The University of Pennsylvania Health System, like various large health agencies, has poured huge resources into building an enterprisewide data infrastructure. With the foundation in place, the health network usually called as Penn Medicine is embarking on a big data project to expand its data horizons and establish predictive analytics to diagnose deadly sicknesses before they occur.
Penn Medicine is a leading academic medical organization. Deployed in Philadelphia, it contains Raymond and Ruth Perelman School of Medicine and the University of Pennsylvania Health System. The health system involves the Hospital of the University of Pennsylvania and Penn Presbyterian Medical Center, Chester County Hospital, Lancaster General Heal, Penn Wissahickon Hospice and Pennsylvania Hospital as well as a number of inpatient and other care facilities.
The open-source, big data initiative, known as Penn Signals, is concentrated on building out an enterprise warehouse and enabling the data science group to establish learning models from historical, at-rest data and then position those models into a real-time information stream, states Corey Chivers, a data scientist at Penn Medicine who is included as one of the leads on the project.
“Our target is to create an infrastructure that can scale up to control a huge variety of data sources within our network that contain data about the health of our sufferer population,” Chivers claims. “We initiated with the obvious candidates—our EHRs and labs—to try to establish predictive models for extreme sepsis and heat failure. But we have policies to increase the utilization of predictive models on the Penn Signals platform and make them present outside the agency.”
The backbone is a homegrown enterprise data warehouse, known as Penn Data Store. The warehouse consists of information from clinical and administrative systems, involving the 3 major clinical information networks at Penn Medicine: an outpatient EHR from Epic, which is utilized by 1,800 affiliated physicians; an inpatient EHR from Allscripts, which is utilized within Penn Medicine’s 5 hospitals; and an enterprise laboratory information network from Cerner. In all, the warehouse stores over 4 billion rows of clinical information, with 2 million being added per day.
For the big data attempt, Chivers and the Penn Medicine’s data science group utilized an ETL procedure to pull information from the enterprise warehouse into an open source database from MongoDB that offers flexibility for the machine-leaning applications the data science team uses. From there, analysis made by the clinical staff is converted into time-series formats, or events, that can be observed by machine learning applications, Chivers states.
The group utilizes Python programming language, ZeroMQ messaging and the iPython Notebook computational atmosphere to pull data sets and explore that information utilizing dimensional reduction and machine learning. They then can protect predictive models they have established and ship them up into the real-time data stream as operational models, Chivers elaborates.
For many health cases, timing is everything. In the situation of sepsis, every hour a sufferer goes undiagnosed increases the mortality amount by more than 7%, in accordance to clinical studies, which also estimate that only 50% of septic shock sufferers acquire effective therapy on time.
At Penn Medicine, the algorithm to track when a sufferer was slipping into severe sepsis depended on analysis of 6 key sign measurements and lab values with threshold rules. The Penn Signals predictive model takes into account more than 200 clinical variables. It has enabled Penn Medicine to track 80% of extreme sepsis cases within 30 hours of the typical onset of symptoms, Chivers claims.
The heart failure predictive algorithm is enabling the Penn Medicine to track 20% more sufferers who are trending toward cardiac failure, and recognizing a group of sufferers that is 5 times more likely to be readmitted after heart failure. Having that predictive data on hand means Penn Medicine clinicians can intervene initially with at-risk groups and concentrate resources on those sufferers more likely to have ongoing heart problems.
Conveying the output from the predictive algorithms is done through text messages that alert particular clinical staff when a sufferer’s condition is heading in a dangerous direction. Penn Medicine also has established a mobile app, known as Caroline, which offers clinicians with a pared-down version of a sufferers’ EHR containing clinical information related to the alert.
The data science group also utilizes the online visualization site Plotly, as well as visualization tools within the iPython Notebook atmosphere, to give clinic department heads and clinical floor leaders with aggregate looks at the predictive information. “Our visualizations are under constant development because we need the clinicians to have a thorough view of how these predictive models work and the data they utilize,” Chivers states. “We have worked closely with clinical staff to comprehend how we as information scientists can interact with them.”
Penn Medicine now finds itself at a crossroad: it is constructed an open-source framework that can control the influx of health information and utilize predictive algorithms in real-time, but the volume, velocity and variety of information are ramping up rapidly, Chivers states. “We are planning to use new data streams—from wearable tools, telemetry devices and ICU monitors—and as we move toward that machine-generated data that is coming in at much higher amount, we have to concentrate on scalability.”
Penn Signals plans to take the infrastructure to that next stage through an agreement with Intel to partner in the establishment of the company’s Trusted Analytics Platform, or TAP.
TAP is an open-source infrastructure constructed on a data layer that involves Apache Hadoop, Spark and other information components, as well as an analytics layer that involves a information science tool kit to simplify model development and an extensible design to produce predictive approaches.
Penn Medicine policies to deploy a 100-terabyte data stack through the TAP framework, Chivers states. The health network also plans to market Penn Signals to other health care agencies, he adds. “We need to get it out to other contributors and find out what is most precious to them—is it something they would like to deploy themselves, or would it be more useful as a stage for service? We do not have an answer to that, but we constructed the platform utilizing open-source devices so that it could be utilized beyond Penn Medicine.”
Your email address will not be published. Required fields are marked *
Nice post, Thank you for sharing valuable information. I enjoyed readi ...
Any study that compares opioids to other pain-relief medicines? ...
Hey, very nice site. I came across this on Google, and I am stoked tha ...
Thanks for the information. I attended it and it was nice to learn the ...
Oh my goodness! Awesome article dude! Thank you so much. Thanks!! ...
Aetna Announces The Completion of $1 Billion Bond Public Offering
Aetna Declares A Brighter Experience For Entire Members of Aetna Dental Team
Urgent care chain utilizes patient feedback to empower performance
Patrick Conway is quitting CMS to supervise BCBS North Carolina
Copyright© 2015 Healthcare insurance News All Right Reserved