Objective To explore an individualized prediction scheme for sepsis patients with moderate to severe acute gastrointestinal injury based on clinical characteristics.

Methods A retrospective analysis was conducted on the clinical data of 316 sepsis patients admitted to Henan Traditional Chinese Medicine Hospital from August 2021 to August 2024, and they were designated as observation group. According to the occurrence of moderate to severe acute gastrointestinal injury during hospitalization, patients in the observation group were divided into complication group (n=165) and non-complication group (n=151). Multivariate logistic regression analysis was used to screen the influencing factors for sepsis patients with moderate to severe acute gastrointestinal injury. An individualized nomogram prediction model was constructed, and its predictive efficacy was internally validated. Additionally, the clinical data of 158 sepsis patients admitted to our hospital from January 2020 to July 2021 were retrospectively analyzed and used as external validation group to externally validate the nomogram model. A random forest prediction model was constructed using the data from the observation group, and its predictive efficacy was evaluated using the data from the external validation group.

Results The results of multivariate logistic regression analysis showed that mechanical ventilation (OR=2.472), multiple organ dysfunction syndrome (MODS) (OR=4.023), high Sequential Organ Failure Assessment (SOFA) score at admission (OR=3.083), high Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score (OR=2.835), high white blood cell (WBC) count (OR=1.610), high blood lactate level (OR=1.893), high C-reactive protein (CRP) level (OR=2.036), high D-lactate level (OR=2.620), high endotoxin level (OR=3.834), high diamine oxidase (DAO)level (OR=3.916), high intestinal fatty acid-binding protein (I-FABP) level (OR=4.175), and high intra-abdominal pressure (OR=3.511) were all risk factors for sepsis with moderate to severe acute gastrointestinal injury (P < 0.05). High levels of glucagon-like peptide-2 (GLP-2) (OR=0.825), high gastric motility index (MI) (OR=0.485), and high superior mesenteric artery end-diastolic velocity (VPd) (OR=0.559) at admission were all protective factors (P < 0.05). Based on the above factors, a nomogram prediction model for sepsis with moderate to severe acute gastrointestinal injury was constructed. After internal and external validation, the concordance indices were 0.862 and 0.858, respectively, and the calibration curves closely fitted the ideal curves. The receiver operating characteristic (ROC) curve showed thatthe sensitivity, specificity, and area under the curve (AUC) of the model's predictions were 88.48%, 86.09%, 0.889 and 87.50%, 85.90%, 0.884, respectively. The decision analysis curve showed that the model had a high net benefit when the threshold probability was in the ranges of 0.25—0.94 and 0.31—0.98. According to the ranking based on the average decrease in Gini index in the random forest prediction model, MODS, SOFA score, I-FABP, DAO, and endotoxin were the top five indicators, which had a significant impact on the prediction of moderate to severe acute gastrointestinal injury. The ROC curve assessment showed that the sensitivity, specificity, and AUC of the random forest prediction model were 78.75%, 83.33%, and 0.816, respectively. The sensitivity and AUC of the individualized nomogram model for predicting sepsis with moderate to severe acute gastrointestinal injury were higher than those of the random forest prediction model, and their specificities were basically equivalent.

Conclusion Mechanical ventilation, MODS, and SOFA score, APACHE Ⅱ score, WBC count, blood lactate, CRP, GLP-2, D-lactate, endotoxin, DAO, I-FABP, intra-abdominal pressure, MI, and VPd at admission are all influencing factors for sepsis with moderate to severe acute gastrointestinal injury. The constructed individualized nomogram prediction model has important guiding value for guiding the early clinical screening of high-risk patients and the timely formulation of appropriate intervention plans.