TY - GEN
T1 - Generating Longitudinal Synthetic EHR Data with Recurrent Autoencoders and Generative Adversarial Networks
AU - Sun, Siao
AU - Wang, Fusheng
AU - Rashidian, Sina
AU - Kurc, Tahsin
AU - Abell-Hart, Kayley
AU - Hajagos, Janos
AU - Zhu, Wei
AU - Saltz, Mary
AU - Saltz, Joel
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021
Y1 - 2021
N2 - Synthetic electronic health records (EHR) can facilitate effective use of clinical data in software development, medical education, and medical research without the concerns of data privacy. We propose a novel Generative Adversarial Network (GAN) approach, called Longitudinal GAN (LongGAN), that can generate synthetic longitudinal EHR data. LongGAN employs a recurrent autoencoder and the Wasserstein GAN Gradient Penalty (WGAN-GP) architecture with conditional inputs. We evaluate LongGAN with the task of generating training data for machine/deep learning methods. Our experiments show that predictive models trained with synthetic data from LongGAN achieve comparable performance to those trained with real data. Moreover, these models have up to 0.27 higher AUROC and up to 0.21 higher AUPRC values than models trained with synthetic data from RCGAN and TimeGAN, the two most relevant methods for longitudinal data generation. We also demonstrate that LongGAN is able to preserve patient privacy in a given attribute disclosure attack setting.
AB - Synthetic electronic health records (EHR) can facilitate effective use of clinical data in software development, medical education, and medical research without the concerns of data privacy. We propose a novel Generative Adversarial Network (GAN) approach, called Longitudinal GAN (LongGAN), that can generate synthetic longitudinal EHR data. LongGAN employs a recurrent autoencoder and the Wasserstein GAN Gradient Penalty (WGAN-GP) architecture with conditional inputs. We evaluate LongGAN with the task of generating training data for machine/deep learning methods. Our experiments show that predictive models trained with synthetic data from LongGAN achieve comparable performance to those trained with real data. Moreover, these models have up to 0.27 higher AUROC and up to 0.21 higher AUPRC values than models trained with synthetic data from RCGAN and TimeGAN, the two most relevant methods for longitudinal data generation. We also demonstrate that LongGAN is able to preserve patient privacy in a given attribute disclosure attack setting.
KW - Deep learning
KW - Electronic health records
KW - Generative models
KW - Machine learning
KW - Synthetic data generation
UR - https://www.scopus.com/pages/publications/85122565661
U2 - 10.1007/978-3-030-93663-1_12
DO - 10.1007/978-3-030-93663-1_12
M3 - Conference contribution
AN - SCOPUS:85122565661
SN - 9783030936624
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 153
EP - 165
BT - Heterogeneous Data Management, Polystores, and Analytics for Healthcare - VLDB Workshops, Poly 2021 and DMAH 2021, Revised Selected Papers
A2 - Rezig, El Kindi
A2 - Gadepally, Vijay
A2 - Mattson, Timothy
A2 - Stonebraker, Michael
A2 - Kraska, Tim
A2 - Wang, Fusheng
A2 - Luo, Gang
A2 - Kong, Jun
A2 - Dubovitskaya, Alevtina
PB - Springer Science and Business Media Deutschland GmbH
T2 - VLDB workshops: International Workshop on Polystore Systems for Heterogeneous Data in Multiple Databases with Privacy and Security Assurances, Poly 2021 and 7th International Workshop on Data Management and Analytics for Medicine and Healthcare, DMAH 2021
Y2 - 20 August 2021 through 20 August 2021
ER -