The return of “negative” trials?

Mario de Lemos raises several points concerning the trials of atrial fibrillation management that I discussed in my commentary.1

First, he asks whether a type II error might have occurred in the AFFIRM trial,2 given the difference between the planned sample size and study enrolment.3 The selection of appropriate sample size always involves estimating a clinically relevant difference and calculating the sample size needed to detect this difference with acceptable power. The sample size was reduced to 4060 to ensure sufficient power to reliably detect a difference of 30%, which was felt to be a minimally important clinical difference.4 The mortality rate in AFFIRM was marginally higher (by 15%, p = 0.08) in the rhythm-control group. The primary finding of AFFIRM was that a rate-control approach is not inferior to a rhythm-control approach. A larger sample size (and 5300 patients might not have been sufficiently large) might have detected a statistically significant increase in mortality rate with rhythm control; however, the investigators judged that the differential impact of a significant p value for this small effect was not sufficient to justify the substantial additional cost (and the potential detrimental effect of exposing additional patients to nonsuperior and more complex rhythm-control therapy) of extending the trial.

De Lemos also states that the efficacy of rate control was within the upper bound of the 95% confidence limit of that of rhythm control in the trial by Van Gelder and associates.5 However, those authors did not use efficacy as an endpoint. Their primary endpoint was a composite index of cardiovascular death, heart failure, thromboembolic complications, bleeding, pacemaker implantation and severe adverse drug reactions. In fact, the primary endpoint (which was a negative outcome) was more prevalent in the rhythm-control group, with the 90% 2-sided confidence limit barely including a neutral effect.

De Lemos further argues that it is unclear whether the rhythm-control strategy was a suitable active comparator. This statement seems to miss the point of the trials, which was to compare the 2 widely used approaches to therapy for atrial fibrillation: rate versus rhythm control. Both studies used patient populations in which recurrence was deemed likely, so a placebo group might not have been ethical in light of presently accepted medical practice.

De Lemos also criticizes use of an intention-to-treat analysis, rather than a per-protocol analysis (in which only events while the patient is receiving active therapy are analyzed), which he claims “is generally preferred.” In fact, the weight of clinical trials opinion favours intention-to-treat analyses. The simplest way to understand the advantage of an intention-to-treat approach is to imagine a therapy that has a neutral effect on outcome but a high frequency of side effects in high-risk patients. Such a drug would be discontinued in many high-risk patients. With a per-protocol analysis, there would be an appearance of a better outcome among patients maintaining therapy, but this would be due to the drop-out of high-risk patients rather than a direct benefit.

Finally, de Lemos criticizes the AFFIRM2 and Van Gelder and associates5 trials for not defining compliance. Because both trials assessed approaches to therapy (rate versus rhythm control), compliance would have been difficult to define. It would presumably include such standard measures as taking prescribed medication, but also reporting of events, acceptance of cardioversion when prescribed, and even physician-based components such as vigour of pursuit of heart-rate and sinus-rhythm endpoints.

It must be kept in mind that the goal of these studies was to compare 2 widely used strategies in a clinically relevant context, a goal that was largely achieved. It is true that as physicians we prefer “positive” trials because they leave us with a sense of a conclusive message. However, both the AFFIRM2 and Van Gelder and associates5 trials did yield a conclusive and important message, that for presently available approaches to atrial fibrillation therapy, rate control is not inferior overall to rhythm control. It is debatable whether larger studies that achieved a statistically significant p value would have provided any more practical information.

Stanley Nattel Department of Pharmacology and Therapeutics McGill University Montréal, Que.