Abstract

Background

Preference-based health utilities are used in economic analyses of disease burden and health care interventions. When specifically designed instruments cannot be applied, mapping algorithms for non–preference-based instruments can be used for prediction of health utility scores.

Objectives

To develop a mapping algorithm for the Chronic Liver Disease Questionnaire-Hepatitis C Version (CLDQ-HCV), the hepatitis C virus–specific quality-of-life instrument.

Methods

We used a sample of patients with HCV who completed the short form 36 health survey and the CLDQ-HCV in clinical trials; six-dimensional health state short form (SF-6D) utilities were derived from the 36-item short form health survey. Regression models with components of the CLDQ-HCV being predictors and SF-6D being the outcome were developed and tested in an independent testing set and in clinically significant subpopulations.

Results

The sample of 34,822 records was split (4:1) into training and testing set. Simple mixed models had a root mean square error up to 0.088; predicted and observed utilities were highly correlated (Pearson correlation 0.81–0.82) although predicted utilities were underestimated in the range closest to perfect scores. Generalized linear models had better average accuracy (root mean square error up to 0.0839; correlations up to 0.844) and significantly better accuracy in the highest values (median error up to 0.065). Accuracy in the independent testing set was nearly identical, and so was accuracy in patients with compensated and decompensated cirrhosis; the errors of group means were less than 0.015.

Conclusions

A number of linear models for mapping domains or items of CLDQ-HCV to SF-6D health utilities have been developed. The models have excellent accuracy at the group level. Predicted health utility scores can be used in further economic analyses involving patients with HCV.