Korea EHR Data | 2,436 Patients · 76 Hospitals · 1.09M OCR Pages | Raw PDF + Markdown + JSON for Medical AI Training

The provider’s Korea EHR Dataset is a patient-consented, multi-institution collection of real hospital records from South Korea - the only Korean clinical corpus of this depth that can be licensed for commercial AI training down to the raw source document. WHAT IS IN THE DATASET - 618,294 structured clinical records from 2,436 unique patients - 1,094,479 raw medical-document pages (OCR) and 172,294 longitudinal clinical events - 76 tertiary and general hospitals across 12 Korean regions, including Seoul Asan, Samsung Seoul, Seoul National University Hospital, Severance and Seoul St. Mary's - Mean observation window 10.2 years (median 6.9); 58.5% of patients span 5 years or more, and the longest patient timeline runs over 50 years - 33.2% of patients hold records from two or more hospitals (up to 18 institutions), genuine patient-side longitudinality rather than a single-site extract DOCUMENT TYPES Prescriptions 23.0%, outpatient notes 22.1%, lab results 17.8%, inpatient progress notes 15.1%, imaging reads 9.3%, surgery and procedure records 1.8%, discharge summaries 1.7%, emergency 1.7%, admission 1.5%, vital signs 1.2%, pathology 1.1%. COHORT COMPOSITION Cancer 32.7% (breast, gynecologic, hepatobiliary and pancreatic, thyroid, prostate, hematologic, lung, urologic, colorectal, gastric and other sites), digestive 10.1%, neurological 7.1%, rare and intractable disease 6.8% across 29 conditions, musculoskeletal 6.2%, cardiovascular 5.8%, women's health 5.7%, metabolic and endocrine 5