Platform Trials: the Impact of common Controls on Type One Error and Power

TL;DR

This study examines how using a common control in platform trials affects type I error and power, highlighting the trade-offs and conditions under which efficiency gains are maintained or lost, especially with multiplicity adjustments.

Contribution

It provides a simulation-based analysis of the impact of common controls on error rates and power in platform trials, addressing a debated issue in trial design.

Findings

Common controls reduce family-wise error rate inflation compared to separate controls.

Platform trials with common controls generally maintain efficiency after multiplicity adjustment.

In some scenarios, common controls may lead to loss of efficiency gains.

Abstract

Platform trials offer a framework to study multiple interventions in a single trial with the opportunity of opening and closing arms. The use of a common control in platform trials can increase efficiency as compared to individual control arms or separate trials per treatment. However, the need for multiplicity adjustment as a consequence of common controls is currently a controversial debate among researchers, pharmaceutical companies, as well as regulators. We investigate the impact of a common control arm in platform trials on the type one error and power in comparison to what would have been obtained with a platform trial with individual control arms in a simulation study. Furthermore, we evaluate the impact on power in case multiplicity adjustment is required in a platform trial. In both study designs, the family-wise error rate (FWER) is inflated compared to a standard, two-armed randomized controlled trial when no multiplicity adjustment is applied. In case of a common control, the FWER inflation is smaller. In most circumstances, a platform trial with a common control is still beneficial in terms of sample size and power after multiplicity adjustment, whereas in some cases, the platform trial with a common control loses the efficiency gain. Therefore, we further discuss the need for adjustment in terms of a family definition or hypotheses dependencies.