Predicting the severity of COVID-19 using machine learning methods

Abstract Background COVID-19 represents a wide range of clinical severity. Early identification of patients at high risk of severe disease is critical for appropriate clinical management and resource allocation. This study aims to predict COVID-19 severity using machine learning methods. Methods In this retrospective study, laboratory data from 816 hospitalized COVID-19 patients in Hamadan Province, Iran, were analyzed. According to the World Health Organization guideline, patients were classified into two groups: severe and non-severe based on clinical evaluation. Blood parameters including D-dimer, red cell distribution width, mean platelet volume, platelet count, and others were extracted from patient records. The performance of machine learning methods, including support vector machine, least-squares support vector machine, random forest, and extreme gradient boosting, was evaluated in predicting disease severity. Statistical comparisons were conducted using the Mann–Whitney and chi-square tests. Results Levels of D-dimer, red cell distribution width, and mean platelet volume were significantly elevated in the severe group (P-value < 0.0001). Random forest and Extreme Gradient Boosting mean sensitivities were 76% and ≥ 74%, respectively. The area under the curve values for these methods were 0.91 and 0.90, respectively. D-dimer was identified as a strong predictor of COVID-19 severity based on random forest analysis. Conclusions The findings suggest that random forest outperforms support vector machine and least-squares support vector machine in predicting the severity of COVID-19 using routine blood tests. Applying machine learning methods may assist clinicians in identifying high-risk patients and facilitating timely clinical interventions.