3091 — Estimating Preference Weighted Health Status from SF-36 Responses for the ACQUIP (VA) Patients
Derleth AM (Puget Sound HSR&D)
Diehr P (University of Washington)
Patrick D (University of Washington)
Reiber G (Puget Sound HSR&D; University of Washington)
McDonell M (Puget Sound HSR&D)
Fihn SD (Puget Sound HSR&D; University of Washington)
Several methods have been published to transform the SF-36 into a single preference weighted value that includes 0 for death and 1 for best health. The objectives of this study were (1) to calculate and compare 10 methods using 20,000 SF-36’s from veterans with a high prevalence of chronic disease and (2) provide guidance for selection of methods based on type of study being undertaken.
Preference weighted values were calculated for each respondent by each of 10 published transformations. Results were compared across transformations for respondents grouped by socio-demographics and health problems using regression and t-tests.
Preference weighted values were lower in those with worse health, as hypothesized. The average values for the same group of persons varied substantially by method. The preference weighted scores that were based on standard gamble methods were only moderately responsive to differences in level of health. Other transformations less based on expected utility theory were more powerful in detecting differences in health.
Preference weighted values obtained by transforming the SF-36 were very method dependent. In cost-effectiveness analyses, it would be important to be sure that the same transformation method is used for all preferences. If the objective is to obtain a single index from SF 36 responses for comparing outcomes or years of healthy life, other measures showed greater power and discriminative ability. The level of detail of the SF-36 data available may not permit calculation of the preferred index.
The SF-36 is commonly obtained in studies where preferences or other single index outcomes have not been obtained. It is often desired to transform the responses into a single index value for comparing outcomes among groups or for estimating cost effectiveness. This study provides results for respondents grouped by health problem from using 10 transformations in a very large set of SF-36 responses. Transformations are available that use question responses, domain scores or summary scores and the results in this study provide a guidance for selecting among the methods depending on data available and objective of the analysis.