Development of a Machine Learning Algorithm for the Prediction of WHO Grade 1 Meningioma Recurrence

Objective Meningiomas commonly recur following gross total resection (GTR), and the risk of recurrence is difficult to predict using current classification schemes such as the World Health Organization (WHO) tumor grade. This study aimed to create a predictive model of recurrence risk following GTR of WHO grade 1 meningiomas based on histopathological and epidemiological factors.  Methods A retrospective chart review was completed for all patients at our institution who underwent their first surgery for a WHO grade 1 meningioma between 2017 and 2022. Those with genetic predispositions, such as neurofibromatosis type 2, were excluded. Baseline characteristics, including histopathology findings, were obtained, and we used a Risk-calibrated Superspase Linear Integer Model (Risk-SLIM) with a five-fold cross-validation (CV) to create a predictive model of recurrence over an average follow-up of three years.  Results Univariate analysis of our selected variables revealed a significant predictive association between WHO grade 1 meningioma recurrence and subtotal resection but not with any other variable. However, the meningioma recurrence score (MRS) generated by our machine learning algorithm revealed multiple predictive factors of recurrence, including age, female gender, and various histopathologic features, including the Ki-67/MIB-1 index. Conclusions Machine learning algorithms like the one we present here may help identify patients at high risk of recurrence of their WHO grade 1 meningioma, and they are more likely to benefit from closer postoperative surveillance or adjuvant treatment, even when GTR is achieved.