Dataset Overview
Definition：Structured Electronic Medical Record Text Datasetis provided byChangsha Langhui Information Technology Co., Ltd.50 million authentic de-identified electronic medical records obtained through the compliant transfer authorization channel. This is currently the largest on the platform.Maximum amount of dataThe datasets cover all types of hospital documents, including chief complaint, present illness history, past medical history, admission, discharge, surgical records, and medical records.Direct Authorization from Three Grade-A Tertiary HospitalsandCity-Level Comprehensive Authorization Documents。
| Dataset Name | Structured Electronic Medical Record Text Dataset |
| Total Data Volume | 50 Million Cases Electronic medical records (largest platform-wide) |
| Data Type | De-identified text |
| Data Source | Resale (partnerships with more than 50 hospitals, including Fudan Top-100 hospitals) |
| License Type | Direct authorization from the hospital. Direct Authorization from Three Grade-A Tertiary Hospitals + City-Level Comprehensive Authorization Documents |
| Delivery Cycle | 2 weeks - 1 month |
| Document Type | Chief complaint, present illness history, past medical history, admission, discharge, surgical records, medical records |
| De-identification | completed Removal of personal identity information (name/ID number/telephone/relationship) |
| Department Coverage | General departments of medicine, surgery, gynecology and obstetrics, pediatrics, orthopedics, neurology, oncology, cardiology, respiratory, digestive, etc. |
Core Data Fields
The seven core categories of documents covering the full cycle of hospitalization, each containing rich clinical structured fields.
| Document Type | Typical Fields |
|---|---|
| Principal | Description of the main symptoms, duration, causes of visit |
| Present Illness | Incidence, evolution of symptoms, medical treatment, accompanying symptoms, general conditions |
| From the past. | History of disease, history of surgery, history of allergies, history of blood transfusions, history of infectious diseases, history of vaccinations |
| Admission note | Principal, current medical history, past history, medical examination, auxiliary examination, preliminary diagnosis, medical programme |
| Discharge summary | Date of admission, date of discharge, admission diagnosis, discharge diagnosis, medical treatment, discharge |
| Operative note | Date of operation, name of operation, anaesthesia, procedure, condition of operation and treatment |
| Progress note | Daily medical records, ward round notes, consultation notes, phase summaries, rescue records |
Data Sample Showcase
The following is a sample of the stylized electronic patient record text after de-identification, showing the typical format and content of the six core instruments.
{ "anon_id": "EMR-2023-PM-00048291", "record_type": "Admission note", "admission_date": "2023-03-15", "department": "Cardiology", "demographics": { "gender": "male", "age_group": "60-70sample", "admission_type": "sample value" }, "chief_complaint": "sample1sample，sample3sample。", "present_illness": "sample1sample，sample3-5sample，sample，sample。3sample，CNdaily3-4sample，sample、sample，sample，sample。sample，sample"sample unstable angina"sample。", "past_history": { "hypertension": {"duration": "8sample", "treatment": "Nifedipine controlled-release tablets30mg qd"}, "diabetes": {"duration": "3sample", "treatment": "sample0.5g tid"}, "surgery": "sample value", "allergy": "sample value" }, "physical_exam": { "bp_mmhg": "145/90", "hr_bpm": 88, "resp_rate": 20, "temp_c": 36.5 }, "initial_diagnosis": "coronary atherosclerotic heart disease unstable angina", "treatment_plan": "sample、sample、sample、CNCTACN，sample、sample、sample。"
}AI Application Scenarios
CBSS
The CDSS model, based on 50 million real medical records, provides clinical decision support functions such as diagnostic identification, treatment plan advice, pharmacovigilance, etc.
auto-generated medical records
Training in medical text generation models, which are based on basic patient information, automatic generation of admission records, medical records, summary records of discharges, etc., and the production of structured medical records, which significantly improve the effectiveness of doctors' work.
Disease prediction and early screening
Training in disease prediction models using a history of mass disease to achieve precision predictions such as early disease warning, risk prediction for complications, risk assessment for re-hospitalization.
Medical Knowledge Graph Construction
The physical relationship of the disease-symptomology-therapeutic-drugs is drawn from 50 million medical records, and a large-scale medical knowledge map is built to inject structured knowledge into the medical AI.
Medical foundation model pre-training
The 50 million-grade medical text material for continuous pre-training in the medical field (CPT) and supervised fine-tuning (SFT) for base models such as GPT/Llama/Qwen, and the development of a large model for specialized medical care.
Medical records quality control and DRG/DIP
Trained in medical history quality control AI, auto-detecting medical history writing deficiencies; supporting medical history coding and compliance review under the payment of DRG/DIP.
Project value
50 million records for large-scale medical NLP research
Fifty million cases of volume are in absolute leadership in the current medical NLP training data market. Fifty million grade text language versions are decisive for pre-training and fine-tuning large language models — with each quantitative increase in the volume of data, the model's performance in the field of health can lead to a qualitative leap.
End-to-end coverage across seven clinical document types
The document is fully covered for each link from admission to discharge, ensuring that the AI model understands the full clinical process. A single data set supports the full life cycle of AI training in hospitalization without having to combine multiple sources of data.
Data governance: de-identification and documented authorization
100% de-identification processes remove all personal identity information, plus three direct authorization and city-level authorization documents for tertiary hospitals. In an increasingly stringent regulatory environment for medical data compliance, this depth of compliance safeguards is a central competitive barrier for AI enterprises.
Specialized data for medical language models
High-quality domain data are at the core of differential competition in the immediate context of increased competition for generic megamodel homogenization. Fifty million medical texts can significantly enhance the medical specialization of the large model and the strategic data asset for building large medical-specific models.
Frequently asked questions
Build your medical model? Starting with 50 million medical records.
Changsha Langhui Information Technology Co., Ltd. DataAssetsAPI platform provides you with a one-stop data solution from big medical text to structured knowledge.
Data sample preview • Section customization programme • Compliance review support