Warfarin dosing policy website

This is a simple Flask application which exposes the learned warfarin policy from our research as a human-usable web form. We created it to give researchers the ability to interrogate the dosing algorithm. It is hosted here on Heroku.

Model weights

This repository also provides the final policy neural network weights and learned data preprocessors, which could facilitate large-scale validation on external data. For example, you could clone this repository (which contains the model weights in model.pt and a wrapper in model.py) and do:

>>> from model import predict
>>> predict(inrs=[3.], warfarin_doses=[25.], traj_split_since=[])
3

Running the website locally

Install the requirements under (tested under Python 3.9):

$ python3 -m pip install -r requirements.txt

Launch the app:

$ gunicorn web:app

See also

• Our research repository