A Practical Method for Estimating Generalized Risk-Adjusted Cost-Effectiveness Utilities and Willingness-to-Pay...

Commonly used health-related utility measures fail to account for risk preferences. Generalized Risk-Adjusted Cost-Effectiveness (GRACE) solves this problem, but it currently requires visual analog scale (VAS) measures of health states, which may not be available to the analyst. We develop an empirically based approach for constructing GRACE utilities from existing time trade-off (TTO) utilities.

Using nationally representative patient-level data on both VAS and TTO measures of health states, we estimate a mapping from TTO to VAS. Using published estimates of GRACE utility over the VAS domain, we present estimates allowing analysts to map from existing TTO utility to GRACE and to make corresponding willingness-to-pay threshold corrections.

Compared with expo-power and hyperbolic absolute risk-aversion models, a linear model provides the best mapping from TTO utilities to VAS measures of health state. We find TTO utilities approximate VAS reasonably well for moderate health state utilities but suffer from more approximation error elsewhere. Notably, TTO underestimates VAS health for the sickest health states, eg, for TTO levels below 0.4.

A straightforward linear mapping permits analysts to estimate GRACE using traditional TTO health state utilities. Many existing studies use time-trade-off utilities in place of VAS. Our results suggest that this existing approach underestimates the GRACE value of health improvements for the sickest patients. Our study provides a solution without requiring generation of new data.