SURG-02. SURVIVAL PREDICTION AFTER NEUROSURGICAL RESECTION OF BRAIN METASTASES: A MACHINE LEARNING APPROACH. (9th November 2020)

• Record Type: Journal Article
• Authors: Hulsbergen, Alexander; Lo, Yu Tung; Kavouridis, Vasileios; Phillips, John; Smith, Timothy; Verhoeff, Joost; Yu, Kun-Hsing; Broekman, Marike; Arnaout, Omar
• Is Part Of: Neuro-oncology. Volume 22(2020)Supplement 2
• Page Start: ii203
• Page End: ii203
• Publication Date: 2020-11-09
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noaa215.849
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: INTRODUCTION: Survival prediction in brain metastases (BMs) remains challenging. Current prognostic models have been created and validated almost completely with data from patients receiving radiotherapy only, leaving uncertainty about surgical patients. Therefore, the aim of this study was to build and validate a model predicting 6-month survival after BM resection using different machine learning (ML) algorithms. METHODS: An institutional database of 1062 patients who underwent resection for BM was split into a 80:20 training and testing set. Seven different ML algorithms were trained and assessed for performance. Moreover, an ensemble model was created incorporating random forest, adaptive boosting, gradient boosting, and logistic regression algorithms. Five-fold cross validation was used for hyperparameter tuning. Model performance was assessed using area under the receiver-operating curve (AUC) and calibration and was compared against the diagnosis-specific graded prognostic assessment (ds-GPA); the most established prognostic model in BMs. RESULTS: The ensemble model showed superior performance with an AUC of 0.81 in the hold-out test set, a calibration slope of 1.14, and a calibration intercept of -0.08, outperforming the ds-GPA (AUC 0.68). Patients were stratified into high-, medium- and low-risk groups for death at 6 months; these strata strongly predicted both 6-months and longitudinal overall survival (p < 0.001). CONCLUSIONS: We developed and internally validated an ensemble ML model that accurately predicts 6-month survival after neurosurgical resection for BM, outperforms the most established model in the literature, and allows for meaningful risk stratification. Future efforts should focus on external validation of our model.