METHODS: We used initial data from the ongoing National Health Measurement Study, a random digit dialed telephone interview of 1262 US adults aged 35-89 collected between June 2005 and February 2006. This survey included the EQ-5D with VAS, HALex, HUI2/3, SF-36v2, and QWB-SA. From these HRQoL instruments we calculated 7 HRQoL summary scores: EQ-5D with US weights, HALex, HUI2, HUI3, SF-6D, QWB-SA, and VAS. We used the 1094 respondents who completed all indices to regress each summary score on age, sex, and each of 13 health-related conditions. The health-related conditions included self-reported congestive heart disease or myocardial infarction, stroke, diabetes, arthritis, eye disease, sleep disorder, respiratory disease, anxiety or depression, GI ulcer, thyroid disorder, overweight, obese, and current smoker. Regressions used survey weights for nationally representative estimates in STATA 8. The regression coefficient estimate for each condition is the estimated effect size for the condition. We compare the average effect sizes from each scoring system, the correlation of effect sizes across scoring systems, and the mean absolute differences in effect sizes between scoring systems. We also test a transformation (Franks et al 2006) meant to standardize effect sizes between scoring systems.
RESULTS: The age- and sex- adjusted effect sizes averaged across the 13 health-related conditions ranged from .077 in the SF-6D scoring system to .122 in the HUI3 scoring system. Correlations of effect sizes ranged from .630 (HUI3 and VAS) to .917 (HUI2 and HUI3). The average absolute differences in effect sizes ranged from .022 (|QWB-SA - SF-6D|) to .058 (|HUI3 – VAS|). The Franks et al transformations had inconsistent effects on the relationship of effect sizes between scoring systems.
CONCLUSIONS: The impact of health conditions, as measured by the age- and sex- adjusted effect of health-related conditions, varies by the HRQoL summary score system used. These effect sizes, however, are highly correlated across preference-based summary scores. Our results imply a league table of cost-effectiveness ratios with one summary score system should be similar to one based on another system but the absolute ratios will differ.