As a training set, 191 patients with PTH were treated with an infusion test, followed by ventriculoperitoneal shunting (VP) or lumboperitoneal shunting (LP) (consultant decisions based on clinical, radiological assessment and the infusion results) in our institution (Department of Neurosurgery, The First Affiliated Hospital, Zhejiang University) from August 2015 to July 2018. The evaluable pre-therapeutic Rout and the level of modified Rankin Scale (mRS) after shunting were retrospectively analyzed. These were all patients with ventriculomegaly on CT or MRI (as reported by an experienced consultant radiologist) and with clinical characteristics of PTH.

To confirm the predictive value of Rout and RAP, and to explore the predictive possibility of combining them, patients aged between 16 and 72 years with diagnosed PTH within 2 months were consecutively and prospectively enrolled in our study as a validation data set, from October 2018 to February 2020. Our specific inclusion criteria were as follows:

In our study, all patients with PTH should suffer from a clinical status with a level running from 3 to 6 in modified Rankin Scale (mRS) before enrolling, because we need to find the potential patients who may benefit from shunting and to reduce the baseline error, which can disturb the outcome assessment based on mRS. Then, they, firstly, underwent an infusion test to generate Rout data. Secondly, they received continuous lumbar drainage to collect RAP for consecutive 48 h. Finally, shunt operation without surgical taboos was implemented with them based on clinical, radiological assessments and the test results.

The exclusion criteria should include epilepsy, inflammation, non-communicating hydrocephalus, and other serious diseases, which make patients unable to tolerate invasive diagnosis and chemotherapy. Patients with acute hydrocephalus or intracranial hypertension, who required mannitol or glycerol fructose to reduce intracranial pressure, were also excluded. The patients taking recent antiepileptic drugs, such as sodium valproate and carbamazepine, were also excluded. Finally, normal controls were not available, as all studies in our center are performed on clinical indications.

The study was approved by the Ethical Committee of The First Affiliated Hospital attached to Zhejiang University School of Medicine. All the patients in the validation data set were provided with written informed consent.

Infusion studies were performed via lumbar puncture with the patient in the lateral recumbent position. An 18-G spinal needle was performed below L3. The patient's head was carefully positioned to align the inion with the spine horizontally and parallel to the floor. The needle was connected through a 3-way tap, with a pressure-recording device to one side and with an infusion pump to the other side (pump, Grase by 3100; recorder, Codman, Medtronic, USA). The baseline steady state of CSF pressure was recorded for at least 10 min (range, 10–20 min). To avoid systematical error, we were more inclined to take constant-rate infusion other than bolus injection. Once the intracranial pressure (ICP) reading remained steady, infusion with saline solution was started at a constant rate of 1.5 ml/minute in all patients, and recording was continued for a further 10 min after ICP had reached its plateau but ≤ 15 mmHg. The CSF pressure was recorded continuously for 15–35 min (mean, 25 min). In addition, the infusion was terminated either if CSF pressure increased continuously until reaching a Ro > 18 mmHg/ml/minute or once a steady-state plateau pressure was approached (25). Then, the CSF Rout was calculated according to the equation described by Czosnyka et al. (14).

As the infusion test is accomplished, the next day, a standard lumbar puncture was performed, usually in the lumbar 4–5 interspinous space, with a large-bore Tuohy needle (a 14–16 gauge). As the CSF was encountered, the curve of the needle was directed superiorly, the stylet was removed, and the catheter was advanced into the subarachnoid space for about 20–22 cm. Then, a (a 17–18 gauge) catheter (Medtronic, USA) was slowly inserted with one hand, while the needle was simultaneously extracted. The drain was attached to a three-way tap with a pressure-recording device (Codman, Medtronic, USA) to one side and with an external drainage sterile container to the other side. The catheter was then taped over the patient's flank. The lumbar drain was fixed to the shoulder level, and the patients were advised to have complete bed rest. We paid close attention and monitored the drain to avoid overtraining. Once the monitored ICP had reached its plateau but ≤15 mmHg, we started to open drainage. The lumbar drain was set to drain 10–15 cc per hour and approximately 300–400 cc in 2 days. Data were processed using Neumatic DCR + software (Haoju, Shanghai, China) and saved in our Neumaticstudy database.

Modified Rankin Scale (mRS) has remarkable strengths in evaluating the outcome of spontaneous subarachnoid hemorrhage (SAH) and hydrocephalus (26, 27). More importantly, mRS covers the degrees of gait and defecation disorder, which are two important clinical criteria for hydrocephalus assessment. Therefore, we decided to use mRS as an indicator to judge the prognosis instead of Glasgow Outcome Scale (GOSE). MRS is a 7-level outcome scale, running from perfect health without symptoms to death, which shows as follows:

0: no Symptoms.

We defined mRS of 0–2 as desirable recovery level (dRL) and mRS of 3 to 6 as non-dRL referred to related research. Since dRL was easily communicable, understandable, and desirable to patients and neurosurgeons, it was chosen as a primary study endpoint of our study.

The following CSF dynamics parameters were extracted and calculated from our database: ICP baseline (ICPb), ICP at plateau (ICPp), cranioplasty date (if applicable), resistance to outflow (Rout), and the correlation coefficient between the ICP wave amplitude and the mean ICP level (RAP), which is also called the compensatory reserve index. In our study, to avoid the inaccuracy of instant RAP value in a prognosis, we also designed a maximum RAP value as RAPmax, which stands for the highest RAP parameter during different date segments (RAP1 max represents the baseline RAPmax at the beginning of continuous lumbar drainage, RAP2max represents the highest RAP parameter from the start to the 1st day end, and RAP3 max represents the highest RAP parameter for the 2nd day), and defined ΔRAPmax as (baseline RAPmax – interim RAPmax)/baseline RAPmax × 100 (%). Additional clinical data: patient demographics, date/severity of TBI, date of infusion test, continuous lumbar drainage study, and brain imaging were extracted from the electronic hospital record system.

IBM SPSS 22.0 (IBM Corp. New York, USA) and MedCalc 19.0 (MedCalcSoftware bvba, Ostend, Belgium) were used for statistical analyses. Data were described as numbers (percentages) or means ± standard deviation (SD). The predictive performance of parameters was tested by receiver-operating characteristic (ROC) curve analysis. The logistic regression model was used to evaluate different predictive parameters and calculated predictive probability used for ROC analysis. Optimal cut-off values to predict shunting response were calculated by the Youden index using MedCalc19.0 software. We have used Pearson's or Spearman's correlation when appropriate, depending on how much our data deviated from a bivariate normal distribution or an asymptotically normal distribution, and P-values of less than 0.05 were considered statistically significant.

Table 2 summarizes the characteristics of patients in the training set. About 191 patients with PTH with Rout and the dRL 3 months after shunting were retrospectively analyzed, among whom 134 patients (70.2%) were treated with VP shunting based on clinical and radiological assessment. From the records, 134 (70.1%) had been initially classified as having “severe” and 57 (29.9%) “mild” TBI according to the Glasgow Coma Scale (GCS). The interval time between TBI and infusion test varied between subjects, from 2 weeks to 18 weeks; average time, 8.2 ± 3.1 weeks. Of patients, 50.3% (96/191) achieved dRL 3 months later.

From October 2018 and February 2020, 82 patients were enrolled in our study, and underwent CT scan and Rout assessment, but 12 of them were excluded because of bad physiological conditions, which could not get them through shunting (n = 8) and withdrawal consent with the reason of being short of money or other individual subjective reasons (n = 4). Finally, 70 patients with evaluable Rout, RAP, and outcome data were eventually included for statistical analysis. All patients with PTH were treated with shunting. Of them, 72.9% (51/70) received VP shunting, 27.1% (19/70) received LP shunting. The follow-up time for evaluation based on the mRS was designed at 3 months after shunting, but, finally, 70% (47/70) of them accomplished the follow-up at 3 months, 8.6% (6/70) and 7.1% (5/70) fulfilled ≥ 4 months and 5 months, respectively, due to patients' personal causes. Overall, dRL was achieved in 54.4% of the patients (38/70).

In the training set, Rout was 14.1 ± 3.3 mmHg/min/ml (mean ± SD). We used the ROC curve to assess the value of Rout in predicting the dRL to shunting independently (Figure 1), and then acquired an AUC of 0.686 (95% confidence interval [CI] 0.615–0.751, p < 0.001). It suggested that pretherapeutic Rout was a dRL predictor of shunting in patients with PTH. Cut-off value of 13.6 for Rout offered the best accuracy in predicting dRL with the sensitivity of 61.05%, specificity of 75.00%, positive predictive value (PPV) of 70.7%, and negative predictive value (NPV) of 66.1%.

In the validation set, Rout was 13.7 ± 4.1 mmHg/min/ml. Then, all 70 patients were served with continuous lumbar drainage tests to collect RAP at the baseline (RAPb) and during drainage (RAPdra). The correlation between Rout and RAPb was significant with a moderate correlation coefficient of 0.40 (p < 0.001; Figure 2). Baseline RAPmax (RAP1max, mean ± SD) was 0.70 ± 0.14, the 1st-day end and the 2nd-day end of continuous lumbar drainage, mean RAPmax's (RAP2max and RAP3max) were 0.57 ± 0.10 and 0.34 ± 0.13, respectively (Figure 3). MeanΔRAPmax's% after 1 and 2 days (ΔRAP1max% and ΔRAP2max %) were 40.2 ± 15.1% and 40.4 ± 17.4%, respectively. The correlation coefficients between dRL and ΔRAP1max %, dRL, and ΔRAP2max % were 0.67 (p < 0.001) and 0.69 (p < 0.001), respectively. The correlation between ΔRAP1max% and ΔRAP2max% was significant with a correlation coefficient of 0.59 (p < 0.001; Figure 4). A representative example of a continuous lumbar drainage test performed on a patient with posttraumatic hydrocephalus is shown in Figure 5.

Firstly, we performed univariate logical regression with all parameters. Then, significance threshold of p < 0.10 removed age, shunting strategy, ICPb, ICPdra, RAPb, RAPdra from the comparison. And put the remaining parameters, including Rout (p = 0.06), ΔRAP1max% (p < 0.001), andΔRAP2max% (p < 0.001), into multivariate logistic regression to conduct multiple regression models. Finally, 5 separate regression models (Table 3) were analyzed for their ability to predict the shunting response of patients with PTH. The predictive model of the total index shows the largest AUC = 0.88 and highest significance of the Hosmer–Lemeshow test (0.211) among the 5 models, indicating its better performance in the goodness of fit.

Consequently, we evaluated the predictive value of Rout, ΔRAP1max%, and ΔRAP2max% alone using ROC analysis in the validation set (Figure 6). There was no significant value for Rout in prediction dRL with the AUC of 0.634 (95% CI, 0.511–0.746, p = 0.0613). However, either ΔRAP1max% or ΔRAP2max% could effectively predict the dRL of shunting in the primary stage of PTH, with the AUC of 0.773 (95% CI, 0.657–0.864, p < 0.001) and 0.786 (95% CI, 0.671–0.875, p < 0.001), respectively. At a cut-off of 45%, ΔRAP1max% offered the best accuracy in predicting dRL with the sensitivity, specificity, PPV, and NPV of 89.5%, 56.3%, 70.8%, and 81.8%, respectively; 53% was the optimum cut-off of ΔRAP2max%, and it discriminates between dRL and non-dRL. The sensitivity, specificity, PPV, and NPV were 76.32%, 75.00%, 78.4%, and 72.7%, respectively.

To exploit the predictive value of association of Rout and ΔRAPmax%, we united Rout with ΔRAP1max % and ΔRAP2max % in the logistic analysis model, and the predictive model of total index shows the largest AUC = 0.88 (95% CI, 0.671–0.875, p < 0.001) and highest significance of Hosmer–Lemeshow test (0.211) among the 5 models, indicating its better performance in the goodness of fit (Figure 6). The sensitivity, specificity, PPV, and NPV of the total index model to predict dRL were 81.62%, 90.62%, 91.2%, and 80.6%, respectively.

We also explored the predictive value of RAPmax and found the baseline RAPmax (RAP1max), post-1-day RAPmax (RAP2max) and post-2-day RAPmax (RAP3max) all could not predict dRL (figures were not shown in the article). However, when we changed the target parameter as totally non-dRL (Levels 3 to 6 in modified Rankin Scale), the RAP2max and RAP3max revealed remarkable predictive value with the AUC of 0.891 (95% CI, 0.793–0.953, p < 0.001) and 0.746 (95% CI, 0.628–0.842, p < 0.001; Figure 7), RAP1max still showed no predictive value (AUC of 0.587, p = 0.2058). With the cut-off value at 0.65 of RAP2max, the sensitivity, specificity, PPV, and NPV to predict the non-dRL were 89.5%, 84.4%, 67.2%, and 97.1%, respectively. In addition, 0.51 was the optimum cut-off for RAP3max in predicting the non-dRL with the sensitivity, specificity, PPV, and NPV of 92.11%, 50.00%, 58.6%, and 94.2%, respectively.

In summary, our correlation analysis subsequently revealed a positive correlation between Rout and RAP at the baseline (r = 0.40, p < 0.001) (Figure 2). It indicated that, in patients with higher Rout (suggesting abnormal CSF circulation), the accompanied RAP at the baseline may often reach a higher level approaching 1 (meaning poorer compensatory reserve). Then, we originally hypothesized that the model of Rout united with RAPb could present excellent ability in predicting a prognosis. Unfortunately, in our univariate logical regression, neither RAPb (p = 0.231) nor RAPdra (p = 0.304) showed significance in evaluating the outcome of patients with PTH after shunting. However, the RAP-related parameters, ΔRAP1max (<0.001) and ΔRAP2max (p < 0.001), which weighed the maximum variation of RAP value based on RAPb; both entered the multivariate logistic regression. There was also a strong association between dRL and ΔRAP1max% with a correlation coefficient of 0.67 (p < 0.001), dRL and ΔRAP2max% with 0.69 (p < 0.001). We further combined Rout with ΔRAP1max% and ΔRAP2max%; the AUC in predicting dRL reached 0.879 (p < 0.001). However, excluding Rout from the total index model lowered its effectiveness to an AUC of 0.81.

Therefore, ΔRAP1max% should be taken more attention in clinical observation. Moreover, we found ΔRAP2max% was significantly correlated with ΔRAP1max% when RAPmax during the 1st day had a significant decrease; the decrease would last to the 2nd day. It suggested that the continuous drainage, which imitated shunting, continued to be effective in some patients. Therefore, the ΔRAP1max% after the drainage of the 1st day was efficient to predict the PTH patient's dRL.

There were no complications or side effects related to the lumbar infusion test or continuous lumbar drainage test. The shunting procedure was complicated by a subdural hematoma in two patients: in one of these, the hematoma was solved surgically; in the other one, the hematomas did not require dissipated and were absorbed spontaneously. One patient had a postoperative shunt infection with encephalitis and was treated with antibiotics and removal of the shunt device. After 2 months, a new shunt was implanted. All complications had been resolved before the follow-up assessment.