Method: Individual level data (n = 2,211) collected from 3 different studies were separated into a derivation set (n=1,660) and an internal validation set (n=551). Data of 123 colon cancer patients was analyzed for external validation. Prediction models were analyzed using an ordinary least square regression, a two-part model and a multinomial logit model using 8 scale scores, two summary scores and its interaction terms of SF-36 as independent variables. EQ-5D index from the Korean value set or each dimension of EQ-5D were used for dependent variable on approaches. Performances of models were compared in the perspectives of the mean absolute errors (MAE) and R-square in the derivation, internal validation and external validation dataset.
Result: Our findings offered that three different scoring algorithms have similar performances in respective to MAE and R2. Considering familiarity and parsimony, OLS model including PF, BP, SF, RE and MH could be recommended as the final algorithm in this study. The MAEs for OLS models showed consistent results in both derivation set (0.087 to 0.109) and external validation set (0.082 to 0.097).
Conclusion: This study provided mapping algorithms to estimate EQ-5D index from SF-36 profile using individual based dataset and showed that algorithms had high explanatory power and low prediction errors.
See more of: SMDM Asia-Pacific Conference