Automated Logical Observation Identifiers Names and Codes mapping with biomedical natural language processing models:...

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Clinical Terminology Standardization

Background:

• Efficient health information exchange relies on consistent clinical concept representation across healthcare systems.
• The Logical Observation Identifiers Names and Codes (LOINC) standard facilitates interoperability but faces mapping challenges with diverse, incomplete, or inconsistently formatted datasets.
• Algorithmic performance in controlled settings often differs from real-world deployment due to data heterogeneity.

Purpose of the Study:

• To develop a biomedical natural language processing (NLP) approach for mapping heterogeneous laboratory test strings to LOINC v2.81.
• To compare the performance of the developed NLP approach against established algorithms in the Open Concept Lab (OCL) Mapper.
• To evaluate the utility of NLP for improving LOINC mapping accuracy with diverse clinical data.

Main Methods:

• Implementation of a ScispaCy-based pipeline (ScispaCy-LOINC) involving clinical entity identification, UMLS Concept Unique Identifier linking, LOINC code assembly, and weighted scoring.
• Evaluation of ScispaCy-LOINC performance against two OCL algorithms: Elasticsearch Keyword Retrieval (OCL-Keyword) and MiniLM Semantic Search (OCL-Semantic).
• Testing on two datasets: MIMIC-IV lab_d_items and a LOINC-mapped subset of the CIEL interface terminology.

Main Results:

• ScispaCy-LOINC achieved the highest coverage in the MIMIC-IV dataset (42.3%), outperforming OCL-Keyword (19.5%) and OCL-Semantic (21.4%).
• In the CIEL dataset, OCL-Semantic demonstrated the highest coverage (54.4%), followed by OCL-Keyword (46.9%), with ScispaCy-LOINC at 28.4%.
• Algorithmic performance varied based on dataset characteristics, with ScispaCy-LOINC excelling in noisier or sparser data.

Conclusions:

• ScispaCy-LOINC offers a flexible and effective approach for LOINC mapping, particularly for challenging, real-world clinical datasets.
• OCL-based algorithms perform better with more standardized terminologies, indicating complementary strengths among different mapping strategies.
• An integrated framework combining diverse algorithmic approaches is recommended to enhance robustness and accuracy across various clinical data types.