LOINC-Related Clinical Terminology Management in an Academic Medical Center Using Human in the Loop Machine Learning.

Area of Science:

• Medical Informatics
• Clinical Terminology Management
• Machine Learning in Healthcare

Background:

• Poor mapping of laboratory results to LOINC is a widespread issue in healthcare institutions.
• Accurate LOINC mapping is crucial for data interoperability and clinical decision-making.

Purpose of the Study:

• To describe the long-term application of a human-in-the-loop machine learning system for clinical terminology management.
• To evaluate the effectiveness of the TruData system in improving LOINC assignments for laboratory results.

Main Methods:

• Utilized TruData, an application developed at Weill Cornell Medicine since 2002, integrating reference laboratories and electronic health records (EHR).
• Employed Bayesian networks and other machine learning techniques within TruData to predict LOINC codes for laboratory results.
• Incorporated a human-in-the-loop approach for staff to adjudicate predictions and manage EHR clinical content.

Main Results:

• Over 20 years (2002-2025), TruData identified 244,000 unique laboratory result concepts, achieving LOINC determination for 193,000 (79%).
• By 2025, 84% (42,000 of 50,000) of unique laboratory result components in the EHR had LOINC assignments.
• Identified reasons for missing LOINC codes, including new results and non-LOINC warranting components like "Comment:" and "Reflex Information".

Conclusions:

• Presents one of the first case reports of human-in-the-loop machine learning applied to clinical terminology management for over two decades.
• Highlights the potential value of the TruData system and similar human-in-the-loop machine learning approaches for improving clinical terminology management in other academic medical centers.