The Cerebrospinal System Failure Score (CSF-Score): A Multidimensional Prognostic Tool for Leptomeningeal Metastasis from Non-Small Cell Lung Cancer

BACKGROUND Leptomeningeal metastasis (LM) from non-small cell lung cancer (NSCLC) has entered a new therapeutic era, yet no validated composite prognostic score jointly captures functional, molecular, anatomical, and cerebrospinal fluid (CSF) dynamic dimensions. We developed and internally validated the Cerebrospinal System Failure Score (CSF-Score), a pre-specified five-dimensional prognostic score for NSCLC-LM. METHODS We retrospectively analyzed 123 consecutive patients with cytology-confirmed NSCLC-LM between January 2022 and June 2025. The CSF-Score (0–10 points) was constructed a priori and integrates host resilience, molecular engine, anatomical burden, fluid dynamics, and biological toxicity, with a Karnofsky Performance Status <60 gatekeeper directing patients to the high-risk stratum. Patients were classified as low- (0–2), intermediate- (3–6), or high-risk (≥7 or gatekeeper). Discrimination was assessed using Harrell’s C-index with 1,000-resample bootstrap internal validation and temporal split validation. RESULTS Median OS was 20.5 months (95% CI 16.6–23.9), 12.6 months (95% CI 8.9–15.8), and 4.5 months (95% CI 3.2–5.7) in the low- (n = 32), intermediate- (n = 49), and high-risk (n = 42) groups, respectively (log-rank p < 0.001). The C-index was 0.741 (95% CI 0.705–0.779), with negligible optimism after bootstrap correction and stable temporal split validation (training 0.742 vs testing 0.733). Four of five dimensions retained independent significance on multivariable analysis; fluid dynamics showed a mediation-consistent attenuation pattern. The total score remained independently associated with OS after adjustment for treatment covariates (HR 1.28 per point, p < 0.001). CONCLUSIONS The pre-specified CSF-Score achieved clear three-tier prognostic stratification in NSCLC-LM while incorporating dimensions relevant to CSF system status. External multicenter validation is required before clinical adoption.