EBM+: Advancing Evidence-Based Medicine via two level automatic identification of Populations, Interventions, Outcomes in medical literature

This repository hosts the implementation described in our paper EBM+: Advancing Evidence-Based Medicine via two level automatic identification of Populations, Interventions, Outcomes in medical literature. This repository is based on the EBM-NLP as described in the publication A Corpus with Multi-Level Annotations of Patients, Interventions and Outcomes to Support Language Processing for Medical Literature, which we are extending by the following contributions. Hence, similarities to the original code and repository are expected.

In our work, we present

1. A neural end-to-end PICO Entity Recognizer that identifies Population, Intervention/Comparator and Outcome entities in medical publications.
2. Novel Neural Network architecture for Entity Recognition involving a self-attention mechanism, a 2D Convolution feature extraction from character embeddings and a Highway residual connection.
3. A PICO Statement classifier that identifies sentences containing all the PICO Entities and answering clinical questions.
4. A high quality, manually annotated by medical practitioners dataset for PICO Statement classification.

The PICO Statements Dataset can be found here.

Requirements/Setup

The code runs under Python 3.6 or higher. The required packages are listed in the requirements.txt, which can be directly installed from the file:

pip install -r /path/to/requirements.txt

ELMo Weights and options files should be downloaded from AllenNLP.

Citation

If you find our work interesting, please cite using the following:

Stylianou, Nikolaos, et al. "EBM+: Advancing Evidence-Based Medicine via two level automatic identification of Populations, Interventions, Outcomes in medical literature." Artificial Intelligence in Medicine 108 (2020): 101949.