vickyvamsi22/Clinical_Notes_benchmark_2025

Clinical Notes Dataset (Multidomain, De-identified)

Dataset Overview

This dataset consists of de-identified clinical notes written in formal medical language, covering a wide range of healthcare domains including general medicine, pediatrics, gynecology, cardiology, neurology, endocrinology, respiratory, gastrointestinal, musculoskeletal, and dermatological contexts.

The notes are structured to reflect realistic outpatient, follow-up, telemedicine, and review-style clinical encounters, while avoiding any personally identifiable information or treatment directives.

Data Characteristics

• —Fully de-identified textual clinical narratives
• —No personal identifiers (names, ages, locations, IDs)
• —No explicit diagnoses or medication prescriptions
• —Medically plausible symptom descriptions and clinical reasoning
• —Consistent internal logic between symptoms, risk level, and duration

Dataset Structure

Each record is provided in JSON format with the following fields:

• —doctor_note: Detailed clinical narrative written in professional medical language
• —visit_context: Type of healthcare encounter
• —key_clinical_findings: Semicolon-separated clinical observations
• —risk_level: Clinical risk assessment (Low / Moderate / High)
• —symptom_duration: Temporal pattern of symptoms
• —reference_explanation_en: Medically detailed explanatory text for trained readers
• —source_reference: Public medical knowledge source referenced for context

Intended Use

This dataset is intended for:

• —Clinical natural language processing research
• —Medical text understanding and generation tasks
• —Risk stratification modeling
• —Educational and benchmarking purposes

The dataset is not intended for clinical decision-making or patient care.

Ethical Considerations

All records are constructed to ensure privacy preservation and compliance with ethical guidelines for medical data sharing. The dataset does not contain protected health information (PHI) and does not correspond to identifiable individuals.

Limitations

• —No diagnostic labels are included
• —No therapeutic recommendations are present
• —Notes are standardized for operational healthcare

License

CC BY 4.0