Purpose: To compare four previously-published methods of transforming Short Form 36 and 12 Item Health Surveys (SF-36 /SF-12) data into utilities, using survey responses from veterans with diabetes (DM) and chronic kidney disease (CKD); to determine if these transformations are valid for discriminating utility losses (disutilities) as CKD severity increases; and to estimate the disutility associated with progressive CKD.
Methods: Veterans with DM were selected who responded to the Large Veterans Health Survey in 1999 and divided into those with recent-onset DM (duration of ≤3 years) and prevalent DM (duration >3 years). Surveys were merged with data from the Diabetes Epidemiology Cohort, a well-established longitudinal cohort of veterans with diabetes. ICD-9 and procedure codes determined if respondents were on dialysis or had end-stage renal disease (ESRD). If subjects did not have ESRD/dialysis, serum creatinines were used to stage CKD. Four previously-published SF-36 /SF-12-to-utility transformations (A = SF-12 to SF-6D, B = SF-36 to SF-6D, C = SF-36 to HUI2, D= SF-12 to VR-6D) were used to estimate utilities (U) for each respondent. Generalized linear regression models estimated the disutility associated with each CKD stage, after adjustment for demographics, socio-economics, and co-morbidities.
Results: Of 67,694 diabetic patients, 22,273 had recent-onset and 45,691 patients had prevalent DM. The figure gives mean utilities by each method for recent-onset DM patients; results were similar for prevalent DM. Method A did not discriminate utility by CKD stage, among either recent-onset or prevalent diabetics. The remaining three methods showed a stepwise decline in utility as CKD stage increased. The rank order was consistently U(A)>U(C)>U(B)>U(D). In recent-onset DM, mean disutilities associated with increasing CKD stage differed significantly by transformation method (p<0.0001) and ranged between 0.0017 - 0.0042, -0.0067 - -0.0019, -0.0256 - -0.0041, and -0.0116 - -0.0091 for CKD stages 2, 3, 4/5, and ESRD/dialysis respectively; results were similar for prevalent DM.
Conclusions: In a cross-sectional analysis of diabetic veterans, systematic differences were found in utilities estimated using four transformations of SF-36 /SF-12 data. In particular, method A may not capture all available SF-36 information, resulting in inconsistent utility estimates relative to other methods. CKD-associated disutility values differed significantly between methods at each CKD stage, suggesting that selection of transformation method requires careful consideration of potential floor and ceiling problems.