phantomDB

We are sharing diganostic history of  2,000,000  phantom patients (1M phantoms generated from a national model, and 1M generated from Chicagoland African Americans trreated at the University of Chicago Medical Center), comprising time-stamped diagnostic and procedural codes, along with the generative model and software (python package pypi.org, all with a permissive license. We also share the generative algorithm and results that demonstrate via critical stress-test comparisons that it is difficult  to differentiate between the training database and generated phantom patients. The research community will now have access to a high quality EHR database, validated to closely replicate real data without licensing constraints. The generative AI models for the national cohort and teh African-American cohort are also included.