| Quality of life (QOL) is now routinely assessed as an outcome in cancer clinical trials. The analysis of longitudinal QOL study is often confronted with the issue of missing data. Serious biases can occur when missing data mechanism is missing not at random (MNAR), which is often plausible in QOL studies. Under MNAR, a model which relates the missing data process to longitudinal QOL data is required.; In this study, a joint model for the longitudinal QOL data and event time (drop-out time), which assumes the random parameters of a subject's underlying response profile and the event time (or its transformation) follow the multivariate normal distribution, is applied to estimate and compare the average initial treatment effect and the mean slopes of QOL change from baseline over time in two treatment groups in order to deal with the non-ignorable missing data. (Abstract shortened by UMI.)... |
