MSc Stats Project Presentation "Joint Modelling for Longitudinal and Time-to-Event Data with Applications to Bipolar Disorder Data"

CANDIDATE:   RUOXI DONG

ABSTRACT:

Data with both longitudinal measurements and time-to-event outcomes is increasingly common in medicine and other fields. Statistical analysis is usually carried out using survival models with the longitudinal measurements specified as time-varying covariates. However that approach can be subject to bias due to measurement error. In recent years, joint model approaches have gained popularity as they can eliminate bias. In this project, we explore the joint modelling framework with a freely available software R package, JM. A Cox regression model with time-dependent covariate is compared to a joint model, with applications to data collected from a bipolar disorder study. For our dataset, regression coefficients from the Cox regression model with time-varying covariate are smaller than coefficients from the joint model. This may be caused by inappropriate treatment of measurement error in the Cox model.