Development of a Prognostic Model for the Outcome of Viral Lung Disease Using Machine Learning Algorithms

Background : During periods of increasing respiratory infections, stratifying patient care for virus-associated lung injury is crucial. Objective : To develop a prognostic model for the outcome of viral lung injury using machine learning algorithms. Materials and methods : The study included 295 patients with COVID-associated pneumonia, divided into two groups: 1) deceased (n=78) and 2) survivors (n=217). Data processing was performed using IDE R Studio (Version 4.3.1). The prognostic model was created using potential predictors and machine learning algorithms: multivariate logistic regression, random forest, and stochastic gradient boosting. A 5-fold cross-validation method (KFold) was used. Prediction models for mortality were implemented in Google Colaboratory with Python. Model quality was assessed by metrics such as ROC-AUC and confusion matrix, which provided measures like precision, recall, sensitivity, specificity, and F1-score. Results : The prognostic outcome model was based on mortality predictors, demonstrating a high level of statistical significance between the predicted groups, with significant predictors including the presence of two or more comorbidities, respiratory failure above grade 2, pronounced myalgia and dyspnea, neutrophil-to-lymphocyte ratio >5.1, a critical decrease in eosinophil count accompanied by elevated ESR, glucose, AST, CRP, urea levels, and decreased hemoglobin ( p