Artificial Intelligence-Based Prediction of Sarcopenia Risk in Intensive Care Unit Patients With Intracranial Pathology

This study is designed as a prospective observational study. Patients admitted to the Level III Intensive Care Units of Trabzon University Faculty of Medicine, Kanuni Training and Research Hospital, Kaşüstü Campus, due to intracranial pathologies between January 1, 2026, and June 30, 2026, will be included. Approximately 100-150 patients are planned to be evaluated.

Demographic data of the enrolled patients will be recorded, and the modified Nutrition Risk in Critically Ill (mNUTRIC) score will be calculated on the first day of intensive care unit admission. Rectus femoris muscle thickness will be evaluated by ultrasonography on Day 0 and Day 7 of ICU admission. All ultrasonographic measurements will be performed using the same ultrasound device and by the same investigator according to a standardized protocol. During the measurements, the patient will be positioned supine, the knee will be kept in extension, and the muscle will be evaluated in a relaxed position. Three repeated measurements will be obtained at each assessment, and the mean value will be recorded.

As part of the laboratory assessment, prealbumin levels will be measured on Days 0, 3, and 7. Biochemical parameters evaluated during routine clinical follow-up will be recorded from the hospital information system.

No intervention, additional procedure, or treatment modification will be performed as part of this study. All data will consist of observational data obtained during routine clinical follow-up. Data collection will be conducted by a resident physician from the Department of Anesthesiology and Reanimation with experience in intensive care.

The collected clinical, laboratory, and ultrasonographic data will be provided to different artificial intelligence models, and their accuracy and performance in predicting sarcopenia development on Day 7 will be evaluated. The primary objective of the study is to assess the predictive performance of artificial intelligence models, including ChatGPT, Gemini, and Claude, for Day 7 sarcopenia development in intensive care patients with intracranial pathologies. Secondary objectives include comparing artificial intelligence predictions with clinical assessments, comparing predictive performance among different artificial intelligence models, and evaluating the potential usability of artificial intelligence models as clinical decision-support tools in intensive care practice.