Photo Credit: metamorworks
The following is a summary of “Development and External Validation of a Detection Model to Retrospectively Identify Patients With Acute Respiratory Distress Syndrome,” published in the April 2025 issue of Critical Care Medicine by Levy et al.
Researchers conducted a retrospective study to create and externally validate a machine-learning model that identified individuals with acute respiratory distress syndrome (ARDS) using electronic health record (EHR) data.
They identified ARDS through physician-adjudication in 3 cohorts of individuals with hypoxemic respiratory failure (training, internal validation, and external validation). Machine-learning models were instructed by vital signs, respiratory support, laboratory data, medications, chest radiology reports, and clinical notes. The models’ performance was evaluated and validated internally and externally using area under receiver-operating curve (AUROC), area under precision-recall curve, integrated calibration index (ICI), specificity, sensitivity, positive predictive value (PPV), and ARDS timing. The study focused on individuals with hypoxemic respiratory failure receiving mechanical ventilation across 2 distinct health systems.
The results showed that the internal validation cohort included 556 individuals, and the external validation cohort consisted of 199 individuals, the ARDS prevalence was 17% and 31%, respectively. The best-performing models were the regularized logistic regression models using structured data (EHR model) and a combination of structured data and radiology reports (EHR-radiology model). During internal and external verification, the EHR-radiology model achieved an AUROC of 0.91 (95% CI, 0.88–0.93) and 0.88 (95% CI, 0.87–0.93), respectively. Externally, the ICI was 0.13 (95% CI, 0.08–0.18). At the specified threshold, sensitivity was 80% (95% CI, 75%–98%), specificity was 64% (95% CI, 58%–71%), and the model identified individuals with a median of 2.2 hours (interquartile range 0.2–18.6) after meeting Berlin ARDS criteria.
Investigators concluded that machine-learning models analyzing EHR data had the capability to identify patients with ARDS across various institutions.
Create Post
Twitter/X Preview
Logout