Bayesian Reanalyses of the Trials TOMAHAWK and COACT

Abstract

Background: The timing of coronary angiography in patients with successfully resuscitated out-of-hospital cardiac arrest and missing ST-segment elevations on the electrocardiogram has been investigated in 2 large randomized controlled trials, TOMAHAWK (Angiography After Out-of-Hospital Cardiac Arrest Without ST-Segment Elevation) and COACT (Coronary Angiography After Cardiac Arrest Trial). Both trials found neutral results for immediate vs delayed/selective coronary angiography on short-term all-cause mortality. The TOMAHAWK trial showed a tendency towards harm with immediate coronary angiography, though not statistically significant with traditional frequentist methods. Probabilistic analyses of both trials may enable greater clinical understanding of the trial findings.

Objectives: The purpose of this study was to perform reanalyses of both trials within a Bayesian framework.

Methods: Post hoc analyses of both multicenter randomized controlled trials were performed in both cohorts separately and combined. The primary endpoint, 30-day all-cause mortality, was analyzed using Bayesian logistic regression. A spectrum of priors included "flat," "neutral," "optimistic," and "pessimistic" priors based on assumptions made when designing both trials.

Results: In the TOMAHAWK trial, immediate coronary angiography showed a very high posterior probability of increased mortality between 90% and 97% across all priors. The ORs across all priors were directed towards harm. Similarly, COACT showed odds ratios ranging from 0.98 to 1.11 for the 30-day mortality endpoint. When combining both trials, immediate coronary angiography showed a high probability of increased mortality between 83% and 95%, again with ORs across all priors indicating a direction towards harm.

Conclusions: Bayesian reanalyses showed a very high probability of increased 30-day mortality risk with immediate compared with delayed/selective coronary angiography in the TOMAHAWK trial and combined trial cohort. These findings may shift the current understanding of both trials from a "neutral" towards a likely "harmful" effect of immediate coronary angiography after successfully resuscitated out-of-hospital cardiac arrest without ST-segment elevations. Therefore, adoption of a delayed strategy of coronary angiography might be preferred in clinical practice until the results of the DISCO (Direct or Subacute Coronary Angiography in Out-of-Hospital Cardiac Arrest) trial become available.

Keywords: Bayesian analysis; cardiac arrest; coronary angiography.