Boosting Patient Outcomes with Big Data

Healthcare turns to analytics platforms and data mining for deeper insights to patient care and outcomes.
Customer Experience

Leveraging data to gain insights into customer behaviors and provide better services is the cornerstone of the retail and marketing industries, but healthcare is still catching up to these strategies. Data mining, predictive analytics, and trends analyses for improving patient outcomes are some of the technologies that are beginning to take hold in healthcare.

And even though healthcare companies are only beginning to catch up to companies in other industries in leveraging data, it's already a crowded space, notes Munzoor Shaikh, a senior manager at consulting firm West Monroe Partners.

"There are a lot of players out there, everybody's experimenting, everybody's trying something; everybody's got a success story," Shaikh says. "What we find in the midst of this chaosstrategy is key. That's the first thing that we ask people to do. The other thing is it takes time because there's lots of historical data that you need to be able to prove out."

Bringing a solution to market, even with a strategy, is a complicated process that can leave companies in limbo. Three years ago, Chris Poulin, principal partner at predictive analytics provider Patterns and Predictions, set out to develop a tool that could identify U.S. veterans who were likely to be suicidal using text-mining and predictive analytics.

In 2010, Poulin founded the Durkheim Project with funding from the Defense Advanced Research Agency (DARPA). He also brought together medical and artificial intelligence experts from Dartmouth Engineering, Dartmouth Medical School, and the U.S. Veterans Administration to develop the research.

The project's first goal was to prove that text-mining methods could provide statistically significant predictions of suicidal tendencies. Working with a control group of about 100 veterans, the researchers developed linguistics-driven prediction models to estimate suicide risk, based on unstructured clinical notes.

In 2011, Patterns and Predictions began building out the foundational infrastructure and predictive modeling that would support the project's extensive data collection and analysis with enterprise analytic data management provider Cloudera. The researchers trained the system by feeding it keyword combinations, patterns, and other linguistic clues based on data analyzed from a variety of veterans' database sources. Once trained, the machine learning could identify useful clues in real data, and establish a risk "score."

In early 2013, the platform achieved a 65 percent accuracy rate in predicting suicide risk among the veteran control group. The next goal was to scale the project by testing it with at least 100,000 veterans who opt into the project and receive a unique Facebook app that is designed to capture posts, Tweets, mobile uploads, and location data. Additional profile data was captured as well, such as physician information and clinical notes.

However, the project has stalled, according to Poulin, as he looks for a clinical group to sponsor the research and provide feedback. "Many groups have expressed interest in testing the system in a clinical field study, but we're not sure who's going to ultimately pay for it," he says. "Maybe we need to compensate clinical groups on integration and technical support, but this experience has been eye-opening; I would have thought there's more flexibility."

Timing may also be in an issue. After the Facebook debacle in which the social network was severely criticized by the media for using users as unknowing test subjects, Poulin says he has had to draw clear distinctions between his project and Facebook's own tests.

"There has been some confusion about whether Facebook is using my academic team to do social-engineering research," Poulin notes. "We are only doing clinical research; we're not using the data for marketing at all." For other companies that are considering launching their own research, it is critical to "maintain a high degree of transparency," Poulin adds. "You need to be clear about who's doing what and why."

Reducing wait times in emergency rooms and providing care more efficiently are other challenges that companies are trying to solve. In 2013, Emory University Hospital partnered with IBM on a research project to advance predictive care for critical patients in the ICU. The experimental system uses IBM's streaming analytics platform with Excel Medical Electronics' bedside monitor data aggregation application to collect and analyze more than 100,000 real-time data points per patient per second. The software then identifies patterns that could indicate serious complications like sepsis, heart failure, or pneumonia, to provide real-time medical insights to clinicians.

"We have oceans of data," says Dr. Tim Buchman, founding director of the Emory Critical Care Center, "and in order for us to deliver better health, better care, and lower costs, we have to get smarter about the way we use all that data."

IBM's technology provides a visual display of patterns in a patient's vital signs and other data points that serves as an advanced "air traffic control view of patients in an ICU" and alerts providers to critical changes in a patient's condition and allows them to be more efficient in allocating their attention, Dr. Buchman notes.

"By using the data wisely and foreseeing Big Data analyzed in real time, we can actually do a better allocation of our human resource so that at any given moment, we've got the right people at the right bed," he says.

When asked about the impact of the research, Dr. Buchman demurred, noting that the research is ongoing and it's too early to measure the value of the technology and its effect on patient outcomes. But more tools and research are needed, Dr. Buchman added, as the needs of aging baby boomers and other patients outpace the ability of providers to give adequate care.

"The fact of the matter is we're running out of doctors and healthcare is very late to the game in harvesting data in real time and using analytics and predictive modeling to figure out what's going to happen next," Dr. Buchman says. "So we have to rearrange the labor force and use appropriate technologies that are going to enable people to deliver the best care today and tomorrow."

Dr. Gilanthony Ungab, a cardiac electrophysiologist who practices in National City, Calif., agrees that data mining tools offer opportunities to better serve patients. People who have pacemakers and defibrillators, for example, must wait for a vendor or trained representative to collect the data and produce a report about the device's status during an examination. This process can be time-consuming, especially if the reports must be printed or faxed and can be easily misplaced, according to Dr. Ungab.

After observing the long wait periods in emergency rooms, three years ago Dr. Ungab teamed up with the CEO of a healthcare product design firm to develop a platform for managing data from cardiac devices.

They founded Geneva Healthcare, which offers a cloud-based solution that collects and standardizes a patient's cardiac device data according to guidelines established by the Heart Rhythm Society. The Geneva Healthcare Suite then organizes the data on a dashboard.

When a patient enters an emergency room, for example, a nurse or technicians collects the device data by waving a wand over the implant site. The data is entered in the platform and clinicians can review the condition of the device and the patient remotely or within a provider's electronic medical record and/or a health information exchange in the hospital. "We're using real-time physiological data and allowing clinicians to make decisions quickly," Dr. Ungab says.

The UCSD Medical Center has been using the Geneva Healthcare Suite in its Hillcrest and Thornton Emergency Departments since September 2012. Since it implemented the platform, the UCSD Medical Center has reduced waiting periods in its emergency rooms by 92 minutes, according to a study published in the Academic Emergency Medicine Journal.

Sharp Healthcare, a non-profit healthcare delivery system based in San Diego, is in the process of implementing the Geneva Healthcare Suite across its hospitals. And Paradise Valley Hospital, a subsidiary of Prime Healthcare Service, is launching a pilot program of Geneva's technology.

"Using Geneva's technology platform we have been able to quickly train our emergency department staff on how to collect information about a patient's device regardless of the device manufacturer or type of device. We do not have to rely on each manufacturer's device representative or on different proprietary reporting systems," comments Dr. Theodore Chan, UCSD Chair of the Emergency Medicine, in a statement.

Other healthcare players see opportunities in using data to address health conditions before they flare up. Dr. Scott Rifkin is certified in internal medicine and is the founder and CEO of Mid-Atlantic Health Care, a provider of nursing care and post-acute services. Dr. Rifkin oversees the operations of 18 nursing homes in Maryland, Delaware, and Pennsylvania. He also founded Real-Time Medical Solutions, a company that offers data mining software that detects warning signs in nursing home patients' electronic medical records and alerts nurses to check on patients.

The software compares information from patients' medical records against 130 questions about their vital signs, daily activities (did the patient walk today, move his or her bowels, etc.), and other data points to identify changes in the patients' health and behavior.

"Every hour of every day we run 130 questions against our data to decide if there are critical issues where if we intervene today, the patient is less likely to go to the hospital," Dr. Rifkin says. The RTMS data mining technology was implemented across the Mid-Atlantic Health Care facilities nearly three years ago and has contributed to a two-third reduction in hospital readmission rates among nursing home patients, according to Dr. Rifkin. More than 50 other nursing homes across the U.S. have also implemented the RTMS software.

More work is needed to better understand a patient's conditions, Dr. Rikfin adds. "There's still a lot of room for improvement," he notes, "Even in terms of communications between the physician and nursing facility and an out-patient and provider, but this is a start."