Prediction of Need for ICU Admission and Mortality of Covid-19 patients using Machine Learning: A Comparative Analysis

Abstract

The novel coronavirus disease (COVID-19) has caused severe damage worldwide, affecting the lives of millions of people and destroying the global infrastructure and health systems. The timely prediction of a patient's mortality risk can facilitate the health care systems to learn about the patients that are going to become severe and offer timely medical care to those patients, thereby reducing mortality and the burden on the health systems. This will also ensure the optimal allocation of resources in hospitals. Machine Learning can prove very helpful in this prediction of mortality. We have evaluated five different machine learning algorithms to predict the need for ICU admission and mortality of Covid-19 patients using two different datasets and identified the most significant features. This identification of significant features among an array of available features helps identify the patients at higher risk of severity and mortality. We have also compared the significant predictors of mortality from two datasets from the US and Mexico to analyze the effect of the infection on different populations. It was found that Random Forest achieves the best performance in the classification task, followed by Logistic Regression. Therefore, Random Forest's predictive model can be helpful for clinicians to prioritize patients appropriately.