Improving information retrieval from electronic health records using dynamic and multi-collaborative filtering

Area of Science:

• Health Informatics
• Information Retrieval
• Artificial Intelligence

Background:

• Physicians experience significant information overload with the growing volume of patient data in health information technology systems.
• Inefficiencies in reviewing electronic health records (EHRs) can hinder effective patient care.
• Current systems often struggle to present the most relevant information to clinicians at the point of care.

Purpose of the Study:

• To introduce a novel hybrid dynamic and multi-collaborative filtering method for enhanced information retrieval from EHRs.
• To assist physicians by recommending pertinent information during patient visits.
• To address the challenge of information overload in clinical decision-making.

Main Methods:

• Developed a hybrid approach combining dynamic filtering (Markov model) and multi-collaborative filtering.
• Modeled information search dynamics using a Markov model to capture temporal aspects of information seeking.
• Incorporated collaborative filtering principles to prioritize information based on similarities between physicians, patients, and data items.
• Utilized EHR data from the Indiana Network for Patient Care for testing and validation.

Main Results:

• The hybrid method achieved 46.7% accuracy in top-5 recommendations for predicting physician information interests.
• The method demonstrated 24.7% accuracy for top-1 recommendations.
• The new approach showed a 22.3% improvement over the conventional Markov model for top-1 recommendations.

Conclusions:

• The proposed hybrid dynamic and multi-collaborative filtering method effectively improves information retrieval from EHRs.
• This approach offers a promising solution to mitigate physician information overload and enhance clinical efficiency.
• The findings suggest significant potential for recommender system techniques in improving healthcare information systems.