SurvivalGAN: Generating Time-to-Event Data for Survival Analysis

This repository contains the experimental code of SurvivalGAN, a generative model that handles survival data firstly by addressing the imbalance in the censoring and time horizons, and secondly by using a dedicated mechanism for approximating time-to-event/censoring. For more details, please read our AISTATS 2023 paper: 'SurvivalGAN: Generating time-to-event Data for Survival Analysis'.

The implementation of the method is included in the synthcity library, in the SurvivalGAN plugin.

Installation

Install synthcity and other depends

pip install -r requirements.txt

For more tutorials and examples, checkout the Synthcity tutorials section.

Datasets

Add the data in the experiments/data folder.

Dataset	No. instances	No. censored instances	No. features	Access
ACTG 320 clinical trial dataset	1151	1055	11	Link
METABRIC	1093	609	689	Link
CUTRACT	10086	8881	6	private
PHEART	40409	25664	29	private
SEER prostate cancer	171942	167568	6	private

Reproducing results

Result	Source notebook
Figure 1	experiments_00_km_plots_tte_models
Table 1,2,9,10,15	experiments_01_benchmark_synthetic_survival_data
Table 3	experiments_02_sources_of_gain_parametric
Table 11	experiments_04_loglikelihood
Table 12, 13, 14	experiments_05_predicting_censoring
Table 16	experiments_03_gmm_test_perf
Figure 4,5,8	plots_00_data_fidelity
Figure 6,7	plots_02_benchmark_gain_of_function

Citing

@misc{https://doi.org/10.48550/arxiv.2302.12749,
  doi = {10.48550/ARXIV.2302.12749},
  url = {https://arxiv.org/abs/2302.12749},
  author = {Norcliffe, Alexander and Cebere, Bogdan and Imrie, Fergus and Lio, Pietro and van der Schaar, Mihaela},
  keywords = {Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {SurvivalGAN: Generating Time-to-Event Data for Survival Analysis},  
  year = {2023},
  copyright = {Creative Commons Attribution 4.0 International}
}