ECMO, ARDS and meta-analyses: Bayes to the rescue?

Area of Science:

• Critical Care Medicine
• Respiratory Medicine
• Biostatistics

Background:

• A meta-analysis by Munshi et al. (2019) suggested extra corporeal membrane oxygenation (ECMO) improves mortality in acute respiratory distress syndrome (ARDS).
• The original meta-analysis was based on a very small number of studies (2 RCTs, 3 observational studies).

Purpose of the Study:

• To comprehensively re-analyze the data from Munshi et al.'s meta-analysis on ECMO for ARDS.
• To investigate the certainty of ECMO's claimed mortality treatment efficacy in ARDS.

Main Methods:

• Data were re-analyzed using various frequentist (Der-Simonian & Laird, REML, Paul-Mandel) and Bayesian models.
• Fragility indices were calculated to assess the robustness of the findings.

Main Results:

• Only uncorrected frequentist estimators showed significant risk ratios (RR) in both the 2- and 5-study meta-analyses.
• Most models, except for the beta-binomial model in the 2-study analysis, produced confidence intervals including the null value.
• Both meta-analyses demonstrated statistical fragility, indicating sensitivity to small changes in data.

Conclusions:

• The re-analysis found no certainty regarding the efficacy of ECMO in treating ARDS.
• Alternative statistical methods challenge the initial conclusions of Munshi et al.