U-statistics-based Pairwise Discordance Regression for Joint Mean-Variability Inference in Longitudinal Studies with Missing Data

A semiparametric pairwise discordance regression for longitudinal data estimates joint mean and variability trajectories using weighted U-statistic-based generalized estimating equations, including missing-at-random dropout.

• Why it matters: Mean-only longitudinal analyses treat variability as nuisance, but temporal variability and heterogeneity can carry scientific signal. Existing mixed-effects location-scale models depend on strong parametric assumptions and can be computationally intensive.
• What they did: The method models pairwise contrasts to characterize mean trajectories, between-subject heterogeneity, and within-subject temporal variability, using conditional moment restrictions. It fits via (weighted) U-statistic-based GEE with a sandwich variance estimator and extends to missing-at-random dropout.
• The result: Asymptotic distributions are derived under mild regularity conditions, and simulations show reliable finite-sample performance under non-Gaussian errors and misspecified working correlations. A longitudinal blood pressure during pregnancy application demonstrates feasibility on real data.

AI summary of the abstract · may contain errors · not medical advice · read the paper.