EHR-ML: A data-driven framework for designing machine learning applications with electronic health records

Area of Science:

• Healthcare Analytics
• Machine Learning in Medicine
• Artificial Intelligence in Healthcare

Background:

• Integration of AI into healthcare analytics faces generalisability challenges due to local data variations and suboptimal ML strategies.
• Electronic Health Record (EHR) data presents unique biases and temporal complexities that hinder traditional AI model development.
• Lack of cross-institutional data validation further complicates the reliable application of AI in diverse clinical settings.

Purpose of the Study:

• To introduce EHR-ML, a structured framework for data-driven design of optimal machine learning applications.
• To address challenges in AI generalisability within healthcare by standardizing processes and incorporating local context.
• To facilitate the development of high-performance, accurate, and locally relevant predictive models.

Main Methods:

• EHR-ML framework supports ingestion and standardization of local EHR data from diverse systems.
• Employs a fully data-driven, evidence-based approach for study design and parameter optimization.
• Utilizes customizable ensemble models to handle unique EHR data characteristics and integrates with quality control tools.

Main Results:

• Case studies demonstrate EHR-ML's capability for fully automated, high-performance model development.
• EHR-ML consistently surpasses traditional methodologies in predictive model performance.
• Models developed using EHR-ML exhibit strong generalisability across varied healthcare settings.

Conclusions:

• EHR-ML enhances clinical relevance and accuracy of predictive models by integrating local context.
• The user-friendly, automated framework accelerates hypothesis testing for localized biomedical knowledge generation.
• EHR-ML offers a robust solution for overcoming AI generalisability issues in healthcare analytics.