Machine learning for endoscopic third ventriculostomy success prediction-a systematic review and meta-analysis

Endoscopic third ventriculostomy (ETV) is a common treatment for pediatric obstructive hydrocephalus, but predicting its success remains challenging. Traditional predictive tools, such as the Endoscopic Third Ventriculostomy Success Score (ETVSS) and logistic regression (LR) models, are widely used; however, recent advancements in machine learning (ML) have shown promise in improving prediction accuracy. This systematic review and meta-analysis aim to evaluate the effectiveness of ML models in predicting ETV success and compare them to traditional models.

A systematic search across five databases was performed. Authors searched for studies, which used ML algorithms to predict ETV success. This review included studies included studies that reported the area under the receiver operating characteristic curve (AUC) for model performance. ETV success was considered the absence of ETV failure criteria in 6 months after procedure: either recurrence of hydrocephalus symptoms, repeated surgery, or mortality.

A total of four studies involving 3087 pediatric patients were included. The overall pooled AUC for ML models was 0.63 (95% CI 0.56-0.70), with significant heterogeneity (I² = 96%). Subgroup analysis revealed that models including imaging data had a slightly higher AUC (0.74, 95% CI 0.61-0.88). No significant differences were found between ML models and traditional tools like ETVSS or LR models.

ML models show moderate potential for predicting ETV success but do not outperform traditional tools like ETVSS and LR models in clinical application. High heterogeneity and methodological limitations suggest that further research is needed to optimize and validate ML algorithms.