Your Heart is Big Data

Main page content

UW Tacoma’s Center for Data Science collaborates with MultiCare and Microsoft to improve outcomes for hospital patients with heart disease.

The Center for Data Science Risk-o-Meter team, from left: David Hazel, Si-Chi Chin, Ankur Teredesai, Senjuti Basu Roy, Vivek Rao, Kiyana Zolfaghar, Rui Liu. Photo by Cody Char.

When heart failure patients leave the hospital after treatment, some continue to heal, resuming at least a semblance of their normal routines. But others don’t stay on the recovery track, suffering from acute problems that require readmission soon after their initial release.

Recently, health care organizations have begun to look at ways to reduce these readmissions. With advances in electronic medical records and the development of sophisticated analysis techniques that didn’t even exist five years ago, hospitals can now gain insights into patient-discharge procedures and how they are linked to readmissions.

MultiCare Health System, the South Sound’s largest hospital and health care organization, asked researchers at UW Tacoma’s Center for Data Science to analyze its readmission cases for heart failure patients. By looking for patterns in massive amounts of multi-faceted data related to those patients, could algorithms be developed to help clinicians at MultiCare predict which patients would be most susceptible to readmission in fewer than 30 days? Could adjusting their care management and treatment improve patients’ overall health and reduce costs of care?

Big Data

The Center for Data Science (CDS), part of UW Tacoma’s Institute of Technology, was founded in 2013 to bring together expertise in “big data,” the tsunami of information generated by today’s computer systems.

The Center’s researchers and students address real-world challenges using a variety of data—from bioinformatics to social networks. These data sources share three common traits--known as the "three Vs" that define big data. They produce huge quantities of data (volume) very quickly (velocity) with many different attributes (variety). The sources are rich in information, but poor in the tools needed to analyze that information. Finding the silver lining of this data cacaphony is the mission of CDS. The students and researchers search for patterns that indicate deep connections between what may appear, on the surface, to be isolated phenomena. They seek to bring scientific rigor to this pattern analysis, allowing mathematical models to be developed and predictions to be made.

Heart Failure

MultiCare Health System Risk-o-Meter team, from left: Dr. Lester Reed, Albert Marinez, Dr. Tony Kim, Yoshi Williams, Holly Burke. Photo by Cody Char.

The Center’s work with MultiCare started as “sort of a skunk works,” according to Center director Ankur Teredesai, who studied computer science at University at Buffalo and started his academic career at Rochester Institute of Technology, before coming to UW Tacoma in 2007.

“The motivation to use data mining on health care data emerged from a series of discussions in the classroom with students who were already working or interning at MultiCare. We started exploring predictive modeling on clinical data and soon had breakthrough ideas,” said Teredesai.

Dr. Lester Reed, MultiCare’s senior vice-president for quality, helped connect the researchers’ ideas to the clinical side of the hospital system. Teredesai says, “Thanks to Dr. Reed, we were able to spend a lot of time with the clinical staff, learning their clinician’s-view of heart failure. We had to understand what congestive heart failure means and how it is different from cardiac heart failure, for example. We were able to develop a set of qualitative and quantitative measures that are important for diagnosing and understanding the disease.”

Dr. Reed was also the conduit to the technology staff at MultiCare, starting with Dr. Paul Amoroso, medical director of MultiCare’s Institute for Research and Innovation. Albert Marinez, MultiCare’s director of information intelligence, provided critical insights into the types of data on heart failure patients being collected by the hospitals’ IT systems.

Senjuti Basu Roy, assistant professor in UW Tacoma’s Institute of Technology, is one of the lead researchers on the MultiCare project. “Real world clinical data is messy. It has significant anomalies and missing values, and often we are integrating data from multiple sources. We worked closely with Albert and his team on data exploration, in which we devise techniques to extract intelligence from the data.” Marinez and his team helped with data pre-processing, such as removing irrelevant and redundant information.

Microsoft’s Azure for Research

In the early stages, the work was very labor-intensive. David Hazel, managing director of the Center for Data Science, said, “If we go back to the very beginning of the effort, we had students trying to build models on their laptops—completely underpowered and dealing with information systems that weren’t designed for the integration effort they’re working on. So, the result was the data experiments took a really long time.”

“The turning point was in late 2013, when Microsoft awarded us an Azure for Research grant. The access to cloud-computing infrastructure, running on Azure, allowed us to quickly scale to run multiple experiments simultaneously. This greatly accelerated our effort,” said Hazel.

Microsoft’s cloud platform, Azure, is certified to meet federal regulations surrounding patient medical records privacy.  (The Healthcare Insurance Portability and Privacy Act of 1996, or HIPAA, establishes rules for the handling of protected health information.) “Having that HIPAA compliance saves us a huge amount of time and effort,” said Teredesai.


Screenshot of Risk-o-Meter web interface.

The CDS team and their MultiCare colleagues now had plenty of data, and were able to use it to predict which heart failure patients would be readmitted to the hospital less than 30 days after an earlier release. They realized, however, that the ultimate benefit of the insights from the data would be to let patients themselves know what their readmission risk is, and to provide suggestions for treatments and choices the patients and their doctors could make that would lessen that risk.

Thus was born the idea of the Risk-o-Meter.

The idea is to give patients access to a simple interface (doctors have a separate interface with a more complex set of attributes) via the web or apps for Android and Windows phones. The patient or a caregiver enters demographic and basic clinical data; for example: age, gender, ethnic or racial background, length of previous hospital stay, blood pressure, existence of other diagnosed diseases such as diabetes or history of stroke, and clinical measures of current heart function, if known.

One feature that distinguishes the Risk-o-Meter from other readmission-risk calculators is its ability to work with incomplete data. If a patient doesn’t know his or her blood pressure, for example, that submission can be skipped.

The Risk-o-Meter compares the patient’s submitted information against millions of records of other patients. Through a process called ‘machine learning,’ the Risk-o-Meter is constantly sifting through these millions of records and building predictive models so that when a new patient comes along, he or she can be assigned a likelihood of an outcome that parallels that of patients with similar characteristics. All this happens constantly, at lightning speed, and over massive volumes of patient records.

The Risk-o-Meter then displays to the user a single number, which represents the percentage chance the patient will be readmitted to the hospital in less than 30 days. The number is color-coded: green for low risk, yellow for medium risk, red for high risk. The patient also can review the leading risk factors that yield that prediction, and suggestions of actions to take that would lead to reducing the risk.

The power behind the Risk-o-Meter is the huge number of patient case histories that go into building the predictions, and the fact that the data is always growing and the predictive model is always getting tuned.


The Risk-o-Meter project started with MultiCare, but from the beginning UW Tacoma and MultiCare both realized the innovative data analytics behind it could benefit any hospital. “We are launching what we call ‘readmission score as-a-service’,” says Teredesai, riffing off the term ‘software-as-a-service’ used widely to describe cloud-based computing power. “This would enable anyone from health care providers to insurers to mobile app developers to connect to our service and use the bank of predictive models.” Ultimately, the readmission score service concept could work for many chronic conditions, not just heart failure.

The Center for Data Science is enlisting the support of UW’s Center for Commercialization (C4C), an office that helps researchers take their innovations out of the lab. Says CDS’s David Hazel, “C4C’s expertise is very important as we explore commercialization opportunities, either through a licensing agreement or a spinoff.” Last year, C4C successfully spun off 18 companies based on research from UW.

Said Teredesai, “Our vision is to expand this research to attack other chronic conditions, integrate more varied data sources, and make the cloud the de-facto research infrastructure for health care.”

More information

Center for Data Science web site

Risk-o-Meter project presentation to Microsoft Research

UW Tacoma Risk-o-Meter video

Written by: 
John Burkhardt / August 1, 2014
Media contact: 

John Burkhardt, UW Tacoma Communications, 253-692-4536 or