Machine Learning Models for Predicting Mortality in Pneumonia Patients.

Area of Science:

• Medical Informatics
• Clinical Medicine
• Artificial Intelligence in Healthcare

Background:

• Pneumonia is a leading cause of hospital mortality.
• Accurate mortality prediction is crucial for patient management.
• Existing prediction methods may lack precision.

Purpose of the Study:

• To systematically review Machine Learning (ML) predictors for pneumonia mortality.
• To develop and validate an ML model for predicting mortality in hospitalized pneumonia patients.
• To compare ML model performance against traditional severity scores.

Main Methods:

• Systematic literature review of 16 studies (313,572 patients) to identify ML-based mortality predictors.
• Development of a Random Forest (RF) model using clinical data from 343 hospitalized pneumonia patients.
• Validation of the RF model using accuracy and Area Under the Curve (AUC) metrics.

Main Results:

• Systematic review identified age, oxygen levels, and albumin as common predictors.
• The developed RF model achieved 99% accuracy and 0.99 AUC.
• Key predictors in the local cohort included worsening chest X-ray, ventilator use, age, and oxygen support.

Conclusions:

• Machine learning demonstrates high potential for accurate pneumonia mortality prediction.
• ML models show superior performance compared to traditional clinical scores.
• The findings highlight the practical clinical utility of ML in managing pneumonia patients.