Utility score mapping from FACT-G to EQ-5D-5L and SF-6Dv2 in breast and colorectal cancer patients: a focus on beta mixture models.

objectiveTo develop algorithms mapping the Functional Assessment of Cancer Therapy-General Scale (FACT-G) onto the EuroQol 5-Dimension 5-level (EQ-5D-5 L) and the Short-Form Six-Dimension version 2 (SF-6Dv2) for patients with breast or colorectal cancers.

methodsAn online survey was conducted to collect responses to FACT-G, EQ-5D-5L, and SF-6Dv2 from cancer patients in Quebec, Canada (N = 202). Linear models including ordinary least squares (OLS), Censored Least Absolute Deviations (CLAD), the robust MM-estimator model (MM), as well as mixture models including two-part model (TPM), and beta-based mixture (betamix) model were used. Mean absolute error (MAE), root mean squared error (RMSE), R2, Bayesian information criteria (BIC), and limits of agreement (LOA) calculated using the five cross-validation to assess the predictive ability of the models. Furthermore, the distribution of observed versus predicted values was assessed using Bland-Altman plot.

resultsBased on RMSE and MAE, mixture models better performed than linear models. The betamix model with truncation that included domains and squared terms was the best-performing algorithm for EQ-5D-5 L (MAE = 0.0518, RMSE = 0.0744, R2 = 46.40%, and LOA=-0.166 to 0.165) and SF-6Dv2 (MAE = 0.1375, RMSE = 0.1764, R2 = 35.32%, and LOA=-0.356 to 0.337). EQ-5D-5 L and SF-6Dv2 utility scores for better health states and more severe health states were underestimated and overestimated, respectively.

conclusionThis study developed robust algorithms to estimate EQ-5D-5 L and SF-6Dv2 utilities from FACT-G. Consistent with the recent literature, the betamix model outperformed all other econometric models considered in this study. This suggests that mixture models generally exhibit higher performance and are the best choice for mapping.