Prediction of Heart and Liver Iron Overload in β-Thalassemia Major Patients Using Machine Learning Methods.

Area of Science:

• Hematology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Beta-thalassemia major (β-TM) patients experience severe complications due to iron overload in vital organs.
• Accurate assessment of heart and liver iron levels is crucial for managing β-TM.

Purpose of the Study:

• To identify the optimal machine learning (ML) model for predicting cardiac and hepatic iron overload in β-TM patients.
• To evaluate the predictive performance of Random Forest (RF), Gradient Boost Model (GBM), and Logistic Regression (LR) for iron overload.

Main Methods:

• Utilized data from 624 β-TM patients.
• Applied three ML models: RF, GBM, and LR, analyzed using R software.
• Assessed predictive performance using sensitivity, specificity, accuracy, and Area Under the Curve (AUC).

Main Results:

• For cardiac iron overload, Logistic Regression (LR) showed the highest AUC (0.68), sensitivity (75.0%), while Gradient Boost Model (GBM) achieved the highest specificity (69.0%) and accuracy (67.0%).
• For liver iron overload, Random Forest (RF) demonstrated the highest AUC (0.68) and accuracy (66.0%), with LR achieving the highest sensitivity (84.0%).
• Ferritin, transfusion duration, and age were identified as key predictors for both heart and liver iron overload.

Conclusions:

• Logistic Regression (LR) is the most effective model for predicting cardiac iron overload.
• Random Forest (RF) is the most effective model for predicting liver iron overload.
• Older patients with higher serum ferritin and longer transfusion history are at increased risk of iron overload.