A heart attack used to mean almost certain death. Thanks to advances in the prevention and treatment of coronary artery disease (CAD), that is no longer the case. These improvements in management, however, make trials assessing death as a sole outcome not feasible due to the length of time or size of trials that would be required.
In the early 1990s, Braunwald and colleagues argued that adding nonfatal complications to the outcome of death would improve trial efficiency and accelerate the pace of treatment innovations for the management of acute myocardial infarction (MI). Interestingly, the original article by Braunwald argued for using nonfatal MI only in small preliminary trials, with interventions that proved promising needing to be validated in larger trials that did not use nonfatal MI as an outcome. Since the inception of composite outcomes, however, nonfatal MI has been included as an endpoint in all landmark trials of interventions for acute coronary syndromes despite a lack of any formal assessment of its validity as a surrogate for all-cause or cardiovascular mortality. A recent systematic review and meta-analysis set out to evaluate the suitability of nonfatal MI for surrogacy.
Researchers included 144 randomized trials evaluating interventions for CAD that reported all-cause mortality and nonfatal MI outcomes in more than 1,200,000 adults with over 5,000,000 patient-years of follow-up. The authors used the coefficient of determination (R2) to assess how well nonfatal MI correlated with either overall mortality or cardiovascular mortality. R2 ranges from zero-to-one, with a value of at least 0.8 indicating correlation high enough to prove surrogacy. Prespecified subgroup analyses included intervention type (primary prevention, secondary prevention, mixed primary/secondary, or revascularization), era of the trial (before 2000, from 2000 to 2010, and after 2010, distinguished for availability and sensitivity of troponin tests), duration of follow-up, and active vs. placebo control.
For nonfatal MI, R2 values were not nearly sufficiently correlated with all-cause mortality (R2 = 0.02; 95% CI, 0.00-0.08) or cardiovascular mortality (R2 = 0.11; 95% CI, 0.02-0.27) to demonstrate validity as a surrogate. Nonfatal MI likewise did not meet the threshold for surrogacy for any of the subgroup analyses.
As the dictum goes, correlation does not mean causation. Establishing that one endpoint can serve as a surrogate, i.e., can substitute for another, requires a higher level of “proof” and nonfatal MI does not make the cut. With more than 5,000,000 patient-years assessed in this study, the argument that the threshold might have been met if the study had been bigger or longer won’t work. Using nonfatal MI as a surrogate for death assumes that interventions that prevent nonfatal MI also prevent death. We now know that is not sufficiently the case. Going forward, we should take a more skeptical view of large trials that continue to use surrogate outcomes as part of composite outcomes.