Journals / Kocatepe Tıp Dergisi / 2026 / Cilt: 27 Sayı: 3

BETWEEN ARTIFICIAL INTELLIGENCE AND CLINICAL INTUITION: EVALUATING ANESTHESIOLOGIST AGREEMENT WITH VARIOUS AI MODELS IN ASA SCORING

Muhammed Emin Zora(UŞAK ÜNİVERSİTESİ, TIP FAKÜLTESİ)0000-0001-7919-3578
Pages
362–369
Publication Date
2026-07-20
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

Further multicenter studies involving real patient data and multiple expert evaluations are needed before clinical implications can be established.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:OBJECTIVE: The American Society of Anesthesiologists (ASA) Physical Status classification is widely used for preoperative risk assessment, but its subjective nature may lead to variability amog practitioners.
📊
Key Findings & Data:Their scores were compared to those of an anesthesiologist with 15 years of experience.
💡
Conclusion & Impact:Further multicenter studies involving real patient data and multiple expert evaluations are needed before clinical implications can be established.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Anesteziyoloji · Klinik Karar Destek Sistemleri · Makine Öğrenmesi

Abstract

OBJECTIVE: The American Society of Anesthesiologists (ASA) Physical Status classification is widely used for preoperative risk assessment, but its subjective nature may lead to variability amog practitioners. This study evaluated the agreement of between large language model (LLM)–based artificial intelligence (AI) systems and a human anesthesiologist in ASA scoring. MATERIAL AND METHODS: One hundred simulated patient cases, balanced across ASA classes (I–V), were developed by two expert anesthesiologists. Five current LLMs (ChatGPT, Copilot, Grok, DeepSeek, and Gemini) assigned an ASA class for each case. Their scores were compared to those of an anesthesiologist with 15 years of experience. Statistical analysis included Friedman and Wilcoxon signed-rank tests with Bonferroni correction, and Cohen’s Kappa. RESULTS: The Friedman test revealed significant differences among the models (p < 0.001). Copilot showed the highest agreement with the anesthesiologist (Kappa = 0.82), followed by ChatGPT and Grok. Deepseek and Gemini demonstrated significantly lower agreement (p < 0.001). Misclassifications were predominantly observed between adjacent ASA classes, particularly between ASA II–III and ASA III–IV. CONCLUSIONS: Specific LLM-based AI systems, particularly Copilot and ChatGPT, demonstrated high agreement with expert opinion in ASA classification under controlled and simulated conditions. This study is based on simulated cases, and the findings should be considered hypothesis-generating. Further multicenter studies involving real patient data and multiple expert evaluations are needed before clinical implications can be established.

Özet

AMAÇ: Amerikan Anesteziyologlar Derneği (ASA) Fiziksel Durum sınıflaması, preoperatif risk değerlendirmesinde yaygın olarak kullanılmaktadır. Ancak öznel yapısı, uygulayıcılar arasında farklılıklara yol açabilmektedir. Bu çalışmada, büyük dil modeli (LLM) tabanlı yapay zeka sistemlerinin ASA skorlamasında deneyimli bir anesteziyolog ile uyumu değerlendirildi. GEREÇ VE YÖNTEM: ASA sınıflarını (I–V) dengeli şekilde temsil eden 100 simüle hasta vakası, iki uzman anesteziyolog tarafından oluşturuldu. Beş güncel LLM (ChatGPT, Copilot, Grok, Deepseek, Gemini) her bir vaka için ASA sınıfı belirledi. Sonuçlar, 15 yıllık deneyime sahip bir anesteziyoloğun skorları ile karşılaştırıldı. İstatistiksel analizde Friedman ve Wilcoxon signed-rank testleri (Bonferroni düzeltmesi ile) ve Cohen’s Kappa katsayıları kullanıldı. BULGULAR: Friedman testi modeller arasında anlamlı farklılık gösterdi (p < 0,001). Copilot, anesteziyoloğun skorları ile en yüksek uyumu sağladı (Kappa = 0,82), onu ChatGPT ve Grok izledi. Deepseek ve Gemini’nin uyumu anlamlı derecede düşüktü (p < 0,001). Hatalı sınıflandırmalar çoğunlukla komşu ASA sınıfları arasında görüldü (özellikle ASA II–III ve ASA III–IV). SONUÇ: Copilot ve ChatGPT gibi bazı LLM tabanlı yapay zeka sistemleri, ASA sınıflamasında uzman görüşü ile yüksek uyum göstermektedir. Bununla birlikte, çalışma simüle vakalara dayanmakta olup sonuçlar hipotez üretici niteliktedir. Klinik kullanıma yönelik çıkarımlar için çok merkezli, gerçek hasta verilerine dayalı ve çoklu uzman değerlendirmelerini içeren ileri çalışmalara ihtiyaç vardır.

🔖

Cite This Article

Zora, M. E. (2026). BETWEEN ARTIFICIAL INTELLIGENCE AND CLINICAL INTUITION: EVALUATING ANESTHESIOLOGIST AGREEMENT WITH VARIOUS AI MODELS IN ASA SCORING. Kocatepe Tıp Dergisi, Cilt: 27 Sayı: 3, 362–369. https://doi.org/10.18229/kocatepetip.1794838

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (10)

  1. Yoon SB, Lee J, Lee H, et al. Comparison of NLP machine learning models with human physicians for ASA Physical Status classification. NPJ Digit Med. 2024;7:259.
  2. Elhaddad M, Hamam S. AI-driven clinical decision support systems: an ongoing pursuit of potential. Cureus. 2024;16(4):e57728.
  3. Turan Eİ, Baydemir AE, Gümüş Özcan F, et al. Evaluating the accuracy of ChatGPT-4 in predicting ASA scores: a prospective multicentric study. J Clin Anesth. 2024;96:111475.
  4. Cheng T, Li Y, Gu J, et al. The performance of ChatGPT in day surgery and preanesthesia risk assessment: a case-control study of 150 simulated patient presentations. Perioper Med (Lond). 2024;13(1):111.
  5. Gencer K, Gencer G. Tıp Fakültesi Öğrencilerinin Tıbbi Yapay Zeka Hazır Bulunuşluluğunun İncelenmesi. Kocatepe Tıp Dergisi. 2024;25(2):143-9.
  6. Sankar A, Johnson SR, Beattie WS, et al. Reliability of the American Society of Anesthesiologists physical status scale in clinical practice. Br J Anaesth. 2014;113(3):424–32.
  7. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56.
  8. Hashimoto DA, Witkowski E, Gao L, et al. Artificial Intelligence in Anesthesiology: current techniques, clinical applications, and limitations. Anesthesiology. 2020;132(2):379-94.
  9. Riley RH, Holman CDJ, Fletcher DR. Inter-rater reliability of the ASA Physical Status Classification in a sample of anaesthetists in Western Australia. Anaesth Intensive Care. 2014;42(5):614-8.
  10. Chen L, Zaharia M, Zou J. How is ChatGPT’s behavior changing over time? arXiv. 2023;1:2307.09009.