The Health Informatics Revolution
When your doctor diagnoses a condition and recommends a course of treatment, she relies on her extensive training, guidelines from professional medical organizations, and previous experience with thousands of other patients.
But what if your diagnosis and treatment could be further informed by the experience of millions of other patients, including those who not only had similar symptoms, but perhaps also were your age, gender, ethnicity — and with similar medical history? That’s among the benefits coming soon from health analytics and informatics.
Using massive data sets, machine learning, and high-performance computing, health analytics and informatics is drawing us closer to the holy grail of health care: precision medicine, which promises diagnosis and treatment tailored to individual patients. The information, including findings from the latest peer-reviewed studies, will arrive on the desktops and mobile devices of clinicians in health care facilities large and small through a new generation of decision-support systems.
Margaret Wagner Dahl, Georgia Tech’s associate vice president for information technology and analytics, says health informatics requires an integrated approach. She’s shown in the radiology room at Georgia Tech’s Stamps Health Services.
“There are massive implications over the coming decade for how informatics will change the way care is delivered, and probably more so for how care is experienced by patients,” said Jon Duke, M.D., director of Georgia Tech’s Center for Health Analytics and Informatics. “By providing data both behind the scenes and as part of efforts to change behavior, informatics is facilitating our ability to understand patients at smaller population levels. This will allow us to focus our diagnostic paths and treatments much better than we could before.”
At Georgia Tech, health informatics researchers are partnering with both public- and private-sector organizations to develop and apply transformative technology that will connect incompatible systems and analyze vast data sets. This technology also will help clinicians track the latest research, potentially shortening the time required to move health care advances into practice.
“Our goal is to be directly involved with that health care transformation and to be one of the contributors focusing on what technology can do well,” said Steve Rushing, senior strategic advisor for health extension services at Georgia Tech. “Technology has to be leveraged in a way that will meet the goals of improving the quality of care, bettering the patient experience, and addressing the rising cost of health care.”
Georgia Tech’s health informatics effort combines academic researchers in computing and the biosciences, practitioners familiar with the challenges of the medical community, extension personnel who understand the issues private companies face, and engineers and data scientists with expertise in building and operating secure networks tapping massive databases.
“It takes all of these components to really make a difference in an area as complex as health informatics,” said Margaret Wagner Dahl, Georgia Tech’s associate vice president for information technology and analytics. “This integrated approach allows us to add value to collaborators as diverse as pharmaceutical companies, health care providers, large private employers, and federal agencies.”
ADVICE FROM MILLIONS OF PATIENTS
The availability of millions of claims records produced by doctors, clinicians, hospitals, pharmacies, and others presents a unique opportunity for providing decision support to individual physicians, who often have less and less time to treat patients whose conditions are more and more complicated.
Finding information useful to doctors within claims records — and electronic health records — offers a challenge much greater than the proverbial needle in a haystack. For one thing, the decision-support system must find that needle quickly enough — in a matter of seconds — to give doctors the information they need while they’re working with patients.
To address that challenge, Associate Professor Jimeng Sun and his graduate students in Georgia Tech’s School of Computational Science and Engineering are using advanced algorithms, machine learning, and high-performance computing to find subtle signals in the data sets. But before they can apply these powerful tools, they must confront the fact that the data they’re using was gathered to help medical providers gain reimbursement — not to aid in diagnosis.
“Every clinic and every hospital has this data, so the benefit of using electronic health records and claims records is that you don’t have to put in extra effort to collect it,” Sun noted. “But it’s very difficult to get the data into a high-quality state that will allow you to use it.”
Once the data is clean, Sun’s lab uses neural networks — modeled on the brain — to teach computers to find useful similarities among patient records. Finding those connections requires access to millions of records: some to train the system, and the rest to analyze. The researchers use high-performance computers based on graphics processing units (GPUs), originally developed for the game industry, to run their algorithms.
Based on factors such as a patient’s age, gender, medical history, previous medication results, and other information, the system may suggest a treatment — perhaps a set of drugs to try — supplementing the doctor’s own training and experience. “Even for experienced clinicians, it can be quite challenging to figure out which treatment or drug is likely to work best for a given patient,” Sun said. “Using the signals we find in the data sets, we are able to improve the clinician’s accuracy. There can be a dramatic improvement in the disease risk prediction and treatment recommendations.”
Sun’s team has worked with Children’s Healthcare of Atlanta to develop models for treating asthma and medically complex patients using electronic health record information. Sun’s team has also worked with Sutter Health in predicting heart failure, and with Vanderbilt University and Northwestern University on automatic phenotype discovery.
WHAT WORKS FOR WORKERS
Maintaining a healthy and productive workforce is a top priority for many major companies today. One of the leaders of that effort is Southwire, North America’s foremost manufacturer of wire and cable used in electricity distribution and transmission.
With the title “Director of Living Well,” Lisa Evans is responsible for everything that relates to health and wellness for Southwire’s more than 7,000 employees at locations around the country.
EmployersLikeMe, facilitated by Georgia Tech, helps major employers identify trends and best practices for the health care services they provide their employees. The company offers broad-based programs for its employees, and its headquarters in Carrollton, Georgia, includes a pharmacy, wellness center, and fully certified primary care medical facility.
For Evans, the challenge is to measure what works so she can guide the company’s investment into new initiatives. That’s one reason she is a member of EmployersLikeMe, a Georgia Tech-facilitated initiative that brings together the state’s major employers to discuss common issues such as health care – and share data that could help participants identify trends and best practices.
“Our people are what differentiates us in the marketplace, and it’s important for us to create and sustain an environment that allows people to be the best they can be,” Evans said. “We want to make sure we can glean information from our health care data to help us take action or create awareness that could facilitate more healthy behaviors in the people we serve.”
EmployersLikeMe (ELM) began as a roundtable for representatives from top Georgia companies to share concerns and best practices. Because health care is a major cost today — one that is rising faster than other costs of doing business — ELM has moved into health care informatics.
“We are working to get data from dozens of private companies, all of whom have different pieces of data in different formats with different third-party administrators,” said Don Betts, an extension specialist who leads ELM in Georgia Tech’s Enterprise Innovation Institute. “Employers have a lot of data, and their data is from the private-pay, real-world that’s very different from Medicaid or Medicare information.”
Bringing that information together could help employers identify trends affecting their work force and get feedback on programs designed to help prevent and control such chronic issues as high blood pressure, diabetes, heart disease, and obesity — medical concerns that affect most companies across the board.
“The potential and the power that this knowledge could unleash is amazing,” Betts said. “Most of the ELM employers have self-funded health care programs, so they have an interest in looking at how they can improve access, ensure the best quality of care, and test new approaches for having healthy and productive people.”
In non-metropolitan areas of Georgia, finding and retaining the best workers is a challenge, and access to health care can be among the reasons. Wellness initiatives — sometimes as basic as teaching nutrition and exercise — can be part of attracting people and helping them remain productively on the job. The cost of health problems goes beyond medical claims; costs related to lost productivity, retraining, and other issues also add up, Betts noted.
“If there were any other cost that was going up 10 to 15 percent a year, the company would wrestle it to the ground and figure out what to do,” he added. “But health care is seen by many companies as something they can’t do anything about. The ability to get good data could change that. We want to give executive leadership the information they need to make good decisions.”
Besides Southwire, EmployersLikeMe has representation from major Georgia employers including Fieldale Farms, the Georgia Power Company, The Langdale Company, Mohawk Industries, Newell Brands, Procter & Gamble, Rollins, Inc., Sunset Foods, and The Savannah Business Group. In addition to facilitating sharing among the companies, Betts helps them collaborate with outside initiatives in areas such as telemedicine and addressing chronic health issues.
With its statewide extension presence, Georgia Tech is in a unique position to provide real value for these companies, noted Dahl. “What these companies are most interested in is Georgia Tech’s ability, as a neutral and trusted source, to take their patient data and claims data and aggregate it in a way that will inform them about trends and what works,” she said.
DEATH RECORDS HELP THE LIVING
About 2.6 million people die each year in the United States, according to the Centers for Disease Control and Prevention (CDC). For each one, a death certificate is issued listing the conditions and causes of death. Aggregating and analyzing that data not only provides information about the state of public health, but also offers clues that could help clinicians treat the living.
With support from the CDC’s National Center for Health Statistics (NCHS), Georgia Tech researchers are looking at death certificate data both to see what can be learned and to support efforts to create an improved and uniform reporting system. While there is a national standard death certificate, 57 different reporting jurisdictions — corresponding to each state and some municipalities — may collect additional data elements.
“Most physicians do a good job of ascertaining the immediate cause of death, but for public health and planning purposes, we need information about the underlying causes,” said Paula Braun, entrepreneur-in-residence at the CDC. “We are really excited to be working with Georgia Tech to understand the nuances and components that need to come together to make this information more useful.”
May Dongmei Wang, a professor in the Department of Biomedical Engineering at Georgia Tech and Emory University, is leading an effort on using up to 40 years of National Vital Statistics System (NVSS) public use death records data published by NCHS, along with data provided by states, to assist in the design of a next-generation electronic death registration system. The system will be based on Fast Healthcare Interoperability Resources (FHIR), the emerging health data interoperability standard.
The project will advise the CDC on what kinds of data should be gathered in future electronic death records and ways in which technology could assist doctors with a task many of them don’t do often.
“We are looking at novel ways to use technology to inform and simplify this very complicated process,” Braun said. “Death certificate data is one of the best sources of health information at the population level, and it is used in many different policy-setting purposes, so we want to improve this process.”
Researchers also will look into possible disease pathways leading to death. That information could be useful to clinicians in intensive care units, for instance, who often must prioritize care for patients with complex medical conditions, balancing treatment for conditions like congestive heart failure, kidney disease, heart valve problems, and loss of lung function.
Using the public use mortality data from the NVSS, researchers have identified the top 10 causes of death and the many issues associated with the direct causes. “We want to see the associations between the patient’s condition and their final outcome,” Wang said. “The insights we can gain from this data could go back into helping physicians.”
DATA DRIVES HEALTH CARE POLICY DECISIONS
Individual states have significant independence in setting health care priorities and determining how those goals will be met through specific service offerings. But how do policymakers evaluate those decisions?
That’s a question Nicoleta Serban and Julie Swann are addressing by studying billions of transaction records from the nation’s Medicaid program. Medicaid largely addresses the needs of children, whose issues differ dramatically from those of private-sector workers and older Americans. The Georgia Tech researchers have access to eight years of complete records from 50 million patients, 38 million of them children.
“People want to make informed decisions, and we can help them make decisions based on data,” said Serban, an associate professor in Georgia Tech’s Stewart School of Industrial and Systems Engineering. “Health care policies have often been set by clinicians or public health agents who are making decisions based on their experience. We can add a layer of data and health analytics to that.”
For example, in a project supported by Children’s Healthcare of Atlanta, Serban looked at how states in the Southeast were managing asthma in children — a condition identified by the CDC as a priority. With proper diagnosis and management, acute asthma attacks can be minimized.
Based on the Medicaid data, however, she found significant differences among the states in the degree to which the disease was being managed in accordance with guidelines. In the best-performing state, 80 percent of children studied received the recommended care, while in the worst-performing state, only 30 percent of the children studied did.
The implications of care quality are significant and can be measured in costs of emergency treatment, chronic diseases carried into adulthood, time away from work for parents, and the quality of life for affected children. “If we can provide the best care now for these children, we can rely on them as adults to be healthier and more productive,” Serban said.
The findings could spur underperforming states to re-examine their policies, perhaps by looking at states that have done a better job. In some states, care for children with asthma is limited by a shortage of medical providers and practices that accept Medicaid. “We cannot charge either patients or clinicians for lack of adherence,” Serban said. “We are simply looking at compliance within the system and how to intervene.”
Serban and Swann — also a professor in the Stewart School of Industrial and Systems Engineering — have also looked at access to mental and oral health care, access to primary care, the cost-effectiveness of telemedicine, and at the potential return on investment for expanding Medicaid services to cover more preventive care for children. The Medicaid data includes personal information about patients, so Georgia Tech has an extensive protocol for securing it. For instance, the data is maintained on networks not connected to the internet, and can be accessed only in secure data review rooms from which cellphones and other electronic devices are excluded. Only Georgia Tech employees can see the information, and then only for research permitted by the agreement with the U.S. Department of Health and Human Services.
Working out the security infrastructure and agreements for using the data took nearly two years, including the construction of secure data rooms housed at the Institute for People and Technology, a Georgia Tech interdisciplinary research institute.
“When we started, I was a bit naïve about the level of security that would be required to protect these data,” Serban said. “Now that we have established the infrastructure for this database, it is a proof-of-concept for how Georgia Tech can deal with that level of data security and data complexity.”
STUDENTS MOVE INNOVATION INTO THE WORLD
Pushing health care informatics into the nation’s health care system will require more than research. In Georgia Tech’s School of Interactive Computing, faculty members are working toward a course concentration in health informatics, building a group of students who are highly sought-after by organizations now developing and implementing real-world health informatics systems and tools.
“We are well on our way to having a significant concentration with a heavy emphasis on analytics and big data,” said Mark Braunstein, M.D., a professor of the practice who teaches health informatics as an elective graduate seminar in the college and in the Online Master of Science in Computer Science (OMSCS) program.
Using FHIR, Georgia Tech students developed an application to assist the Emory Electronic ICU Center, which monitors intensive care units at multiple hospitals. Shown are Tim Buchman, M.D., Emory’s Critical Care Center director; Cheryl Hiddleson, MSN, director of the Emory Electronic ICU Center, and Georgia Tech graduate student Robert Allen.
Students in Braunstein’s course are the first in any university to work on real-world problems applying HL7’s Fast Healthcare Interoperability Resources (FHIR), a new standard for describing data formats and elements for exchanging electronic health records. FHIR addresses the challenges of interoperability among health information systems, and Braunstein says training students in the new standard will help transfer a new way of thinking into the industry. “The development of FHIR is transforming the whole field because, for the first time, if you can use analytics to develop rules that can help doctors and patients, you can deliver them to where they can be used,” he said. “FHIR allows us to develop applications that can operate essentially as built-in enhancements to the electronic medical records systems that doctors and health care organizations are using.”
One example is a FHIR application developed in collaboration with Emory University clinicians to support advanced patient monitoring at the Emory Electronic ICU Center. The center is linked to intensive care units at Emory Saint Joseph’s Hospital, Emory University Hospital, Emory University Hospital Midtown, East Georgia Regional Medical Center, and Emory Johns Creek Hospital. Braunstein’s students worked with Tim Buchman, M.D., Emory’s Critical Care Center director, to develop and test the application.
At the end of each semester, student projects from Braunstein’s class are presented in a briefing attended by health care industry representatives from as far away as California. “People want to partner with Georgia Tech because we have the knowledge and skills, our faculty are willing to work on practical problems, and we have great students,” Braunstein said.
Health informatics applications couldn’t operate without extensive and secure infrastructure. Myung Choi and Richard Starr work behind the scenes for all these projects as part of Georgia Tech’s Institute for People and Technology (IPaT) and the Interoperability and Integration Innovation Lab (I3L). They make sure data storage is available, provide security, help faculty members navigate approval processes, and make sure research projects comply with data use agreements and federal rules such as HIPAA.
“We must have infrastructure to connect systems, a protocol that allows them to communicate with each other, and data content that is in the same language,” said Choi, a senior research engineer in the Georgia Tech Research Institute (GTRI). “My part is to make sure that all those systems work together.”
Humans can understand data structures that are similar, but machines must have exact matches. Part of Choi’s job is to help researchers structure the data they’re using to meet the needs of individual projects. “We are trying to free the data so researchers can do what they need to do,” he said.
Starr, a research scientist in IPaT, considers himself an enabler for researchers. He helps them work through institutional review board regulations, data use agreements, and protection requirements. Most data has been de-identified to remove information that could lead directly to patients, but savvy programmers might be able to combine information from different sources to restore identities. Part of Starr’s job is to make sure that doesn’t happen.
“We can take the steps required to get data in less time if we prepare in time,” he said. “From the time we get data from our research partners until the time it’s ready for researchers to use, there’s a lot of work to be done. Every data set we get has to be maintained in its own environment.”
Starr and Choi operate a health informatics data hub in which information can be stored centrally. That lifts the data management and security burden from researchers, and allows them to know about specific data sets and the restrictions on using them, potentially accelerating projects.
Before becoming involved in health informatics, Choi worked with secure computer network communication systems and simulations for Department of Defense customers. Like many others in the group, he heard a presentation made by a patient who described her struggle to overcome a disease.
“This helped us to understand that we were doing a good thing,” he said. “I feel like I am making an impact.”
The growth and application of health informatics faces challenges as great as its promises. For instance, most data consists of information collected for other purposes, either for submitting claims to insurance companies or maintaining health records to help clinicians treat patients.
In addition, the data contains errors — known as “noise” to researchers. Data fields may not match, and there may be duplicate information that could skew results. And the data is maintained in perhaps hundreds of different proprietary software systems that aren’t designed to talk with one another. “Taking these disparate information sources and systems, and putting them together in a way that is computable, is one of the challenges we are tackling,” Dahl said.
Beyond working with existing information, Georgia Tech researchers are looking ahead to integrate genomics data, which could provide a more complete description of each patient. Doing that will require a new set of collaborators at Georgia Tech, Emory University, and other leading institutions.
“THE DEVELOPMENT OF FHIR IS TRANSFORMING THE WHOLE FIELD BECAUSE FOR THE FIRST TIME, IF YOU CAN USE ANALYTICS TO DEVELOP RULES THAT CAN HELP DOCTORS AND PATIENTS, YOU CAN DELIVER THEM TO WHERE THEY CAN BE USED.”
One major project underway, led by Rushing, supports the U.S. Department of Veterans Affairs (VA) as it plans a new digital health platform (DHP) to replace its pioneering electronic health records system — the country’s first. As a neutral third party, Georgia Tech is building the VA a scalable, proof-of-concept DHP model.
According to Rushing, the architecture for the new platform relies on FHIR as well as other Internet of Things standards within an open application program interface gateway. This architecture opens the platform for care delivery and operations process innovations made possible by flexible and agile interoperability with future technology innovations. The DHP is intended to be as groundbreaking now as the VA’s VistA electronic health records system was 40 years ago.
“The VA is looking for a flexible, future-focused health platform and architecture focused on a services-based model,” said Duke, whose research center is funded jointly by GTRI and the College of Computing. “They are taking a really ambitious approach to it, and this could have a tremendous impact on care for veterans as well as on health systems more broadly.”
The project illustrates the role Georgia Tech can play because it doesn’t have a medical school or hospital and isn’t tied to any specific technology platform.
“When we are working with health systems, we’re neither a competitor nor a vendor,” Duke noted. “We approach each problem from a ‘white hat’ perspective, working to find the right data and infrastructure needed, often using open-source platforms.”
Health informatics is also bringing Georgia Tech units together with industry and government to work on common issues in unique ways. The new Coda building, a mixed-use facility featuring a high-performance computing center scheduled for construction in Georgia Tech’s Technology Square, will bring industry and academics together to help spur the health care revolution many are expecting.
“I see that building as a hub, with the spokes extending out to all the major care delivery systems in the region that are part of a collaborative health research network,” Braunstein said. “You cannot find a more multidimensional space than health care.”
John Toon is editor of Research Horizons magazine and director of research news at Georgia Tech.