eXtreme Gradient Boosting-based method to classify patients with COVID-19.

Area of Science:

• * Computational biology and bioinformatics
• * Medical informatics and machine learning applications in healthcare

Background:

• * COVID-19 (Coronavirus Disease 2019) mortality is influenced by various demographic, clinical, and laboratory factors.
• * Previous studies have utilized traditional statistics and some machine learning (ML) methods to analyze these factors.

Purpose of the Study:

• * To comparatively analyze five ML algorithms for predicting mortality in hospitalized COVID-19 patients.
• * To identify the ML method with the highest accuracy in classifying patients at increased risk of death.
• * To determine the key variables contributing to mortality prediction in COVID-19.

Main Methods:

• * Single-center observational study of 203 adult patients admitted with SARS-CoV-2 infection.
• * Comparison of four supervised ML algorithms (KNN, DT, GNB, SVM) against eXtreme Gradient Boosting (XGB).
• * Data extracted from electronic medical records, including demographic, clinical, and laboratory variables.

Main Results:

• * eXtreme Gradient Boosting (XGB) demonstrated superior prediction accuracy (92%), precision (>0.92), and recall (>0.92).
• * K-nearest neighbors (KNN), Support Vector Machine (SVM), and Decision Tree (DT) showed moderate performance (>80%).
• * Gaussian Naive Bayes (GNB) exhibited lower classification performance.
• * Key mortality predictors identified include C-reactive protein, procalcitonin, liver enzymes, neutrophils, D-dimer, creatinine, lactic acid, ferritin, ventilation days, septic shock, and age.

Conclusions:

• * The eXtreme Gradient Boosting (XGB) algorithm is a highly accurate and effective tool for predicting COVID-19 patient mortality.
• * XGB's performance suggests its utility in clinical settings for risk stratification of COVID-19 patients.
• * The identified variables provide crucial insights into the multifactorial nature of COVID-19 severity and mortality.