PMOA-TTS: PubMed Open Access Textual Time Series

📄 This repository accompanies the paper: PMOA-TTS: Introducing the PubMed Open Access Textual Times Series Corpus.

📚 Overview

PMOA-TTS is the first large-scale, openly available dataset of 124,699 PubMed Open Access (PMOA) case reports, each converted into structured (event, time) timelines via a scalable LLM-based pipeline.

Our approach combines heuristic filtering with Llama 3.3 to identify single-patient case reports, followed by prompt-driven extraction using Llama 3.3 and DeepSeek R1, resulting in over 5.6 million timestamped clinical events.

This repository provides the codebase for timeline extraction, forecasting experiments, survival analysis, and evaluation.

📂 Repository Structure

• make_tts/: Scripts and notebooks for constructing the PMOA-TTS dataset from raw case reports.
• forecasting_results/: Code and results for forecasting experiments on the textual time series data.
• survival_analyses/: Tools and analyses for survival prediction tasks using the extracted timelines.
• TTS_evaluation/: Evaluation scripts assessing the quality of the extracted timelines.

Each subdirectory contains its own README.md with detailed instructions and information.

🛠️ Getting Started

1. Clone the repository:

   bash
   git clone https://github.com/jcweiss2/pmoa_tts.git
   cd pmoa_tts
   

2. Set up the environment:

   Ensure you have Python 3.10 or higher installed. It's recommended to use a virtual environment.

   conda create -n pmoa_tts python 3.10 
   conda activate pmoa_tts 
   conda install pandas scikit-learn numpy tensorboard -c conda-forge 
   conda install pytorch torchvision torchaudio -c pytorch 
   pip install transformers 
   pip install argparse 
   conda install jupyter -c conda-forge 
   pip install sentencepiece
   

3. Access the PMOA-TTS dataset The dataset is available on Hugging Face. You can download it using the datasets library.

from datasets import load_dataset
dataset = load_dataset("snoroozi/pmoa-tts")

📄 Dataset Details

• Number of Records: Approximately 174,000 case reports
• Data Format: Parquet files, compatible with Hugging Face Datasets and pandas
• Languages: English
• License: CC BY-NC-SA 4.0
• Tags: clinical, time-series, biomedical, text
• Tasks Supported: Text Classification, Time Series Forecasting

Each data point includes:

• Textual Time Series: A sequence of timestamped clinical events
• Demographics: Age, sex, and ethnicity (when available)
• Diagnoses: Extracted diagnoses from the case report
• Death Information: Observed time and death event indicator

For more details, visit the Hugging Face dataset page.

📜 License

The dataset is derived from publicly available PubMed Open Access case reports.
All annotations and metadata are released under the CC BY-NC-SA 4.0 license.

🙏 Acknowledgments

This dataset was generated as part of research at the National Library of Medicine (NLM) at the National Institutes of Health (NIH) and Carnegie Mellon University (CMU).
We thank the PubMed Open Access initiative and the authors of the case reports that made this work possible.

📄 Citation

If you use PMOA-TTS in your research, please cite our preprints:

@article{noroozizadeh2025pmoa,
  title={PMOA-TTS: Introducing the PubMed Open Access Textual Times Series Corpus},
  author={Noroozizadeh, Shahriar and Kumar, Sayantan and Chen, George H. and Weiss, Jeremy C.},
  journal={arXiv preprint arXiv:2505.20323},
  year={2025}}

@article{noroozizadeh2025forecasting,
  title={Forecasting from Clinical Textual Time Series: Adaptations of the Encoder and Decoder Language Model Families},
  author={Noroozizadeh, Shahriar and Kumar, Sayantan and Weiss, Jeremy C},
  journal={arXiv preprint arXiv:2504.10340},
  year={2025}
}