Rationalisation of a thrombophilia panel using laboratory medicine Delphi-like consensus evaluation and secondary...

Area of Science:

• Clinical Laboratory Medicine
• Medical Diagnostics
• Artificial Intelligence in Healthcare

Background:

• Optimizing laboratory test panels is essential to prevent overdiagnosis and improve diagnostic accuracy.
• Artificial intelligence (AI) is being explored as a potential tool to assist in clinical decision-making for laboratory testing.
• Thrombophilia testing requires regular evaluation to ensure its clinical utility and efficiency.

Purpose of the Study:

• To revise an existing thrombophilia testing panel using a modified Delphi consensus approach.
• To evaluate the capability of large language models (LLMs) to replicate expert decision-making in revising laboratory test panels.
• To assess LLMs as a potential support tool for laboratory physicians in test panel optimization.

Main Methods:

• A modified Delphi consensus process involving six experts was used to evaluate thrombophilia parameters based on reliability and clinical significance.
• Selected large language models (Elicit, Consensus, STORM) were tested with specific prompts related to thrombophilia testing.
• Agreement between LLM outputs and expert consensus was quantified using percentage concordance and Cohen's kappa (κ).

Main Results:

• Several thrombophilia tests, including PAI-1, MTHFR genotyping, homocysteine, APC resistance, anti-annexin-V, and anti-phosphatidylserine/prothrombin antibodies, were removed or eliminated from the standard panel.
• LLMs showed slight agreement (κ ≈ 0.25) with expert consensus using open prompts, but higher agreement (κ 0.50-0.52) with specific questions.
• LLMs did not consider crucial analytical and technical aspects in their decision-making process.

Conclusions:

• Regular review and optimization of thrombophilia panels are necessary to maintain high diagnostic quality and avoid unnecessary tests.
• The Delphi consensus method is effective for revising and optimizing laboratory test panels.
• LLMs can serve as a supplementary resource for expert decision-making but cannot replace the experience and consensus of medical professionals.