Mapping EORTC-QLQ-C30 scores to EQ-5D-5L utility values for gynecological cancer survivors in follow-up care.

PURPOSE: To develop mapping algorithms from the EORTC-QLQ-C30 to EQ-5D-5L utilities for gynecological cancer survivors in follow-up care, enabling cost-effectiveness analyses when EQ-5D-5L utilities are unavailable. METHODS: We used data from a Norwegian multicentre longitudinal study of gynecological cancer survivors in post-treatment follow-up (LETSGO), including 663 patients with 2,624 observations. Ten model types were estimated for three EQ-5D-5L value sets (Norwegian, UK, US): ordinary least squares, Tobit, beta and fractional logistic regression, linear mixed models, an adjusted limited dependent variable mixture model, and two-part models combining logistic regression for the probability of perfect health with OLS, mixed, beta or fractional logistic regression for utilities below 1. Each was fitted with three prespecified covariate sets comprising all EORTC-QLQ-C30 scales, with and without age, comorbidities and treatment type. Stratified five-fold cross-validation at the patient level was used for internal validation. Performance was assessed using mean absolute error (MAE), root mean squared error (RMSE), proportion of predictions with absolute error (AE) ≤ 0.05 and Lin's concordance correlation coefficient (CCC) and summarized using an average ranking value. RESULTS: A two-part fractional logistic model with all covariates ranked highest, with MAE 0.0539-0.0660, RMSE 0.0806-0.0960, AE ≤ 0.05 for 54.8-67.1% of observations and CCC 0.7559-0.8034. Differences between the best-performing specifications were small. Accuracy declined at low utility levels, though overall mean bias was minimal. CONCLUSIONS: The algorithm enables estimation of EQ-5D-5L utilities from EORTC-QLQ-C30 scores in populations with severity profiles similar to the LETSGO cohort. Accuracy is reduced in poor health states.