---
license: apache-2.0
task_categories:
- question-answering
- text-generation
language:
- en
tags:
- medical
- ehr
- synthetic
- dspy
- gepa
size_categories:
- n<1K
dataset_info:
  features:
  - name: query
    dtype: string
  - name: expected_answer
    dtype: string
  - name: expected_strategy
    dtype: string
  - name: expected_snomed_codes
    list: string
  - name: expected_demographics
    struct:
    - name: age_filter
      dtype: string
    - name: filter
      dtype: string
    - name: gender_filter
      dtype: string
  - name: expected_neo4j_count
    dtype: int64
  - name: query_complexity
    dtype: string
  - name: medical_category
    dtype: string
  - name: expected_excluded_conditions
    list: 'null'
  splits:
  - name: train
    num_bytes: 35562
    num_examples: 210
  - name: validation
    num_bytes: 8680
    num_examples: 45
  - name: test
    num_bytes: 8629
    num_examples: 46
  download_size: 28142
  dataset_size: 52871
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

# Medical EHR Training Dataset

Training dataset for Medical EHR GEPA-optimized module.

## Dataset Description

This dataset contains **382 medical EHR query examples** for training DSPy GEPA optimization.

### Dataset Structure

```
{
  "query": "Show me diabetic patients",
  "expected_strategy": "ENRICHMENT",
  "expected_snomed_codes": ["73211009", "44054006"],
  "expected_neo4j_count": 15,
  "query_complexity": "simple",
  "medical_category": "endocrine"
}
```

### Splits

| Split | Examples |
|-------|----------|
| train | 267 |
| validation | 57 |
| test | 58 |
| **Total** | **382** |

### Complexity Distribution

- **Simple**: 58% (basic demographic or single-condition queries)
- **Moderate**: 40% (multi-filter or combined queries)
- **Complex**: 2% (multi-condition with exclusions and lab filters)

### Medical Categories

17 categories covering:
- Demographic filters (38%)
- Cardiovascular (12%)
- Neurological (11%)
- Respiratory (8%)
- Endocrine (8%)
- And 12 more...

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("dafesmi/medical-ehr-training-data")

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

print(f"Training examples: {len(train_data)}")
```

## License

Apache 2.0