Issue 5 covered what a reviewer checks in your SAP. This one goes back a step, to the decision the SAP is built around.
During my years reviewing NDAs and BLAs at FDA, the primary endpoint was always chosen before the trial began. My question was no longer which endpoint to use. It was what the endpoint results could support. Sometimes the sponsor had run the trial well, the data were clean, and the analysis followed the SAP. Even so, the central question of FDA's review, clinical as well as statistical, came back to the endpoint itself: did it measure what it was intended to measure, and could the results justify the claim the sponsor wanted in the label?
A good analysis can't fix that. The primary endpoint sets the limits of what your trial can show. The sample size is built on it, the SAP is written around it, and the label discussion often starts with it. Ideally, it is fixed before the first patient is enrolled, because it is the hardest part of the design to change once the trial starts.
✅ What a Reviewer Needs to See in the Protocol and SAP
In protocols and SAPs that moved through FDA review without avoidable questions about the primary endpoint, I consistently found the following:
A precise definition. Endpoints are derived from assessments. Temperature measured weekly is an assessment; the endpoint might be the average temperature over three months, or the change from baseline at Month 3. The protocol stated which assessment was used, how and when it was measured, and how the endpoint was derived from it. For a responder endpoint, the thresholds were stated and justified. I could derive the endpoint from its definition in the protocol and in the SAP.
A defined estimand. The protocol and SAP defined the population, the treatments being compared, the patient-level endpoint, how intercurrent events such as treatment discontinuation or use of rescue medication would be handled, and the difference to be estimated (for example, the difference in mean change from baseline between the treatment and placebo groups).
Agreement with FDA before Phase 3. The endpoint and its definition were discussed at the End-of-Phase 2 meeting, with specific questions FDA could answer, and the meeting minutes recorded whether FDA agreed. Where FDA did not agree, the sponsor still had time to change the endpoint before the trial began. I wrote about framing those questions in Issue 4.
⚠️ What Leads to Questions Before the Trial Begins
These came up when FDA reviewed the protocol and SAP, and in meetings with FDA. That is the time to resolve them.
Precedent as the only justification. "This endpoint supported an earlier approval" is a starting point, not a rationale, and FDA already knows the precedent. The earlier approval may have been in a different population, at a different disease stage, or before the relevant FDA guidance was updated. Precedent offered as the only justification was something FDA raised often in meetings with sponsors.
A time point chosen for convenience. An earlier time point can make the trial shorter. It can also measure the change before the drug has had time to work, and miss a treatment effect that takes longer to emerge.
Responders defined by improvement alone. As a primary endpoint, turning a continuous measure into "responders" throws away information and costs power. Responder analyses are more useful for interpreting the primary results, and then they need two thresholds that define three categories: improved, no clinically meaningful change, and worsened. Too often, responder definitions count improvement and ignore worsening. An empirical cumulative distribution function (eCDF) plot by treatment group shows the whole range of responses without choosing a threshold at all.
Vague wording. "Improvement," "stabilization" or "clinical benefit" without a definition doesn't tell a reviewer how the endpoint will be computed. Each undefined term became a review question.
🔍 What I Checked at NDA or BLA Review
Toward the end of my review, I looked at whether the results could support the claims the sponsor wanted in the Clinical Studies section of the label. These were claims about what the trial showed, separate from the indication.
A claim broader than the endpoint. Results on an endpoint that measures symptom frequency support a claim about symptom frequency, not a broader claim about the disease. When the sponsor asked for a broad claim on the basis of results on a narrow endpoint, the gap between the two became a labeling issue.
A composite endpoint driven by a subset of its components. When only a few of the components account for most of the result, the composite says less than it appears to. That is why FDA's guidance on the Clinical Studies section of labeling recommends presenting every component, so readers can see which ones drive the result.
💡 What to Settle Before the Protocol Is Final
Most endpoint questions can be answered before the first patient is enrolled. Before the protocol is final, ask your team:
Does this endpoint measure what it is intended to measure, in this population?
What do you want the Clinical Studies section of your label to include, and could results on this endpoint support it?
Could a statistician outside your team derive this endpoint from the protocol and SAP alone?
Has FDA agreed to the endpoint and its definition for this population?
If any answer is "not yet," that is the work to do now. Sponsors who settle the endpoint with FDA before Phase 3 go into the NDA or BLA with the endpoint questions already addressed, and the review can focus on what the results show. Otherwise, those questions come back during the review, when the endpoint can no longer be changed.
Surrogate endpoints raise their own questions, and I'll cover them separately.
One quick question for you:
How often would you like to hear from me?
If your team is still deciding on a primary endpoint for Phase 3, I'd welcome the conversation. Book a free 30-minute call.
Thank you for reading!
Lisa