Hydrocephalus management pathways in pediatric medulloblastoma: a retrospective cohort study

To compare recorded postoperative intracranial infection and recurrent hydrocephalus across three hydrocephalus management pathways in children with medulloblastoma and to conduct an exploratory internal prediction analysis for postoperative infection after ventriculoperitoneal (VP) shunting.

This retrospective single-center cohort and prediction-model study included 420 children with medulloblastoma and preoperative hydrocephalus treated at Beijing Tiantan Hospital from January 1, 2014, through March 1, 2018. Patients were grouped by initial definitive hydrocephalus pathway: preoperative VP shunt, upfront tumor resection without permanent preoperative diversion, or preoperative endoscopic third ventriculostomy (ETV). Pairwise propensity score matching was performed for VP shunt versus upfront resection, ETV versus upfront resection, and ETV versus VP shunt. The minimum planned postoperative follow-up was 6 months. Candidate VP-shunt infection models were evaluated by repeated stratified 5-fold cross-validation, with imputation, standardization, and sampling fitted within training folds only.

The cohort included 227 VP-shunt, 140 upfront-resection, and 53 ETV patients. Postoperative infection occurred in 55/227 patients (24.2%) after VP shunting, 31/140 (22.1%) after upfront resection, and 5/53 (9.4%) after ETV. Recurrent hydrocephalus or rescue diversion occurred in 5/227 patients (2.2%) after VP shunting, 34/140 (24.3%) after upfront resection, and 5/53 (9.4%) after ETV. In matched ETV versus VP-shunt pairs, infection occurred in 5/50 ETV patients (10.0%) versus 12/50 VP-shunt patients (24.0%; matched odds ratio [OR], 0.36; 95% confidence interval [CI], 0.12-1.14; P = 0.118), and recurrent hydrocephalus occurred in 5/50 ETV patients (10.0%) versus 4/50 VP-shunt patients (8.0%; matched OR, 1.25; 95% CI, 0.34-4.66; P = 1.000). The internally selected L2 logistic VP-shunt infection model achieved repeated cross-validation area under the receiver operating characteristic curve (AUC) 0.627 and Brier score 0.175. ETV recurrence was not modeled because only five events occurred.

The pathway comparisons are exploratory observational associations rather than causal effects. Both matched ETV comparisons were non-significant, with wide confidence intervals compatible with clinically opposite directions and important residual imbalance. The internally evaluated VP-shunt infection model is not independently validated and must not be used for individual treatment decisions. Larger prospective, anatomy-enriched, multi-institutional studies with complete follow-up are required.