TRIDE: A Temporal, Robust, and Informative Data Augmentation Framework for Disease Progression Modeling

Hyunwoo Sohn, Kyungjin Park, Baekkwan Park, Min Chi

Keywords: Temporal robustness, data augmentation, representation learning, language modeling, electronic health records

Abstract: Modeling the progression of a target disease using electronic health records (EHRs), especially early predicting the onset of a disease, is critical for timely and accurate clinical interventions. While numerous deep learning-based prediction models have shown great success in handling sequential multivariate data such as EHRs, they often lack temporal robustness. This is problematic because they may not perform consistently well across different early prediction hours as training data become scarce upon targeting further future. Indeed, having even one weak point of time can significantly restrict the reliability of the models. In this work, we present TRIDE, a temporal, robust and informative data augmentation framework that can learn temporal representations of EHRs and use them to generate diverse and meaningful training samples by optimizing the level of data transformation. We validate TRIDE on modeling the progression of an extremely challenging disease, septic shock, by using real-world EHRs collected from two different medical systems. Our results show that TRIDE significantly outperforms strong baseline models across different prediction times and datasets, and thus enhances the temporal robustness. Further, we provide in-depth analyses of the generated samples and estimated model parameters to clarify the processes.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)