Machine Learning Based Clinical Decision Support System for Early COVID-19 Mortality Prediction.

Area of Science:

• Computational biology
• Medical informatics
• Epidemiology

Background:

• COVID-19 pandemic poses significant challenges to healthcare systems due to high mortality.
• Accurate mortality prediction is crucial for optimizing patient treatment and resource allocation.
• Routine blood tests offer a readily available data source for predictive modeling.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting COVID-19 mortality risk using blood test data.
• To identify key hematological and biochemical biomarkers associated with COVID-19 mortality.
• To assess the early predictive performance of machine learning models for patient outcomes.

Main Methods:

• Trained and compared various machine learning models including neural networks, logistic regression, XGBoost, random forests, SVM, and decision trees.
• Utilized a feature set comprising neutrophils, lymphocytes, lactate dehydrogenase (LDH), high-sensitivity C-reactive protein (hs-CRP), and age.
• Evaluated model performance based on accuracy and predictive timing relative to patient outcomes.

Main Results:

• A combination of five features achieved 96% accuracy in predicting mortality.
• The best performing model, XGBoost with neural network classification, predicted outcomes with 90% accuracy up to 16 days prior.
• Key biomarkers identified include neutrophils, lymphocytes, LDH, hs-CRP, and age.

Conclusions:

• Machine learning models utilizing routine blood tests can accurately and early predict COVID-19 mortality.
• The identified biomarkers and predictive models offer practical tools for healthcare decision-making.
• This approach facilitates timely and targeted medical interventions, improving patient management during the pandemic.