Context: Evaluative health-related quality-of-life instruments used in clinical trials should be able to detect small but important changes in health status. Several approaches to minimal important difference (MID) and responsiveness have been developed.
Objectives: To compare anchor-based and distributional approaches to important difference and responsiveness for the Wisconsin Upper Respiratory Symptom Survey (WURSS), an illness-specific quality of life outcomes instrument.
Design: Participants with community-acquired colds self-reported daily using the WURSS-44. Distribution-based methods calculated standardized effect size (ES) and standard error of measurement (SEM). Anchor-based methods compared daily interval changes to global ratings of change, using: (1) standard MID methods based on correspondence to ratings of "a little better" or "somewhat better," and (2) two-level multivariate regression models.
Participants: About 150 adults were monitored throughout their colds (1,681 sick days.): 88% were white, 69% were women, and 50% had completed college. The mean age was 35.5 years (SD = 14.7).
Results: WURSS scores increased 2.2 points from the first to second day, and then dropped by an average of 8.2 points per day from days 2 to 7. The SEM averaged 9.1 during these 7 days. Standard methods yielded a between day MID of 22 points. Regression models of MID projected 11.3-point daily changes. Dividing these estimates of small-but-important-difference by pooled SDs yielded coefficients of .425 for standard MID, .218 for regression model, .177 for SEM, and .157 for ES. These imply per-group sample sizes of 870 using ES, 616 for SEM, 302 for regression model, and 89 for standard MID, assuming alpha = .05, beta = .20 (80% power), and two-tailed testing.
Conclusions: Distribution and anchor-based approaches provide somewhat different estimates of small but important difference, which in turn can have substantial impact on trial design.