Assessing the accuracy of ChatGPT in interpreting blood gas analysis results ChatGPT-4 in blood gas analysis.

Engin İhsan Turan, Abdurrahman Engin Baydemir, Anıl Berkay Balıtatlı, Ayça Sultan Şahin

Journal: Journal of clinical anesthesia 2025;102():111787

PMID: 39986120

Full Text on Elsevier Science View on PubMed

Abstract

BACKGROUND

Arterial blood gas (ABG) analysis is a critical component of patient management in intensive care units (ICUs), operating rooms, and general wards, providing essential information on acid-base balance, oxygenation, and metabolic status. Interpretation requires a high level of expertise, potentially leading to variability in accuracy. This study explores the feasibility and accuracy of ChatGPT-4, an AI-based model, in interpreting ABG results compared to experienced anesthesiologists.

METHODS

This prospective observational study, approved by the institutional ethics board, included 400 ABG samples from ICU patients, anonymized and assessed by ChatGPT-4. The model analyzed parameters including acid-base status, oxygenation, hemoglobin levels, and metabolic markers, and provided both diagnostic and treatment recommendations. Two anesthesiologists, trained in ABG interpretation, independently evaluated the model's predictions to determine accuracy in potential diagnoses and treatment.

RESULTS

ChatGPT-4 achieved high accuracy across most ABG parameters, with 100 % accuracy for pH, oxygenation, sodium, and chloride. Hemoglobin accuracy was 92.5 %, while bilirubin interpretation showed limitations at 72.5 %. In several cases, the model recommended unnecessary bicarbonate treatment, suggesting an area for improvement in clinical judgment for acid-base balance management. The model's overall performance was statistically significant across most parameters (p < 0.05).

DISCUSSION

ChatGPT-4 demonstrated potential as a supplementary tool for ABG interpretation in high-demand clinical settings, supporting rapid, reliable decision-making. However, the model's limitations in interpreting complex metabolic markers highlight the need for clinician oversight. Future refinements should focus on enhancing AI training for nuanced metabolic interpretation, particularly for markers like bilirubin, to ensure safe and effective application across diverse clinical contexts.

Address: Department of Anesthesiology, Istanbul Health Science University Kanuni Sultan Süleyman Education and Training Hospital, Istanbul, Turkey. Electronic address: [email protected].; Department of Anesthesiology, Basaksehir Cam ve Sakura City Hospital, Istanbul, Turkey.; Department of Anesthesiology, Istanbul Health Science University Kanuni Sultan Süleyman Education and Training Hospital, Istanbul, Turkey.

DOI: 10.1016/j.jclinane.2025.111787

Link outs

Subscription / membership required

Full Text Sources:

MeSH Terms: Humans, Blood Gas Analysis, Prospective Studies, Female, Acid-Base Equilibrium, Male, Middle Aged, Oxygen, Hemoglobins, Intensive Care Units, Aged, Anesthesiologists, Adult, Feasibility Studies, Bilirubin, Critical Care, Acid-Base Imbalance, Generative Artificial Intelligence