Machine Learning Based Semi-Automatic Iterative Annotation Method of Similar Laboratory Test Item Combination Between...

Area of Science:

• Clinical Informatics
• Machine Learning in Healthcare
• Big Data Analytics

Background:

• Secondary use of clinical datasets is crucial for data-driven research.
• Integrating multi-facility Electronic Health Records (EHRs) presents significant data management challenges.
• Establishing a unified laboratory test master across institutions is a major hurdle.

Purpose of the Study:

• To propose a sustainable, evolving method for generating a common laboratory test master.
• To leverage machine learning to reduce manual mapping efforts in clinical data integration.
• To enhance the efficiency of building large-scale, multi-facility clinical databases.

Main Methods:

• Developed a classification machine learning model using XGBoost.
• Utilized test result statistics and metadata similarity scores as classification parameters.
• Trained the model on data from 21 facilities and tested on data from 10 facilities.

Main Results:

• The XGBoost model demonstrated high performance in mapping laboratory test items.
• Achieved a high Mean Success Rate (MSR), indicating robust mapping capabilities.
• Showcased substantial efficiency gains with a high Productivity Improvement Ratio (PIR) in the initial iteration.

Conclusions:

• The proposed machine learning model offers an effective solution for automating laboratory test master creation.
• The model is designed to iteratively improve with the incorporation of more facility data.
• This approach facilitates the construction of large clinical datasets for secondary use, advancing data-driven research.