Construction and performance evaluation of a prediction model for postoperative poor in-hospital prognosis in patients with traumatic brain injury
10.3760/cma.j.cn501098-20250613-00336
- VernacularTitle:创伤性脑损伤患者术后院内不良预后预测模型构建与效能评估
- Author:
Tao MEI
1
;
Zheyong JIA
1
;
Lie CHEN
1
;
Peng CAO
1
;
Wei XIAO
1
;
Weiqiang MAO
1
;
Jianwu GONG
1
;
Lixin XU
1
Author Information
1. 中南大学湘雅医学院附属常德医院（常德市第一人民医院）神经外科，常德 415100
- Publication Type:Journal Article
- Keywords:
Brain injuries;
Prognosis;
Risk factors;
Nomogram;
Logistic models
- From:
Chinese Journal of Trauma
2025;41(11):1048-1058
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To construct a prediction model for postoperative poor in-hospital prognosis in patients with traumatic brain injury (TBI) and evaluate its predictive performance.Methods:A retrospective case control study was conducted to analyze the clinical data of 1 120 TBI patients admitted to Changde Hospital Affiliated to Xiangya Medical College of Central South University from May 2019 to December 2024. The patients were divided into the training set ( n=784) and verification set ( n=336) at a ratio of 7∶3. Based on the Glasgow outcome scale-extended (GOS-E) at discharge, the training set was stratified into favorable prognosis group ( n=335, GOS-E 5-8 points) and poor prognosis group ( n=449, GOS-E 1-4 points). The two groups in the training set were compared in terms of general baseline indicators, TBI-related clinical indicators, and admission laboratory blood test results. Univariate analysis and Lasso regression analysis were employed to screen risk factors associated with postoperative poor in-hospital prognosis in TBI patients. Multivariate Logistic regression analysis was used to determine independent risk factors and construct a regression equation. The regression equation was presented using R language to create a visual nomogram for predicting postoperative poor in-hospital prognosis in TBI patients. In both the training set and verification set, the predictive performance of the model was evaluated by calculating the area under the receiver operating characteristic (ROC) curve (AUC), plotting calibration curves, and performing decision curve analysis (DCA). Results:The results of the univariate analysis indicated that the age, Charlson complication index (CCI), time from trauma to admission, time from trauma to operation, cause of injury, abbreviated injury scale (AIS) (head and neck), injury severity score (ISS), admission Glasgow coma scale (GCS), admission pupil responsiveness, multiple craniocerebral injuries, subdural hematoma, intracerebral hematoma, intraventricular hemorrhage, subarachnoid hemorrhage, decompressive craniotomy, intraoperative blood loss, intraoperative blood transfusion, traumatic cerebral infarction, postoperative delayed bleeding, epilepsy seizures, as well as the following admission tested results including red blood cell count, white blood cell count, platelet count, neutrophil percentage, percentage of lymphocytes, albumin, total bilirubin, urea nitrogen, thrombin time (TT), prothrombin time (PT), international standardized ratio (INR), glutamic aminotransferase, alanine aminotransferase, creatinine, and blood glucose were statistically different between the two groups in the training set ( P<0.05). Lasso regression analysis suggested 14 risk factors of age, CCI, cause of injury, head and neck AIS, ISS, admission GCS, admission pupil responsiveness, multiple craniocerebral injuries, subdural hematoma, intracerebral hematoma, intraoperative blood loss, admission platelet count, admission albumin, admission blood glucose for postoperative poor in-hospital prognosis. The results of the multivariate Logistic regression analysis showed that age ( OR=1.02, 95% CI 1.00, 1.03, P<0.01), CCI ( OR=1.46, 95% CI 1.02, 2.09, P<0.05), head and neck AIS ( OR=1.43, 95% CI 1.11, 1.85, P<0.01), ISS ( OR=2.16, 95% CI 1.39, 3.35, P<0.01), admission GCS ( OR=1.59, 95% CI 1.19, 2.13, P<0.01), intracerebral hematoma ( OR=4.41, 95% CI 2.15, 9.44, P<0.01), intraoperative blood loss ( OR=1.05, 95% CI 1.00, 1.09, P<0.05), admission platelet count ( OR=0.98, 95% CI 0.97, 0.99, P<0.01), admission blood glucose ( OR=1.08, 95% CI 1.02, 1.15, P<0.05) could be the main risk factors to construct a prediction model for postoperative poor in-hospital prognosis in TBI patients. Meanwhile, a regression equation was constructed: Logit[ P/(1- P)]=-2.4+ 0.02×"age"+0.38×"CCI"+0.36×"head and neck AIS"+0.77×"ISS"+0.47×"admission GCS"+1.48×"intracerebral hematoma"+0.05×intraoperative blood loss-0.02×admission platelet count+0.08×admission blood glucose. In the training set, the predictive model for poor postoperative in-hospital prognosis in TBI patients achieved an AUC of 0.87 (95% CI 0.84, 0.89), with a Youden′s index of 0.57, sensitivity of 73.70%, and specificity of 83.00%. In the verification set, the model showed an AUC of 0.80 (95% CI 0.76, 0.85), with a Youden′s index of 0.63, sensitivity of 65.20%, and specificity of 77.90%. In the training set, the Brier score for the calibration curve was 0.14 (95% CI 0.13, 0.16). In the verification set, the Brier score for the calibration curve was 0.18 (95% CI 0.15, 0.20). The DCA diagram indicated that the nomogram prediction model provided high clinical net benefit for predicting postoperative poor in-hospital prognosis in TBI patients. Conclusion:The prediction model for postoperative poor in-hospital prognosis in TBI patients, constructed based on age, CCI, head and neck AIS, ISS, admission GCS, intracerebral hematoma, intraoperative blood loss, admission platelet count, and admission blood glucose, exhibits good predictive performance.
