README.md

---
dataset_info:
  features:
  - name: nct_id
    dtype: string
  - name: updated_at
    dtype: timestamp[us]
  - name: brief_title
    dtype: string
  - name: official_title
    dtype: string
  - name: acronym
    dtype: string
  - name: study_type
    dtype: string
  - name: overall_status
    dtype: string
  - name: study_first_submit_date
    dtype: timestamp[ms]
  - name: start_date
    dtype: timestamp[ms]
  - name: primary_completion_date
    dtype: timestamp[ms]
  - name: completion_date
    dtype: timestamp[ms]
  - name: phases
    sequence: string
  - name: enrollment_count
    dtype: float64
  - name: minimum_age
    dtype: float64
  - name: maximum_age
    dtype: float64
  - name: sex
    dtype: string
  - name: healthy_volunteers
    dtype: bool
  - name: brief_summary
    dtype: string
  - name: detailed_description
    dtype: string
  - name: eligibility_criteria
    dtype: string
  - name: lead_sponsor_name
    dtype: string
  - name: lead_sponsor_class
    dtype: string
  - name: org_study_id_info
    struct:
    - name: id
      dtype: string
    - name: link
      dtype: string
    - name: type
      dtype: string
  - name: why_stopped
    dtype: string
  - name: expanded_access_info
    struct:
    - name: hasExpandedAccess
      dtype: bool
    - name: nctId
      dtype: string
    - name: statusForNctId
      dtype: string
  - name: last_update_submit_qc_date
    dtype: timestamp[ms]
  - name: last_update_post_date_struct
    struct:
    - name: date
      dtype: string
    - name: type
      dtype: string
  - name: study_first_post_date_struct
    struct:
    - name: date
      dtype: string
    - name: type
      dtype: string
  - name: std_ages
    sequence: string
  - name: study_population
    dtype: string
  - name: sampling_method
    dtype: string
  - name: oversight_has_dmc
    dtype: bool
  - name: design_info
    struct:
    - name: allocation
      dtype: string
    - name: interventionModel
      dtype: string
    - name: interventionModelDescription
      dtype: string
    - name: maskingInfo
      struct:
      - name: masking
        dtype: string
      - name: maskingDescription
        dtype: string
      - name: whoMasked
        sequence: string
    - name: observationalModel
      dtype: string
    - name: primaryPurpose
      dtype: string
    - name: timePerspective
      dtype: string
  - name: conditions
    sequence: string
  - name: keywords
    dtype: string
  - name: interventions
    dtype: 'null'
  - name: locations
    list:
    - name: city
      dtype: string
    - name: country
      dtype: string
    - name: facility
      dtype: string
    - name: geoPoint
      struct:
      - name: lat
        dtype: float64
      - name: lon
        dtype: float64
    - name: state
      dtype: string
  - name: collaborators
    list:
    - name: class
      dtype: string
    - name: name
      dtype: string
  - name: arm_groups
    dtype: 'null'
  - name: outcomes
    struct:
    - name: other
      list:
      - name: description
        dtype: 'null'
      - name: measure
        dtype: string
      - name: timeFrame
        dtype: string
    - name: primary
      list:
      - name: description
        dtype: 'null'
      - name: measure
        dtype: string
      - name: timeFrame
        dtype: string
    - name: secondary
      list:
      - name: description
        dtype: 'null'
      - name: measure
        dtype: string
      - name: timeFrame
        dtype: string
  - name: overall_officials
    list:
    - name: affiliation
      dtype: string
    - name: name
      dtype: string
    - name: role
      dtype: string
  - name: study_references
    dtype: string
  - name: misc_info_module
    dtype: string
  - name: condition_browse_module
    struct:
    - name: ancestors
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
    - name: browseBranches
      list:
      - name: abbrev
        dtype: string
      - name: name
        dtype: string
    - name: browseLeaves
      list:
      - name: asFound
        dtype: string
      - name: id
        dtype: string
      - name: name
        dtype: string
      - name: relevance
        dtype: string
    - name: meshes
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
  - name: intervention_browse_module
    struct:
    - name: ancestors
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
    - name: browseBranches
      list:
      - name: abbrev
        dtype: string
      - name: name
        dtype: string
    - name: browseLeaves
      list:
      - name: asFound
        dtype: string
      - name: id
        dtype: string
      - name: name
        dtype: string
      - name: relevance
        dtype: string
    - name: meshes
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
  - name: mesh_terms
    struct:
    - name: conditions
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
    - name: interventions
      list:
      - name: id
        dtype: string
      - name: term
        dtype: string
  splits:
  - name: train
    num_bytes: 3939617129
    num_examples: 541897
  download_size: 1773167837
  dataset_size: 3939617129
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: apache-2.0
task_categories:
- question-answering
- feature-extraction
- text-classification
language:
- en
- fr
- es
tags:
- biology
- medical
- clinical
- trials
- scien
pretty_name: Clinical trials dataset
size_categories:
- 100K<n<1M
---


# Clinical Trials Dataset

A comprehensive dataset of clinical trials sourced from ClinicalTrials.gov, featuring structured metadata, detailed study information, and pre-computed semantic embeddings for machine learning applications in biomedical research.

## Dataset Description

This dataset provides a rich collection of clinical trial information systematically collected from the official ClinicalTrials.gov database. Each record contains detailed study metadata, eligibility criteria, intervention descriptions, outcome measures, and organizational information. The dataset is enhanced with semantic embeddings generated using specialized biomedical language models, making it immediately ready for downstream ML tasks.

### Key Features

- **Comprehensive Coverage**: Complete clinical trial records with 30+ structured fields
- **Rich Metadata**: Study phases, enrollment data, eligibility criteria, and outcome measures
- **Temporal Data**: Complete timeline information from submission to completion
- **Medical Ontology**: MeSH terms and condition/intervention classifications

## Dataset Statistics

| Metric | Value |
|--------|-------|
| Total Studies | 500,000+ |
| Data Source | ClinicalTrials.gov Official API |
| Languages | English |

## Data Collection Methodology

The dataset is built using a robust data pipeline that ensures data quality and consistency:

1. **API Integration**: Direct connection to ClinicalTrials.gov API v2
3. **Data Validation**: Pydantic models ensure schema compliance and data integrity
4. **Type Conversion**: Automatic parsing of dates, numbers, and JSON structures
5. **Embedding Generation**: Semantic embeddings computed using domain-specific models

### Data Quality Assurance

- **Schema Validation**: All records validated against comprehensive Pydantic models
- **Type Safety**: Proper data type conversion with null handling
- **Deduplication**: Unique constraint on NCT ID prevents duplicates
- **Temporal Consistency**: Date validation and chronological ordering
- **Text Normalization**: Whitespace cleanup and encoding standardization

## Schema Overview

### Core Study Information
- `nct_id`: Unique study identifier (NCT########)
- `brief_title`: Concise study title
- `official_title`: Complete formal study title
- `study_type`: Study design type (Interventional, Observational, etc.)
- `phases`: Clinical trial phases (Phase I, II, III, IV)
- `overall_status`: Current study status

### Study Design & Population
- `enrollment_count`: Target or actual enrollment number
- `minimum_age` / `maximum_age`: Age eligibility bounds
- `sex`: Gender eligibility (All, Male, Female)
- `healthy_volunteers`: Whether healthy volunteers accepted
- `eligibility_criteria`: Detailed inclusion/exclusion criteria
- `study_population`: Target population description

### Clinical Context
- `conditions`: Medical conditions studied
- `keywords`: Study-related keywords
- `brief_summary`: Study purpose and rationale
- `detailed_description`: Comprehensive study description
- `primary_outcomes` / `secondary_outcomes`: Measured endpoints

### Organizational Information
- `lead_sponsor`: Primary study sponsor
- `collaborators`: Additional supporting organizations
- `locations`: Study sites with geographic coordinates
- `overall_officials`: Principal investigators and study officials

### Temporal Data
- `study_first_submit_date`: Initial submission to ClinicalTrials.gov
- `start_date`: Study start date
- `primary_completion_date`: Primary endpoint completion
- `completion_date`: Overall study completion
- `last_update_submit_date`: Most recent data update

### Enhanced Features
- `brief_summary_embedding`: 768-dim semantic embedding of study summary
- `eligibility_criteria_embedding`: 768-dim embedding of eligibility text

## Usage Examples

### Basic Data Loading

```python
from datasets import load_dataset

# Load the complete dataset
dataset = load_dataset("louisbrulenaudet/clinical-trials")

# Access train split
train_data = dataset["train"]

print(f"Dataset size: {len(train_data)}")
print(f"Features: {train_data.features}")
```

### Advanced usage and embeddings

```python
from datasets import load_dataset
from sentence_transformers import SentenceTransformer

model = SentenceTransformer(
  "thomas-sounack/BioClinical-ModernBERT-base", device="cuda", model_kwargs={"torch_dtype": "float16"}
)

dataset = load_dataset("louisbrulenaudet/clinical-trials", split="train")

columns_to_embed = [
  "brief_summary",
  "eligibility_criteria"
]

def embed_texts(batch):
  for col in columns_to_embed:
    embeddings = model.encode(batch[col], convert_to_numpy=True, normalize_embeddings=True)
    batch[f"{col}_embedding"] = embeddings.tolist()
  return batch

embedded_dataset = hf_dataset.map(
  embed_texts,
  batched=True,
  batch_size=512,
  desc="Embedding columns"
)
```

## Applications

This dataset enables various research and development applications:

### Clinical Research
- **Study Landscape Analysis**: Understanding research trends and gaps
- **Protocol Optimization**: Learning from successful study designs
- **Site Selection**: Geographic analysis for multi-center trials
- **Regulatory Intelligence**: Phase progression and approval patterns

### Machine Learning
- **Text Classification**: Categorizing studies by therapeutic area
- **Information Extraction**: Parsing eligibility criteria and outcomes
- **Similarity Search**: Finding related studies using embeddings
- **Trend Prediction**: Forecasting research directions

### Healthcare Analytics
- **Disease Burden Analysis**: Understanding research investment by condition
- **Innovation Tracking**: Monitoring emerging therapies and interventions
- **Collaboration Networks**: Analyzing sponsor and investigator relationships
- **Geographic Health Mapping**: Regional research activity patterns

## Data Limitations & Considerations

### Coverage Limitations
- **Source Dependency**: Limited to studies registered on ClinicalTrials.gov
- **Registration Bias**: Not all studies worldwide are required to register
- **Temporal Scope**: Historical data quality varies by submission period
- **Language**: Primarily English-language studies

### Data Quality Notes
- **Self-Reported**: Information accuracy depends on sponsor reporting
- **Update Lag**: Some studies may have outdated status information
- **Completeness**: Optional fields may be sparse for older studies
- **Standardization**: Free-text fields may lack consistent formatting

### Ethical Considerations
- **Privacy**: No individual participant data included
- **Transparency**: Enhances clinical research transparency
- **Research Bias**: May reflect existing healthcare disparities
- **Access**: Promotes equitable access to clinical research information

## Licensing & Attribution

### Dataset License
This dataset is released under the MIT License, allowing for both academic and commercial use with proper attribution.

### Source Attribution
- **Primary Source**: ClinicalTrials.gov (https://clinicaltrials.gov/)
- **API**: ClinicalTrials.gov API v2

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

```bibtex
@dataset{louisbrulenaudet2025,
  title={Clinical Trials Dataset: Comprehensive ClinicalTrials.gov Data for Semantic analysis},
  author={[Louis Brulé Naudet]},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/louisbrulenaudet/clinical-trials}
}
```

*This dataset aims to accelerate biomedical research by providing easy access to comprehensive clinical trial information. We encourage responsible use that advances medical knowledge and improves patient outcomes.*

## Feedback

If you have any feedback, please reach out at [[email protected]](mailto:[email protected]).