We are at the design phase of a randomized open-label study to test the efficacy of an intervention. Apparently, the sponsor wants to perform an interim analysis, with the intention to stop the study due to cost-effectiveness consideration. The decision will be based on descriptive data only. If the decision is stop the study, we want to test the study hypotheses, and if the decision is to continue will not perform the statistical hypotheses testing at that point but only at the end of the study as planned. My question is whether there are alpha spending considerations we should take into account. In other words, can we test the hypotheses at the overall significance level (5%) in case of early stopping?
What happens if the interim decision based upon the cost effectiveness data is to stop the study, your primary endpoint is favorable in direction, and clinically relevant in the magnitude of treatment effect, but is not statistically significant at that time? Say the p value is between 0.1 and 0.2, for the sake of discussion.
Since your study is presumably powered based upon the full enrollment cohort, your primary endpoint will be underpowered at the interim analysis. Would you continue the study to enroll additional subjects to meet the original power target, or would you still stop it?
Consider that, generally speaking, if a study is stopped at an interim analysis for efficacy, the magnitude of the treatment effect is typically biased towards being overly optimistic.
That being said, this design, conceptually, is somewhat similar to one for which I was on the trial DSMB, where the interim analyses and stopping rules were based upon a surrogate measure of efficacy. Since we never analyzed the primary endpoint at the interim analyses, there was no need to adjust the p value of that endpoint at the final analysis in the SAP.
Thus, IMHO, if your decision to stop the study at the interim analysis is truly a final decision, and there are no scenarios under which you would continue the study given the results of that primary endpoint analysis, then I do not see a need to adjust the p value of your primary endpoint.
Thank you, that’s what I thought