Predicting Risk of Stroke in Hospitalized Patients with COVID-19: A Machine Learning Approach

ABSTRACT

Objective: The aim of this study is to leverage advanced analytics to develop an accurate, individualized risk prediction paradigm for stroke in hospitalized patients with COVID-19. Background: Patients hospitalized with COVID-19 have significantly higher risk of developing stroke compared to the general population. Predicting stroke in these patients, particularly those who are encephalopathic or ventilator-dependent, remains a challenge. Design/Methods: Patients admitted to a network of hospitals in the Austin metropolitan area with laboratory-confirmed COVID-19 were analyzed from March 2020 to June 2021. Using each patient's demographic, medication, laboratory and clinical assessment data, several machinelearning algorithms were developed to predict probability of stroke. Results: A total of 8,183 patients hospitalized with COVID-19 met our inclusion criteria. Among this cohort, 174 patients experienced a stroke. Of the machine learning algorithms that were developed and evaluated for stroke prediction, the random forest model achieved the highest prediction accuracy with a strong overall discriminatory performance. The features with the most significant prognostic value include ventilator dependence, history of hyperlipidemia, history of cancer and administration of aspirin. Conclusions: Machine learning models are capable of multi-dimensional analysis and can be leveraged to predict risk of stroke in patients hospitalized with COVID-19. Further studies are underway to incorporate a final model into the clinical pipeline for automated, patient-specific risk stratification.