Integration of Next Generation Sequencing Data to Inform Survival Prediction of Patients with Spine Metastasis

Background/Objectives: Spinal metastatic disease is a life-altering problem for individuals with cancer. Prognostication is key for tailored treatment of spinal metastases. This manuscript provides a comprehensive overview of the genomic profiles of metastatic spine tumors and investigates the potential of mutational data to stratify overall survival (OS) across various histologies. Methods: This is a cohort study of consecutive patients with spine metastatic disease whose tumors were sequenced on a next generation sequencing platform; a machine learning (ML) algorithm was used to stratify OS risk. Results: Targeted sequencing and stratification of OS risk of 282 spine metastases (breast (84), non-small cell lung (56), prostate (49), other (93)) was performed. TP53 (HR 1.80; 95% CI 1.26, 2.56) and KEAP1 (HR 3.95, 95% CI 2.24, 6.98) mutations were associated with poor survival across the entire cohort in univariate Cox proportional hazards models. The ML algorithm categorized breast cancer metastasis into low- and high-risk groups, revealing a median OS of 71 compared to 22 months (HR 3.3, p < 0.001). TP53 mutations and ESR1 mutations conferred poor prognosis. In lung cancer, low- and high-risk groups with median OS of 30 and 6 months (HR 8.3, p < 0.001), respectively, were identified with poor prognosis linked to MET amplification. No significant prognostic associations were identified for spinal prostate metastases. Conclusions: Metastatic spine tumor molecular data allows for the identification of prognostic groups. We present an open-source machine learning algorithm utilizing genomic mutational data that may aid in prognostication and tailored decision making.