Examining the Influence of Component Outcomes on the Composite at the Design Stage
This article requires a subscription to view the full text. If you have a subscription you may use the login form below to view the article. Access to this article can also be purchased.
While the literature simultaneously promotes and condemns composite end points, they are often used as the primary end point in clinical trials. The selection of the individual components to form the composite may be arbitrary, and the various algorithms suggested for combining component outcomes differ in their attempt to maximize clinical meaning while retaining statistical power. Thus, they lead to differential weighting of the component outcomes. We introduce the concept of “influence” for evaluating the extent to which individual component outcomes may be expressed or suppressed in the overall composite.
The need for Fiber Addition in Symptomatic Heart Failure (FEAST-HF) trial—a single-center clinical trial in ambulatory patients with chronic heart failure—is to evaluate whether dietary supplementation with acacia gum reduces heart failure–related biomarkers, clinical end points, exercise, quality of life, and how the gut microbiome responds to dietary supplementation with acacia gum. For this study, we contemplated using a composite end point, that is, an overall end point derived from multiple outcomes, and are viewing this as a phase 2 type of trial.
We favor the global rank composite here because it handles missing data and censoring and multiple variable types.1 To derive the composite, patients are ranked from the most adverse response (rank=1) to the most favorable response (rank=N, if no ties), while taking into account multiple data sources. Outcomes are prioritized in a hierarchy as follows: mortality, cardiovascular readmission, 6-minute walk test (6MWT), Kansas City Cardiomyopathy Questionnaire (quality of life), and NT-proBNP (N-terminal pro-B-type natriuretic peptide). We move down the hierarchy until we reach an outcome on which the patient “failed”. Then the patient is assigned a rank according to their response on that outcome (Figure I in the Data Supplement).
Software and Methods
The probability index (PI) is a measure of the effect size (acacia gum versus control), …