Early COVID-19 respiratory risk stratification using machine learning

Abstract

Background COVID-19 has strained healthcare systems globally. In this and future pandemics, providers with limited critical care experience must distinguish between moderately ill patients and those who will require aggressive care, particularly endotracheal intubation. We sought to develop a machine learning-informed Early COVID-19 Respiratory Risk Stratification (ECoRRS) score to assist in triage, by providing a prediction of intubation within the next 48 hours based on objective clinical parameters. Methods Electronic health record data from 3447 COVID-19 hospitalizations, 20.7% including intubation, were extracted. 80% of these records were used as the derivation cohort. The validation cohort consisted of 20% of the total 3447 records. Multiple randomizations of the training and testing split were used to calculate confidence intervals. Data were binned into 4-hour blocks and labeled as cases of intubation or no intubation within the specified time frame. A LASSO (least absolute shrinkage and selection operator) regression model was tuned for sensitivity and sparsity. Results Six highly predictive parameters were identified, the most significant being fraction of inspired oxygen. The model achieved an area under the receiver operating characteristic curve of 0.789 (95% CI 0.785 to 0.812). At 90% sensitivity, the negative predictive value was 0.997. Discussion The ECoRRS score enables non-specialists to identify patients with COVID-19 at risk of intubation within 48 hours with minimal undertriage and enables health systems to forecast new COVID-19 ventilator needs up to 48 hours in advance. Level of evidence IV. ©