19, 20, 77
The single most usable statement of the randomised ECPR evidence. It is about nine to one that ECPR helps; about seven to three that it helps by at least five absolute percentage points (the trialists' own minimal clinically important difference); a coin-toss whether it helps by ten. The heterogeneity is greatest in the shockable subgroup — the population everyone agrees is the best candidate — which is what a systemic rather than biological explanation predicts. Chapter 19 §19.8.
Conventional CPR
DOI 10.1186/s13054-024-05008-9
ECPR-based treatment
Three trials only, of markedly different design and size (ARREST n=30). Not all patients randomised to ECPR received it. Wide credible intervals. Wide variability in how the minimal clinically important difference is defined in ECPR research. No formal sceptical- or enthusiastic-prior sensitivity analysis reported.
6-month neurologically favourable survival, ALL RHYTHMS: median RR 1.47 (95% CrI 0.73–3.32); mean absolute risk difference 8.7% (−5.0 to 42.7); NNT 11. Posterior probability that the absolute benefit exceeds 0% = 91.0%, 5% = 71.1%, 10% = 43.7%, 15% = 23.5%, 20% = 13.4%. SHOCKABLE RHYTHMS: median RR 1.54 (0.79–3.71); ARD 10.8% (−4.2 to 73.9); NNT 9; probabilities 92.4%, 75.8%, 50.4%, 29.8%, 17.6%. Assumed control risk 18.4% (11.8–27.7), I-squared 33%, tau-squared 0.077 for all rhythms; 19.9% (8.3–40.7), I-squared 76%, tau-squared 0.513 for shockable.
The three randomised ECPR trials in refractory out-of-hospital cardiac arrest — ARREST, PRAGUE-OHCA and INCEPTION. 209 ECPR and 211 conventional CPR patients. Bayesian hierarchical random-effects model under vague priors (mean log RR 0, SD 2). Preregistered INPLASY2023120060. Written by authors of all three trials.