Simultaneous inference for multiple marginal generalized estimating equation models

Area of Science:

• Biostatistics
• Statistical Inference
• Longitudinal Data Analysis

Background:

• Small-sample studies in fields like ophthalmology and dermatology often involve multiple endpoints and repeated observations.
• Simultaneous inference for multiple endpoints with repeated measures presents statistical challenges, particularly in small samples.

Purpose of the Study:

• To develop a robust statistical framework for simultaneous inference on multiple endpoints in the presence of repeated observations.
• To improve the performance of statistical methods in small-sample settings.

Main Methods:

• Utilizing generalized estimating equation (GEE) models for marginal analysis of each endpoint.
• Deriving Wald-type simultaneous confidence intervals and hypothesis tests using asymptotic joint normality.
• Implementing bias adjustment for the covariance matrix estimate and employing a multivariate t-distribution for small samples.
• Developing a generalized score test based on stacked estimating equations.

Main Results:

• The proposed methods demonstrate strong control of the family-wise type I error rate, even with small sample sizes.
• The approach offers increased statistical power compared to traditional Bonferroni-Holm multiplicity adjustments.
• Simulation studies confirm the effectiveness and suitability of the methods for small-sample longitudinal data.

Conclusions:

• The developed framework provides an efficient way to leverage information from repeated observations of multiple endpoints in small-sample studies.
• The methods are suitable for applications in ophthalmology, dermatology, and other fields requiring simultaneous inference on multiple correlated outcomes.
• This approach enhances statistical rigor and power in challenging small-sample research scenarios.