WEIGHTING CLINICAL TRIAL ENDPOINTS: AN APPLICATION OF CONJOINT ANALYSIS TO EVIDENCE-BASED MEDICINE
Ahmed M. Bayoumi, MD, MSc, St. Michael's Hospital, University of Toronto, Toronto, ON, Canada and Joel G. Ray, MD, MSc, St. Michael's Hospital, University of Toronto, Toronto, ON, Canada.
Purpose: Clinical trials often use composite or multiple endpoints to evaluate treatment effects, but few studies consider the relative importance of individual components. We used utility theory to derive a weighting function for heart failure outcomes. Methods: We designed a computer-based survey of heart failure related outcomes reported in the Digoxin Intervention Group (DIG) Trial (death, worsening dyspnea, hospitalization, dizziness, fatigue, diarrhea, and nausea). For each outcome, we defined 1 to 2 levels and developed 16 scenarios using a fractional factorial design. The scenarios were randomly paired and presented as dichotomous choices. Participants were allowed to change answers and skip questions. The surveys were pilot tested in 25 individuals and administered by 2 interviewers. We analyzed responses using a random effects probit regression model and rescaled responses so that having no adverse events or outcomes was associated with a score of 100 and being dead was associated with a score of 0. We re-analyzed the DIG trial by assigning utility weights to each interval between study visits and calculating quality-adjusted life days by multiplying utility weights by interval duration; in the current analysis we focused only on deaths, hospitalizations, and worsening dyspnea. We report the average utility for the time participants were in the study to account for differential rates of follow-up and censoring. Results: We evaluated 2410 responses from 301 participants. Having moderate to severe dyspnea was rated about half as bad as being dead (score 49.6, 95% confidence interval 40.8 to 56.5). The corresponding scores for moderate to severe dizziness, fatigue, and diarrhea were 59.5 (52.0 to 65.4), 77.3 (72.4 to 81.2), and 76.0 (71.1 to 79.9). Each hospitalization was associated with a 5.4 unit (5.2 to 5.6) decrease in the utility score. Moderate to severe nausea was not associated with a significant change in utility. The average treatment outcomes utility for patients treated with digoxin was 84.4 (95% CI 83.8 to 85.0) and for patients treated with placebo was 85.7 (85.0 to 86.3). The difference was clinically small but statistically significant (1.3, 95% CI 0.4 to 2.1). Conclusions: Conjoint analysis offers a feasible and transparent method of deriving weights for clinical study endpoints, reflecting their varying importance to patients. This method can be applied to clinical trials reporting composite or multiple outcomes.