Abstract
Purpose :
Group sequential (GS) methods are the most commonly used tools for studies with interim monitoring plans, which control type I error rate, protect statistical power, and avoid false positive/negative findings. However, to the best of our knowledge, when both eyes (clustered) from at least some patients are included in an eye study, specific GS methods accounting for the inter-eye correlation do not exist. We develop statistical tools for the GS designs for randomized eye trials accommodating all possible design options with various types of endpoints.
Methods :
We propose a unified GS design for vision research that accounts for the inter-eye correlations when both eyes of a patient are included in the study. The unified design accommodates all possible options, including the scenarios of only one eye eligible and/or both eyes on the same or different treatment arms. We investigate the design properties of the proposed GS methods and derive sample size/power calculation results and interim monitoring boundaries while accounting for the inter-eye correlation.
Results :
Simulation studies demonstrate that the proposed methods perform well in practical settings. Using the Age-Related Eye Disease Study (AREDS) as an example, we calculate the sample sizes for comparing the change of IOP from baseline to year 5 in a future two-arm trial; see Figure 1, which shows that ignoring the inter-eye correlation may lead to incorrect sample size calculations. A demonstration of using the R package iTrial that implements the newly proposed methodology will be presented.
Conclusions :
Our novel unified GS design accommodates all scenarios in eye trials and protects type I error rates and study power in the presence of inter-eye correlations. Successful implementation of the proposed methodology in randomized trials involving both eyes of a patient is important for evaluating new treatments. Proper study design and interim monitoring using these methods will help improve the study quality and achieve successful study conduct.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.