Generating Electronic Health Records with Multiple Data Types and Constraints

Sharing electronic health records (EHRs) on a large scale may lead to privacy. Recent research has shown that risks may be mitigated by simulating through generative adversarial network (GAN) frameworks. Yet the methods to date are limited because they 1) focus on generating data of a type (e. g. , diagnosis codes), neglecting other data types (e. g. , , procedures or vital signs) and 2) do not represent constraints features. In this paper, we introduce a method to simulate EHRs of multiple data types by 1) refining the GAN model, 2) accounting for constraints, and 3) incorporating key utility measures for such tasks. Our analysis with over 770, 000 EHRs from Vanderbilt Medical Center demonstrates that the new model achieves higher in terms of retaining basic statistics, cross-feature correlations, structural properties, feature constraints and associated patterns from data, without sacrificing privacy.