PREDICTING THE HUI3 AND EQ-5D FROM THE MOS-HIV IN PATIENTS WITH ADVANCED HIV
Method: We developed and validated mapping algorithms using data from two separate HIV clinical trials. We divided data from the first trial into estimation (n=294 patients) and internal validation (n=73) datasets; data from the second trial formed the external validation dataset (n=168). We compared ordinary least squares (OLS) with the more flexible beta regression method; the 10 MOS-HIV domain scores served as predictor variables. We assessed model performance using mean absolute error (MAE) and root mean square error (RMSE).
Result: Both the OLS and beta regression models accurately predicted the mean HUI3 and EQ-5D scores in the external validation sample. The mean observed HUI3 score was 0.837, while the predicted score from the OLS model was 0.817; the mean observed EQ-5D score was 0.897, while the OLS-predicted score was 0.883. Model fit, in terms of mean absolute error values in the external validation sample, ranged from 0.068 to 0.104. Both the OLS and beta regression models predicted HUI3 and EQ-5D values that were too high for patients in poor health. For the sickest tertile, the mean observed HUI3 was 0.13; both the OLS and beta regression models predicted a mean of 0.38.
Conclusion: The proposed mapping algorithms can be used to predict HUI3 and EQ-5D health state values from the MOS-HIV, with the caveat that overprediction may pose a problem in samples where a substantial proportion of patients are in poor health. These algorithms may be useful for estimating health state values for cost-effectiveness studies when HUI3 or EQ-5D data are not available.