This retrospective study included patients admitted to the neurosurgery ward of our institute from January 2015 to December 2016. All enrolled patients suffered from severe disturbance of consciousness with the GCS scores ≤ 12 before shunt implantation. Glasgow Outcome Scale Extend (GOS-E) score and Coma Recovery Scale-Revised (CRS-R) were used to evaluate their clinical status. We assessed the prognostic value of several clinical characteristics to formulate a comprehensive predictive model. And the model was developed after statistical analysis. Then the verification of this model was tested in a validation cohort of patients with PTH and disturbance of consciousness.

The inclusion criteria were as follows: (1) diagnosis of PTH; (2) complete medical records including history of treatment, personal characteristics, radiographs, and surgical information; (3) persistent unconsciousness after TBI (with no other causes, no lucid interval, and GCS score ≤ 12); and (4) ventriculoperitoneal shunt procedure performed in our institute. Exclusion criteria: (1) patients with known severe neural deficits before the TBI, and (2) those who died because of any other diseases before the end of our two-year follow-up.

Each patient underwent a brain Computed Tomography (CT) scan upon arrival at our department, as well as a follow-up scan after VPS surgery. The CT images evaluated the diagnosis of PTH, the severity of PTH, and the location of the proximal catheter. In all patients, intracranial pressure (ICP) was measured by lumbar punctures before and after the shunt surgery. VPS was performed, and the Miethke proGAV or Medtronic Strata II with the delta chamber was inserted. For proGAV values, the initial pressure was set at the level that about 20 mmH₂O lower than the ICP. And for Medtronic Strata II, the valves were set one level lower than the measured ICP. Subsequent adjustments were performed according to the alteration of patients’ clinical symptoms and imaging examinations.

The diagnosis of PTH was based on clinical symptoms and radiographic evidence. The detailed standards were as follows: (1) the value of Evans’ index greater than 0.3; (2) the enlargement of the third ventricle, the temporal and anterior horns of lateral ventricles, and the periventricular interstitial edema without brain atrophy; (3) clinically progressive degradation of neurological function or no evidence of significant clinical improvement; and (4) development of hydrocephalus within 12 months of trauma. The ratio of the cerebral ventricle diameter (the maximum transverse diameter in the middle of the ventricle) to the biparietal diameter (V/BP) was used to evaluate the severity of hydrocephalus; 26-40% indicated mild, 4-60% indicated moderate, and 61-90% indicated severe hydrocephalus.

Data were collected retrospectively from our electronic medical record system. A team of radiologists reviewed all previous imaging data. Our medical team compiled the following clinical data: necessary information (age and gender), cause of injury, severity of hydrocephalus, status of the cisterna ambiens, GCS scores at the time of injury and before shunt placement, ICP measured by lumbar puncture, skull defect (unilateral or bilateral), time interval from injury to cranioplasty, and the time interval from injury to shunt implantation. The GOS-E and Coma Recovery Scale-Revised (CRS-R) scores were evaluated before shunt implantation and at every follow-up appointment. The causes of injury included falls, road traffic accidents, violence, and other reasons. For convenience in establishing the prognostic model, we converted some of the variables into dumb variables (0, 1, 2…). The details of the patients’ characteristics and variable transformation are presented in Table 1.

The neurological outcomes were evaluated at 6, 12 months, and 2 years after shunt implantation using the GOS-E (Teasdale et al., 1998) and CRS-R scores. An increase in the GOS-E score after shunt implantation was considered as an improvement. For a patient whose GOS-E score remained unchanged, an increase in the score of any single subscale of the CRS-R was recognized as an improvement in the patient’s outcome. We obtained the outcome assessment from the patient’s medical examination in our institution or medical records from other rehabilitation centers. To those who received rehabilitation treatment at home, the referent physician or legal guardian was contacted. GOS-E scores, part of the CRS-R scale (motor scores and arousal scores), were evaluated when performed by a legal guardian.

The predictive model was developed using binary logistic regression, with the outcome after shunt implantation at the end of the two-year follow-up as the dependent variable and the analyzed clinical characters as independent variables. The weighted scores were assigned to each factor based on their β-regression coefficient values and then were transformed to the nearest integer to develop a practical and uncomplicated prognostic score. The prognostic score was calculated by summing all the scores for each patient, and then the patients were grouped into three groups based on their score percentile: low probability (0-33rd percentile), intermediate probability (34th-67th percentile), and high probability (68th-100th percentile) of improvement after shunt implantation. The validation was then launched to assess the accuracy of this scoring model. The β-regression coefficient values and the transformation of the weighted scores are presented in Table 2.

We used the SPSS (Version 17.0, Chicago, IL, USA) for analyses. Pearson’s chi-square or Fisher’s exact test was used for univariate analysis. Variables associated with the outcome of shunt implantation (P < 0.05) were included in the logistic regression model to identify the independent predictors of outcome and were retained in the final model if the P-value was < 0.05. We used the chi-squared test to determine if there was an increased probability of improvement in the patients’ outcome among the three categories. The receiver operator characteristic curve (ROC) and the area under the receiver operating characteristic curve (AUC) were used to access the discriminatory power of the scoring system. The AUCs was classified into three levels: good (AUC > 0.8) predictive ability, moderate (AUC = 0.7-0.8) predictive ability, and low (AUC = 0.6-0.7) predictive ability. And then, the H-L test was used to evaluate the model’s calibration ability.

A total of 66 patients with PTH combined with disturbance of consciousness were included. Two patients were excluded due to missing data during the 2-year follow-up period. Five patients experienced severe uncontrolled infection after the VPS surgery, and the tubes were finally removed. Of the 59 patients, 12 had a GOS-E score of 2, and the remaining 47 received a score of 3 before the VPS surgery. By the end of the two-year follow-up period, clinical improvements were observed in 31 (52.5%) patients. Among them, 28 patients showed an improved GOS-E score (GOS-E increased from 3 to 4 in 21 patients, 3 to 5 in 4 patients, and 2 to 3 in 3 patients), while the other 3 patients demonstrated an increased CRS-R score (1 showed improvement in motor scores and 2 in arousal scores). No significant difference was observed between the improved and non-improved groups on gender, ICP, skull defect, the status of cisterna ambiens, the time interval from injury to cranioplasty, and GCS score before shunt implantation.

By univariate analysis, correlations were revealed between the outcome of shunt implantation and the patients’ age, cause of injury, the severity of hydrocephalus, time interval from injury to shunt surgery, and GCS score at the time of injury. Subsequently, a binary logistic regression analysis was performed. After eliminating variables with poor predictive ability, the following factors were included in the final model: age < 50 years, mild hydrocephalus, the time interval from injury to shunt implantation < 3 months, and GCS scores 9-12 at the time of injury. A predictive scoring system was developed according to the logistic regression model, and each predictor was assigned a single score based on its regression coefficient value (Table 2). The total score was calculated by summing the scores of the corresponding predictive factors for each patient. And then, the patients were divided into three groups based on the total score: low-probability (0-10 points), intermediate-probability (11-16 points), and high-probability (17-30 points). The probabilities of improvement after shunt implantation in these three groups were 14.3%, 52.6%, and 94.7%, respectively (Table 3).

Patients admitted to our institute from January 2014 to December 2015 were included in the validation cohort. Thirty-two patients with PTH and disturbance of consciousness were added according to the same inclusion and exclusion criteria. There was no statistically significant difference in the patients’ characteristics between the derivation and validation cohorts (Table 4). In the validation cohort, 15 patients showed improvement in GOS or CRS-R score by the end of the 2-year follow-up period, while the remaining 17 patients show no improvement. Using our predictive scoring system, the rates of improvement in the three groups were 21.4%, 54.5%, and 85.7%, respectively. The probability of improvement after shunt surgery significantly increased in patients with higher scores in both the derivation and validation cohorts (both chi-square for trend P values < 0.05, Table 3). A ROC curve was created to evaluate the accuracy of the predictive model. The ROC and the AUCs were shown below (Fig. 1). The AUC of the derivation data was 0.869 (95% confidence interval, 0.755-0.942), with a specificity of 64.29% and a sensitivity of 90.32%, indicating that the model performed good predictive power for the outcome. The analysis of the Hosmer-Lemshow (H-L) test (P = 0.391) shows a good calibration.

Figure 1.

Receiver-operating characteristics curve for the scoring system in the derivation cohort (area under the receiver operating characteristic curve = 0.869).