COMP-13. EARLY TUMOR VOLUME EVOLUTION ADAPTIVELY PREDICTS PATIENT-SPECIFIC RESPONSE AND PROGRESSION RISK TO RADIOTHERAPY, PEMBROLIZUMAB, AND BEVACIZUMAB IN RECURRENT HIGH-GRADE GLIOMA IN A PHASE I TRIAL. (11th November 2019)

• Record Type: Journal Article
• Authors: Glazar, Daniel; Brady, Rene; Howard, Rachel; Grass, Daniel; Arrington, John; Yu, Michael; Raghunand, Natarajan; Sahebjam, Solmaz; Enderling, Heiko
• Is Part Of: Neuro-oncology. Volume 21(2019)Supplement 6
• Page Start: vi63
• Page End: vi64
• Publication Date: 2019-11-11
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noz175.256
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: PURPOSE/OBJECTIVES: We set out to predict response and progression risk in recurrent high-grade glioma patients treated with hypofractionated stereotactic radiation plus pembrolizumab and bevacizumab (NCT02313272). At present RANO criteria define progression as 25% increase in sum of the products of perpendicular diameters of enhancing lesions compared with the smallest tumor measurement (either at baseline or best response) or significant increase in T2/FLAIR non-enhancing lesion on stable or increasing doses of corticosteroids. To this extent, a predictive model is needed to adaptively learn and forecast individual response to therapy. We evaluate if early tumor volume evolution can train a mathematical model to predict subsequent response to therapy. MATERIALS/METHODS: We develop a mathematical model that describes patient-uniform tumor growth rate and initial pembrolizumab and bevacizumab responses, and patient-specific treatment response dynamics. A total of 154 radiology scans were delineated to derive longitudinal tumor volumes of 26 patients. In a leave-one-out study, patient-uniform model parameters are derived and then applied to the left-out patient to adaptively learn treatment response dynamics to forecast tumor volume evolution and progression risk. Model prediction performance is evaluated based on classification accuracy, including sensitivity and specificity. RESULTS: Patient-uniform glioma growth rate and initial treatment response rates are estimated to achieve acceptable fits to the longitudinal data of all untrained patients with R 2 = 0.81 [0.75, 0.89]. Response dynamics are predicted with high accuracy (0.78 [0.76, 0.79]), with positive and negative predictive values of 0.90 [0.86, 0.92] and 0.81 [0.78, 0.86], and sensitivity and specificity of 0.71 [0.69, 0.73] and 0.88 [0.83, 0.91] respectively. CONCLUSIONS: Two patient-specific parameters in a mathematical model can be adaptively learned from early tumor volume evolution to predict subsequent response to therapy and progression risk for individual patients. Future validation is required in an independent dataset and prospective evaluation in another clinical trial.