OBJECTIVES: To refine Health-related Quality of Life (HRQL) Physical and Mental Comorbidity Indices into one single composite index, which can be feasible for risk adjustment in the studies of HRQL METHODS: The 2007 Medical Expenditure Panel Survey (MEPS), a nationally representative database in the United States, was used for refining the comorbidity index. The SF-12 in the MEPS was the measure of HRQL and Clinical Classification Codes in the MEPS indicated individual disease conditions. The least absolute shrinkage and selection operator (LASSO) was applied to select disease conditions significantly associated with the SF-6D scores which were converted from SF-12 physical and mental data. Confirmatory Factor Analysis (CFA) was used to identify the best disease dimensional structure in the comorbidity index. The validation in terms of prediction accuracy for HRQL was evaluated in the 2009 MEPS. There is no overlap of people between the 2007 and 2009 databases ensuring that the validation set is independent of the data used for refining the index. Prediction errors and model R2were compared between our refined index and Charlson Comorbidity Index, a widely used risk adjustment tool today. RESULTS: Fourteen clinical conditions were identified by LASSO and then categorized into 9 disease dimensions by CFA. Statistical weights for disease dimensions were derived based on the regression model predicting HRQL. The R2in the model using our refined index was higher than that in the model using Charlson index (0.21 vs. 0.08). The average prediction error from the model using our index was smaller than that obtained from the model using Charlson index (0.018 vs. 0.021). CONCLUSIONS: In the context of HRQL outcomes, comorbidity index scores estimated by our refined index had better prediction than those from Charlson index. Further validation of our refined index is needed in other measures of HRQL and health care settings.