Separating Gains and Losses in Health When Calculating the Minimal Important Difference for Mapped Utility Measures

Abstract

Rationale: The minimally important difference (MID) represents the smallest change on a patient reported outcome which can be considered important. Currently, only Walters and Brazier (2005) have published MID data for health utilities. Their results suggested that the MID preference-based measures were not equal, however these investigators combined patients reporting minimal gains and losses in health when determining the MID.

Objectives: The objective of this study was to determine whether patients perceiving minimal gains and losses in health should be treated equally when calculating the MID for a variety of mapped utility measures.

Methodology: A retrospective analysis of a longitudinal study in a Western US managed care population was conducted. Mapped utilities from the SF-36 and SF-12 were derived using validated of methods (Brazier, Lundberg, Nichol, Sengupta, and Shmueli) for baseline, year one, and year two assessments. The self-reported general health question of the SF-36 was used as an anchor to identify patients who considered their current health as 'somewhat better' or 'somewhat worse' than their health one year prior. The magnitude of utility change for these patients was compared to determine if gains and losses can be combined to identify one MID for change or if separate MIDs are required for each situation. The mean utility change was then calculated for patients who experienced an MID in health using the anchor. Effect sizes were calculated in accordance to Norman et al (2001) as a distribution-based method to determine the MID. Results were compared for each mapped utility method over both time periods.

Results: Of the 6,932 patients, 2,100 (33.33%) and 2,093 (30.23%) reported an MID using the anchor-based approach for years one and two, respectively. When combining patients who reported minimal gains and losses in health the mean absolute utility change was 0.032 (range = 0.020 to 0.047, sd = 0.113) over both time periods. The average effect sizes was 0.285 which supported that these MID changes were small according to Cohen's classification system. However, when separated, the mean MID utility change for those reporting 'somewhat better' and 'somewhat worse' health was 0.014 (range = -0.026 to 0.0302, sd = 0.106) and -0.064 (range = -0.049 to -0.103, sd = 0.116), respectively and these were significantly different (p<0.001). The average effect sizes were 0.131 and 0.551 indicating 'no' and 'moderate' change according to Cohen's classification system, respectively. These results were robust in subgroup analysis of chronic disease cohorts.

Conclusions: Convergent validity between anchor and distribution-based approaches for identifying the mean MID for these mapped utility measures was originally suggested when gains and losses were combined. The 0.03 estimate was in agreement with previous work which identified 0.04 as the SF-6D MID. Differences between mapped utility measures were statistically significantly as suggested by the same research showing differences between the SF-6D and EQ-5D MID (p<0.0001). But the principal differences occurred in the comparison of gains and losses. Caution must be employed if researchers consider combining minimal gains and losses in future utility MID calculations.

Keywords: Utility, Mapping, MID

JEL Classification: Z00