Comparison of primary analysis strategies of randomized controlled trials with multiple endpoints with application to kidney transplantation

TL;DR

The paper compares different statistical methods for analyzing clinical trials with multiple endpoints, showing that some methods are more powerful but may be harder to interpret.

Contribution

The study evaluates three primary analysis strategies for RCTs with multiple endpoints, highlighting their trade-offs in power and interpretability.

Findings

Global testing strategies like composite endpoints and generalized pairwise comparisons showed higher power than multiple testing with correction.

Global methods can produce significant results even with heterogeneous treatment effects across endpoints.

The study emphasizes the need for careful interpretation and component-wise assessment when using global procedures.

Abstract

Relying on a single primary endpoint in randomized controlled trials (RCTs) is often infeasible, for example due to rare or heterogeneous events. Regulatory guidance therefore allows multiple endpoints, but different analytical strategies address different scientific questions and null hypotheses, even when applied to the same set of variables. We explored three approaches to consider multiple endpoints in the primary analysis of RCTs, as stated in the FDA and EMA guidelines on multiplicity: (i) a composite endpoint (CE), (ii) multiple testing and multiplicity correction (MTMC), and (iii) a hierarchical non-parametric procedure, called generalized pairwise comparisons (GPC). Using clinical trial simulations, we compared these strategies’ power in two-arm RCTs perform when testing strategy-specific hypotheses across a range of scenarios reflecting endpoint prioritization, correlation between endpoints, and opposing treatment effects. When testing time-to-event endpoints, global testing strategies (CE and GPC) generally achieved higher power than MTMC. However, we also demonstrate that global procedures may yield statistically significant results even when treatment effects are heterogeneous across endpoints, underscoring the importance of careful interpretation and component-wise assessment. As trials increasingly use multiple endpoints, understanding the trade-off between statistical efficiency and interpretability, and provide practical guidance for choosing endpoint definitions and primary analysis strategies in future trials.

Figures