Photo Credit: wmf
The following is a summary of “Subphenotyping prone position responders with machine learning,” published in the March 2025 issue of Critical Care by Fosset et al.
Researchers conducted a retrospective study to identify subphenotypes in individuals acute respiratory distress syndrome (ARDS) in undergoing prone positioning using machine learning and evaluated association with mortality and treatment response.
They analyzed 353 mechanically ventilated individuals with ARDS who underwent at least 1 prone positioning cycle. Unsupervised machine learning identified subphenotypes using respiratory mechanics, oxygenation parameters, and demographic data collected in the supine position. The primary outcome was 28-day mortality, and secondary outcomes were changes in respiratory system compliance, driving pressure, PaO2 /FiO2 ratio, ventilatory ratio, and mechanical power.
The results showed 3 distinct subphenotypes. Cluster 1 (22.9%) had a higher PaO2 /FiO2 ratio and lower Positive End-Expiratory Pressure (PEEP) and Cluster 2 (51.3%) had more individuals with COVID-19, lower driving pressure, higher PEEP, and better respiratory system compliance. Cluster 3 (25.8%) had lower pH, higher PaCO2, and an elevated ventilatory ratio. Mortality varied significantly across clusters (P = 0.03), with Cluster 3 having the highest rate (56%). No significant differences were found in prone positioning response across studied parameters. Transpulmonary pressure measurements in a subcohort did not enhance subphenotype characterization.
Investigators concluded that ARDS subphenotypes with different mortality responses to prone positioning were found, predictive models using current data were insufficient.
Source: ccforum.biomedcentral.com/articles/10.1186/s13054-025-05340-8
