We recently discussed composite outcomes on our podcast in the context of the Andromeda-Shock-2 trial (listen here: https://youtu.be/gQpXct08AmU?si=JNnH8gbUaOxFJ9VM). In this trial, the primary outcome was a hierarchical composite of mortality, duration of organ support, and hospital length of stay.
What are composite outcomes?
As the term suggests, composite outcomes combine two or more clinical endpoints into a single primary outcome. In critical care trials, common examples include days alive and free of hospital and days alive and free of mechanical ventilation (or more broadly, days alive and free of organ support).
Why use composite outcomes?
One of the main reasons researchers use composite outcomes is to improve statistical power. Simply put, increasing the number of events increases the likelihood of detecting a difference between groups.
For example, if death alone were used as the primary endpoint—and death was a relatively rare event—a much larger sample size would be required. In contrast, using a composite endpoint such as cardiovascular death, stroke, or myocardial infarction (a classic approach in cardiovascular trials) increases event rates and reduces the required sample size.
The trade-off: interpretational challenges
Despite these advantages, composite outcomes introduce important challenges in interpretation.
Consider the SMART trial, which compared balanced crystalloids with saline in critically ill patients (https://www.nejm.org/doi/full/10.1056/NEJMoa1711584). The primary outcome was a composite of death from any cause, new renal replacement therapy, or persistent renal dysfunction (defined as a ≥200% increase in creatinine from baseline).
These components are clearly not equivalent from a patient’s perspective. Although the trial demonstrated a statistically significant difference in the composite outcome (14.3% vs. 15.4%), mortality itself was similar between groups. The observed difference was driven by small differences in the components—differences that, when combined, reached statistical significance. This raises the question: is the result clinically meaningful?
When components move in different directions
Interpretation becomes even more complex when individual components of a composite outcome move in opposite directions.
A classic example is the POISE trial (https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(08)60601-7/fulltext), which evaluated beta-blockers in patients undergoing non-cardiac surgery. The composite outcome (death, stroke, or myocardial infarction) was lower in patients receiving metoprolol. However, a closer look revealed a trade-off: while myocardial infarction was reduced, the risks of death and stroke were increased!!
Do hierarchical composites solve the problem?
Pre-specifying a hierarchy of outcomes—as done in the Andromeda-Shock-2 trial—can address some of these limitations by prioritizing clinically important events. However, this approach introduces new complexities, particularly in interpretation. We explore this further in our blog on win ratios: https://ccmevidence.blog/2026/03/18/win-ratios/
Take-home message
Composite outcomes offer clear statistical advantages, particularly in improving power and reducing sample size. However, they come with important interpretational challenges. As readers and consumers of clinical research, it is essential to look beyond the composite and examine the individual components to understand what truly drives the results—and whether they matter clinically.
Suggesting reading:
ccmevidence 
Leave a comment