Methods for Addressing Missing Patient-Reported Outcome Data in Longitudinal Clinical Trials: A Secondary Analysis of the Frequent Hemodialysis Network Trials

Abstract

Patient-reported outcomes from longitudinal clinical trials on hemodialysis typically have missing data, yet there is a lack of empirical evidence to guide appropriate analysis and interpretation. To compare approaches for analysis and interpretation of patient-reported outcomes in the Frequent Hemodialysis Network (FHN) Trials. Secondary analysis of two multicentre randomized controlled trials. Hemodialysis centres and home hemodialysis programs in North America. Participants in the FHN Daily (n = 245) and Nocturnal (n = 87) Trials. RAND-36 physical health composite scores at baseline, months 4 and 12. Complete case analysis (CCA), mixed model for repeated measures (MMRM), and multiple imputation (MI) were used to estimate mean differences in scores at Month 12. To relax the normality assumption and enhance interpretation of mean differences, rank-based analyses were performed to estimate the win probability (WinP) that a treated participant has a better score than (or wins over) a control participant at Month 12. Data complete at both post-randomization timepoints were available for 74% to 75% of participants. In the Daily Trial, estimates of Month 12 mean difference (95% CI) were 3.04 (0.66 to 5.42, p = 0.012) by MMRM, 2.97 (0.42 to 5.52, p = 0.023) by MI, and 2.76 (0.31 to 5.21, p = 0.027) by CCA. Estimates of WinP were 0.59 (0.52 to 0.65, p = 0.007) by MMRM and 0.59 (0.52 to 0.65, p = 0.011) by CCA. In the Nocturnal Trial, treatment effects were small, imprecise, and not statistically significant across all methods. This was a secondary analysis and rank-based analyses were exploratory. Approaches that retained participants with partially observed data can be more efficient than complete-case analyses. Win probability analyses provided an alternative to mean differences in patient-reported outcome measure scores.