Robust statistical inference for the matched net benefit and the matched win ratio using prioritized composite endpoints

TL;DR

This paper proposes a robust statistical inference method for the net benefit and win ratio in composite endpoints, addressing limitations of large-sample assumptions especially in small samples or extreme proportions.

Contribution

It introduces a new test statistic, sample size formula, and MOVER-based confidence intervals for NB and WR, improving accuracy in small or boundary cases.

Findings

MOVER confidence intervals perform well in small samples.

Proposed methods outperform existing ones near boundary proportions.

Simulation studies validate the robustness of the new approach.

Abstract

As alternatives to the time-to-first-event analysis of composite endpoints, the {\it net benefit} (NB) and the {\it win ratio} (WR) -- which assess treatment effects using prioritized component outcomes based on clinical importance -- have been proposed. However, statistical inference of NB and WR relies on a large-sample assumptions, which can lead to an invalid test statistic and inadequate, unsatisfactory confidence intervals, especially when the sample size is small or the proportion of wins is near 0 or 1. In this paper, we develop a systematic approach to address these limitations in a paired-sample design. We first introduce a new test statistic under the null hypothesis of no treatment difference. Then, we present the formula to calculate the sample size. Finally, we develop the confidence interval estimations of these two estimators. To estimate the confidence intervals, we use the {\it method of variance estimates recovery} (MOVER), that combines two separate individual-proportion confidence intervals into a hybrid interval for the estimand of interest. We assess the performance of the proposed test statistic and MOVER confidence interval estimations through simulation studies. We demonstrate that the MOVER confidence intervals are as good as the large-sample confidence intervals when the sample is large and when the proportions of wins is bounded away from 0 and 1. Moreover, the MOVER intervals outperform their competitors when the sample is small or the proportions are at or near the boundaries 0 and 1. We illustrate the method (and its competitors) using three examples from randomized clinical studies.