DOI: 10.1093/noajnl/vdag164 ISSN: 2632-2498

Prediction of meningioma shrinkage after cyproterone acetate cessation

Annabelle Collin, Virginie Montalibet, Paul E Constanthin, Olivier Saut, Julien Engelhardt

Abstract

Background

Progestin-induced meningiomas may stabilize or regress after discontinuation of cyproterone acetate (CPA), but the determinants of this volumetric response remain unclear. Identifying predictors of tumor shrinkage could improve patient management and prognosis.

Methods

We retrospectively analyzed the longitudinal volumetric evolution of 137 meningiomas diagnosed during ongoing CPA treatment in 52 consecutive patients. Tumor volumes were measured from all available imaging studies. Four mathematical models (linear, exponential, power, Gompertz) were compared to describe tumor growth trajectories. Response patterns were derived from Gompertz modeling. Predictive models were constructed using random forests, with leave-one-out cross-validation. Baseline clinical and radiological variables, as well as early volumetric response at 3 months, were evaluated as predictors.

Results

The Gompertz model best described tumor volume evolution. Three patterns of tumor response after CPA withdrawal were identified: response with low limit volume (47%), response with high limit volume (41%), and non-response (12%). At the patient level, analogous clusters were found: low limit response (22%), high limit response (64%), and non-response (14%). Random forest models predicted cluster membership with error rates of 24.8% at the tumor level and 23.1% at the patient level. The most informative predictors were initial tumor burden and patient age at CPA exposure and withdrawal. Incorporating early volumetric response at 3 months reduced the prediction error to 9.5% for tumors and 3.8% for patients.

Conclusions

Tumor burden and age at CPA exposure are major predictors of meningioma shrinkage after treatment discontinuation. Early volumetric changes provide additional prognostic value and significantly improve predictive accuracy.

More from our Archive