A Logistic Regression Model for Detecting the Presence of Malignant Progression in Atypical Meningiomas

To develop a method to distinguish atypical meningiomas (AMs) with malignant progression (MP) from primary AMs without a clinical history.

The clinical, radiologic, and pathologic data of 33 previously Simpson grade I resected (if any) as well as no radiotherapy treated intracranial AMs between January 2008 and December 2015 were reviewed. Immunohistochemical staining for connexin 43 (Cx43) and Ki-67 was performed. Descriptive analysis and univariate and multivariate logistic regression analyses were used to explore independent predictors of MP. A multivariable logistic model was developed to estimate the risk of MP, and its diagnostic value was determined from a receiver operating characteristic curve.

There were 11 AMs (33.3%) with histopathologically confirmed MP from benign meningiomas. The other 22 (66.7%) were initially diagnosed AMs with no histopathologically confirmed MP during a median 60.5 months (range, 42-126 months) of follow-up. Univariate and multivariate logistic analyses showed that irregular tumor shape (P = 0.010) and low Cx43 expression (P = 0.010) were independent predictors of the presence of MP, and the predicted probability was calculated by the following formula: P = 1/[1+exp.{1.218-(3.202×Shape)+(3.814×Cx43)}]. P > 0.5 for an irregularly shaped (score 1) AM with low Cx43 expression (score 0) indicated a high probability of MP. The sensitivity, specificity, positive predictive value, negative predictive value, and overall predictive accuracy were 63.6, 95.6, 87.5, 84.0, and 84.8%, respectively.

Low Cx43 expression and irregular tumor shape were independent predictors of the presence of MP. The relevant logistic regression model was found to be effective in distinguishing MP-AMs from primary AMs.