A scoring system categorizing risk factors to evaluate the need for ventriculoperitoneal shunt in pediatric patients after brain tumor resection

Objectives To develop a scoring system based on independent predictors of the need for ventriculoperitoneal (VP) shunt after brain tumor resection in pediatric patients. Methods A total of 416 pediatric patients (≤ 14 years old) with brain tumors who underwent surgery were randomly assigned to the training (n = 333) and validation cohorts (n = 83). Based on the implementation of VP shunt, the training cohort was divided into the VP shunt group (n = 35) and the non-VP shunt group (n = 298). Univariate and multivariate logistic analyses were performed. A scoring system was developed based on clinical characteristics and operative data, and scores and corresponding risks were calculated. Results Age < 3 (p = 0.010, odds ratio [OR] = 3.162), blood loss (BL) (p = 0.005, OR = 1.300), midline tumor location (p < 0.001, OR = 5.750), preoperative hydrocephalus (p = 0.001, OR = 7.044), and total resection (p = 0.025, OR = 0.284) were identified as independent predictors. The area under the curve (AUC) of the scoring system was higher than those of age < 3, BL, midline tumor location, preoperative hydrocephalus, and total resection (0.859 vs. 0.598, 0.717, 0.725, 0.705, and 0.555, respectively; p < 0.001). Furthermore, the scoring system showed good performance in the validation cohort (AUC = 0.971). The cutoff value for predictive scores was 5.5 points, which categorized patients into low risk (0-5 points) and high risk (6-14 points) groups. Conclusions Our scoring system, integrating age < 3, BL, midline tumor location, preoperative hydrocephalus, and total resection, provides a practical evaluation. Scores ranging from 6 to 14 points indicate high risk.


Introduction
Brain tumors have the highest morbidity and mortality among pediatric patients with solid tumors throughout all stages of childhood (1,2).The rapid growth and development of the nervous system in childhood make radiotherapy and chemotherapy relatively contraindicated for children with brain tumors (3).As a result, surgery remains the predominant treatment for pediatric brain tumors (1,3,4).
Hydrocephalus is a serious postoperative complication in children with brain tumors, characterized by pathological ventricular expansion and increased intracranial pressure.Its pathogenesis may be related to an imbalance between the production and absorption of cerebrospinal fluid (CSF) (5)(6)(7).The incidence of preoperative hydrocephalus in children with brain tumors is approximately 50%, while postoperative hydrocephalus can range from 16% to 35% (7) (8,9).Hydrocephalus can cause many symptoms and sequalae depending on the age of the child such as speech impairment, neuropsychiatric disorders and life-threatening events (7,10,11).Prompt ventriculoperitoneal (VP) shunt placement is typically necessary since hydrocephalus tends to be progressive (7).Therefore, it is crucial to identify the risk factors for postoperative hydrocephalus and provide appropriate treatment.
Several studies have emphasized the significant role of age, preoperative hydrocephalus, total resection, and tumor pathologies to predict postoperative hydrocephalus in children with brain tumors, while inconsistent findings prevented from comprehensively evaluating risks (8,12,13).Factors such as limited sample sizes, variations in variables, different tumor locations (supratentorial or infratentorial), varying age definitions (ranging from < 16 to < 20 years old), and differences in statistic methods (univariate or multivariate), might contribute to these inconsistencies (9,14).Moreover, Hu et al. has made a novel discovery regarding the blood loss (BL) as an independent predictor for hydrocephalus in the children with infratentorial tumors (15).In the present work, we comprehensively involved related variables using multivariate analysis and developed a scoring system to assess the occurrence or progression hydrocephalus that needed a VP shunt in children with brain tumors.

Patients and data
The flowchart of patient selection is presented in Figure 1.The study was approved by Tongji hospital's institutional ethics committee (TJ_JRB20211271), and data were collected after obtaining consent from the patients' parents or guardians.From November 2020 to January 2021, a total of 436 patients under 14 years of age were diagnosed with brain tumors and underwent tumor resection at our hospital.Twenty patients were excluded as follows: Hydrocephalus was diagnosed using magnetic resonance imaging, symptoms, and an Evans' ratio > 0.3 (16) (Figure 2F).VP shunts were

Statistical analysis
Statistical analysis was performed using SPSS 26.0 (IBM Inc, Chicago, IL).Continuous variables were presented as median ± interquartile range, while categorical variables were expressed as frequencies (percentages).The normal distribution of the parameter dataset was assessed using the Kolmogorov-Smirnov test.Univariate logistic analysis was performed to analyze all variables between two groups.Using the stepwise method, significant variables in univariate analysis (operative time, BL, age < 3, Ki-67 index, midline tumor location, infratentorial tumors, WHO grade, preoperative hydrocephalus, total resection, ASA scale, and pathology) were then entered into a multivariate logistic regression (19).A logistic model (Model-Logit) was constructed based on independent risk factors.Risk factor categories were employed to develop a scoring system.Receiver operating characteristic (ROC) curves were generated to calculate significant variables of areas under the curve (AUCs) and cutoffs.The Delong test was performed to compare the AUCs of scoring system in training cohort with in validation cohort.In accordance with the literature, predictive scores and corresponding risk estimate were calculated (20,21).Differences with p < 0.05 were considered statistically significant.

Predictive factors for VP shunt and scoring system
The univariate logistic regression results of the predictive factors for VP shunt are shown in the Table 1.To further explore the independent predictors, we used the stepwise forward method to incorporate significant variables in univariate analysis into multivariate analysis, as presented in Table 2.The age < 3 (p = 0.010, OR = 3.162, CI = 1.314 -7.608), BL (p = 0.005, OR = 1.300,CI = 1.084 -1.560), midline tumor location (p < 0.001, OR = 5.750, CI = 2.406 -13.742), preoperative hydrocephalus (p = 0.001, OR = 7.044, CI = 2.120 -23.405), and total resection (p = 0.025, OR = 0.284, CI = 0.095 -0.855) were the independent predictors.Based on these findings, we established the Model-Logit and developed a corresponding scoring system, which is presented in Table 3.The scoring system provides the corresponding points and risk estimates, as outlined in Table 4.

Midline tumor location
Based on a cutoff value of 5.5 points, the predictive scores classified patients into low-risk (0-5 points) and high-risk (6-14 points) categories.Furthermore, the scoring system demonstrated excellent performance in an independent dataset consisted of 83 pediatric patients with brain tumors (AUC = 0.971) (Figure 3B).

Discussion
Brain tumors are commonly diagnosed in pediatric patients, and surgical resection is the primary treatment (22).However, postoperative hydrocephalus can significantly increase mortality and morbidity, especially in children (15).Previous studies investigating predictors of postoperative hydrocephalus in children with brain tumors have yielded inconsistent findings due to variations in inclusion criteria, statistical methods, limited variables, and sample sizes (1,9,12,23).Most previous studies enrolled patients aged between 16 and 20 years old.However, we believe that including pediatric patients under 14 years old is justified as it allows for a more representative reflection of the patient population, considering tumor spectrum and CSF pathophysiology (15,24,25).
In our study, we conducted a comprehensive analysis of correlated parameters in pediatric patients with brain tumors using multivariate analysis.We included patients under 14 years old and analyzed various factors associated with postoperative hydrocephalus.We observed that most cases of postoperative hydrocephalus progression occurred within two weeks.Additionally, we developed a scoring system based on independent risk predictors, including ag e< 3, BL, preoperative hydrocephalus, midline tumor location, and tumor resection.The scoring system exhibited an AUC comparable to that of the Model-Logit (0.859 vs. 0.856, p = 0.486) and outperformed any single variable in both the training and validation cohorts.
Consistent with our findings, previous analyses have shown that younger age is associated with a higher risk of postoperative or progressive hydrocephalus requiring a VP shunt (9,15,23).The incidence of preoperative and postoperative hydrocephalus is significantly higher in younger children compared to adults (26).Approximately 50% of children are reported to have hydrocephalus at the time of diagnosis, which aligns closely with the rate observed in our training cohort (9).Preoperative hydrocephalus has been found to be significantly associated with the need for VP shunt implementation following tumor resection in children with brain tumors.Surgical trauma, combined with the immature function of CSF circulation, exaggerated intracranial hypertension, and ventricular dilatation, contribute to the increased formation or acute progression of hydrocephalus.Furthermore, the unique ROC curves analyzing scoring system and independent predictors in training cohort (A) and validation cohort (B).A, AUCs of the scoring system, Model-Logit, BL, Midline tumor location, age < 3, preoperative hydrocephalus, and total resection are 0.859, 0.856, 0.717, 0.725, 0.598, 0.705 and 0.555, respectively in the training cohort.B, AUCs of scoring system, BL, Midline tumor location, age < 3, preoperative hydrocephalus, and total resection are 0.971, 0.841, 0.774.0.675, 0.811, and 0.512, respectively in the validation cohort.ROC, receiver operator characteristic; AUC, area under the curve, BL, blood loss.
anatomical structure of posterior cranial fossa has led to increased interest in exploring the influence of preoperative hydrocephalus on postoperative hydrocephalus in children with infratentorial tumors (9,23).It is noteworthy that while the incidence of preoperative hydrocephalus is higher in infratentorial tumors compared to supratentorial tumors, the location itself is not significantly associated with postoperative hydrocephalus.It implied another category (midline tumor location and others) may better explain the cause of postoperative hydrocephalus.Midline tumor location have been identified as an independent predictor of postoperative hydrocephalus formation or progression, which is consistent with previous studies (9,22,23).This association may be attributed to the inflammatory reaction of surrounding tissues caused by surgical resection adjacent to the midline.Consequently, adhesion and obstruction of the interventricular foramen, third ventricle, midbrain aqueduct, and fourth ventricle can exacerbate hydrocephalus (27).Additionally, surgical damage to the ventricular zone, blood-brain barrier, and subarachnoid space may contribute to the development or progression of hydrocephalus (27).While the supratentorial or infratentorial categories did not yield significant results, this lack of significance can be partly attributed to the proportion of cerebellar hemisphere tumors.In contrast, the significance of midline tumors underscores its close proximity to the ventricles, subsequently influencing CSF.This observation aligns with existing literature that has emphasized the proximity of infratentorial tumors to the fourth ventricle as a notable risk factor (15).Besides, our data showed that the histology was not an independent predictor of postoperative hydrocephalus.This finding may be explained by the correlation observed between histology and typical midline tumors such as medulloblastoma and ependymoma.Therefore, we recognized the impact of surgical procedures on postoperative hydrocephalus and included the extent of tumor resection in our analyses.Analysis of postoperative images in the VP group revealed that 7 children had not undergone total resection, and among them, 5 cases, including 4 with midline tumors, developed postoperative obstructive hydrocephalus.This underscores the clinical importance of total resection.Furthermore, total resection played an independent protective role, which is consistent with certain studies (14,22) although some studies have reported conflicting results (9,15,23).This discrepancy could potentially be attributed to the limited number of cases involving incomplete total resections.
We introduced a novel predictive variable, BL, which we estimated using intraoperative blood transfusion volume to mitigate the subjective bias of the operator and anesthetist in calculating BL during surgery (15).This approach provides a relatively objective reflection of intraoperative blood volume and maintenance of blood circulation in children.Evaluating intraoperative BL indirectly provides insight into the blood supply, tumor size, and the difficulty of resection, thus offering predictive value for the prognosis of children with brain tumors.Intraoperative hemorrhage can induce an inflammatory reaction and local tissue adhesion in the surgical area.Consequently, this can disrupt the connections between choroid plexus cells and corresponding cells, leading to impaired CSF flow and decreased ventricular volume maintenance function (28).Karimy et al. explored the pathophysiological mechanisms of intraoperative hemorrhage leading to the progression of hydrocephalus and found that hemorrhage can stimulate choroid plexus epithelial cells to produce an inflammatory response through factors like Toll-like receptor 4 and nuclear factor-kB (29).Thus, timely control of bleeding and blood loss management are crucial in children with brain tumors.

Rationale for scoring system
Although any single markers presented good predictive performance, they were only highlighted by their significance and applied thresholds.Numerous factors contributed to the results.Our scoring system showed the better predictive performance than any single marker both in the training cohort and validation cohort.The risk estimate corresponding to the total point could also be used in future studies.The optimal cutoff value of the scoring system was 5.5 points, which defined patients with low risks (0-5 points) and high risks (6-14 points).

Limitations of the study
There were several limitations in our study that should be acknowledged.It was retrospective and conducted in a single center, potentially limiting generalizability.Validation using data from other centers would have been preferable.We did not include postoperative CSF tests, and surgical position and imaging characteristics were not accounted for in our analysis.Additionally, BL calculation was challenging due to intraoperative factors, leading us to estimate BL based on transfusion volume.Further research is necessary to validate the findings and address the limitations.

Conclusions
Most postoperative hydrocephalus progresses within two weeks.The scoring system integrating age < 3, midline tumor location, preoperative hydrocephalus, total resection, and BL could apply practical evaluations.Children with total scores from 6 to 14 points had a high-risk level and need careful attention after surgery.conducted in accordance with the local legislation and institutional requirements.The human samples used in this study were acquired from primarily isolated as part of your previous study for which ethical approval was obtained.Written informed consent for participation was not required from the participants or the participants' legal guardians/next of kin in accordance with the national legislation and institutional requirements.
(a) refusal to undergo surgery or opting for biopsy only (n = 5); (b) previous history of VP shunt treatment (n = 7); (c) poor postoperative outcome, such as death or coma lasting over 2 weeks (n = 8).Next, 416 patients were randomly categorized into the training cohort (n = 333) and validation cohort (n = 83) based on 4:1 ratio.Based on the implementation of VP shunt, the training cohort were divided into a VP group (n = 35) and a non-VP group (n = 298).

FIGURE 1 The
FIGURE 1The flow chart of patient selection.VP, ventriculoperitoneal.

FIGURE 2
FIGURE 2 Images of pediatric patients with brain tumors.(A), a 10 years old female with supratentorial tumors in the T2 FSE of MRI before surgery; (B), enhanced MRI T1 FSE+C showed bilateral supratentorial tumors before surgery; (C), CT image of the supratentorial tumor with incomplete total resection; (D), a 3 years old male with infratentorial tumors in the enhanced MRI T1 FSE+C before surgery; (E), CT image of the infratentorial tumor after tumor resection; (F), Postoperative hydrocephalus with Evan's index > 0.3.FSE, fast spin echo; CT, compute tomography; MRI, magnetic resonance imaging; Evan's index = L1L2/L3L4.
points represented higher risks.The distance (D) was calculated based on the equation: D = b i* (W ij − W REF ).We set the constant B change of each risk factor for each point in the model.We regarded every increase of 2 U of BL as one point, as follows: B = 2 * b BL , Points j = D i /B.Finally, the risk estimate corresponding to the total score was based on the following equation: P = 1 1+exp(− o p i=0 b i c i ) ; o p i=0 b i c i = b constant + b Age* W 1REF + b BL* W 2REF + b TM * W 3REF + b PH * W 4REF + b TR* W 5REF + B * Total score = 0:524 * Total score − 5:337.Total cores ranged from 0 to 14 points.The total point and risk estimates are displayed in the Table

2 W
ij , reference value; W REF , the basic risk value; D, distance, D = b*(W ij -W iREF ); Points i = D i /B.

TABLE 1
Univariate analysis of the predictive factors for VP shunt.
Significance level where p < 0.05 were in bold.ASA, American Society of Anesthesiologists; WHO, World Health Organization; VP, ventriculoperitoneal.

TABLE 2
Results of multivariate logistic regression.

TABLE 3
Predictive model using risk factor categories.

TABLE 4
Estimate of risk corresponding to total scores.