A Personalized and Evidence-Based Clinical Decision Support System Using Ensemble Learning.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Digital Health

Background:

• Clinical Decision Support Systems (CDSS) enhance patient care using evidence-based data.
• Digital Twins offer holistic patient representations for predictive decision-making.
• Current systems require improvement in adaptability and integration of medical guidelines.

Purpose of the Study:

• To develop a modular, informed, interpretable, personalized, and evidence-adaptive CDSS using ensemble learning.
• To evaluate the CDSS's performance on diverse datasets, including handling feature separations and missing data.
• To demonstrate the system's utility as a foundational component for patient-centered Digital Twins.

Main Methods:

• Ensemble Learning approach for CDSS development.
• Validation on the Cleveland Heart Disease and TCGA Glioma datasets.
• Incorporation of medical guidelines through Informed Machine Learning.
• Simulation of performance under missing data conditions.

Main Results:

• Achieved Area Under the Curve (AUC) of 0.94 on the Cleveland Heart Disease dataset without extensive preprocessing.
• Demonstrated consistent performance across information-based and procedure-based data subsets.
• Improved AUC to 0.93 on the TCGA Glioma dataset by integrating guidelines, outperforming existing methods.
• Showcased enhanced accuracy, precision, and recall compared to purely data-driven models, especially with missing data.

Conclusions:

• The developed CDSS is a robust building block for Digital Twin architectures.
• The system effectively integrates evidence-based guidelines and handles data complexities.
• Achieved over 90% AUC on two independent tasks, highlighting clinical relevance and adaptability.