1074 — Estimating Quality-Adjusted Life Years Using SF-36 Questionnaire Responses: Challenges in a Population of Veterans with Diabetes
Sinha A (REAP - East Orange, NJ), Rajan M
(REAP - East Orange, NJ), Pogach L
(REAP - East Orange, NJ)
Health economic analyses of new medical interventions require preference-weighted measures of quality of life, utilities to estimate quality-adjusted life years (QALYs). Using existing quality of life data, e.g. SF-36, to estimate utilities has the potential to efficiently provide investigators with utility information, based on large samples of responses.
Our objective was to compare three previously published methods of transforming SF-36 data into utilities, using available survey responses from veterans with recent-onset diabetes.
We utilized the Large Veterans Health Survey (LVHS) data from 1999 to select respondents with recent-onset diabetes. We used three previously published regression-based transformations (A = Nichol et al 2001, B = Brazier et al 2004, C = Brazier et al 2002) to estimate utilities, on a scale of 0 to 1, associated with specific health states and combinations of health states.
47,779 LVHS respondents had recent-onset diabetes. Utilities estimated by method A were systematically found to be less than utilities estimated by B or C. For example, among patients with chronic renal failure, utility [U, mean(sd)] was 0.596 (0.149), 0.694 (0.100), 0.649 (0.106) using A, B, and C respectively. The rank order remained consistently U(B) > U(C) > U(A) when patients were stratified by GI, pulmonary, musculoskeletal, neurological, or substance abuse conditions. Pearson’s correlation showed that U(B) was modestly correlated to U(A) and to U(C) (r = 0.51, 0.49 respectively). U(A) and U(C) were highly correlated (r = 0.84) for respondents with < = five health conditions. U(A) and U( C) were moderately correlated (r = 0.73) for those with > = six health conditions.
Because utilities factor into the denominator of cost-effectiveness ratios, small differences in utility values can have a large impact on cost-effectiveness results. Among diabetic veterans, we found systematic differences in the utilities estimated using three transformations of SF-36 data. In particular, method B may not capture all available SF-36 information, resulting in inconsistent utility estimates relative to methods A and C.
The use of SF-36 transformations to estimate utilities needs further development using empirically collected data before routine incorporation into cost-effectiveness analyses.