Construction and performance evaluation of a predictive model for post-traumatic hydrocephalus in patients with severe traumatic brain injury
10.3760/cma.j.cn501098-20250618-00347
- VernacularTitle:重型创伤性脑损伤患者并发创伤后脑积水预测模型构建与效能评估
- Author:
Bin XU
1
;
Xin WANG
;
Jiahao LIAO
;
Yuhai WANG
;
Jinxu ZHOU
Author Information
1. 安徽医科大学无锡临床学院，无锡 214040
- Publication Type:Journal Article
- Keywords:
Brain injuries;
Hydrocephalus;
Forecasting;
Nomogram;
Model
- From:
Chinese Journal of Trauma
2025;41(11):1059-1069
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To develop a predictive model for the risk of post-traumatic hydrocephalus (PTH) in patients with severe traumatic brain injury (sTBI) and validate its predictive performance.Methods:A retrospective case control study was conducted to analyze the clinical data of 580 sTBI patients admitted to the 904th Hospital of the Joint Logistics Support Force of the PLA between January 2016 and December 2023, including 413 males and 167 females, aged 18-88 years [(54.3±14.6)years]. Patients were stratified into PTH group ( n=195) and non-PTH group ( n=385), based on the presence of PTH within 6 months after injury. Data collected from the two groups such as general baseline indicators, TBI-related clinical indicators (including surgical data), laboratory findings, and radiological features. Except for the data collected during the operation, all the above data are the results of the first examination at admission. Univariate analysis and Lasso regression analysis were used to screen predictors for the risk of PTH in sTBI patients. Subsequent multivariate Logistic regression was employed to identify predictors and construct a regression equation. Based on this equation, a nomogram prediction model was developed using the R language. Model discrimination was estimated through the receiver operating characteristic (ROC) curve, and calibration performance via the Hosmer-Lemeshow (H-L) goodness-of-fit test and calibration curve. Moreover, decision curve analysis (DCA) and clinical impact curve (CIC) were used for evaluating the clinical utility of the model. Results:Univariate analysis revealed statistically significant differences in 37 variables between the two groups, including age, age group, heart rate, oxygen saturation, Glasgow coma scale (GCS) score, left pupil size, right pupil size, pupillary light reflex, intracranial pressure (ICP) monitoring, type of decompressive craniectomy, neutrophil count, lymphocyte count, monocyte count, red blood cell count, platelet count, neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), neutrophil-lymphocyte-platelet ratio (N/LP), lymphocyte-to-monocyte ratio (LMR), systemic inflammation response index (SIRI), systemic immune-inflammation index (SII), fibrinogen (FIB), D-dimer, D-dimer-to-fibrinogen ratio (DFR), serum albumin, prognostic nutritional index, blood glucose, status of basal cisterns, midline shift, degree of midline shift, cerebral herniation, epidural hematoma (EDH), subdural hematoma (SDH), intraventricular hemorrhage (IVH), subarachnoid hemorrhage (SAH), modified Fisher grade, and skull fracture ( P<0.05). Lasso regression analysis identified 24 potential predictors for PTH, including age, GCS score, pupillary light reflex, type of decompressive craniectomy, monocyte count, platelet count, NLR, PLR, N/LP, LMR, SII, D-dimer, DFR, serum albumin, prognostic nutritional index, blood glucose, status of basal cisterns, degree of midline shift, cerebral herniation, EDH, SDH, IVH, modified Fisher grade and skull fracture. Multivariate Logistic regression analysis demonstrated that age, unilateral pupillary light reflex, absent pupillary light reflex, bilateral decompressive craniectomy, monocyte count, PLR, cerebral herniation, SDH, IVH, linear skull fracture and depressed skull fracture were independent risk factors for PTH. In contrast, serum albumin was identified as an independent protective factor for PTH ( P<0.05). The regression equation derived from these factors was: Logit[ P/(1- P)]=0.05×"age"+1.65×"unilateral pupillary light reflex"+2.79×"absent pupillary light reflex"+1.60×"bilateral decompressive craniectomy"+1.90×"monocyte count"+0.02×"PLR"-0.12×"serum albumin"+2.07×"cerebral herniation"+2.59×"SDH"+2.23×"IVH"+1.24×"linear skull fracture"+ 1.66×"depressed skull fracture"-22.61. The prediction model built upon this equation achieved an area under the ROC curve (AUC) of 0.95(95% CI 0.93, 0.97), with a sensitivity of 91.79%, specificity of 85.97%, and Youden′s index of 0.78. The H-L goodness-of-fit test indicated good calibration ( χ2=7.90, P=0.545). DCA results showed that the bias-corrected curve closely aligned with the actual curve and approximated the ideal curve, indicating a high clinical net benefit. Furthermore, CIC results demonstrated that with threshold probabilities greater than 60%, the number of patients identified as high-risk by the model highly corresponded with the actual number of patients who developed PTH. Conclusion:The prediction model incorporating age, unilateral pupillary light reflex, absent pupillary light reflex, bilateral decompressive craniectomy, monocyte count, PLR, serum albumin, cerebral herniation, SDH, IVH, linear skull fracture and depressed skull fracture exhibits robust predictive performance for PTH in sTBI patients.
