Development and Prospective Validation of a Transparent Deep Learning Algorithm for Predicting Need for Mechanical Ventilation.

摘要: Importance Objective and early identification of hospitalized patients, particularly those with novel coronavirus disease 2019 (COVID-19), who may require mechanical ventilation is great importance aid in delivering timely treatment. To develop, externally validate prospectively test a transparent deep learning algorithm for predicting 24 hours advance the need patients COVID-19. Design Observational cohort study SETTING: Two academic medical centers from January 01, 2016 to December 31, (Retrospective cohorts) February 10, 2020 May 4, (Prospective cohorts). Participants Over 31,000 admissions intensive care units (ICUs) at two hospitals. Additionally, 777 COVID-19 were used prospective validation. Patients placed on within four their admission excluded. MAIN OUTCOME(S) MEASURE(S): Electronic health record (EHR) data extracted an hourly basis, set 40 features calculated passed interpretable deep-learning predict future advance. commonly clinical criteria (based heart rate, oxygen saturation, respiratory FiO2 pH) was assess ventilation. Performance algorithms evaluated using area under receiver-operating characteristic curve (AUC), sensitivity, specificity positive predictive value. Results After applying exclusion criteria, external validation included 3,888 general ICU 402 patients. The performance model (AUC) 24-hour prediction horizon site 0.882 population 0.918 In comparison, ROX score achieved AUCs range 0.773 - 0.782 0.768 0.810 COVID-19, respectively. Conclusions relevance A generalizable improves traditional including Such help clinicians optimizing timing tracheal intubation, better allocation resources staff, improve patient care.