Statistical supervised meta-ensemble algorithm for medical record linkage.

Area of Science:

• Health Informatics
• Machine Learning
• Data Science

Background:

• Patient record linkage across multiple healthcare facilities is a significant challenge for continuous care and health research.
• The increasing volume and complexity of patient data in the big data era necessitate scalable and accurate linkage methods.
• Privacy and security concerns are paramount when identifying and linking patient records.

Purpose of the Study:

• To develop and evaluate an ensemble classification method for linking patient records across different service locations.
• To assess the performance of the ensemble method compared to individual supervised learning algorithms.

Main Methods:

• An ensemble classification method was developed by combining bagging and stacking techniques.
• The ensemble model utilized three base learners: support vector machines, logistic regression, and feed-forward neural networks.
• Hyperparameters for base learners were optimized using a grid search technique on two synthetic datasets (FEBRL and ePBRN).

Main Results:

• The ensemble method demonstrated superior performance over individual base learners across all evaluation metrics on the ePBRN dataset.
• Precision improved from 90.70% to 94.85% (FEBRL) and 62.17% to 99.28% (ePBRN).
• F-score increased from 94.92% to 98.18% (FEBRL) and 72.99% to 91.72% (ePBRN).

Conclusions:

• Ensemble strategies can significantly enhance the performance of individual patient record linkage algorithms.
• The proposed ensemble method offers a scalable and accurate solution for patient identification challenges in healthcare.
• This approach holds promise for improving data quality in health research and patient care continuity.