Improving the Analysis of KCCQ Endpoints in Heart Failure Clinical Trials

Background The Kansas City Cardiomyopathy Questionnaire (KCCQ) is frequently used in heart failure (HF) clinical trials to evaluate treatment effects on function and symptoms. However, due to the 0–100 boundedness in KCCQ scores, conventional mean change from baseline analysis can underestimate effects for lower—and overestimate effects for higher—baselines. This study demonstrates key issues with conventional statistical methods for analyzing treatment effects on KCCQ in randomized trials and evaluates alternative statistical methods that appropriately account for 0–100 boundedness of scores. Methods We conducted clinical trial simulations and re-analyzed a real HF randomized trial—the PRIORITIZE-HF phase II trial of sodium zirconium cyclosilicate. KCCQ change from baseline was analyzed with conventional ANCOVA models, and compared to methods that account for the 0–100 score boundedness: ANCOVA interaction models, Tobit regression, and Beta regression. Mean treatment effects and extent of bias are summarized by method. Results There were clear baseline-dependencies in mean effects for KCCQ score change, both in simulated trials and in PRIORITIZE-HF. In the real trial, the conventional ANCOVA model mean effect on KCCQ-overall summary score was + 2.58 overall while for methods allowing effects to depend on baseline – ANCOVA interaction models and Beta regression – mean effects were over twice as large at baseline = 30 (+ 6.33 to + 6.75) and less than half at baseline = 70 (− 0.31 to + 0.89). Conclusions Conventional analyses of treatment effects on overall mean KCCQ score changes lead to misinterpretations in clinical trials, but this can be mitigated by using methods allowing for baseline-dependent effects.