MD2K Center Gathers Reliable and Consequential Health Information Using Mobile Sensors

The National Institutes of Health signaled a role for mobile devices in health research when, in September 2014, it awarded a $10.8 million, four-year grant to establish the Center of Excellence for Mobile Sensor Data-to-Knowledge—or MD2K—for tools to gather and interpret health data generated by wearable sensors.

Mobile sensor data sources used by MD2K researchers. Credit: MD2K, University of Memphis

MD2K, headquartered at the University of Memphis, is a consortium of 12 universities and university medical centers—with 20 investigators whose research spans computing, engineering, and statistics and biomedical research.

The center is one of 13 created by the NIH Big Data to Knowledge (BD2K) program, and one of four supported through the National Institute for Biomedical Imaging and Bioengineering (NIBIB). BD2K is a trans-NIH initiative to harvest the wealth of information contained in biomedical big data. Included in this sweep of information are vast sets of genomic code, clinical data accumulated from such cohort studies as the 65-year-long Framingham Heart Study, and immense sets of data points from imaging scans. Also included are data available through new and rapidly evolving wearable sensors that MD2K is poised to turn into reliable and consequential health information.

“The MD2K center sprang out of previous NIH funding to develop a chest-based sensor system,” said Richard Conroy, Ph.D., director of NIBIB’s Division of Applied Science and Technology, who manages the grant issued to MD2K. The sensor system development was funded by NIH’s Genes, Environment and Health Initiative (GEI), launched by NIH in 2006. “They showed that they could develop that technology as well as the software to go with it to collect relevant physiological measures for stress and addictive behavior assessment in the mobile environment.”

“MD2K is about developing the technology, capability, and science for undertaking large-scale clinical trials involving mobile sensors,” said Santosh Kumar, Ph.D., center director and the Lillian and Morrie Moss Chair of Excellence Professor at the University of Memphis. “By releasing our software and models, we are enabling others to build on what we have done rather than having to start from scratch.”

An open-source software approach can accelerate progress in the data science community, according to Kumar. He also says that open-source software for handling high-frequency mobile sensor data—which is free for use in commercial products—can help the industrial sector to reduce the time to market, improve future products, or create new products, all of which benefits end users.

The four-V nature of MD2K data

MD2K is collecting data that has high velocity, variety, volume, and veracity. Velocity pertains to the speed of information being streamed and transmitted. The MD2K researchers must decide whether the system will process data on the mobile device or at the warehouse, for instance. They must factor security into each step in the process.

Data variety is achieved when multiple sensors measure different output. “That could be everything from the radar system that measures fluid in your lungs, to skin temperature, your GPS location, and how to align all of those into the system,” Conroy said. “Results vary if you measure blood pressure in your arm, leg, or neck.”

Continuously generated measures produce high data volume. “At this point, MD2K will err on the side of measuring everything and will work out later what’s the most important,” Conroy said.

Veracity is a characteristic of data that acknowledges uncertainties that must be accounted for, including biases and other types of data noise that must be filtered as part of analysis.

The four-Vs apply broadly across all data collected through the 13 BD2K centers, not just to MD2K data. “While there are challenges with such data, there are also unprecedented opportunities for improved healthcare, which MD2K is on the leading edge of exploiting,” says Phil Bourne, Ph.D., associate director for data science at NIH, and the lead on the BD2K project.

Smoking cessation and congestive heart failure serve as use cases to pilot MD2K methodologies

The first areas targeted for MD2K investigation use sensors to measure behaviors and symptoms in both cardiovascular health and smoking cessation. Five types of mobile sensors are used for data collection. A chest strap, called AutoSense, measures heart rhythm, respiration and physical activity. Hearth rhythm and respiration data are indicators of stress; respiration data can detect inhalation. A motion sensor worn on the arm tracks movements typical of either smoking or eating. A third mobile device non-invasively measures heart activity and volume of lung-fluid, indicators of congestive heart failure. Smart eyeglasses record an individual’s surroundings, which might signal visual exposure to smoking triggers, such as cigarette advertising. The fifth source of sensor data in the MD2K toolkit is the global positioning system in a smart phone, which might indicate proximity to particular vendors of tobacco or fast food.

The researchers have conducted preliminary tests and will enroll 225 study participants in separate three-year studies to gain improved understanding of congestive heart failure and smoking cessation. Different teams of researchers will contribute to processing, representing, and analyzing the health data. Their goal is to measure physical, biological, behavioral, social, and environmental factors that together are reliable in predicting either risk of heart failure or resumption of smoking behavior.

To reach that goal, MD2K researchers must develop and test software systems for handling the wealth of information embedded within the sensor feeds. “The question is then, with all this data, what do we do with it and how can we learn from it?” said Conroy, adding that MD2K is pursuing a dual approach of information-capture and data warehousing, all of which is built into a modular system that others can use in the future. “The two use cases will demonstrate the versatility of software they’re trying to design,” Conroy said. “The software development side is the core of this center and will be open access and have a modular design.”

Another innovation to be pursued by the researchers is just-in-time intervention. MD2K’s smoking cessation study will take next step beyond collecting data and will try to learn when an individual is about to smoke or not, so researchers can help a participant abstain from smoking.

Attempting intervention is unusual for this type of research, according to Conroy. “Typically, the way we set up clinical trials is to administer a pill once a day,” Conroy said. Observations of clinical outcomes of the intervention are recorded at that point. “We don’t think about clinical trials in terms of an intervention being triggered by an event.” The just-in-time intervention clinical trial design will be a key contribution of the MD2K center’s innovative approach.

Smoking is the leading cause of preventable death in the world and has been one of the hardest to solve because the cause and effect are not that immediate, according to Kumar. “If someone smokes a cigarette today it has a cumulative effect; the person doesn’t see the adverse effect right away,” he said. “There are reasons to believe that providing timely, sensor-driven interventions may make a difference to outcomes to this deadly disease. Those just-in-time, momentary interventions have a potential to not only reduce smoking lapses or improving smoking cessation, but also be applicable to other adverse behaviors such as reducing impulsive eating or drinking.”

During the upcoming smoking cessation study, researchers will make use of a combination of smartphone apps for just-in-time stress intervention. Thought Shakeup is an app that researchers at the University of Michigan designed and that MD2K researchers developed to improve the participant’s thought processes by asking several guided questions. Mood Surfing, also designed by the University of Michigan collaborators and developed by MD2K researchers, guides the participant through three brief exercises to counter the effects of stress. Users are asked questions following the exercise about whether it helped alleviate their stress. During the study, participants will receive three or four stress interventions each day, depending on their stress level. MD2K researchers may add to the number of just-in-time interventions and are fine tuning the timing of interventions to assist the participant in efforts to avoid smoking relapse.

Early MD2K milestones

One year into its work, MD2K has achieved several milestones along its ambitious road to the use of mobile sensors in health research and practice. This past October the researchers published a study of 61 smokers who wore sensors for three-days. A computational model they had created, called puffMarker, collected arm gesture and breathing pattern data from separate wearable sensors that accurately detected the first smoking lapse in 28 of the 32 participants who resumed smoking.

As outreach to other mobile sensor investigators, MD2K launched a website this past November called mHealthHub to exchange research information; meeting and event notices; releases of software, data, or devices; and funding opportunities. Found at, the site includes a discussion forum for registered users on deploying mobile sensors and developing open-source software, along with training materials. In future, it will offer curricular materials for use in classrooms, in both data science and biomedical disciplines.

Each year, a select group of postdoctoral candidates are trained in mHealth during a one-week summer offering, called the mHealth Training Institute. The goal of these efforts is to build the next generation of scientific workforce that is capable of using mobile health technologies in increasingly sophisticated biomedical applications. The videos of the mHealth Training Institute courses are available via mhealthHUB for broad dissemination.

In planning for the future of mobile big data, MD2K is developing a collection of online tutorials, training videos, virtual seminars and a comprehensive web-based resource library all of which will be made available via mHealthHub.

In April 2016, MD2K plans to release mobile apps to familiarized researchers with the mobile sensor software code that others in the data science and health communities can adapt it for their particular system and application. MD2K researchers will release the source code for their software platform that can collect data from sensors and the companion software to do the analysis and interpretation, and to carry out just-in-time interventions. The free, open-source code ( will make it easier for researchers to collect, analyze, visualize, interpret, integrate, and share mobile health data.

Collaboration within MD2K and among BD2K Centers

“As a center, they’ve done a really good job,” Conroy said of MD2K’s ability to assemble and coordinate a large team of diverse researchers. Data science investigators are developing software and coming up with algorithms that measure behaviors. Engineers are focused on technology development for devices that are biocompatible and durable. Behavioral scientists are considering behavior modification paradigms. “Each has their way of doing things; they all speak different languages,” Conroy said. “It all ties together into thinking about how to use these technologies, whatever the behavior, for just-in-time interventions and how to intelligently analyze the information.”

MD2K, like all the BD2K centers, are encouraged to collaborate. With the ENIGMA Center for Worldwide Medicine, Imaging, and Genomics, also funded through NIBIB, the focus is on brain imaging. Together, the two centers study what brain imaging, when combined with mobile sensor data, can reveal about human behaviors. With the MOBILIZE National Center for Mobility Data Integration to Insight, MD2K is collaborating on sensors to measure mobility. According to Grace Peng, the NIBIB program officer who manages the grant for MOBILIZE, “They are trying to combine physiological measures and movements associated with particular activities with general physical activity.” MD2K also collaborates with the Center for Big Data Translational Genomics to use sensors to measure everyday activities of a large number of subjects whose genomes are sequenced as part of the 1000 Genomes project.

The MD2K team consists of researchers from Cornell Tech, New York City; Georgia Institute of Technology, Atlanta; Northwestern University, Evanston, Illinois; The Ohio State University, Columbus; Rice University, Houston; the University of California, Los Angeles; University of California, San Diego; University of California, San Francisco; the University of Massachusetts, Amherst; the University of Memphis, Tennessee; the University of Michigan, Ann Arbor; West Virginia University, Morgantown; and Open mHealth, a nonprofit organization.


Tell Us What You Think

Do you have a review, update or anything you would like to add to this news story?

Leave your feedback