A commentary on: statistical inference problems in sequential parallel comparison designs

A Sequential Parallel Comparison Design has two stages, the first comparing drug to placebo and the second comparing drug to placebo among patients who did not respond to placebo in the first stage. The paper, Statistical Inference Problems in Sequential Parallel Comparison Designs, claims that the estimate of the treatment difference in the second stage is biased and that under certain circumstances, a suggested hypothesis test will reject the null hypothesis when it should be accepted. This rejoinder argues that the estimate in the second stage is not biased when the true target of estimation (estimand) is properly understood. Further, the null hypothesis that the authors posit is not the correct null hypothesis for clinical trials, and in the situation, they describe that the treatment should be considered to be effective.