Regression analysis of asynchronous longitudinal data with informative dropout and dependent observation

Abstract

Informative dropout occurs when the subject’s follow-up time depends on the response process, even conditional on the covariates process. Most existing works for the regression analysis of asynchronous longitudinal data presume that dropouts are virtually non-informative. In this paper, we propose kernel-weighted estimating equations to accommodate asynchronous measurement and informative dropout simultaneously. We specify a semiparametric linear regression model for the response process and an accelerated failure time model for the dropout process. Unlike the fully specified parametric or nonparametric model in most existing literature, the proposed method does not require model specification for the joint distribution. To deal with the informative dropout, an artificial censoring device is employed. Besides, the observation process is allowed to depend on the covariate process. The proposed estimators are shown to be consistent and asymptotically normally distributed. We conduct a series of simulation studies to assess the finite sample performance of the proposed estimators. Applying the suggested approaches to the Study of Women’s Health Across the Nation indicates a significant negative association between follicle-stimulating hormones and triglycerides.

Keywords:

• Artificial censoring
• Asynchronous longitudinal data
• Dependent observation
• Informative dropout
• Kernel weighted estimating equation

Disclosure statement

No potential conflict of interest was reported by the author(s).