Partial credit model with random effects for longitudinal analysis of quality of life in oncology
Résumé
The health-related quality of life (HRQoL) has become one of priority objectives of clinical trials in cancer research to evaluate efficiency of care. The EORTC has developed the cancer-specific QLQ-C30 self-questionnaire evaluating five functional scales, nine symptom scales and global health-status/QoL scale. Commonly used models have statistical and practical limitations. The aim of this study was to propose an alternative method for longitudinal analysis of HRQoL, without data modification. Built on the modern Item Response Theory (IRT), our approach considers the HRQoL as a latent variable directly estimated from raw data. For polytomous items, we extended the partial credit model to a longitudinal analysis (LPCM), thereby modeling the latent variable as a function of time and other covariates. LPCM can be seen as a generalized linear mixed model with a multinomial logit link. The main objective is to apply LPCM and then to compare results with the usual methods. We propose an alternative approach to the commonly used linear mixed model. Our model directly takes into account the ordinal data coming from self-reported questionnaires. LPCM was used to analyze the 15 HRQoL scales and to compare the impact of both treatments. LPCM allowed a more precise analysis of each scale or each item, which could be easily plotted for a clearer interpretation.