BACKGROUND: Several prognostic scores have been developed to estimate survival of breast cancer (BC) patients with brain metastases (BM). Modified Breast Graded Prognostic Assessment (GPA), based on a single-institution cohort of 1552 patients, has been proposed as refinement of Breast-GPA. In addition to age, tumour subtype and KPS, Modified Breast-GPA comprises number of BM. This study was designed to validate Modified Breast-GPA. PATIENTS AND METHODS: Clinical data of 668 BC patients diagnosed with BM at four institutions between 1996 and 2016 were reviewed. Patients were classified by Breast-GPA and Modified Breast-GPA. Overall survival (OS) was calculated from time of BM diagnosis to death or last follow-up. Cox proportional models were used to calculate hazard-ratios and their 95% CI. The performances of Breast-GPA and Modified Breast-GPA were compared using Harrell's concordance index. RESULTS: Median age at BM diagnosis was 56 years (range 24-85). At last follow-up, 632 patients (94.6%) had died. Median OS was 8.1 months (95% CI 6.9-9.4). The number of BM (1-3 vs. >3) was significantly associated with OS in univariate analysis (p < 0.001) and having >3 BM was identified as a negative prognostic factor in multivariate analysis. Both Breast-GPA and Modified Breast-GPA accurately predicted OS (p < 0.001 for both scores). Performance of Modified Breast-GPA was better: concordance indices were 0.6390 (95% CI, 0.6381 to 0.6399) and 0.6647 (95% CI, 0.6639 to 0.6655) for Breast-GPA and Modified Breast-GPA, respectively (p < 0.001). CONCLUSIONS: This work provides the first external independent validation of Modified Breast-GPA and confirms its better performance as compared to Breast-GPA.