Development of a clinical-radiological nomogram for predicting severe postoperative peritumoral brain edema following intracranial meningioma resection.

Objective: The goal of this study was to develop a nomogram that integrates clinical data to predict the likelihood of severe postoperative peritumoral brain edema (PTBE) following the surgical removal of intracranial meningioma. Method: We included 152 patients diagnosed with meningioma who were admitted to the Department of Neurosurgery at the Affiliated People's Hospital of Jiangsu University between January 2016 and March 2023. Clinical characteristics were collected from the hospital's medical record system. Factors associated with severe postoperative PTBE were identified through univariate and LASSO regression analyses of clinical, pathological, and radiological features. A multivariate logistic regression analysis was then performed incorporating all features. Based on these analyses, we developed five predictive models using R software: conventional logistic regression, XGBoost, random forest, support vector machine (SVM), and k-nearest neighbors (KNN). Model performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC) and conducting decision curve analysis (DCA). The most optimal model was used to create a nomogram for visualization. The nomogram was validated using both a validation set and clinical impact curve analysis. Calibration curves assessed the accuracy of the clinical-radiomics nomogram in predicting outcomes, with Brier scores used as an indicator of concordance. DCA was employed to determine the clinical utility of the models by estimating net benefits at various threshold probabilities for both training and testing groups. Results: The study involved 151 patients, with a prevalence of severe postoperative PTBE at 35.1%. Univariate logistic regression identified four potential risk factors, and LASSO regression identified four significant risk factors associated with severe postoperative PTBE. Multivariate logistic regression revealed three independent predictors: preoperative edema index, tumor enhancement intensity on MRI, and the number of large blood vessels supplying the tumor. Among all models, the conventional logistic model showed the best performance, with AUCs of 0.897 (95% CI: 0.829-0.965) and DCA scores of 0.719 (95% CI: 0.563-0.876) for each cohort, respectively. We developed a nomogram based on this model to predict severe postoperative PTBE in both training and testing cohorts. Calibration curves and Hosmer-Lemeshow tests indicated excellent agreement between predicted probabilities and observed outcomes. The Brier scores were 10.7% (95% CI: 6.7-14.7) for the training group and 25% (95% CI: 15.2-34.8) for the testing group. DCA confirmed that the nomogram provided superior net benefit across various risk thresholds for predicting severe postoperative PTBE, with a threshold probability range from 0 to 81%. Conclusion: Utilizing conventional logistic regression within machine learning frameworks, we developed a robust prediction model. The clinical-radiological nomogram, based on conventional logistic regression, integrated clinical characteristics to enhance the prediction accuracy for severe PTBE in patients following intracranial meningioma resection. This nomogram showed promise in aiding clinicians to create personalized and optimal treatment plans by providing precise forecasts of severe PTBE.