19, 20, 77
The observational body is more favourable than the randomised body, and the centre-volume signal supports Chapter 19's argument that the programme is the unit of intervention. Use the concordance of direction with the randomised evidence as reassurance — not the GRADE rating as certainty.
Conventional CPR
DOI 10.1186/s13054-024-04830-5
ECPR
Ten of thirteen studies are propensity-score-matched cohorts. The randomised trials do not individually reach significance while the matched cohorts do, which is the classic signature of residual confounding by indication: patients selected for ECPR differ by the act of selection, in ways the dataset cannot capture. The authors list residual confounding first among their own limitations. Two randomised trials were stopped early. No quality-of-life data. THIS BOOK GRADES THE BODY MODERATE AT BEST, NOT HIGH.
In-hospital mortality OR 0.63 (0.50–0.79), rated high certainty, with the cumulative Z-curve passing the required information size and the TSA-adjusted boundary for benefit. Out-of-hospital subgroup OR 0.62 (0.45–0.84) — newly significant where the same group's earlier analysis found none. Short-term favourable neurological outcome OR 1.57 (1.14–2.15). 30-day survival OR 1.70 (1.29–2.26). The group's 2023 Lancet Respiratory Medicine analysis additionally reported a meta-regression coefficient of −0.17 (−0.32 to −0.017, P = 0.030) for odds of mortality per doubling of annual centre volume.
13 studies, 14 pairwise comparisons — 3 randomised trials and 10 propensity-score-matched cohorts — covering both out-of-hospital and in-hospital cardiac arrest. 6,336 ECPR and 7,712 conventional CPR patients. Searched 1 January 2000 to 1 November 2023.