Triage of patients with COVID19-: Using Ensemble learning method for risk factor analysis and death prediction

Early identification of high-risk patients with COVID19- using non-laboratory data at the time of admission may help the effective use of limited healthcare resources and improve clinical decision-making which reduces cost and time, and consequently the death of patients.

This study aims to provide an intelligent approach to triaging COVID19- patients by analyzing effective risk factors and predicting the risk of death due to COVID19- using an ensemble learning method.

This is a descriptive-applied study that was conducted in 2021. Non-laboratory data were used during the admission of 4558 confirmed patients with COVID19- referred to the Shohaday-e-Khalij-e-Fars Hospital in Bushehr registered in the medical care information system. After data preprocessing, the risk factors affecting the death were identified and ranked in importance. The ensemble learning (voting) method was used to develop the death prediction model, and the confusion matrix criteria evaluated its performance.

From a total of 5433 patient records, the data of 4558 cases were included in the study, of which %45.5( 2222) were women and %64.5( 2663) were men. The average age of the patients was 47.6. Out of all the investigated factors, 17 characteristics were identified as effective in predicting the death of patients, and their ranking indicated that the first six factors for predicting the death of patients are state of blood oxygen, age, state of consciousness, fever, cough, and body pain, respectively. The proposed model estimated the risk of death with an accuracy of %85, sensitivity of %85, specificity of %0.85, and AUROC of 0.91.

The results of the study provide a low-cost, fast, and innovative solution for the early identification of patients with COVID19- at risk of death to triage them more effectively, which can be used by the managers of healthcare systems to manage resources by continuously updating new data and providing appropriate treatments for high-risk patients.