Predicting mortality, duration of treatment, pulmonary embolism and required ceiling of ventilatory support for COVID-19 inpatients: A Machine-Learning Approach

Introduction Within the UK, COVID-19 has contributed towards over 103,000 deaths. Multiple risk factors for COVID-19 have been identified including various demographics, co-morbidities, biochemical parameters, and physical assessment findings. However, using this vast data to improve clinical care has proven challenging. Aims to develop a reliable, multivariable predictive model for COVID-19 in-patient outcomes, to aid risk-stratification and earlier clinical decision-making. Methods Anonymized data regarding 44 independent predictor variables of 355 adults diagnosed with COVID-19, at a UK hospital, was manually extracted from electronic patient records for retrospective, case-controlled analysis. Primary outcomes included inpatient mortality, level of ventilatory support and oxygen therapy required, and duration of inpatient treatment. Secondary pulmonary embolism was the only secondary outcome. After balancing data, key variables were feature selected for each outcome using random forests. Predictive models were created using Bayesian Networks, and cross-validated. Results Our multivariable models were able to predict, using feature selected risk factors, the probability of inpatient mortality (F1 score 83.7%, PPV 82%, NPV 67.9%); level of ventilatory support required (F1 score varies from 55.8% “High-flow Oxygen level” to 71.5% “ITU-Admission level”); duration of inpatient treatment (varies from 46.7% for “≥ 2 days but < 3 days” to 69.8% “≤ 1 day”); and risk of pulmonary embolism sequelae (F1 score 85.8%, PPV of 83.7%, and NPV of 80.9%). Conclusion Overall, our findings demonstrate reliable, multivariable predictive models for 4 outcomes, that utilize readily available clinical information for COVID-19 adult inpatients. Further research is required to externally validate our models and demonstrate their utility as clinical decision-making tools.