DIAGNOSTIC ACCURACY AND THERAPEUTIC RATIONALITY IN ACID-BASE IMBALANCES: A COMPARATIVE ANALYSIS BETWEEN PHYSICIANS AND ARTIFICIAL INTELLIGENCE

S.T. Amrin, A.S. SATBAYEVA, A.A. Abdussemyatova, Y.A. DUISEN

Vestnik, 2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.53065/kaznmu.2026.77.2.005
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Introduction. Acid-base balance disorders are critical conditions in intensive care units requiring rapid and accurate management. The study explores the potential of large language models to serve as accessible clinical decision support systems to reduce iatrogenic errors. Aim. To compare the accuracy of diagnosing acid-base balance disorders and the rationality of therapeutic recommendations proposed by artificial intelligence based on ChatGPT and by intensive care physicians. Materials and Methods. Study design: retrospective, single-center, comparative study. The analysis included 302 clinical and laboratory cases of patients treated in an intensive care unit between 2024 and 2025. Using prompt engineering techniques, an adapted ChatGPT model named “ReanimatorKZ” was developed. A comparative expert evaluation was conducted to assess the conclusions of ChatGPT and intensive care physicians regarding acid–base disorders. Statistical analysis was performed using StatTech v.4.12.7 and SPSS Statistics 27.0.1. Results. In the group of doctors, diagnostic accuracy was 71.2% correct, 24.8% partially correct, and 4.0% incorrect conclusions. The AI demonstrated a lower rate of completely incorrect diagnostic conclusions (2.0%), while completely correct diagnoses accounted for 64.9%. The physicians’ treatment strategies were completely correct in 60.9% of cases, whereas the AI’s recommendations were completely correct in 89.7% of cases, with no completely incorrect therapeutic recommendations classified for the AI. Statistically significant differences were confirmed using paired tests (p < 0.05). Conclusion. An adapted version of ChatGPT demonstrated a high level of diagnostic accuracy in identifying acid–base disorders, comparable to that of intensive care physicians, and superior accuracy in formulating therapeutic recommendations for these conditions. Our study supports the potential for developing effective and readily scalable clinical decision support systems based on widely available artificial intelligence models. However, additional prospective validation is required before such systems can be implemented in routine clinical practice. Keywords: artificial intelligence, intensive care, clinical decision support systems, ChatGPT, diagnostic accuracy.