Analysis of Survival Data with Missing Measurements of a Time-Dependent Binary Covariate

Abstract

The objective of this study was to investigate the influence of the number and timing of a binary time-dependent covariate on the bias using the last-observation carried forward in the proportional hazards model. Under various assumptions of censoring rates, transition probabilities of the time-dependent covariate, sample size, and the log hazard-ratio for the covariate, we empirically examined the impact that the number and timing have on the bias of the estimator of the covariate. An example from the Systolic Hypertension in the Elderly Program was used. Inference on the effect of systolic blood pressure on survival is strongly affected by the number and timing of systolic blood measurements.