Joint modeling of longitudinal and time-to-event data with missing time-varying covariates in targeted therapy of oncology

Abstract

In many clinical studies, interest lies in predicting a time-to-event outcome on the basis of longitudinal measurements. These types of studies are called joint modeling. The presence of missing values in the response as well as in covariates, which is very common due to dropout of patients from the study, influences the necessary inferences. This article presents an effective way to handle the missing values in the covariates and responses. Different imputation techniques are compared to showcase an efficient way of generating missing data and to study their impacts on joint modeling. A simulation study is carried out to replicate the complex data structure and conveniently perform our analysis to show its efficacy. We considered different correlation structures like Auto-regressive structure of order 1, independent structure, exchangeable correlation structure, and arbitrary correlation structure among the longitudinal history to obtain the most efficient strategy for imputing missing observations in joint modeling scenario. Dataset consisting of longitudinal outcome and survival history of polymerase chain reaction test is conveniently used to show the effectiveness of our study on real data. Extensive use of R programming language has been assessed to perform the necessary analysis and estimation of parameters.