Covariate-adjusted generalized pairwise comparisons in small samples.

Area of Science:

• Biostatistics
• Statistical Modeling
• Clinical Trial Analysis

Background:

• Semiparametric probabilistic index models are used for comparing two groups while adjusting for covariates within generalized pairwise comparisons (GPC).
• Traditional regression methods face challenges with limited data, leading to invalid inference due to unmet asymptotic normality assumptions and potential separation issues in small samples.

Purpose of the Study:

• To address the limitations of current probabilistic index models in small sample settings.
• To propose a method for valid statistical inference in probabilistic index models, even with limited data and potential separation.

Main Methods:

• Utilized generalized estimating equations (GEE) for parameter estimation in probabilistic index models.
• Employed adjustments to sandwich variance-covariance matrix estimators to improve finite sample properties and handle bias from separation.
• Conducted extensive simulation studies to validate the proposed methodology.

Main Results:

• Demonstrated that GEE can effectively estimate parameters of the probabilistic index model.
• Showcased improved finite sample properties and bias correction for variance-covariance estimators.
• Confirmed the ability to perform appropriate statistical inference through simulations.

Conclusions:

• The proposed GEE approach provides valid statistical inference for probabilistic index models, overcoming limitations of small sample sizes and separation.
• This method enables accurate calculation of GPC statistics, including net treatment benefit and success odds, enhancing clinical trial analysis.