Three subtypes of postoperative ARDS that showing different outcomes and responses to mechanical ventilation and fluid management: A machine learning and latent profile analysis

Abstract

BackgroundARDS is a heterogeneous clinical syndrome, and operation and trauma are common indirect etiologies. The identification of postoperative ARDS subtypes may optimize individualized clinical management. ObjectivesTo identify the subtypes of postoperative ARDS and explore the impact of therapy on outcomes. MethodsThis retrospective study used data obtained from a database. Patients diagnosed with ARDS who underwent surgical procedures within 7 days were included in the study. Laboratory and clinical variables were used for latent profile analysis (LPA). XGBoost and multivariable logistic regression models were used to explore the association between therapy and outcomes. ResultsA total of 1065 patients were included. The LPA identified three subtypes of postoperative ARDS: Patients in profile 1 were mainly accepted neurosurgery, while those in profile 2 and 3 were treated with orthopedic and vascular or thoracic surgery, respectively. The XGBoost model effectively predicted mortality with an AUC of 0.935, which was higher than SOFA (0.622), APACHE 2 (0.629), SLIP (0.579), and SLIP-2 (0.550). ConclusionsThis study identified three subtypes of postoperative ARDS with different clinical characteristics, mechanical support, and fluid resuscitation responses.