Combining events of similar severity such as cause-specific mortality, non-fatal myocardial infarction and stroke is generally accepted. In addition, the diagnostic criteria for these events are well defined and can be validated. Problems emerge when events of varying severity are combined. Adding self-reported angina, vascular procedures, and hospitalizations to major cardiovascular events is debatable. Whether a patient is hospitalized or has a costly procedure could be seen as a marker of disease severity, but it could also be influenced by whether the patient has health insurance coverage.
Experience has shown that the more subjective and the least serious events that represent components of a composite outcome are the most likely to respond favorably to treatment, compared to events that are more objective and serious. In MIRACL, more than 3,000 patients with unstable angina or non-Q-wave acute myocardial infarctions were randomized to 80 mg of the lipid-lowering drug atorvastatin or placebo and followed for 16 weeks.5 The risk of the composite outcome, which included all-cause mortality, non-fatal myocardial infarction, cardiac arrest and hospitalization for recurrent ischemic symptoms, was lower in the statin group, although the p-value was only borderline (0.048).
The overall difference between the active and placebo groups was driven by a 26% reduction in hospitalized angina (p=0.02), which comprised 45% of all events. For the harder outcomes, the treatment group differences were smaller and none reached statistical significance. A recent meta-analyses1 of 12 statin trials in this population (n = 13,024) confirmed that initiation of statin therapy within two weeks of acute coronary syndrome does not reduce death, recurrent infarction or stroke within four months.
When faced with a composite outcome that is statistically significant, always consider 1) why each individual component was selected and 2) its contribution to the overall outcome. The components of a composite endpoint should make clinical sense. Ideally, the most important components should show individual statistical significance, or very strong and consistent trends.
Was this article helpful?