Construction and validation of a prediction nomogram model for acute gastrointestinal failure in patients with severe traumatic brain injury.

Abstract

This study aims to establish and validate the prediction model of acute gastrointestinal failure (AGF) in patients with severe traumatic brain injury. A total of 665 inpatients from Shaoxing People's Hospital from January 2018 to January 2024 were admitted and randomly divided into training group (466 cases) and validation group (199 cases). Data were collected by general situation questionnaire and AGF assessment tool. According to the results of multivariate logistic regression analysis, the prediction nomogram model was established with R software. Bootstrap method was used for internal verification of the model, and verification group was used for external verification. The area under receiver operating characteristic (ROC) curve, Hosmer-Lemeshow test and calibration curves were used to evaluate the differentiation and calibration degree of the model. Multivariate Logistic regression analysis showed that pulmonary infection, hypoxemia, glasgow coma scale (GCS) score ≤ 8 on admission, hyponatremia and metabolic acidosis were independent risk factors for AGF in patients with severe traumatic brain injury (P < .05). On this basis, a new prediction model was constructed, as follows: logit P = -4.998 + 0.858 × pulmonary infection + 0.923 × hypoxemia + 1.488 × GCS score ≤ 8 + 1.274 × hyponatremia + 1.020 × metabolic acidosis. The area under ROC of the new model was 0.787 (95% CI: 0.831-0.909), and the cutoff point was 0.4589. The sensitivity and specificity of the model were 69.74% and 76.15%, respectively. Hosmer-Lemeshow goodness of fit test showed that the prediction model had a good fitting effect (χ2 = 4.828, P = .681). External verification showed that the Hosmer-Lemeshow goodness of fit test showed that the prediction model had a good fitting effect (χ2 = 12.712, P = .122). Calibration curves showed the nomogram established fits well with the real data. The prediction model constructed in this study has good differentiation and calibration degree, which can intuitively and easily select high-risk patients, and provide reference for early screening and gastrointestinal nursing intervention.