Abstract
The health-related quality of life is a major objective in oncology clinical trials to improve patients’ care and better evaluate the impact of the treatments on their everyday life. Auto-questionnaires are usually used to measure this endpoint. In this work, different statistical models for the longitudinal analysis of health-related quality of life in oncology are proposed and applied to clinical trial data. First, we present different models derived from the item response theory (IRT) to achieve a longitudinal analysis directly on raw data (multi-response outcomes) for each dimension. Within the generalized linear mixed model background, a conceptual selection of the IRT models concluded that the graded response model seems to be the most suitable. Then, we propose a structural equation model which allows taking into account the multidimensional nature of data at each time and the longitudinal aspect induced by the repeated measurements. At each measurement time, the model allows to link all the observed variables issued from the questionnaire given explanatory variables. Two factors are estimated, each summarizing a set of observed variables. The longitudinal analysis is performed on the global health status and on the factors, thus reducing the number of tests. Finally, an approach based on a mixture of mixed models is used to obtain latent classes from quality of life trajectories. The approach has resulted in the identification of homogeneous subpopulations and their latent trajectory according to specific patient profiles.