Comparing Traditional Risk Scores and an AI-Based Multimodal Model for Predicting Cardiovascular Events After Gastrointestinal Surgery

Overview

The goal of this observational study is to develop and evaluate an artificial intelligence (AI)-based multimodal model for predicting major cardiovascular events within 30 days after gastrointestinal surgery in adults at Bach Mai Hospital. The study will also compare the predictive performance of this AI-based model with commonly used traditional risk scores. The main questions it aims to answer are: Can an AI-based multimodal model predict major cardiovascular events within 30 days after gastrointestinal surgery? Does the AI-based model show better predictive performance than the Revised Cardiac Risk Index (RCRI), the American College of Surgeons National Surgical Quality Improvement Program Myocardial Infarction or Cardiac Arrest calculator (ACS NSQIP MICA), and the ACS NSQIP Surgical Risk Calculator (ACS NSQIP SRC)? Researchers will compare the AI-based multimodal model with traditional risk scores using measures of predictive performance, including discrimination, calibration, net reclassification improvement, and integrated discrimination improvement. Participants will be adults undergoing gastrointestinal surgery. Researchers will review medical record data from patients treated in 2025 and will also collect the same types of clinical data prospectively in 2026. The clinical outcome being predicted is the occurrence of major cardiovascular events within 30 days after surgery. The study will not change routine clinical care.

Major cardiovascular events after gastrointestinal surgery remain an important cause of early postoperative complications and poor outcomes. Traditional perioperative cardiac risk scores, including the Revised Cardiac Risk Index (RCRI), the American College of Surgeons National Surgical Quality Improvement Program Myocardial Infarction or Cardiac Arrest calculator (ACS NSQIP MICA), and the ACS NSQIP Surgical Risk Calculator (ACS NSQIP SRC), are widely used in clinical practice. However, their performance may be limited in specific surgical populations and may not fully capture complex interactions among clinical, laboratory, physiologic, and procedural variables. This observational study aims to develop and evaluate an artificial intelligence (AI)-based multimodal model for predicting major cardiovascular events within 30 days after gastrointestinal surgery and to compare its predictive performance with traditional risk scores. The study will be conducted at Bach Mai Hospital and will include adult patients undergoing gastrointestinal surgery. The study uses a mixed retrospective-prospective design, with retrospective data collection from patients treated in 2025 and prospective data collection in 2026. The target clinical outcome for prediction is the occurrence of major cardiovascular events within 30 days after surgery. These events include cardiovascular death, nonfatal myocardial infarction, cardiac arrest with return of spontaneous circulation, new stroke, and clinically significant arrhythmias requiring treatment. Data used for model development and comparison may include demographic characteristics, medical history, cardiovascular comorbidities, surgical characteristics, anesthetic information, preoperative laboratory results, electrocardiographic findings, biomarkers when available, and functional or risk assessment variables. The primary outcome of the study is the discrimination performance of the AI-based multimodal model compared with traditional risk scores, measured by the area under the receiver operating characteristic curve for predicting 30-day major cardiovascular events after gastrointestinal surgery. Secondary outcomes include calibration performance, net reclassification improvement, and integrated discrimination improvement of the AI-based multimodal model compared with traditional risk scores, including RCRI, ACS NSQIP MICA, and ACS NSQIP SRC. The study is observational and will not alter routine perioperative management. Data will be obtained from existing medical records and prospective clinical collection, coded for confidentiality, and analyzed to support risk stratification and model comparison in patients undergoing gastrointestinal surgery.