Predicting mortality risk and determining critical factors in intensive care patients: A preliminary study on covid-19 patients

Abstract

Aim: To predict the mortality risk of COVID-19 patients in the intensive care unit (ICU) using clinical parameters and machine learning approaches.

Methods: Data from 307 ICU patients at Erciyes University Hospital (2021–2022) were analyzed. Particle swarm optimization (PSO) and least absolute shrinkage and selection operator (LASSO) methods were utilized for feature selection. Four machine learning algorithms support vector machine (SVM), K-nearest neighbors (KNN), ensemble methods, and artificial neural networks (ANN) were applied to the selected parameters.

Results: The top 10 predictive parameters, common to both LASSO and PSO, included sodium, nucleated red blood cell (NRBC) count, magnesium, mean corpuscular hemoglobin (MCH), and lymphocyte count. The best prediction performance was achieved using PSO feature selection and ANN (AUC: 86.77%, sensitivity: 85.12%, specificity: 77.44%, F1-score: 81.10%).

Conclusions: This study identifies critical parameters for predicting ICU COVID-19 patient mortality risk, employing two feature selection methods and comparing their performance with four machine learning algorithms. These results offer valuable insights for specialized physicians regarding disease progression and mortality risk prediction, but further research is needed.