Response by Packer to Letter Regarding Article, “Development and Evolution of a Hierarchical Clinical Composite End Point for the Evaluation of Drugs and Devices for Acute and Chronic Heart Failure: A 20-Year Perspective”
The hierarchical clinical composite end point (HCCE) was developed to address 2 important goals in the analysis of clinical trial data: (1) to minimize the bias inherent in the censoring of important adverse events when the primary end point was focused entirely on a patient benefit, and (2) to expand the range of responses when the primary end point was focused entirely on the risk of an adverse event. Over the past 20 years, the hierarchical clinical composite has succeeded with respect to both goals, and yet, as I described in my article,1 the HCCE still has important limitations and remains a work in progress.
However, I disagree with the points raised by Brown and Ezekowitz in their letter. The clinical composite does not rely on any arbitrary weighting of outcomes because the definition of ranks is always clinically determined. The only requirement for the HCCE is a fixed and predefined period of observation. Brown and Ezekowitz correctly observe that when the period of evaluation is short, events of high importance (eg, deaths) are rare. Yet when the number of events is sparse, estimates based on apparent imbalances in the number of deaths are inherently unreliable. Brown and Ezekowitz would improperly assign disproportionate weight to such unreliable estimates simply because they deem them to be clinically important. Such an approach seems ill advised.
I agree with Brown and Ezekowitz that the components of the HCCE should never be based on physiological measurements of questionable clinical importance (eg, small changes in serum creatinine). This point was made in the original article.
Brown and Ezekowitz wrongly believe that the HCCE requires the comingling of disparate adverse events into a single category, but the HCCE does not have such a requirement. In fact, recent refinements of the HCCE have relied on expansion of the number of ranks, as long as the ranking is clinically justified and prespecified. This approach builds on the principles described by Finkelstein and Schoenfeld2 in 1999 and expanded on by Pocock et al in 2012.3
Last, it is not true that the HCCE requires that data be discarded; if a clinical investigator wishes to create additional ranks to discriminate between responses, he or she is free to do so. Brown and Ezekowitz worry that the use of a conventional HCCE in phase II trials would have low statistical power and discourage drug development in acute heart failure. Yet given the enormous failure rate of acute heart failure drugs in phase III, it seems clear that current approaches that allow investigators to declare success in phase II are not terribly helpful. In contrast, the widespread adoption of the HCCE in the evaluation of devices has strongly encouraged their development with highly predictable outcomes.
Milton Packer, MD
Circulation is available at http://circ.ahajournals.org.
- © 2017 American Heart Association, Inc.
- Packer M
- Pocock SJ,
- Ariti CA,
- Collier TJ,
- Wang D