TrialPanorama/TrialPanorama-database

---
license: apache-2.0
language:
- en
tags:
- medical
size_categories:
- 10M<n<100M

config_names:
- studies
- conditions
- drugs
- disposition
- outcomes
- results
- biomarkers
- endpoints
- relations
- drug_moa
configs:
- config_name: studies
data_files:
- split: all
path: studies*.parquet
- config_name: conditions
data_files:
- split: all
path: conditions*.parquet
- config_name: drugs
data_files:
- split: all
path: drugs*.parquet
- config_name: disposition
data_files:
- split: all
path: disposition*.parquet
- config_name: outcomes
data_files:
- split: all
path: outcomes*.parquet
- config_name: adverse_events
data_files:
- split: all
path: adverse_events*.parquet
- config_name: results
data_files:
- split: all
path: results*.parquet
- config_name: biomarkers
data_files:
- split: all
path: biomarkers*.parquet
- config_name: endpoints
data_files:
- split: all
path: endpoints*.parquet
- config_name: relations
data_files:
- split: all
path: relations.parquet
- config_name: drug_moa
data_files:
- split: all
path: drug_moa.parquet
---

## Quick start

The easiest way to download the dataset to your local is to use `huggingface-cli`. The specific command you can use is

```
huggingface-cli download zifeng-ai/TrialPanorama-database --local-dir LOCAL_DIR --repo-type dataset
```
where `LOCAL_DIR` should be replaced with the target directory you want to save your dataset to.

# Update history

- **Aug.4 2025**: updated tables with the full set of studies

Dataset website: https://ryanwangzf.github.io/projects/trialpanorama

---

# TrialPanorama

**TrialPanorama** is a large-scale, structured database and benchmark designed to support AI-driven tasks in clinical trial workflows—including systematic review and trial design. This repo hosts the database of clinical trials. To conduct benchmarking experiments on trial design and systematic review asks, check this dataset instead: https://huggingface.co/datasets/zifeng-ai/TrialPanorama-benchmark

## Dataset Overview

* Aggregates over 1M clinical trial records from ClinicalTrials and PubMed
* Captures standardized elements such as trial setups, interventions, conditions, biomarkers, outcomes, and links to biomedical ontologies (e.g. DrugBank, MedDRA).
* Structured into multiple conceptual clusters and tables (e.g. trial-level attributes, protocol design, results, links)

## Benchmark Suite

The dataset supports a suite of **8 benchmark tasks** across two domains:

* **Systematic Review**:

* Study search
* Study screening
* Evidence summarization

* **Trial Design**:

* Arm design
* Eligibility criteria
* Endpoint selection
* Sample size estimation
* Trial completion assessment

## Data Schema (Major Tables)

* **`studies`** — core metadata (e.g. study\_id, intervention type, sponsor type, start year, trial phase, recruitment status)

* **`protocols`**, **`interventions`**, **`conditions`**, **`biomarkers`**, **`outcomes`**, **`results`** — each standardized to a unified schema

## &#x20;Use Cases

* Developing and evaluating LLMs and ML models for clinical trial-related tasks
* Benchmarking AI capabilities in trial search, screening, design, and summarization
* Conducting meta-analyses and exploring evidence synthesis across disease areas
* Enabling interoperability with biomedical ontologies to support richer clinical trial reasoning

## Getting Started

## &#x20;Citation

If you use TrialPanorama, we appreciate your citation:

```
@article{wang2025trialpanorama,
title={TrialPanorama: Database and Benchmark for Systematic Review and Design of Clinical Trials},
author={Wang, Zifeng and Jin, Qiao and Lin, Jiacheng and Gao, Junyi and Pradeepkumar, Jathurshan and Jiang, Pengcheng and Danek, Benjamin and Lu, Zhiyong and Sun, Jimeng},
journal={arXiv preprint arXiv:2505.16097},
year={2025}
}
```