Design and monitoring of randomized trials in vision research with binary endpoints.

Abstract

Background: Randomized eye trials often involve both eyes if eligible from a patient. Few studies have investigated the design properties of randomized eye trials with binary endpoints, for instance, the sample size/power determination, leaving a significant gap in the literature. In addition, proper interim data-monitoring plans, which are essential for a successful design, are not available for eye trials with binary endpoints.

Methods: We propose a unified design for vision research with binary endpoints that accommodate all possible design options while accounting for correlations between the two eyes of the same patient. We derive closed-form formulas for the sample size and power calculations under the generalized linear mixed model framework. We also describe how to properly calculate the information fractions during interim monitoring, which is an essential component used to determine the efficacy/futility boundaries or calculate the alpha-spending function, as the standard technique may not apply.

Results: We conduct extensive simulation studies to examine the performance of the proposed unified design for clustered binary data in randomized eye trials and compare it with the existing specific individual design. The simulation results demonstrate that the proposed methods perform well in practical settings. We provide an application to the Age-Related Eye Disease Study.

Conclusion: We propose a unified design that accommodates all scenarios in eye trials with binary outcomes, including one-eye, paired, two-eye, mixture of one-eye and two-eye, and hybrid designs, and develop appropriate interim analysis plans. A proper study design and interim monitoring using the proposed methods in these studies can help study investigators make informed decisions.