OBJECTIVES: FDA accelerated approvals for some chemotherapies used PFS as an OS surrogate. However, one mBC approval was revoked after OS gains were not observed post-approval. We compared PFS with OS by first-line (1L) chemotherapy using real world evidence from US mBC patients. METHODS: IMS Health Comprehensive Disease Records for Breast Cancer link IMS PharMetrics Plus, oncology EMRs, and Social Security Death Index. Female breast cancer patients (ICD-9-CM 174.x, 233.0) were selected with index metastasis (ICD-9-CM 196.x-198.x or EMR Stage IV) 7/1/2006-3/31/2012, age ≥18, no HR overexpression, 1L chemotherapy (anthracyclines or txanes [AT] or new cytotoxic agents [NCA]), enrollment 180 days pre-index and ≥30 days post-index, and no other malignancies. PFS proxy was 1L duration; OS was measured until death or censoring (end of enrollment). Log-rank analysis compared PFS by 1L treatment. A Cox proportional hazards model assessed post-1L OS by 1L chemotherapy and duration, propensity for NCA vs. AT, HER2 status, and patient characteristics. RESULTS: Of 845 mBC patients, 334 met study criteria (mean [SD] age=51.7 [8.7] years, 20.4% HER2+). Propensity for NCA (n=70) vs. AT (n=264) increased with diabetes history (OR=2.79, 95% CI 1.12-6.90) and higher pre-index healthcare expenditures (OR=1.27, 95% CI 1.05-1.54). 1L NCAs were administered a mean (median) 195.2 (73.0) days (anthracyclines 53.5 [44.0], taxanes 122.2 [47.5], -2Log LR 54.3, 2 df, p<.0001). Cox regression estimated that 30 additional days of 1L chemotherapy predicted slightly shorter post-1L OS (HR=1.03, 95% CI 1.00-1.06); NCA (vs. AT) predicted stronger decreases in post-1L OS (HR=2.32, 95% CI 1.11-4.86) despite propensity adjustment. CONCLUSIONS: Duration of 1L chemotherapy, as a proxy for PFS, predicted slightly shorter post-1L OS. Choice of 1L chemotherapy was a stronger predictor of post-1L OS, adjusted for selection bias. Real world data suggests that PFS is a poor OS surrogate in mBC.