臨床研究におけるQALYの利用とその推定に関する統計学的研究

概要

論文の詳細を見る
EQ-5D-3L is a questionnaire commonly used in Japan for measuring HRQoL (health-related quality of life) of different health states, when economic evaluations with QALY (quality adjusted life year) as an outcome measure are performed. To obtain each utility value for 243 health states described by EQ-5D-3L, an appropriate value set is needed. Several statistical models have been proposed for the estimation, including a non-Bayesian model, which was used to estimate the Japanese value set of EQ-5D-3L in 2002, and a new Bayesian model proposed in 2005. In this study, we compare the performances of the non-Bayesian model and the Bayesian model when estimating a value set for EQ-5D-3L. These models were compared in terms of prediction accuracy, systematic bias and violation of monotonicity, using the real data collected in the United States. We also propose an extended Bayesian model, and compared it with the other models. Our results did not show any major differences among predicted value sets of EQ-5D-3L using estimates of those three models, but the extended model had less mean absolute error of predicted utility values with the observed utility values, than that of other models.