Comparison of measured utility scores and imputed scores from the SF-36 in patients with rheumatoid arthritis

Abstract

Purpose: We sought to evaluate 3 methods for imputing utility-based outcomes from clinical trial data measured using the Medical Outcomes Study 36-Item Short Form (SF-36).

Subjects: Our subjects included 131 male and 505 female adults (mean age, 55.42 +/- 12.59 years) who were participating in a randomized clinical trial evaluating a new treatment of adults with rheumatoid arthritis (RA).

Measures: Participants completed the SF-36, 2 versions of the Health Utilities Index (HUI-2 and HUI-3), the EuroQol EQ-5D, and the Health Assessment Questionnaire (HAQ). SF-36 scores were transformed to utility-based scores using 4 methods developed independently by Fryback, Nichol, and Brazier.

Results: All 4 imputed scores were significantly correlated with HUI-2, HUI-3, EQ-5D, and the disease-specific HAQ scores at baseline and at the end of the clinical trial period (P < 0.05). Changes in the imputed scores from baseline to end of study also were significantly correlated with corresponding changes in the measured utility scores and the HAQ score (P < 0.0001). For all imputed and measured scores, changes from baseline were associated with the clinical assessments, ACR20 and ACR50. The associations were stronger for the utility-based measures than the imputed indices. Both imputed and measured scores were sensitive to change in the clinical trial. However, mean scores for the HUI-3 and the Brazier VAS were significantly lower than for the other measures.

Conclusion: Imputed utility-based score estimates are significantly correlated with measured utility outcomes. However, the imputed measures had more constrained variability, showed poorer correspondence to the ACR20 and ACR50 benchmarks, and predicted less than half of the variance in actual utility-based outcomes. Therefore, directly assessed, not imputed, utility-based measures should continue to be favored for cost-effectiveness analysis.