Machine learning model for predicting hearing preservation after vestibular schwannoma surgery: A meta-analysis

Vestibular schwannomas (VS) are benign tumors that can cause hearing loss, tinnitus and balance disorders. Radiosurgery and open surgery remain the primary treatments, but predicting hearing preservation post-surgery remains challenging due to a variety of factors. Machine learning (ML) has emerged as a promising tool for predicting hearing outcomes in VS patients. This meta-analysis aims to systematically evaluate the diagnostic accuracy of ML models in predicting hearing preservation in VS patients undergoing surgery.

We conducted a systematic review and meta-analysis following PRISMA-DTA guidelines, including studies that applied ML to predict hearing preservation in VS patients undergoing surgery. Studies were selected based on predefined inclusion criteria, and diagnostic accuracy metrics, such as sensitivity, specificity and accuracy were extracted.

A total of 15 models from 3 studies were included. The overall pooled sensitivity was 0.856 (95% CI 0.758-0.919), specificity was 0.853 (95% CI 0.713-0.931), and accuracy was 0.839 (95% CI 0.748-0.902). The area under the summary ROC curve was 0.883 (95% CI 0.770-0.910), indicating high diagnostic effectiveness. Significant heterogeneity was observed.

ML models achieve high accuracy in predicting hearing preservation in VS patients undergoing surgery. However, significant heterogeneity exists across studies, indicating the need for further research to optimize model performance and enhance their generalizability across diverse patient populations. ML has the potential to assist clinicians in providing personalized treatment strategies and improving patient outcomes in the management of vestibular schwannomas.