Last July I wrote something about efforts at Penn Medicine to mine its data warehouse to fine-tune algorithms and predictive logic that could identify adverse events such as the onset of sepsis.
Brian Wells, Penn Medicine’s associate vice president of health technology and academic computing, described “Penn Signals” as a platform that provides the tools needed to build, test and deploy predictive applications powered by Penn’s EHR data stream. “We can run an algorithm against real-time data coming out of the EMR and do predictions almost at the point of care,” said Wells. “That is the foundation we have laid. Because of that investment in having the data organized, aggregated, and mapped in a way that is very usable, it enables our team to do some things that are pretty amazing.”
Now that Penn Signals algorithm approach is being applied more broadly. The Pennsylvania Department of Health has given a four-year, $5 million grant to the University of Pennsylvania, led by the Leonard Institute of Health Economics (LDI) and funded through the Commonwealth Universal Research Enhancement (CURE) program to develop and test algorithms that predict adverse health events in real time in the hospital, in the home, and in the community.
Called “Smarter Big Data for a Healthy Pennsylvania: Changing the Paradigm of Healthcare,” the project’s leaders say the results could help transform the healthcare delivery system from one focused on treating expensive clinical events after they occur to one more proactive in targeting disease, community health, and healthcare access to avoid common and high-risk events before they occur.
Dan Polsky, Ph.D., executive director of LDI, told me that one aspect of the project would explore in-hospital prediction. Amol Navathe, M.D., Ph.D., will work closely with the Penn Signals to develop prediction models of patients at risk for in-hospital complications of common surgical care, including gallbladder surgery, colorectal surgery, and total joint replacements. The model will be developed at Penn’s Health System and tested at Temple University Health System.
To start working on predictions in the home, they are going to start looking at linking data about prescriptions filled with the EHR, Polsky said. “We want to use the data we have to predict who is at highest risk for rehospitalization. They also plan to use activity trackers or smartphone apps to track movement in the home, and enroll subjects to track their movement and see if there are signals in the data about rehospitalization. Mitesh Patel, M.D., M.B.A., leads the research team exploring at-home prediction by developing models to dynamically predict changes in out-of-hospital risk for 30-day readmission by monitoring medication adherence and physical activity in the home and through the integration of these home data sources with the data from insurance claims and the electronic health record.
The third leg of the project is exploring whether integrating social media data with public health data will allow researchers to build and validate a tool to monitor and predict high-morbidity health conditions (heart disease, cancer, chronic lung disease, stroke, and unintentional injury) at a community level. It will also build computational models to use Twitter data for monitoring and predicting dynamic public health events in the state (e.g., influenza, food-borne illness, infectious outbreaks, and acute environmental exposures).
Raina Merchant, M.D., director of Penn’s Social Media & Health Innovation Lab, will lead the team. “We are able to geocode about 10 percent of tweets,” Polsky explained. “So knowing the location of the tweeters, we can do sentiment analysis, and show correlation between certain words and sentiments and cardiac risk. We can find words that predict cardiac risk by county.” Most community health surveys take three years to complete, he said. “We are unlikely to track changes in cardiac risk, but things like infectious diseases or depression, violence, addiction, things where sentiment and rapid changes in the health of communities you could pick up quite quickly. You could track sentiment over time and find localized communities where you might want to be concerned about what is going on with drug addiction and intervene much more quickly than waiting for the morgue to tell you about the uptick in addiction.”
I will check back with these researchers and report on the results of their work over the next year or two.
