Comparing causal random forest and linear regression to estimate the independent association of organisational factors with ICU efficiency

Leonardo S. L. Bastos, Safira A. Wortel, Ferishta Bakhshi-Raiez, Ameen Abu-Hanna, Dave A. Dongelmans, Jorge I. F. Salluh, Fernando G. Zampieri, Gastón Burghi, Silvio Hamacher, Fernando A. Bozza, Nicolette F. de Keizer, Marcio Soares

Abstract: Highlights•Causal machine learning models, such as causal random forests (CRF), showed promising results in estimating heterogeneous treatment effects.•Seven organizational factors were analysed in 128 (277,459 admissions) Brazilian/Uruguayan ICUs.•Both regression modelling with interactions and CRF showed comparable average treatment effects.•CRF could predict significant effects in certain ICUs, even when the average effect was nonsignificant.•The use of machine learning methods could assist healthcare managers in further in-dept evaluation of process interventions to improve ICU efficiency.