joeljames270/virtual-patient-profiles-sample
VHS Patient Profiles — Nutrition Education Simulation Dataset Summary A curated collection of 8 richly structured synthetic patient personas designed for healthcare education simulations, with a primary focus on dietetics and nutrition counseling training. Each record represents a complete clinical case with demographic, clinical, psychosocial, and behavioral dimensions, along with structured guidance for educators and AI simulators. Profiles are intended for use… See the full description on the dataset page: https://huggingface.co/datasets/joeljames270/virtual-patient-profiles-sample.
VHS Patient Profiles — Nutrition Education Simulation
Dataset Summary
A curated collection of 8 richly structured synthetic patient personas designed for healthcare education simulations, with a primary focus on dietetics and nutrition counseling training. Each record represents a complete clinical case with demographic, clinical, psychosocial, and behavioral dimensions, along with structured guidance for educators and AI simulators.
Profiles are intended for use in virtual patient simulation platforms where students practice clinical nutrition competencies: applying the Nutrition Care Process (NCP), formulating PES statements, conducting motivational interviewing (MI), integrating medication management with nutrition care, and identifying interprofessional and community resources.
Supported Tasks
- Virtual patient simulation: Each profile supports a full simulated clinical encounter. The
information_disclosuresection defines what the patient volunteers, what must be elicited, what is reluctant, and what is hidden in the chart. - LLM fine-tuning / prompting: Profiles can serve as grounding documents for AI patient roleplay models.
- Curriculum design: The
learning_opportunitiessection maps each patient to specific pedagogical objectives.
Languages
English (all profiles). Two patients have noted bilingual backgrounds (Spanish/English, Hindi/English), which is reflected in the demographics.language field.
Dataset Structure
Data Instances
Each line of patients.jsonl is one self-contained JSON object representing a single patient. Example top-level keys:
{
"patient_overview": {...},
"demographics": {...},
"clinical_profile": {...},
"medication_profile": {...},
"nutrition_profile": {...},
"psychosocial_profile": {...},
"information_disclosure": {...},
"learning_opportunities": {...},
"metadata": {...}
}Data Fields
Data Splits
The dataset ships as a single split (train). All 8 records are in patients.jsonl.
Patient Roster
Dataset Creation
Curation Rationale
The profiles were generated to support a virtual human simulation platform for dietetics and health professions education. Real patient data was not used. Each persona was designed to represent a distinct cluster of clinical and psychosocial challenges that nutrition students commonly encounter, spanning age, ethnicity, health literacy, cultural context, and disease complexity.
Source Data
All profiles are synthetically generated. Clinical values (labs, vitals, medications) are constructed to be internally consistent and clinically plausible but do not represent real individuals.
Annotations
The learning_opportunities section serves as structured annotation for each case, mapping patient behaviors to educational objectives aligned with the Nutrition Care Process and ACEND competencies.
Personal and Sensitive Information
No real individuals are represented. Names, demographics, and clinical details are entirely fictional. The dataset does include synthetic sensitive health information (substance use, mental health, STI risk) within specific profiles, which is intentional for educational fidelity.
Considerations for Using the Data
Social Impact
These profiles are designed to foster cultural humility, empathy, and evidence-based practice in nutrition education. The diversity of patient backgrounds is intentional and meant to prepare students for real-world clinical populations.
Discussion of Biases
- Profiles reflect common clinical presentations and cultural contexts as understood at the time of creation. They may not cover the full range of lived experiences.
- AI models trained or prompted with these profiles should be evaluated for whether they reproduce or amplify stereotypes present in the synthetic personas.
Other Known Limitations
- 8 profiles is a small dataset. It is intended as a seed corpus for a simulation platform, not for large-scale statistical training.
- Some profiles contain minor inconsistencies introduced during generation (e.g., duplicate list entries, abbreviated field values in some records). These are noted in
metadata.assumptions.
Additional Information
Citation
If you use this dataset, please cite the VHS (Virtual Human Simulation) project at Rutgers University.