Analysis and Interpretation of a Hierarchical Outcome in a Small Pilot Randomized Trial: An Example in the CORTAHF Trial

This is the authors' abstract. We don't add key points for this paper.

Aims Hierarchical composite endpoints (HCE) analyzed using generalized pairwise comparisons (GPC) have been increasingly adopted in cardiovascular trials. Stratified, but not covariate-adjusted, win statistics have been frequently presented. We apply methods to a post hoc HCE for the randomized 100-patient CORTAHF pilot trial of short-term corticosteroid therapy in acute heart failure (AHF). Methods and Results The HCE comprised 90-day death, 90-day HF readmission, 30-day worsening HF, and 30-day change in EQ-5D visual analogue scale (EQ-VAS) score from baseline. Win statistics and patient ranks from worst to best outcome were derived from GPC. These ranks were used to produce site-stratified and baseline EQ-VAS-adjusted win odds. Analysis of individual components suggested beneficial prednisone effects on 90-day HF readmission and 30-day WHF adverse events, with little effect on 30-day change in EQ-VAS. The win ratio and win odds were 1.82 and 1.71, respectively (both p<0.05), with 10% ties remaining after most comparisons were resolved at the last tier. HF readmission contributed most to the overall win difference of 26.1%. Stratification and baseline covariate adjustment using patients’ ranks were feasible, but in this example did not substantially alter the estimated win odds. Conclusions Patients’ ranks can be derived from GPC and used to estimate stratified and covariate-adjusted win statistics as well as to visualize the treatment effect. Use of a well-constructed HCE in a small pilot trial should be considered; a positive signal may support further examination of the treatment. A full analysis can aid in interpretation.