Individual-patient prediction of meningioma malignancy and survival using the Surveillance, Epidemiology, and End Results database

Abstract

Meningiomas are known to have relatively lower aggressiveness and better outcomes than other central nervous system (CNS) tumors. However, there is considerable overlap between clinical and radiological features characterizing benign, atypical, and malignant tumors. In this study, we developed methods and a practical app designed to assist with the diagnosis and prognosis of meningiomas. Statistical learning models were trained and validated on 62,844 patients from the Surveillance, Epidemiology, and End Results database. We used balanced logistic regression-random forest ensemble classifiers and proportional hazards models to learn multivariate patterns of association between malignancy, survival, and a series of basic clinical variables-such as tumor size, location, and surgical procedure. We demonstrate that our models are capable of predicting meaningful individual-specific clinical outcome variables and show good generalizability across 16 SEER registries. A free smartphone and web application is provided for readers to access and test the predictive models (www.meningioma.app). Future model improvements and prospective replication will be necessary to demonstrate true clinical utility. Rather than being used in isolation, we expect that the proposed models will be integrated into larger and more comprehensive models that integrate imaging and molecular biomarkers. Whether for meningiomas or other tumors of the CNS, the power of these methods to make individual-patient predictions could lead to improved diagnosis, patient counseling, and outcomes.

Keywords: CNS cancer; Cancer epidemiology; Predictive markers.

Fig. 1. Descriptive statistics for the malignancy outcome variable.

Kernel density plots illustrate the distribution of benign, borderline malignant, and malignant meningiomas according to age at diagnosis (a) and tumor size (b). These kernel density plots are conceptually equivalent to histograms, but illustrate density (i.e., relative number of patients) as a continuous function of age/tumor size. Total number of meningiomas by WHO ICD-O-3 behavior codes are shown in c. Absolute numbers and percentages of patients with benign, borderline malignant, and malignant meningiomas by subgroup are shown for laterality (d), sex (e), race (f), and primary tumor site (g).

Fig. 2. Log hazard ratios for each of the features of the survival model.

Negative values indicate proportionally lower probability of death. Positive values indicate proportionally higher probability of death. Error bars represent 95% confidence intervals.

Fig. 3. Performance of the malignancy classifier.

a Confusion matrix illustrating predicted vs. true labels for the malignancy classifier, as evaluated on the test set. Values are normalized across each row. B Benign, BM/M Borderline malignancy/Malignant. b Drop-column feature importance showing decrease in classifier performance resulting from dropping a given feature, in decreasing order of importance. The red dot indicates the mean and error bars 95% confidence intervals. c Learning curves illustrating training (red line) and cross-validation (blue line) model performance (measured by Area under the receiver operating characteristic curve and Average Precision) as a function of the number of patients used in training the classifier. The point of convergence between the training and cross-validation curves indicates when adding more cases to the training no longer results in an improvement in performance. Shaded outlines represent 1 standard deviation. The gray line represents the performance of a dummy classifier, which randomly generates predictions on the basis of the class distribution in the training set. d Bivariate kernel density plot (can be understood as a “2-dimensional histogram”) of tumor size vs. age at diagnosis. e Calibration plot, as evaluated on the test set. f Precision-recall curve and receiver operating characteristic curve (g) for Benign vs. Borderline Malignant/Malignant meningioma classification. For f, g the gray dashed line indicates chance-level performance and the shaded outline represents the 95% confidence intervals.

Fig. 4. Calibration and performance of the survival model.

a Survival model calibration plot, as evaluated on the test set. b Time-dependent area under the curve (AUCt, yellow line) and average precision (APt, blue line) for the survival model, as evaluated on the test set. The event rate/chance level is represented by the dashed gray line. Shaded outlines represent 95% confidence intervals.

Fig. 5. Example predicted malignancy and survival curves for an insured 56-year-old white man with a unilateral 54 mm wide meningioma localizing to the cerebral meninges.

Try the app at www.meningioma.app.

Fig. 6

Flow diagram illustrating criteria for patient inclusion.