SAN DIEGO – Perioperative vascular complications may be the largest emerging epidemic in cardiology, and tackling it will require a shift in medical culture to demand larger clinical trials, according to Dr. Philip J. Devereaux.
Each year, 500,000-900,000 patients who undergo noncardiac surgery worldwide die from perioperative cardiac complications or develop nonfatal myocardial infarction, or nonfatal cardiac arrest. At the same time, the number of noncardiac surgeries is increasing, according to research by Dr. Devereaux, a clinical epidemiologist and biostatistician at McMaster University, Hamilton, Ont. (CMAJ 2005;173:627-34).
Dr. Philip J. Devereaux
Speaking at the annual meeting of the American Society of Anesthesiologists, he associated these trends with perioperative medicine's overreliance on small clinical trials and lack of scrutiny regarding the large treatment effects reported by such trials.
"I personally am more concerned about small randomized, controlled trials than I am about observational studies," he said. "The reason is, people question the observational studies. People believe randomized, controlled trials. They're powerful when they're large. When they're not large, they're incredibly likely to mislead."
Small clinical trials dominate perioperative medicine, he said. Physicians who provide perioperative medicine should demand large clinical trials to help guide their decision making, just as large studies have informed other areas of cardiovascular medicine.
"This fundamentally is a cultural shift," he noted. "Twenty-five years ago in cardiology there wasn't a single large, randomized, controlled trial. Yet, today, almost every other week in some journal there is a large cardiology trial," said Dr. Devereaux, adding that such a standard could be a reality for perioperative medicine, too, if the demand were there.
In the meantime, however, he and his associates have developed the Absolute Fragility Index (AFI) to help physicians assess trial results. They hope to publish an article about it soon. The AFI is the minimum number of patients in the treatment group who would be required to switch from not having an "event" (such as an MI) to having an event in order for the results to be considered nonsignificant, as opposed to significant.
For example, he compared two hypothetical randomized, double-blind, placebo-controlled trials of a drug to prevent MI. In the first trial of 200 patients, one in the treatment group and nine in the placebo group develop MIs, a significant difference (P = .02). In the second trial of 8,000 patients, 200 in the treatment group and 250 on placebo develop MIs, a significant difference between groups that has the same P value as the first trial (P = .02). In the first trial, however, if just one more patient in the treatment group had had an MI, the difference between groups would have been nonsignificant, whereas it would take nine additional MIs in the treatment group of the second trial to consider the difference nonsignificant. The first trial had an AFI of 1, and the second trial had an AFI of 9.
"If your trial hinges on one or two patients switching events, it doesn't matter what the P value is. You should be extremely cautious about believing it," he said. And yet most trials in perioperative medicine have an AFI of 1 or 2.
Considering the AFI is more than just an academic exercise, he added. Previous studies of the most highly cited randomized trials in leading medical journals have shown that 16% were later substantively contradicted, and another 16% were shown to have extremely exaggerated treatment effects. The one identifiable factor that could explain these errors was the small size of the initial studies, he said.
Dr. Devereaux's interest in an AFI grew from his experience as a co–primary investigator in the POISE-1 (Perioperative Ischemic Evaluation) trial, the largest randomized, controlled study of cardiac outcomes in patients undergoing noncardiac surgery. Guidelines on the use of perioperative beta-blockers in noncardiac surgery had been based on a separate randomized trial of 112 patients in which two patients in the beta-blocker group and nine in the control group died, a statistically significant difference (P = .02). However, in the POISE-1 trial involving 8,351 patients, more patients died in the beta-blocker group (129) than in the placebo group (97), also a statistically significant difference (P = .03) (Lancet 2008;371:1839-47).
"I thought, there is something fundamentally wrong with how we understand statistics and how we're interpreting results" if these studies are considered equally significant, he said. "Some might say our dose was not safe, and I agree. But to believe that the other trial demonstrates that their dose is safe is foolish."
Clinicians and researchers need to start collaborating and thinking big and internationally about perioperative medicine in order to move the medical culture to large trials, Dr. Devereaux said. "We owe it to patients to be confident about our results."