Machine Learned Mapping of Local EHR Flowsheet Data to Standard Information Models using Topic Model Filtering.

Area of Science:

• Health Informatics
• Machine Learning Applications
• Nursing Data Standards

Background:

• Electronic health record (EHR) data requires mapping to standard information models for interoperability and research.
• Challenges exist in mapping organizational flowsheet data to new nursing information models.

Purpose of the Study:

• To describe an automated approach for mapping organizational flowsheet data to standard information models.
• To address the problem of correctly mapping information models to specific flowsheet data implementations.

Main Methods:

• Utilized stacked machine learning models for automated mapping.
• Employed a topic model keyword filter to identify potential flowsheet row matches.
• Used a support vector machine (SVM) as a secondary, more accurate classifier for each concept.

Main Results:

• Achieved an overall f2 score of 0.74 for mapping flowsheets to information models.
• The stacked model combination demonstrated effectiveness in the mapping process.

Conclusions:

• The developed stacked machine learning approach automates the mapping of flowsheet data to nursing information models.
• This method is generalizable for mapping other data types with short text descriptions, enhancing data interoperability.