hART-heart-failure-Attentive-Risk-Trajectory

This repo provides a sample code to experiment with the hART model. The notebook is best run on Google Colab. This repo was made for our study: "hART: Deep Learning-Informed Lifespan Heart Failure Risk Trajectories"

The notebook titled 'hART.ipynb' is divided into sections to ease the interpretability and usage. Here's a short description of each section:

1. Imports: a) Contains all Imports needed to run the notebook b) Upload the data needed to run the notebook, Note: Synthetic data is provided in the GitHub repo ('synthetic.csv') c) Apply the specific exclusion criteria for our study

2. Preprocessing: a) Contains helper functions to turn event-based data into sequential input and HF labels b) Data split

3. Attention Helper Functions: a) Custom function used in the Models

4. MODELS: a) Contains the novel hART model and previously developed DHTM model (https://github.com/li-lab-mcgill/recurrent-disease-progression-networks) b) Allows you to train and evaluate the performance of the models

5. Patient Population Trajectory Sample: a) Provides the ability to experiment with the hART-predicted trajectory at a population level b) Choose a subgroup to test (severe CHD vs. Non Severe CHD) c) Output is HF Trajectory + Attention Matrix (Heatmap) + Actual Distribution of HF for population

6. Individualized Trajectory Sample: a) Provides the ability to experiment with the hART-predicted trajectory at an individualized level b) Choose a baseline to test (non-severe CHD patients) and a specific individual (patient = ) c) Output is HF Trajectory + Patient's Medical Event History + Attention Matrix (Heatmap)

NOTES:

1. The synthetic is NOT the real data used in the study. It is a representation of the actual data
2. Use the 'sample' data frame to help run the code faster (the whole data takes hours to preprocess)
3. The synthetic data is in a zip file due to size (please unzip to use)