Identification of Heterogenous Treatment Response to Corticosteroids Among Patients With Ventilator-Dependent Respiratory Failure

BACKGROUND: Corticosteroids increasingly are recommended as adjunctive treatments in critical illness syndromes; however, the existing literature supporting their benefit remains uncertain. Ventilator-dependent respiratory failure (VDRF) is a critical illness syndrome with low-acuity and high-acuity subphenotypes identifiable with electronic health record (EHR) data. RESEARCH QUESTION: Do conditional average treatment effects (CATEs) from corticosteroids with respect to hospital mortality differ by clinical subphenotypes in VDRF? STUDY DESIGN AND METHODS: In this multicenter, retrospective observational cohort, we measured CATE by clinical subphenotype among patients > 18 years of age receiving > 48 hours of mechanical ventilation with adjusted doubly robust estimator models using EHR data from the first 24 hours after initiation of mechanical ventilation. RESULTS: Low-acuity and high-acuity subphenotypes from the Medical University of South Carolina (MUSC) and Medical Information Mart for Intensive Care III (MIMIC-III) cohorts showed heterogeneous CATEs on mortality resulting from corticosteroid treatments (low-acuity MUSC, -4.1%; low-acuity MIMIC-III, -5.2%; high-acuity MUSC, 9.3%; and high-acuity MIMIC-III, 4.5%) with mean differences of 13.4% (95% CI, 11.8%-14.9%) and 9.7% (95% CI, 9.3%-10.1%), respectively. Protective effects in the low-acuity subphenotype were attenuated among patients with ARDS, pneumonia, or sepsis identified by administrative diagnosis codes; however, the risk of harm was consistent for the high-acuity subphenotype. INTERPRETATION: Our results show that clinical subphenotypes in VDRF can identify heterogeneous treatment effects from corticosteroids. These findings support the usefulness of clinically derived subphenotypes and may inform enrichment strategies for future prospective clinical trials.