Questions and Answers

Studies (usually therapy studies) often create a composite outcome (an outcome in which any one of a number of conditions counts as the outcome) to increase the power of their studies and therefore reduce the sample size required to see a significant difference in that outcome between groups if one exists.

Imagine investigators studying drug X vs. placebo to reduce the risk of atherosclerotic cardiovascular disease. ASCVD can result in a number of different outcomes over time: death, heart failure, heart attack, symptomatic angina, hospitalization for angina, a positive stress test, a need for cardiac catheterization, a need for coronary artery bypass graft surgery, etc. In one way or another, all those outcomes matter to patients, but some are more clinically significant than others. There can be outcomes that are less associated with symptoms - changes in vascular thickness seen on ultrasound, degree of coronary artery stenosis on cardiac catheterization, etc. Studying Drug X's effect on mortality may be a very important question, but these days, with all the treatments we have available, mortality from heart disease is getting less and less common in the short term.

Recall that the way to determine the sample size needed in a study to avoid a type 2 error (not finding a significant difference where one really does exist) involves: 1) setting the alpha level (by convention at 0.05), 2) setting the power of the study (by convention, somewhere between 80 and 90%) and 3) determining the difference you want to be able to detect. Whatever the difference between groups is desired, to see a difference requires a number of events in both groups. So if the study wants to use relatively uncommon mortality as the primary outcome, it will require a large sample size and may have to run for a long time to wait for those events to occur. If the researchers create a composite outcome, where ANY ONE of the list counts as the outcome, then event rates of each of these is conceptually added together (not exactly, but close enough) and the needed sample size and length of the study is reduced.

The key issue for readers of these studies is to look at the outcomes that make up the composite outcome and determine whether these outcomes are equivalent in terms of impact to the patient. A composite outcome of mortality, myocardial infarction, hospitalization for CHF, stroke and need for CABG generally seems reasonable since they are all significant, symptomatic outcomes. Some studies, however, will add more disease-oriented outcomes such as "progression of atherosclerotic change by vascular ultrasound" or "change in ASCVD grade by coronary artery CT". These additional outcomes are more common than the patient-oriented outcomes, but are not as clinically significant. Investigators will use change in the composite outcome to celebrate their intervention's success, but readers are required to do more digging into the article to interpret to what extent each individual outcome is affected by the intervention.

As with most things, it's not that composite outcomes are inherently misleading, but sometimes they are used that way. And, always, it's important to understand the limitations of the conclusions we can draw from them.