Probabilistic mapping of the health status measure SF-12 onto the health utility measure EQ-5D using the US-population-based scoring models

Probabilistic mapping of the health status instrument SF-12 onto the health utility instrument EuroQol-5 dimensions (EQ-5D)-3L using the UK-population-based scoring model showed encouraging results as compared to other mapping methods, although its predictive performance using the US-population-based EQ-5D scoring models has not been investigated. In addition, a new and improved US-population-based EQ-5D scoring method has recently been developed and suggested for use in applications that required US societal health state values. In this study, we assessed predictive performance of the probabilistic mapping approach using the US-population-based scoring models on EQ-5D utility scores based on SF-12 responses and compared the results with those of other mapping methods. Using a sample of 19,678 adults from the 2003 Medical Expenditure Panel Survey, we evaluated the predictive performance of probabilistic mapping using Bayesian networks, response mapping using multinomial logistic regression, ordinary least squares, and censored least absolute deviations models by implementing a fivefold cross-validation method. The EQ-5D utility scores were generated using two US-population-based models: D1 and MM-OC. Overall, the probabilistic mapping approach using Bayesian networks consistently outperformed other mapping methods with mean squared errors (MSE) of 0.007 and 0.007, mean absolute errors (MAE) of 0.057 and 0.039, and overall R (2) of 0.773 and 0.770 for the US-population-based EQ-5D scoring D1 and MM-OC models, respectively. The probabilistic mapping approach can be useful to estimate EQ-5D utility scores from SF-12 responses with better predictive measures in terms of MSE, MAE, and R (2) than other common mapping methods.