Cardiogenic shock is a critical condition that affects up to 10% of patients hospitalized with acute myocardial infarction (AMI), representing the primary cause of death in this group [1, 2]. Recently, frequentist analysis showed no statistically significant treatment effect for the primary outcome in the randomized ECLS-SHOCK trial [3] comparing extracorporeal life support (ECLS) with medical treatment in patients with acute myocardial infarction complicated by cardiogenic shock (AMICS). We performed a Bayesian re-analysis using individual patient data allowing quantification of probabilities of benefits and harms [4].

A total of 417 patients (209 with ECLS, 208 with medical treatment alone) were re-analyzed using Bayesian Poisson regression. The analysis utilized Markov Chain Monte Carlo methods and checked for convergence. The primary outcome was 30-day all-cause mortality, safety outcomes included bleeding, stroke, and peripheral vascular complications. We employed an established framework [5] for post-hoc Bayesian re-analyses, using three priors: a non-informative prior (mean = 0, standard deviation [SD] = 10 on the log scale), a skeptical prior (0, 0.21) centered at 0 with a 5% probability of exceeding the treatment effect (30-day-mortality reduction from 49 to 35%; relative risk [RR] = 0.71) assumed by study investigators [3], and an enthusiastic prior (− 0.34, 0.21) centered at the assumed treatment effect with a 5% probability of no benefit. We calculated three posteriors for the primary endpoint and one for the safety endpoints and the sensitivity analyses using the non-informative prior. These posteriors estimated probabilities of any benefit (RR < 1), a large benefit (RR < 0.71; greater than the effect used for the frequentist sample size calculation), and the effect within the Region of Practical Equivalence (ROPE; RR between 0.9 and 1.1) for the primary endpoint, as well as probabilities of any harm (RR > 1) for the safety endpoints.

The results (Electronic Supplementary Material (ESM) Table S1) showed a median RR of 1.00 (95% highest posterior density (HPD) interval 0.79–1.24) for the skeptical prior, with less than 1% probability of a large benefit, and 62% within the ROPE (Fig. 1). For the enthusiastic prior (RR 0.89; 0.68–1.09), there was a 4% probability of a large benefit and 40% within the ROPE. The probability for a treatment effect greater than a number needed to treat of 10 was low overall. This finding persisted in all sensitivity analyses (ESM Table S2).