A Retrieval-Based Approach for Automatic Interpretation of Multi-Analyte Laboratory Profiles

Area of Science:

• Medical Informatics
• Clinical Laboratory Science
• Artificial Intelligence in Medicine

Background:

• Interpreting multi-analyte laboratory profiles, like lupus anticoagulant (LAC) testing, is complex, time-consuming, and requires significant expertise.
• Existing rule-based systems are rigid and difficult to maintain, while large language models risk inaccuracies.
• A novel retrieval-based approach was developed to automate LAC profile interpretation, aiming for improved efficiency, fidelity, and reliability.

Purpose of the Study:

• To develop and evaluate a novel retrieval-based approach for automating lupus anticoagulant (LAC) profile interpretation.
• To improve the efficiency and scalability of complex laboratory profile interpretation.
• To maintain high fidelity and reliability in clinical decision-making.

Main Methods:

• Laboratory values were processed using ensemble random forest (RF) models.
• The system retrieved similar expert interpretations from a database of 347 previously interpreted LAC profiles.
• Performance was assessed via technical accuracy, workflow efficiency studies, and expert review.

Main Results:

• Technical validation showed high accuracy in retrieving interpretative elements (median 93%).
• Automated interpretation reduced reporting time by 78.6% (median 22s vs. 118s manually).
• Expert review confirmed 100% concordance for LAC presence/absence.

Conclusions:

• The novel retrieval-based approach significantly enhances the efficiency of LAC profile interpretation.
• The method maintains consistency and high fidelity, suitable for clinical use.
• Future work will explore extending this approach to other multi-analyte laboratory profiles.