Additive hazard regression of event history studies with intermittently measured covariates

Abstract

In the analysis of event history data, some covariates may vary over time. Such covariates are typically sparse and measured intermittently in many longitudinal settings. This article considers additive hazard regression models with intermittently measured covariates and proposes kernel weighting methods to estimate the model parameters. Two scenarios are investigated: half kernel estimation for data in which observation ceases at an event, and full kernel estimation in which observation may continue after an event, such as in recurrent event data. The asymptotic properties of the resulting estimators are established. Simulation studies show that the proposed estimators perform satisfactorily. An application to a study on primary biliary cirrhosis is illustrated.