Can you imagine how your own social media chatter filtered through a link to your own electronic medical record could be used to generate a new stream of clinically relevant biometric data?

This futuristic-sounding idea becomes all the more interesting in the wake of the latest study out of the University of Pennsylvania’s Social Media Lab that linked the daily social media output of more than a thousand people to their personal electronic medical records.

Headed by Penn Medicine emergency medicine physician Raina Merchant, MD, the research project just reported in BMJ Quality & Safety was aimed at testing the potential to create a large-scale database linking health-relevant language from an individual’s social media stream with data from their electronic medical record. The participants were recruited from the population of patients visiting a large urban hospital emergency department.

The Lab’s ten-member research team also included study co-author Lyle Ungar, PhD, Professor of Computer and Information Science at Penn’s School of Engineering and Applied Science, and LDI Senior Fellow David Asch, Executive Director of the Penn Medicine Center for Health Care Innovation.

The primary study finding was that “a database that merges social media with EMR data has the potential to provide insights about individual’s health and health outcomes.”

New digital health database
“This was the initial work for building a longitudinal digital health database that will allow us to look at what potential connections can happen,” said Merchant, who is the Director of the Penn Social Media and Health Innovation Lab, an Assistant Professor at Penn’s Perelman School of Medicine and a Senior Fellow at Penn’s Leonard Davis Institute of Health Economics (LDI). “The first question was about feasibility —  ‘would people really be willing link their social media accounts to their electronic medical records?’ We were surprised that so many actually consented to this.”

Of the 2,717 Twitter- and Facebook-using ER patients who were identified as potential study candidates, 1,433 agreed to participate in a study about their social media postings and health. Of those, 1,008 agreed to allow both their postings and personal EMR data to be accessed and analyzed — thus facilitating the first study of its kind. 

“Other research efforts that used social media data to predict things like the spread of the flu were analyzing media postings at a population level,” said Merchant. “We’re looking directly at individuals and their medical records.”

Accuracy of social media comments
Those medical record data can play a potent role in confirming the accuracy of comments in a social media stream. “In our EMR comparisons we actually have validated diagnoses and the dates when those diagnoses occurred,” said Merchant. “If someone is talking about being sick in their social media stream, we may be able to see that they actually DID come to the ER and were admitted to the hospital. So we could identify what was probably commentary about real sickness.”

At the same time, the manner in which the quantity and complexity of a person’s health-related social media language varies over time may itself suggest cognitive changes, the onset of depression, or other mental health issues.

“Our big picture goal is similar to other longitudinal databases,” said Merchant. “We need a large sample of information and we need to be able to collect it over time to be able to identify how a disease may progress or how people may talk about it early on when just diagnosed. We hope to build this database over the next decade and enroll thousands more patients to be able to look at this information.”

Area of future research
“The idea that this merger of social media and medical record data could ultimately generate clinically relevant information is a fascinating one and a very interesting area for further research,” Merchant continued. “The opportunity for the future is figuring out how this information could be made interpretable and actionable for a provider. I think a more immediate step would be to ask ‘How do we make this information more actionable for patients?’ So, our more immediate goal would be to feed these insights back to patients in a way that may positively affect their health and health outcomes.”