Presentation (CTI)

OBJECTIVES: At the population level we evaluate the effect of treatment on patient reported outcomes (PROs) by comparing differences in mean change from baseline (CfB) scores, evaluating if differences are meaningful using thresholds for between-group means (MID). We also conduct responder analysis where the CfB is dichotomised based on individual meaningful within-person change (MWPC) thresholds. Responder analysis may be more intuitive for interpretation, but there is concern regarding lower statistical power. We aimed to explore the relationship between using continuous PROs and responder analysis and the impact on power.
METHODS: We illustrated the relationship in PRO endpoints using EORTC QLQ-C30 physical functioning score for mean CfB and proportion of responders in a phase III clinical trial (treatment (TRT) vs standard of care (SOC)). We assumed CfB scores to be normally distributed, used CfB MIDs from 5 to 15, and dichotomised PRO score using a deterioration MWPC of ≥ -5 points. We calculated statistical power using t-test and binomial test for proportion for range of trial sizes.
RESULTS: There is a direct relationship in the power of CfB means and proportion of responders analyses, which is maintained across ranges of MIDs. For example, mean CfB 0 points (TRT) and -10 (SOC) (MID 10 points) relates to 58% responders (TRT) and 42% (SOC). Power of responder analysis was lower than CfB mean analysis, however, for MIDs > 10 points, in clinical trials over 300 patients the power of both analyses was acceptable at >80%, increasing to >95% for trials with over 500 patients.
CONCLUSIONS: Analysis of the proportion of patients with improved or stable PRO (responders) is a valuable and interpretable PRO endpoint with high statistical power in the majority of settings in phase III clinical trials; analysis of both proportion of responders and CfB means are complementary approaches to holistically interpret PRO endpoints.