Data-driven precision prognostic stratification in breast cancer leptomeningeal metastasis: The BC-LGPA model and treatment refinement.

1114 Background: Breast cancer leptomeningeal metastasis (BCLM) lacks data-driven prognostic models integrating diagnostic certainty with molecular features, resulting in empirical treatment limitations. Methods: We analyzed 403 BCLM patients (Nov 2013-Nov 2025) from four centers, constructing the Breast Cancer Leptomeningeal Graded Prognostic Assessment (BC-LGPA) based on diagnostic criteria (CSF cytology, symptoms, MRI patterns) and molecular subtype, validated externally. Results: The median overall survival (OS) was 7.5 months. As the first data-driven prognostic classification for BCLM, the BC-LGPA integrates diagnostic certainty scores (0–2) and molecular scores (0–1). ECOG performance status and extracranial metastases were excluded owing to collinearity with neurological symptoms and because leptomeningeal metastasis itself constitutes the primary survival-limiting factor. The model stratifies patients into high-risk (0 points, 4.7 months), intermediate-risk (1–2 points, 8.0 months), and low-risk (3 points, 16.6 months) groups (training p=0.001; validation: 4.1 vs. 7.9 vs. 16.0 months, p=0.003). Key innovations include: (1) Granular diagnostic stratification into Class A (cytology-proven), B (clinical-radiological), and C (incidental-radiological), with subclasses A1/A2/A3, B1/B2, and C1/C2/C-special; (2) Ultra-indolent entity identification: Class C-special (localized calvarial-meningeal disease without diffuse LM) shows exceptional median OS of 68.0 months, representing a distinct clinical entity unsuitable for conventional prognostic models; (3) Treatment refinement: Intrathecal therapy improved survival in Class A and Class B-Linear subgroups, whereas focal radiotherapy demonstrated no independent benefit, including in Class C2 (31 patients, p =0.794), challenging empirical RT practices. Conclusions: BC-LGPA establishes the first data-driven BCLM prognostic framework, correcting treatment recommendations through subgroup efficacy analyses to guide precision management and clinical trial design. The scoring criteria of the BC-LGPA. Prognostic Factor Score Criteria Definition Diagnostic Certainty CSF cytology Neurological symptoms MRI Class A(Cytology- proven) 0 + + / - + / - Class B(Clinico-radiological) 1 - / NA + + Class C*(Incidental radiological) 2 - / NA - + Molecular Subtype TNBC 0 ER - PR - HER2 - non-TNBC 1 ER/PR + and/or HER2 + Total score = Diagnostic Certainty score + Molecular Subtype score NA: Not Available; * Localized calvarial-meningeal disease were excluded.