Social Determinants of Health Prediction

A model that predicts Social Determinants of Health for the paper "Integration of Large Language Models and Traditional Deep Learning for Social Determinants of Health Prediction" (see arXiv).

Documentation

See the full documentation. The API reference is also available.

Setup

Create a new Conda environment: conda env create -f src/python/environment-lock.yml
Activate it: activate sdoh
Download the corpus

Rename the corpus:

cd download
mv mimic-iii-clinical-care-database-1.0.1.zip annotation-dataset-of-social-determinants-of-health-from-mimic-iii-clinical-care-database-1.0.1.zip

Installing Lituiev et al. (2023) SDoH Models

Optionally install the Lituiev et al. (2023) SDoH Models.

spaCy hybrid model

Clone the spaCy package: git clone https://github.com/BCHSI/social-determinants-of-health-clbp
Change working directory: cd social-determinants-of-health-clbp/model-hybrid-bow/package/en_sdoh_bow-0.0.2
Build the wheel: pip install wheel ; python setup.py bdist_wheel
Install the wheel: pip install --no-deps dist/en_sdoh_bow-0.0.2-py3-none-any.whl
Install dependencies: pip install lemma_tokenizer
Install NLTK's dependencies: pip install nltk ; python -c "import nltk ; nltk.download('stopwords') ; nltk.download('averaged_perceptron_tagger') ; nltk.download('wordnet')"

spaCy CNN Model

Clone the model repo: git clone https://github.com/BCHSI/social-determinants-of-health-clbp
Change working directory: cd social-determinants-of-health-clbp/model-cnn-ner/packages/en_sdoh_cnn_ner_cui-0.0.0
Build the wheel: pip install wheel ; python setup.py bdist_wheel
Install the wheel: ( cd dist ; pip install en_sdoh_cnn_ner_cui-0.0.0-py3-none-any.whl )

Convert the spaCy model to PyTorch

The HuggingFace 2023 model is for an old version of spaCy (3.2). This converts it to a HuggingFace model from a spaCy source model. It needs a previous version of pip, so install an old version, install the spaCy model, then restore pip. Then the PyTorch model is converted to sdoh-roberta-base.

Remember the old version: OLDVER=$(pip --version | awk '{print $2'})
Compatible pip version for package: pip install --upgrade pip==23.
Install Git Large File System: brew install git-lfs
Download the model: git clone https://huggingface.co/dlituiev/en_sdoh_roberta_cui
Install it: pip install --no-deps en_sdoh_roberta_cui/en_sdoh_roberta_cui-any-py3-none-any.whl
Install dependencies: pip install spacy-transformers
The conversion script needs an older HF package: pip install transformers==4.26
Convert to the PyTorch model: src/bin/topytorch.py
Revert the pip version: pip install --upgrade pip==${OLDVER}
Cleanup: rm -rf en_sdoh_roberta_cui

Training

Set the path to the configuration file:
```
export SDOHRC=etc/model.conf
```
All and testing commands are given with the harness script. See the command line help: ./harness -h

Run the fewshot LLM tests:

for i in mimic synthetic mimthetic ; do
    ./harness fewshot $i
done

Supervise-fine tune the LLM models, then test

for i in mimic synthetic mimthetic ; do
    ./harness train $i
done
for i in mimic synthetic mimthetic ; do
    ./harness test $i
done

Train and ablation test the traditional deep learning models:

./harness binaryabl
for i in mimic synthetic mimthetic ; do
    ./harness multilabelabl
done

Changelog

An extensive changelog is available here.

Community

Please star this repository and let me know how and where you use this API. Contributions as pull requests, feedback and any input is welcome.

License

MIT License

Copyright (c) 2025 Paul Landes

License

sunlabuiuc/sdoh

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Languages

Packages