Hybrid Bayesian Network-Based Modeling: COVID-19-Pneumonia Case.

Area of Science:

• Medical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• COVID-19 pneumonia (CP) presents complex clinical trajectories requiring accurate prediction for effective treatment.
• Existing predictive models may lack the interpretability and accuracy needed for clinical decision support.

Purpose of the Study:

• To develop and validate a hybrid predictive modeling approach for key clinical indicators in COVID-19 pneumonia patients.
• To enhance treatment strategies through accurate prediction of outcomes, length of stay, and medication needs.

Main Methods:

• Utilized dynamic and ordinary Bayesian networks (DBN, OBN), machine learning (ML) algorithms, and a novel hybrid DBN-autoML approach.
• Predicted treatment outcomes, length of stay (LOS), disease severity dynamics, and prescribed drugs using mathematical models.
• Validated models against expert knowledge, clinical guidelines, prior research, and standard predictive metrics.

Main Results:

• Achieved Mean Absolute Error (MAE) of 3.6 days for LOS prediction (DBN + FEDOT autoML).
• Obtained 0.87 accuracy for treatment outcome prediction (OBN).
• Reached an F1 score of 0.98 for predicting prescribed drugs (DBN).
• Demonstrated significant clinical differences between COVID-19 and non-COVID-19 pneumonia.

Conclusions:

• The proposed hybrid Bayesian network-based approach offers interpretable and accurate predictions for COVID-19 pneumonia management.
• Validated models show potential for integration into clinical decision support systems to optimize patient care.
• Identified key predictive indicators that can guide treatment decisions and resource allocation.