Usefulness of interim analyses in portending study results in antipsychotic and antidepressant trials

It is unknown whether interim analyses portend final study results. Fatigue, pressure to complete trials and recruitment differences may mitigate against this. We examined the similarity of efficacy results of the first and second half of recruited patients to complete trials and explore possible intervening variables. Using data from the NewMeds repository of patient level data from placebo-controlled randomized trials of antipsychotics (AP) (22 studies, n=7056) and antidepressants (AD) (39 studies, n=12,217) we compared treatment effect size (placebo vs. active treatment) of the first and second half of patients recruited in completed trials. We found that in AP studies median difference in treatment effect between cohorts was -0.03, indicating that overall first and second cohorts yielded similar results. In AD studies, median difference between cohorts was 0.04, indicating that overall the second cohort had slightly larger active-placebo-difference. Overall, on average there were minimal differences in effect size between the first and the second cohorts, and in 30 of 39 trials interim results were a good estimate of the results on the 2nd cohort. In AD trials first and second cohort results were more similar when the proportion of patients per study centre and recruitment time of the two cohorts was similar. Results suggest that interim analyses in AD and AP studies may reliably serve to estimate ultimate effects and, at least in AD trials, are more accurate when the same sites are used to a similar extent and recruitment time of the two consequent cohorts is similar.