LAUSR

Background: In placebo-controlled clinical trials to develop new drugs for the treatment of psychiatric or neurological disorders, a high and sometimes greater-than-expected placebo response makes it difficult to show the superiority of an investigational drug over a corresponding placebo. To avoid such difficulty, a placebo lead-in design has been presented, but its usefulness has been open to discussion. Although the statistical properties of the placebo lead-in design are investigated in the context of continuous outcomes, whether these properties can be generalized for binary or ordinal cases remains unclear. Methods: We investigate whether the placebo lead-in design is useful in clinical trials with binary outcomes through mathematical formulae and a numerical investigation. Specifically, we compare the proportion of placebo responders, the drug–placebo difference, and the effect size between two populations: one enriched for placebo nonresponders and the other comprising the all-comers. Results: Under positive correlation of the data between the lead-in stage and the randomized stage for both treatment groups, we mathematically show that the proportion of responders in the population enriched for placebo nonresponders is less than that in the all-comers population, and whether the placebo lead-in design increases the drug–placebo difference depends on the variances of outcomes in both treatment groups as well as the correlations of the outcomes between two stages. Further, through a numerical investigation, we show that whether the placebo lead-in design increases the effect size strongly depends on the magnitude of the correlations and their difference. Conclusion: If the correlation of the placebo–placebo group is much higher than that of the placebo–drug group, the placebo lead-in design is advantageous in most cases but has an impact on an estimand in placebo nonresponders. Therefore, we do not recommend using the placebo lead-in design for clinical trials with binary outcomes.