Assessment of Responder Analyses for Patient‐Reported Outcomes With Varying Variability Between Intervention Groups: Maximum Responder Threshold and Effect

Continuous outcomes are frequently dichotomized to classify subjects as "responders" or "nonresponders" with the difference in proportion of response between treatment and control groups defined as the "responder effect," which has been used in regulatory and other settings. Despite the potential benefit of responder analyses in simplifying analysis and lending interpretation of clinically meaningful change between intervention groups, the dichotomization of a quantitative patient-reported outcome for this purpose has come under scrutiny for multiple reasons. Previous research has examined the specific case where treatment and control group outcomes are normally distributed and differ in their means, with a common within-group standard deviation, and found several limitations of conventional single-threshold responder analyses that pose interpretational challenges. The current research complements and supplements previous methodological evidence and conclusions by quantifying and depicting responder analyses for the broader and practical situations of varying group standard deviations. This expanded exposition is grounded in conceptual, theoretical, and illustrative underpinnings. In doing so, new quantitative learnings and findings, with visual insights and enriched understanding, emerge on what threshold yields the maximum responder effect and what is the maximum responder effect. Despite the limitations of responder analyses, which researchers should continue to exercise caution, responder analyses (along with cumulative distribution function) can still be useful to augment interpretation of a patient-reported outcome and an overall clinical benefit of a treatment when applied responsibly and judiciously. As such, this article is intended to provide advanced understanding and additional insights into responder analyses to aid in prudent interpretation.