This repo contains a C++ implementation of our paper, "Hawkes Process Modeling of Adverse Drug Reactions with Longitudinal Observational Data" [PDF].

@inproceedings{hawkes,
title={Hawkes Process Modeling of Adverse Drug Reactions with Longitudinal Observational Data},
author={Bao, Yujia and Kuang, Zhaobin and Peissig, Peggy and Page, David and Willett, Rebecca },
booktitle={Machine Learning for Healthcare Conference},
year={2017} }

Each row in the data file represent the trajectory of one patient, which is a sequence of ordered time-event pairs separated by whitespace:

t_1 m_1 t_2 m_2 ... t_n m_n

where 0 <= t_1 <= ... <= t_n and m_i in {1,...,numOfVariables}. We assume {1,...,numOfOutcomes} to be the indices for the adverse outcomes and {numOfOutcomes+1,...,numOfVariables} to be the idices for the drugs of interest.

The Marshfield Clinic EHR data is not publicly available due to patients' privacy. We provide an example synthetic data generated from a Poisson autoregressive model.

1. Compile

g++ --std=c++14 Hawkes.cpp -o Hawkes.out

2. Example run

./Hawkes.out -f "file_to_data" -k num_of_kernels -w length_of_windows -l lasso_regularizations