AI-Assisted Blood Gas Interpretation: A Comparative Study With an Emergency Physician

Area of Science:

• Emergency Medicine
• Artificial Intelligence in Healthcare
• Clinical Diagnostics

Background:

• Accurate interpretation of arterial blood gases (ABGs) is crucial in emergency medicine.
• Large language models (LLMs) like ChatGPT are being explored for clinical applications.
• Validation of LLM accuracy in interpreting ABGs is necessary.

Purpose of the Study:

• To assess the agreement between ChatGPT and an emergency physician in interpreting 25 theoretical ABG scenarios.
• To evaluate ChatGPT's performance across various respiratory and metabolic emergencies.

Main Methods:

• Analysis of 25 theoretical ABG cases by both ChatGPT and a specialist emergency physician.
• Evaluation based on five criteria: pH, primary disorder, compensation, diagnosis, and clinical recommendation.
• Inclusion of diverse cases such as COPD, DKA, AKI, sepsis, and poisoning.

Main Results:

• High concordance (≥90%) observed in COPD, asthma, and pulmonary edema cases.
• Moderate concordance (80-90%) in DKA, AKI, and lactic acidosis.
• Lower concordance (<70%) in toxicologic and mixed acid-base disorders.
• ChatGPT provided clinically safe recommendations despite diagnostic ambiguity.

Conclusions:

• ChatGPT exhibits strong interpretive concordance for common ABG scenarios in emergency settings.
• Limitations exist in complex or context-dependent diagnoses, particularly mixed acid-base disorders.
• LLMs show potential as supportive tools for emergency medicine clinicians.