{
  "abstract": "Background Acute gastrointestinal injury (AGI) is a common complication in critically ill trauma patients, yet its quantitative assessment remains challenging. This study aimed to investigate the incidence and risk factors of AGI in this population and to develop a predictive model for AGI grading by incorporating objective bowel sound monitoring.Methods We conducted a retrospective cohort study using the MIMIC-IV database. 2,539 critically ill adult trauma patients admitted to the intensive care unit (ICU) were included. AGI was diagnosed and graded (I–IV) according to the 2012 European Society of Intensive Care Medicine (ESICM) consensus definitions. We analyzed clinical characteristics, outcomes, and digitally quantified bowel sound parameters. Univariate analysis and multivariate Generalized Estimating Equations (GEE) were used to identify independent factors associated with higher AGI grades. Two predictive models were constructed and validated: 1) a clinical model using GEE incorporating key clinical variables (e.g., abdominal injury, vasopressor duration, SOFA score, enteral nutrition), and 2) a comprehensive model using a Back-Propagation (BP) neural network, integrating processed bowel sound data (e.g., frequency, amplitude) with the most predictive clinical features identified from principal component analysis.Results The overall incidence of AGI was high, with 51.87% of patients presenting with AGI grade II or higher. Bowel sound reduction (hypoactive) or absence (absent) were the most prevalent signs. Multivariate GEE analysis identified abdominal injury, duration of sedative/analgesic use, ICU length of stay, laxative use, mechanical ventilation, and shock as independent risk factors for higher AGI grades, while early enteral nutrition served as a protective factor. The GEE-based clinical prediction model for AGI grading achieved an overall accuracy of 83%. The BP neural network model, which synthesized bowel sound parameters with clinical data (including lactate, SOFA score, and vasopressor duration), demonstrated a superior predictive accuracy of 84.54%.Conclusions Critically ill trauma patients have a high incidence of significant AGI. We identified distinct modifiable and non-modifiable risk factors. This study underscores the clinical value of objective bowel sound monitoring as a non-invasive marker of gastrointestinal function. The proposed predictive models, particularly the integrated neural network model, offer accurate, quantifiable tools for grading AGI severity.",
  "authors": [
    {
      "affiliations": [
        "The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, China"
      ],
      "name": "Yi Yu"
    }
  ],
  "title": "IDDF2026-ABS-0050 A predictive model for acute gastrointestinal injury grading in critically ill trauma patients based on bowel sound monitoring",
  "uid": "3c2723e7-97e7-5c57-a860-bde0f7683fc7"
}