Joint modeling of repeated ordinal measures and time to event data for CHD risk assessment

Abstract

Current joint models for longitudinal and time to event data are not sufficient for longitudinal ordinal outcomes. Three joint models for longitudinal ordinal variables and time to event data were suggested. In each joint model, cumulative logit, or continuation-ratio logit and or cumulative probit mixed effects models for the longitudinal ordinal outcome is associated with the time to event variable by random effects approach. Joint modeling likelihood approach, including ordinal repeated measure sub-model and time to event sub-model were constructed. A SAS macro on importance sampling procedure utilizing adaptive gaussian quadrature and conjugate gradient technique algorithm was used to obtain the maximum likelihood estimates of the parameters. This joint model provides inferences for both the longitudinal ordinal variables and time to event simultaneously. Also, a simulation study on various scenarios (concerning different sample size and different correlations among ordinal longitudinal variables) was conducted. The method was applied to the Tehran Lipid and Glucose Study (TLGS) data. The results in simulation study and real data set supports the applicability of the model.