Generating Multi-label Discrete Patient Records using Generative Adversarial Networks

By Edward Choi, Siddharth Biswal, Bradley A. Malin, Jon D. Duke, Walter F. Stewart, Jimeng Sun. 2017

725 citations recorded by the source corpus.

Access to electronic health record (EHR) data has motivated computational advances in medical research. However, various concerns, particularly over privacy, can limit access to and collaborative use of EHR data. Sharing synthetic EHR data could mitigate risk. In this paper, we propose a new approach, medical Generative Adversarial Network (medGAN), to generate realistic synthetic patient records. Based on input real patient records, medGAN can generate high-dimensional discrete variables (e.g., binary and count features) via a combination of an autoencoder and generative adversarial networks. We also propose minibatch averaging to efficiently avoid mode collapse, and increase the learning efficiency with batch normalization and shortcut connections. To demonstrate feasibility, we showed that medGAN generates synthetic patient records that achieve comparable performance to real data on many experiments including distribution statistics, predictive modeling tasks and a medical expert review. We also empirically observe a limited privacy risk in both identity and attribute disclosure using medGAN.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Generative Models, Medical & Health AI

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Generative Models

In Medical & Health AI

Citations in the corpus

FairGAN: Fairness-aware Generative Adversarial Networks (2018, methodology): “this limitation, medGAN incorporates an autoencoder in a regular GAN model to generate high-dimensional discrete variables [10].”
Synthesizing Tabular Data using Generative Adversarial Networks (2018, background): “RGAN and RCGAN [13] can generate real-valued time-series data. medGAN [9], corrGAN [33] and several improved models [2, 6, 47, 3, 44] can generate discrete medical records but do not tackle the complexity in generating multimodal continuous variables. ehrGAN [8] generates augmented medical records…”
A Survey on Data Collection for Machine Learning: A Big Data - AI Integration Perspective (2018, background): “Synthetic Data Generative Adversarial Networks [57], [58], [59], [60], [61], [62] Policies [63], [64] Image [65], [66], [67], [68], [69], [70], [71] Text [72], [73], [74]”
Data Synthesis based on Generative Adversarial Networks (2018, positive): “Two recent studies applied GANs based on recurrent neural network (RNN) architectures to synthesize only discrete values in electronic health records (EHR) [15] and patent records [16].”
NetGAN: Generating Graphs via Random Walks (2018, background): “Focusing on non-sequential discrete data, Choi et al. (2017) generate high-dimensional discrete features (e.g. binary indicators, counts) in patient records.”
Synthesis of Realistic ECG using Generative Adversarial Networks (2019, background): “This is also known as membership inference attack [35].”
Modeling Tabular data using Conditional GAN (2019, methodology): “ble 1: The number of wins of a particular method compared with the corresponding Bayesian network against an appropriate metric on 8 real datasets. outperform Method CLBN [7] PrivBN [28] MedGAN, 2017 [6] 1 1 VeeGAN, 2017 [21] 0 2 TableGAN, 2018 [18] 3 3 CTGAN 7 8 Recent developments in deep generati”
Generative Adversarial Networks in Time Series: A Systematic Literature Review (2022, methodology): “In contrast, others utilized a three-sample test on the training, test, and synthetic data to identify if the synthetic data has over-fitted to the training data [17, 31].”
A Survey on Generative Adversarial Networks: Variants, Applications, and Training (2020, methodology): “Furthermore, GANs have also been utilized in many other medical applications such as drug discovery GAN (Chem-GAN) [88] for the determination of internal chemical diversity of drugs, feedback GAN (FB-GAN) [89] for generating and designing deoxyribonucleic acid (DNA), medical image processing [90] fo”
When Machine Learning Meets Privacy (2020, methodology): “ch can learn geographic patterns”. Liu et al. [86] aim at the addition of geo-privacy protection layer for publication of spatiotemporal datasets based on synthetic trajectory generation. Choi et al. [23] proposed an approach for the generation of synthetic patient records based on GANs and autoenco”
Generation and evaluation of synthetic patient data (2020, background): “Attribute disclosure [29] refers to the risk of an attacker correctly inferring sensitive attributes of a patient record (e.”
Private FL-GAN: Differential Privacy Synthetic Data Generation Based on Federated Learning (2020, methodology): “[10] ﬁrst proposed using GANs to generate synthetic data, and made some progress in the ﬁeld of synthetic data.”
Generating Multi-Categorical Samples with Generative Adversarial Networks (2018, methodology): “MedGAN (Choi et al., 2017) also implements an autoen-coder between the generator and the discriminator, but in this case the discriminator receives real data or fake data obtained from decoding fake codes.”
Machine Learning for Synthetic Data Generation: a Review (2023, background): “, binary and count features) based on real patient records [16].”
Conditional Wasserstein GAN-based Oversampling of Tabular Data for Imbalanced Learning (2020, methodology): “On the other hand, if the discriminator is less powerful then the generator can successfully fool it without generating realistic samples.”
Teacher-Student Compression with Generative Adversarial Networks (2018, background): “L G ] 2 0 M ar 2 02 0 electronic health records (Choi et al., 2017).”
Language Models are Realistic Tabular Data Generators (2022, methodology): “More recent methods for probabilistic tabular data modeling utilize generative adversarial networks (Choi et al., 2017; Park et al., 2018; Mottini et al., 2018; Xu et al., 2019; Koivu et al., 2020) or variational autoencoders (Xu et al., 2019; Ma et al., 2020; Vardhan & Kok, 2020; Darabi & Elor,…”
GAN-Leaks: A Taxonomy of Membership Inference Attacks against Generative Models (2019, methodology): “For instance, GANs have been applied to health record data and medical images [12, 19, 69] whose membership is sensitive as it may reveal a patient’s disease history.”