An international consortium involving Penn researchers pools electronic health record data from around the world to discover clinical insights about COVID-19

PHILADELPHIA— To provide greater clinical insight for the fight against COVID-19, a consortium of research scientists that included faculty from the Perelman School of Medicine at the University of Pennsylvania pooled their efforts to create a common data model and a shared analytics framework that will aggregate information from disparate electronic health records (EHR) internationally. With the creation of this model – the Consortium for Clinical Characterization of COVID-19 by EHR (4CE) –clinical teams and researchers will now have a powerful tool available to them to quickly discover trends and provide answers to questions about the virus. A paper on the effort was published this month in NPJ Digital Medicine. 

Like the other sites in the study, clinical data from Penn Medicine was analyzed and provided for the effort. And for future studies with 4CE, PennAI, a free self-service machine learning tool developed at the Institute for Biomedical Informatics, will be available to each member site to power the project.

“We are excited to use our PennAI software for this project,” said paper co-author Jason Moore, PhD, the director of the Institute for Biomedical Informatics and a professor of Informatics. “It can be installed locally at each site and used to generate machine learning models for predicting COVID-19 outcomes such as death or disease severity. This is a critical need that we will contribute to the project.”

The consortium consists of 96 hospitals from around the world and so far, has gathered data on more than 27,000 COVID-19 cases with 187,000 laboratory tests. Previously, because of differences in electronic health records, all of this data would not have been able to “talk” to each other in a way necessary for analysis. But with so many sites putting their data into a common data model and making it available to be processed and analyzed, consortium scientists were able to detect trends and patterns of this new virus that were previously invisible.

“For example, laboratory data were standardized from Penn Medicine's electronic health record to Logical Observation Identifiers, Names, and Codes (LOINC) and shared units of measure before analyzing their change over time. These steps were critical to uncovering initial clinical insights,” said Danielle Lee Mowery, PhD, Penn Medicine’s chief research information officer and an assistant professor of Informatics. “Notable insights include abnormal trends in D-dimer protein, which is a measure of blood clotting, and C-reactive protein, a measure of inflammation, among COVID-19 patients.”

Among other immediate insights were that liver functions initially presented as typical, but worsened over time as patients were hospitalized. White blood cell counts were also typically normal among patients but only elevated among those with the most serious forms of COVID-19. 

“The COVID-19 data warehouse established at Penn Medicine will enable our researchers to access standardized data and generate results which can be replicated at sites around the world,” said John H. Holmes, PhD, IBI’s associate director for Medical Informatics and a professor of Informatics in Epidemiology. “This opens the door to local insights about COVID-19 patients from the Philadelphia area while at the same time contributing to the global battle against this infectious disease.”

While the consortium itself is new and addressing a new threat, it is actually the culmination of years of work in health analytics.

“Our ability to rapidly respond to a global pandemic was made possible by years of institutional investments in health information technology and biomedical informatics expertise and infrastructure.”, said Moore. “We are seeing the value of electronic health records and artificial intelligence in real-time.”