Evaluation of ChatGPT-5 and Anesthesiologist Predictions of Postoperative Intensive Care Unit Admission Using Preoperative Clinical Data

Postoperative Care

Artificial Intelligence, ChatGPT, Anesthesiology, Intensive Care Unit, Postoperative Care, Risk Prediction

Accurate prediction of postoperative intensive care unit (ICU) requirement is essential for patient safety and efficient use of limited ICU resources. In routine clinical practice, decisions regarding postoperative ICU admission are primarily based on anesthesiologists' preoperative clinical judgment, which may vary among clinicians. This prospective, observational study aims to evaluate the agreement between predictions made by ChatGPT-5(Chat Generative Pre-trained Transformer) and anesthesiologists regarding postoperative ICU requirement using routinely collected preoperative patient data, and to compare these predictions with actual postoperative ICU admission outcomes. The study does not involve any intervention, treatment modification, or additional procedures beyond standard clinical care. All patient data are anonymized, and clinical management is not influenced by the model's predictions

Postoperative intensive care unit (ICU) admission is a critical component of perioperative patient management, particularly in patients with increased surgical or anesthetic risk. Accurate preoperative identification of patients who will require postoperative ICU care may improve patient safety and optimize resource allocation. This study is designed as a prospective, non-interventional observational cohort study conducted in adult patients undergoing elective surgical procedures. Routinely collected preoperative clinical data, including demographic characteristics, comorbidities, laboratory results, and anesthesiologists' assessments, are recorded for each participant. For each patient, postoperative ICU requirement predictions generated by ChatGPT-5 using structured preoperative data are documented. These predictions are compared with anesthesiologists' preoperative ICU admission assessments and with actual postoperative ICU admission outcomes. No additional diagnostic or therapeutic interventions are performed as part of the study. Patient care follows standard institutional practice at all times. All collected data are anonymized prior to analysis. Statistical analyses focus on agreement and predictive performance measures, including sensitivity, specificity, and concordance between prediction methods and actual outcomes.

No related papers found yet.

Turkey (Türkiye)