OBJECTIVES:  The promise of electronic health records (EHR) for population health analytics has not been fulfilled in the US partially due to a lack of interoperability across different systems. Distributed data networks (DDN) represent one approach to addressing this limitation, but require complex governance and coordination among participants. We describe an alternative based on a large-scale, centralized, and physically-integrated EHR data. METHODS:  Patient-level data from multiple different electronic medical records systems were staged and standardized within multiple health systems to create Optum’s EHR Database, (Optum, Eden Prairie, MN). Patient identifiers were de-duplicated and clinical data mapped across information systems before being de-identified. We assessed completeness of select variables among 3 groups: (1) All adults (≥ 18 years), (2) Adults with a chronic disease diagnosis, and (3) Adults who have ≥ 12 months of data preceding index diagnosis that includes at least one prescription order, one laboratory result, and one encounter with a general practitioner. RESULTS:  From 2007-2016, there were data relating to 69 million patients (58% women) from 52 health systems. There were, on average, 3.2 record identifiers from different data sources within or across health systems for each unique patient. There were 4 million adults in Group 1 (36% 18-39 years, 34% 40-59 years, and 30% 60+ years). There were 1.9 million patients in Group 2, and 670,926 in Group 3. The median follow-up span was respectively 26, 48, and 56 months. The median number of medical encounters was 2 in Group 1 and 10 in Group 3. In Group 3, BMI, blood pressure, complete blood count, and basic metabolic panel information was available for ≥ 95% of patients. CONCLUSIONS:  A physically integrated EHR system that spans multiple healthcare systems feasibly overcomes issues of interoperability. Advantages for linkage to claims and mining of free-text notes will be discussed.