Targeted BEHRT

Repository for publication: Targeted-BEHRT: Deep Learning for Observational Causal Inference on Longitudinal Electronic Health Records

IEEE Transactions on Neural Networks and Learning Systems; Special Issue on Causality

https://ieeexplore.ieee.org/document/9804397/

DOI: 10.1109/TNNLS.2022.3183864.

How to use:

In "examples" folder, run the "run_TBEHRT.ipynb" file. A test.csv file is provided to test/play and demonstrate how the vocabulary/year/age/etc function (please read full paper linked above for further methodological details).

Furthermoree, in the examples folder to run the CV-TMLE estimator, run the "CVTMLE_example.ipynb" file. A host of fake fold data is provided to test/play and demonstrate how the CV-TMLE algorithm works (please read methods publication of CV-TMLE for further details).

The files in the "src" folder contain model and data handling packages in addition to other necessary VAE relevant files and helper functions.

Requirements:

torch >1.6.0

numpy 1.19.2

sklearn 0.23.2

pandas 1.1.3