README.md

---
license: mit
task_categories:
- question-answering
- multiple-choice
language:
- en
tags:
- medical
- healthcare
- ehr
- diagnosis
- medication
- clinical
size_categories:
- 10K<n<100K
---

# Medical Question Answering Dataset (QA Pairs MVD 10K)

## Dataset Description

This dataset contains medical question-answering tasks based on Electronic Health Record (EHR) data. The dataset focuses on three main prediction tasks in clinical settings:

1. **Missing Medication MCQ**: Predicting which medication should be added to a patient's current regimen
2. **Next Diagnosis MCQ**: Predicting the most likely future diagnosis for a patient  
3. **Next Measurement Value MCQ**: Predicting future laboratory or vital sign values

## Dataset Structure

### Data Splits

| Split | Examples |
|-------|----------|
| Train | 17,158 |
| Validation | 3,754 |
| Test | 3,683 |
| **Total** | **24,595** |

### Data Fields

- `prompt`: The full question prompt including patient medical history
- `prompt_type`: Type of question (missing_medication_mcq, next_diagnosis_mcq, next_measurement_value_mcq)
- `choices`: List of multiple choice options (typically 5 options A-E)
- `answer_idx`: Index of the correct answer (0-based)
- `completion`: The correct answer choice letter (A, B, C, D, or E)
- `id`: Unique identifier for each example

### Example

```json
{
  "prompt": "You are an assistant tasked with analyzing medical histories to determine which medication is missing from the patient's current regimen....",
  "prompt_type": "missing_medication_mcq",
  "choices": ["Cisplatin 50 MG Injection", "Tacrine 10 MG Oral Capsule", ...],
  "answer_idx": 4,
  "completion": "E",
  "id": "2543984390693637980missing_medication_mcq"
}
```

## Task Types

### 1. Missing Medication MCQ
Analyzes a patient's medical history including demographics, visits, measurements, procedures, and current medications to predict which medication should be added to their regimen.

### 2. Next Diagnosis MCQ  
Predicts the most likely future diagnosis based on a patient's medical trajectory and history.

### 3. Next Measurement Value MCQ
Predicts future laboratory values or vital signs based on historical measurement trends.

## Patient Data Structure

Each prompt includes structured patient data with:
- **Demographics**: Race, gender, year of birth
- **Visit History**: Outpatient visits, ER visits with dates
- **Measurements**: Height, weight, BMI, blood pressure, lab values with timestamps
- **Procedures**: Medical procedures performed with dates
- **Medications**: Current and past medications with start/end dates
- **Diagnoses**: Prior medical conditions with dates

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("your_username/qa-pairs-mvd-10k")

# Access different splits
train_data = dataset['train']
val_data = dataset['validation'] 
test_data = dataset['test']

# Example usage
example = train_data[0]
print(f"Question type: {example['prompt_type']}")
print(f"Prompt: {example['prompt'][:200]}...")
print(f"Choices: {example['choices']}")
print(f"Correct answer: {example['completion']}")
```

## Ethical Considerations

This dataset contains synthetic or anonymized medical data. Users should:
- Ensure compliance with healthcare data regulations (HIPAA, etc.)
- Use the dataset responsibly for research and educational purposes
- Not use for actual medical diagnosis without proper validation
- Consider potential biases in the synthetic data generation process

## Citation

If you use this dataset in your research, please cite:

```bibtex
@dataset{qa_pairs_mvd_10k,
  title={Medical Question Answering Dataset (QA Pairs MVD 10K)},
  year={2024},
  url={https://huggingface.co/datasets/your_username/qa-pairs-mvd-10k}
}
```

## License

This dataset is released under the MIT License.