Machine Learning Model for Predicting Mortality Risk in Patients With Complex Chronic Conditions: Retrospective...

Area of Science:

• Health Informatics
• Machine Learning in Healthcare
• Predictive Analytics

Background:

• Healthcare systems are shifting to patient-centered care for chronic conditions.
• Increased health data enables AI for decision support, resource planning, and diagnosis.
• Predicting hospital needs and optimizing resources are key challenges.

Purpose of the Study:

• Develop and evaluate a machine learning model to predict mortality in complex patients.
• Predict 4-year mortality and 6-month early mortality post-diagnosis.
• Utilize accessible variables and healthcare resource utilization data.

Main Methods:

• Implemented and evaluated 6 classification algorithms.
• Used stratified cross-validation with k=10 and a 70/30 train-test split.
• Evaluated models using accuracy, recall, precision, F1-score, and AUROC.

Main Results:

• The Extreme Gradient Boosting (XGBoost) model predicted 4-year mortality with 87% accuracy (AUC=0.88).
• Predicting 6-month early mortality yielded 83% accuracy (AUC=0.88) with the Gradient Boosting (GRBoost) classifier.
• Performance for early mortality prediction was lower in recall (55%) and precision (64%).

Conclusions:

• The study demonstrates promising results in forecasting mortality for complex, persistent health conditions.
• Accessible variables and healthcare resource utilization data show predictive power.
• The model can identify patients needing tailored care and anticipate resource demands.