Using Machine Learning Methods to Predict Hospitalization Based on Brixia Score and Patient Clinical Data (from the...

Area of Science:

• Radiology
• Medical Informatics
• Machine Learning

Background:

• Chest X-rays are standard for lung disease diagnosis.
• The COVID-19 pandemic highlighted challenges in accurate diagnosis and treatment.
• Predicting patient hospitalization is crucial for resource allocation.

Purpose of the Study:

• To correlate radiological findings (Brixia score) and clinical data with hospitalization.
• To develop and evaluate machine learning models for predicting hospitalization.
• To assess the prognostic importance of various clinical variables.

Main Methods:

• Utilized Brixia score and clinical data (gender, age, hypertension, diabetes).
• Employed four machine learning models: Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), and Support Vector Machine (SVM).
• Predicted patient hospitalization outcomes using these models.

Main Results:

• All four machine learning models achieved an Area Under the Curve (AUC) greater than 0.8, indicating good predictive performance.
• The Brixia score emerged as the most significant predictor of hospitalization among the evaluated variables.
• Decision Tree (DT) offered the most balanced performance across AUC, accuracy, sensitivity, and specificity.

Conclusions:

• Machine learning models demonstrate significant potential for improving clinical practice in diagnosis, therapy, and prognosis.
• The Brixia score is a key factor in predicting hospitalization risk.
• Further research can explore the integration of ML models for more precise patient management.