A Multicenter Observational Study to Develop and Validate a Deep Learning Model for Dynamic Assessment of Postoperative Bleeding Risk to Assist Re-operation Decision-Making in Patients With Gastric Cancer

What this Trial Is About

Researchers are investigating a deep learning model to dynamically assess the risk of postoperative bleeding that may require re-operation in adult patients (18 years and older) with primary gastric cancer who undergo radical gastrectomy. This observational study aims to determine if an AI model using perioperative physiological data and precise intraoperative blood loss can accurately predict bleeding risk, improve clinical decision-making, and enhance patient outcomes such as mortality and hospital stay length. The study uses both retrospective and prospective multi-center data for model training and validation without comparing treatment groups. The study collects retrospective data in two parts: one set to build and train the AI model and another to fine-tune parameters to avoid overfitting. Prospective data is then collected for final performance evaluation. Participants receive standard radical gastrectomy and routine postoperative care, with no additional study-specific interventions. The data collected includes demographics, medical history, vital signs, blood gas analysis, surgical details, and exact measurements of intraoperative blood loss. Participants contribute data from surgery through follow-up assessments up to 30 days post-operation (for the prospective phase). Researchers evaluate the AI model's predictive performance, focusing on its ability to identify high-risk patients requiring re-operation within 30 days, using measures such as the area under the curve (AUC-ROC). The study involves collecting comprehensive perioperative information and monitoring patient outcomes to support model validation and clinical utility.

AI-assisted Decision-making of Reoperation for Postoperative Bleeding of Gastric Cancer