Improving Precision through Adjustment for Prognostic Variables in Group Sequential Trial Designs: Impact of Baseline Variables, Short-Term Outcomes, and Treatment Effect Heterogeneity

TL;DR

This paper quantifies how adjusting for baseline and short-term outcomes in group sequential trials improves precision, reduces sample size, and affects power, considering factors like treatment heterogeneity and pipeline participants.

Contribution

It provides an asymptotic formula for precision gain from adjustments, highlighting the greater benefit of baseline variables over short-term outcomes in trial efficiency.

Findings

Adjusting for baseline variables yields at least as much precision gain as short-term outcomes.

In many cases, baseline adjustment offers substantially greater precision improvements.

Precision gains can be translated into meaningful sample size reductions in group sequential designs.

Abstract

In randomized trials, appropriately adjusting for baseline variables and short-term outcomes can lead to increased precision and reduced sample size. We examine the impact of such adjustment in group sequential designs, i.e., designs with preplanned interim analyses where enrollment may be stopped early for efficacy or futility. We address the following questions: how much precision gain can be obtained by appropriately adjusting for baseline variables and a short-term outcome? How is this precision gain impacted by factors such as the proportion of pipeline participants (those who enrolled but haven't yet had their primary outcomes measured) and treatment effect heterogeneity? What is the resulting impact on power and average sample size in a group sequential design? We derive an asymptotic formula that decomposes the overall precision gain from adjusting for baseline variables and a short-term outcome into contributions from factors mentioned above, for efficient estimators in the model that only assumes randomization and independent censoring. We use our formula to approximate the precision gain from a targeted minimum loss-based estimator applied to data from a completed trial of a new surgical intervention for stroke. Our formula implies that (for an efficient estimator) adjusting for a prognostic baseline variable leads to at least as much asymptotic precision gain as adjusting for an equally prognostic short-term outcome. In many cases, such as our stroke trial application, the former leads to substantially greater precision gains than the latter. In our simulation study, we show how precision gains from adjustment can be converted into sample size reductions (even when there is no treatment effect).