DIRECT COMPARISON OF LUNG-GPA AND LABBM PROGNOSTIC SCORES FOR PATIENTS WITH LUNG CANCER BRAIN METASTASES

Abstract BACKGROUND Brain metastases (BM) are the most common malignant CNS tumors in adults, associated with decreased survival and severe impairment to quality of life. The majority of BM originate from lung cancer (40-50%), where availability of new drug therapies calls for personalized risk-benefit calculations. While several prognostic tools were developed to support therapeutic decision-making in the specific context of BM detection, our experience in the clinic suggests none is currently used routinely. MATERIAL AND METHODS We compared the predictive capabilities of two different prognostic scores on a retrospective cohort of 90 lung cancer patients treated for BM at the Tel Aviv Sourasky Medical Center. Both scores were computed for each patient. Lung-GPA score is based on age, Karnofsy performance status (KPS), number of BM, number of extra-cranial metastases and molecular features when relevant (for NSCLC patients, EGFR/ALK mutations, PD-L1 expression). LabBM score is based on levels of hemoglobin, albumin, CRP, LDH and platelets in blood tests taken at BM diagnosis. RESULTS Median age at diagnosis was 67.1 (range 36-88), 43 patients were female, mean number of BM was four (range 1-10), 79 patients had NSCLC, of which 15 (18.9%) had confirmed EGFR/ALK mutations, and 11 had SCLC. Median survival was 6.8 mo (range 0.5mo-103mo). Survival rates at 3mo, 6mo, and 12mo were 73%, 51%, and 40%, respectively. Both scores were significantly predictive of OS at these times, and direct comparison between them using ROC analysis revealed no differences in predictive ability for survival at 3mo (0.7 for LabBM and 0.67 for Lung-GPA); at 6mo (0.7 for LabBM and 0.74 for Lung-GPA) and at 12mo (0.61 for LabBM and 0.71 for Lung-GPA). CONCLUSION In a real-world cohort of patients with lung cancer BM we find that Lung-GPA score and LabBM score both show a similar prognostic ability. Future studies should examine whether predictive scores could improve by use of patient data currently not captured by either score, such as treatment response or neuroimaging-based features.