresses confounding in a way fixed before the comparison is run.
The external control uses the trial's eligibility criteria
Does the outline say the external control patients were selected by the trial's own eligibility rules?
highA fair comparisonD2The endpoint and its ascertainment are the same in both arms
Are the endpoint definition, measurement and follow-up window identical in the trial and the external control?
highA fair comparisonD3Confounding is addressed by a pre-specified method, with diagnostics
Does the outline name the confounders, the adjustment method and the diagnostics that will show balance?
highA fair comparisonD4The control is contemporaneous and reflects current care
Does the outline show the external control patients were treated in the same period and under the current standard of care?
highA fair comparisonD5The sites and countries have a care standard comparable to the reference market
Does the outline justify that patients at the chosen sites were treated to a standard comparable with the market you are filing in?
mediumA fair comparisonD6The single-arm design is justified and the external control is fixed before unblinding
Does the outline justify why a single-arm design is appropriate, and state that the external control was fixed before the trial result was known?
highA fair comparisonRights and verification
The sponsor needs the right to use the data in the submission, not only for the research, and the source records have to exist and be open to verification. Data use agreements, privacy and human subject protections carry that chain.
The right to use the data covers the submission, not only the research
Does the outline say the sponsor has the right to use the data in the regulatory submission?
highThe right to use it, and to prove itE2A data use agreement is in place, with its scope written down
Does the outline name the data use agreement and the uses it permits?
highThe right to use it, and to prove itE3The source records exist and can be verified, including consent and the protocol
Does the outline confirm that the source records, consent forms and protocol can be produced for verification?
highThe right to use it, and to prove itE4Privacy and human subject protections are built into the plan
Does the outline say how consent, ethics review and privacy law are handled for the data used?
highThe right to use it, and to prove itE5The data partner is obliged to give access, retain records and allow inspection
Does the outline say the data partner must give access, keep the records and allow inspection?
mediumThe right to use it, and to prove itE6The traceability chain is papered through every hand
Does the outline show a documented chain from the original record through each partner to the analysis?
mediumThe right to use it, and to prove itHow to use the rule set
Read the check, then read your outline against it. If the outline cannot be quoted for the element the check names, the check is not met, however familiar the heading looks. The free check does that reading for you and quotes the line.
The checks are not a substitute for the protocol, the statistical analysis plan or the agreements. They are the questions to ask of the outline before those documents are sent.