An Electronic Health Record-Based Classifier for Moderate-to-Severe Acute Respiratory Distress Syndrome With Persistent Hypoxemia.

Abstract

To characterize the performance of an electronic health record (EHR) data-based classifier of persistent moderate-to-severe acute respiratory distress syndrome (ARDS). Retrospective observational study. Six ICUs from two health systems. We included adults receiving greater than or equal to 24 hours of invasive mechanical ventilation (IMV) with a Pa o2 /F io2 of less than or equal to 150 mm Hg on F io2 greater than or equal to 0.6 in the first 72 hours of IMV. We evaluated classifier performance in two temporally and geographically distinct cohorts: a development cohort including patients in one of two academic medical or three mixed community ICUs in the first 3 months of 2021 or in 2022-2023, and a validation cohort from a different academic medical ICU in 2017. None. In both cohorts, study physicians assessed patients for clinical ARDS criteria. We retrospectively applied the EHR classifier, which required a persistent Pa o2 /F io2 of less than or equal to 150 mm Hg or receipt of interventions to address severe hypoxemia (i.e., prone positioning, neuromuscular blockade, inhaled pulmonary vasodilators) in the 24 hours after initially meeting hypoxemia criteria. We also evaluated classifier definitions that used peripheral oxygen saturation (Sp o2 )/F io2 of less than or equal to 162 to indicate persistent hypoxemia. Of 924 patients in the development cohort, 504 (55%) had clinician-adjudicated ARDS. Of 90 patients in the validation cohort, 48 (53%) had clinician-adjudicated ARDS. In the development and validation cohorts, the primary EHR-based classifier identified 382 and 38 patients as having persistent moderate-to-severe ARDS, respectively (positive predictive value [PPV], 71%; 95% CI, 66-75% and PPV, 66%; 95% CI, 50-81%). When Sp o2 /F io2 was used as the second hypoxemia marker more patients were classified as ARDS in both development and validation cohorts but with lower PPVs (67% and 62%, respectively.). An EHR-based classifier using readily available data had acceptable performance for identifying patients with persistent moderate-severe ARDS. This open-source tool could be used for retrospective or prospective identification of this vulnerable population for research and quality improvement initiatives.