Joint survival and longitudinal modelling for combined cohorts

Abstract

A primary goal of survival analysis is modelling the time from an initial event to a failure event and the factors which affect the hazard rate. In this thesis, we consider data drawn from an incident cohort study with follow-up and a prevalent cohort study with follow-up, yielding right-censored failure time data and left-truncated right-censored failure time data, respectively. In addition to the failure/censoring times, we assume the survival data include other measured covariates, for example, longitudinal measurements collected throughout the failure/censoring time durations. In this thesis, we review the methods of estimation for the proportional hazards model, the linear mixed effects model, and the joint model, to determine the association between the survival and the longitudinal processes. We introduce novel estimation methods for the joint model when incident and prevalent cohort data are combined. Using simulated data, we compare the various individual and combined cohort estimation procedures. We also apply the combined cohort joint model to examine the relationship between the mortality of mature female baboons based on their longitudinally measured glucocorticoid levels.