Exhaustively Characterizing a Patient Cohort by Prevalence of EMR Facts: a Generalized, Vendor-Agnostic Method for Quality Control and Research

We present a method for rapidly ranking all distinct facts in an electronic medical record (EMR) system by howover-represented or under-represented they are in a patient cohort of interest relative to some larger referencepopulation of patients in the same EMR. We have implemented this method as a plugin for i2b2, the open sourcedata warehouse platform widely used in research health informatics. Our method is highly flexible in terms of whatmedical terminologies it supports and is vendor-independent thanks to leveraging the i2b2 star schema rather thanany one specific EMR. It can be applied to a wide range of informatics problems including finding healthdisparities, searching for variables to include in a risk calculator or computable phenotype, detection ofcomorbidities, discovery of adverse drug reactions. The case study we present here uses this software to findunlabeled flowsheets for patients suffering from amyotrophic lateral sclerosis.