Validation and Utility of ARDS Subphenotypes Identified by Machine Learning Models Using Clinical Data: An Observational Multi-Cohort Retrospective Analysis.
Background: Two acute respiratory distress syndrome (ARDS) subphenotypes with distinct clinical and biological features and differential treatment responses have been identified using latent class analysis (LCA) in seven individual cohorts. To facilitate bedside identification of subphenotypes, clinical-classifier models using readily available clinical variables have been described in five randomized-controlled trials. Performance of these models in observational cohorts of ARDS is unknown. Methods: We evaluated the performance of machine learning clinical-classifier models for assigning ARDS subphenotypes in two observational cohorts of ARDS: EARLI (n=335) and VALID (n=452), with LCA-derived subphenotype as the gold-standard. We also assessed model performance in EARLI using data automatically extracted from the electronic health record (EHR). In LUNG SAFE (n=2813), a multinational observational ARDS cohort, we applied the model to determine the prognostic value of the subphenotypes and tested their interaction with PEEP strategy, with mortality as the dependent variable. Findings: The clinical-classifier models had an area under receiver operating characteristic curve (AUC) of 0·92 (95% CI: 0·90–0·95) in EARLI and 0·88 (0·84–0·91) in VALID. Model performance was comparable when using exclusively EHR-derived predictors. In LUNG SAFE, 90-day mortality was higher in the Hyperinflammatory subphenotype (57% [414/725] vs. 33% [694/2088]; p<0·0001). There was a significant treatment interaction with PEEP strategy and ARDS subphenotype (p=0·041), with lower mortality in the high PEEP group in the Hyperinflammatory subphenotype, following similar patterns to those observed in prior analyses of the ALVEOLI trial. Interpretation: Classifier models using clinical variables alone can accurately assign ARDS subphenotypes in observational cohorts. Application of these models can provide valuable prognostic information and may inform management strategies for personalised treatment, including application of PEEP, once prospectively validated. Funding: National Institutes of Health (PS: GM142992, CSC: HL140026, LBW: HL103836, HL135849), European Society of Intensive Care Medicine.
Paper
Full text
Validation and Utility of ARDS Subphenotypes Identified by Machine Learning Models Using Clinical Data: An Observational Multi-Cohort Retrospective Analysis.
Semantic Scholar · Medicine · 2022
Abstract
Background: Two acute respiratory distress syndrome (ARDS) subphenotypes with distinct clinical and biological features and differential treatment responses have been identified using latent class analysis (LCA) in seven individual cohorts. To facilitate bedside identification of subphenotypes, clinical-classifier models using readily available clinical variables have been described in five randomized-controlled trials. Performance of these models in observational cohorts of ARDS is unknown. Methods: We evaluated the performance of machine learning clinical-classifier models for assigning ARDS subphenotypes in two observational cohorts of ARDS: EARLI (n=335) and VALID (n=452), with LCA-derived subphenotype as the gold-standard. We also assessed model performance in EARLI using data automatically extracted from the electronic health record (EHR). In LUNG SAFE (n=2813), a multinational observational ARDS cohort, we applied the model to determine the prognostic value of the subphenotypes and tested their interaction with PEEP strategy, with mortality as the dependent variable. Findings: The clinical-classifier models had an area under receiver operating characteristic curve (AUC) of 0·92 (95% CI: 0·90–0·95) in EARLI and 0·88 (0·84–0·91) in VALID. Model performance was comparable when using exclusively EHR-derived predictors. In LUNG SAFE, 90-day mortality was higher in the Hyperinflammatory subphenotype (57% [414/725] vs. 33% [694/2088]; p<0·0001). There was a significant treatment interaction with PEEP strategy and ARDS subphenotype (p=0·041), with lower mortality in the high PEEP group in the Hyperinflammatory subphenotype, following similar patterns to those observed in prior analyses of the ALVEOLI trial. Interpretation: Classifier models using clinical variables alone can accurately assign ARDS subphenotypes in observational cohorts. Application of these models can provide valuable prognostic information and may inform management strategies for personalised treatment, including application of PEEP, once prospectively validated. Funding: National Institutes of Health (PS: GM142992, CSC: HL140026, LBW: HL103836, HL135849), European Society of Intensive Care Medicine.