COVID-19 severity analysis for clinical decision support based on machine learning approach

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Epidemiology

Background:

• The COVID-19 pandemic strained global healthcare systems, highlighting the need for efficient patient management.
• Inadequate ward allocation for severe COVID-19 cases leads to resource wastage and suboptimal patient care.
• Early prediction of disease severity is crucial for timely intervention and resource optimization.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting intensive care unit (ICU) prioritization in COVID-19 patients.
• To compare model performance using two distinct severity criteria: clinical interventions and national policy definitions.
• To identify key clinical predictors for COVID-19 patient severity.

Main Methods:

• Analysis of electronic health records from 806 COVID-19 patients admitted to Chungbuk National University Hospital.
• Development of single machine learning models (logistic regression, random forest, SVM, LightGBM, XGBoost) and ensemble models (voting classifiers).
• Evaluation of models based on recall rates for two different COVID-19 severity criteria.

Main Results:

• The ensemble learning model demonstrated superior performance, achieving recall rates of 96.2% (Criterion I) and 88.2% (Criterion II).
• Key predictive features identified include glucose level, neutrophil count, high sensitivity C-reactive protein (hsCRP), and albumin level.
• Model interpretability was enhanced by identifying significant clinical features.

Conclusions:

• Machine learning, particularly ensemble methods, can effectively predict the need for ICU admission in COVID-19 patients.
• The identified clinical features provide valuable insights for early risk stratification and clinical decision-making.
• This approach supports optimized ward allocation and timely treatment strategies for COVID-19 patients.