Linac-Based Stereotactic Radiosurgery for Brain Metastases: Survival and Salvage Treatment in a Contemporary Cohort of 3,893 Patients

Abstract Background Real-world outcomes after stereotactic radiosurgery (SRS) for brain metastases across the full burden spectrum are limited. We tested whether overall survival (OS) in higher-burden lesion groups is non-inferior to a 5–10 lesion reference and characterized salvage patterns including whole-brain radiotherapy (WBRT). Methods 3,893 consecutive patients received linac-based SRS (1–5 fractions) at seven sites of an NCI-designated cancer center network (2017–2026), stratified by lesion count (1, 2–4, 5–10, 11–20, >20). Multivariable Cox regression used 5–10 as reference; pre-specified non-inferiority testing (margin HR 1.30, one-sided α = 0.025) compared higher-burden groups. Cumulative incidences of salvage interventions used death as the competing event. Results Median age 65 years; median follow-up 41.2 months. 77% had 1–4, 17% had 5–10, and 6% had >10 lesions (maximum 140). Median OS was 16.2, 11.0, 8.2, 8.7, and 6.5 months for the 1, 2–4, 5–10, 11–20, and >20 groups. OS was significantly better for 1 (HR 0.62, 0.55–0.69) and 2–4 (HR 0.81, 0.73–0.90) versus 5–10; non-inferiority was met for 11–20 (p < 0.001) but not >20 (n = 41, p = 0.20). CNS-directed systemic therapy was strongly protective (HR 0.62, 0.57–0.68); 21.7% of patients with ≤20 lesions survived ≥30 months. 12-month salvage WBRT incidence was 2.9–6.3%; 24-month incidence was 6.7% versus 41.7% for a third SRS course. Conclusions OS plateaued from 5 through 20 lesions, with non-inferiority to the 5–10 reference established for 11–20; the >20 group (n = 41) was underpowered and showed a directionally worse estimate that could not exclude a clinically meaningful decrement. Salvage WBRT was uncommon across all lesion-count groups (24-month cumulative incidence ≤7.3%; >90% of patients received no salvage WBRT), even after multiple SRS courses. These data support an SRS-first strategy guided by clinical factors rather than fixed lesion-count thresholds.