Combined imputation, inverse propensity weight, and bootstrap approach to estimate treatment effects on in-hospital mortality: an ELSO registry analysis

While multicenter patient registries are a valuable tool for understanding diseases and exposures that affect small patient populations, retrospective registry studies are particularly prone to imbalances in baseline covariates, confounding by indication, and exposure misclassification due to missing data. This paper describes an analysis performed on the Extracorporeal Life Support Organization (ELSO) registry to evaluate the comparative effectiveness of two treatments – centrifugal versus roller blood pumps – to reduce in-hospital mortality among pediatric extracorporeal membrane oxygenation (ECMO) patients. We outline a combined imputation, inverse propensity weight, and bootstrap approach to address potential sources of bias and obtain a valid estimate of treatment effectiveness.