Can Prospect Theory Corrections Improve Consistency of Estimated Health Utilities?

Presenter: Michael Nichol, USC

Abstract

Background: Inconsistency in measurements from different health utility methods within an individual makes medical decision-making difficult. The inconsistent measurements may due to some level of loss aversion, scale compatibility, and probability weighting. Objectives: 1) To derive seven estimated utilities from the SF-12v2™, and apply prospect theory to generate corrected probability equivalence (PE) estimated utilities. 2) To test the inconsistency among seven estimated utilities before and after PE corrections. Method: We analyzed the data from 2005 Household Component Full-Year Files in the 2005 Medical Expenditure Panel Survey (MEPS). Seven estimated utilities were derived from the SF-12v2, including HUI-3/VAS item models (IM) and categorical models (CM) from the Sengupta-Nichol, Brazier SF-6D, Lundberg VAS, and Sullivan EQ-5D algorithms. Corrected PE estimated utilities were calculated. Paired t-tests were used to compare the estimated utilities before and after PE correction. Paired t-tests or effect sizes were used to compare SF-6D and other six estimated utilities for both PE corrected and uncorrected measures. The effect sizes were calculated from the mean differences between SF-6D and other six estimated utilities divided by standard deviation of SF-6D. Results: The analysis data included a total of 19,475 individuals who completed the SF-12v2 survey. Mean age was 45.5 years (range 18 to 85), 45% were male. Mean estimated utilities ranged from 0.71 (VAS-IM and VAS-CM) to 0.88 (EQ-5D). Mean differences between SF-6D and other six methods ranged from 0.01 (VAS-CM/VAS-IM) to 0.16 (EQ-5D) (all p<0.0001). Overall, the seven PE corrected utilities were consistently and significantly lower than uncorrected utilities, the differences ranged from 0.188 (EQ-5D) to 0.259 (VAS-CM and SF-6D) (all p < 0.001). VAS-IM (ES=0.05) and VAS-CM (ES=0.06) showed no effects; HUI3-IM (ES=-0.36) and HUI3-CM (ES=-0.41) showed small effect; Lundberg VAS (ES=-0.53) showed median effect; and EQ-5D (ES=-1.16) showed large effect as compared to SF-6D. Mean PE corrected estimated utilities ranged from 0.45 (VAS-CM) to 0.69 (EQ5D). Mean differences between PE corrected SF-6D and other six methods ranged from 0.01 (VAS-CM/VAS-IM) to 0.23 (EQ-5D) (all p<0.0001). The differences slightly increased when compared with those for uncorrected estimated utilities except for two methods (VAS-CM and VAS-IM). Effect sizes for PE corrected estimated utilities also slightly increased when compared with the ES for uncorrected estimated utilities. Conclusions: Corrected PE on estimated utilities under prospect theory can not correct the inconsistencies among different estimated utility measurements in a U.S. national population. Developing different correction formulas based on particular populations should be considered.