Ensemble machine learning methods in screening electronic health records: A scoping review.

Area of Science:

• Computational biology
• Medical informatics
• Machine learning

Background:

• Electronic health records (EHRs) offer potential for identifying undiagnosed individuals using machine learning (ML).
• Ensemble ML models, combining multiple predictions, are often superior to single models for predictive tasks.
• A literature review on ensemble ML for EHR-based medical pre-screening is lacking.

Purpose of the Study:

• To conduct a scoping review of literature on ensemble ML models for EHR screening.
• To analyze the performance and application of different ensemble ML types in medical pre-screening.

Main Methods:

• Searched EMBASE and MEDLINE databases using terms for medical screening, EHRs, and ML.
• Included 145 articles meeting inclusion criteria from 3355 retrieved.
• Followed PRISMA scoping review guidelines for data collection and analysis.

Main Results:

• Ensemble ML models are increasingly used across medical specialties and often outperform non-ensemble methods.
• Complex combination strategies and heterogeneous classifiers in ensemble models showed superior performance but were less frequently used.
• Methodologies, processing steps, and data sources for ensemble ML models were often poorly described.

Conclusions:

• Comparing diverse ensemble ML model performances is crucial for effective EHR screening.
• Standardized and comprehensive reporting of ML methodologies in clinical research is essential for reproducibility and advancement.