SMART Scatter Fuses Local Data to Identify Domestic Violence Risk

February 2019
Topics: Health, Health IT, Economics, Social and Behavioral Sciences, Social Behavior, Sociology, Mathematics, Probability and Statistics, Modeling and Simulation
Our health is driven largely by local behavioral, social, and environmental factors. MITRE and Virginia Tech are exploring whether linking local, state, and federal government administrative data can identify the risk of domestic violence.
Woman alone outside

"Think globally, act locally." That has been a longstanding concept for improving the environmental health of our planet. Yet, according to MITRE's Dawn Heisey-Grove, many aspects of the same concept apply to our personal health, as well.

"The vast majority of a person's health is driven by local behavior, social, and environmental factors, but most local governments cannot analyze the relevant data effectively," Heisey-Grove says. She points out that while there may be global factors involving everything from alcoholism to tick-borne diseases, people suffer these ills within their local communities—and the local communities are responsible for helping them.

Heisey-Grove is the principal investigator of a MITRE innovation project that fuses local and national data to help local communities identify potential health risks, with the long-term goal of helping them take preventative steps. This concept is in keeping with MITRE's mission of solving problems for a safer world—including our belief that data is the next medical innovation in health.

She and simulation modeling engineer Kevin Gormley are working with Virginia Tech's Social and Decision Analytics Laboratory to explore how local administrative data from multiple entities within Arlington County, Virginia, can be used to enhance public health surveillance and inform policy decision-making. They are first focusing on domestic violence, which is typically a highly underreported problem.

How "SMART Scatter" Can Identify Potential Risks

Gormley explains the concept of "Simulated Multivariate Adaptive Regression Technique (SMART) Scatter" as a way of imputing (estimating missing data) risk factors using data from multiple geographic levels.

"You may have some data at the county level, other data at the ZIP code or Census Block Group level, and other data at the household level," he says. "By leveraging all this information on how risk factors vary together, better estimates can be made for risk factors including unemployment, substance abuse, income, crime rate, and more—enabling more accurate predictions at levels within a community that are more relevant to policy makers, such as school districts."

Heisey-Grove says, "We are careful to always recognize these indicators do not suggest that any particular individual is at greater risk. Rather, that these data points show risk potentials within small populations, possibly within a few blocks or a zip code."

Being Preventive Rather than Reactive

Heisey-Grove believes these tools can help local governments expend their limited resources and make better decisions by being preventative rather than reactive to these health problems.

"Armed with this kind of information, communities could focus on programs shown to be effective in reducing the odds of domestic violence and other problems."

Heisey-Grove presented this project as part of a panel about social determinants of health at the 2018 Health Datapalooza in Washington, D.C., to a standing-room audience. "Clearly people were very interested in figuring out ways to leverage the social determinants data to improve public health surveillance and population health."

Virginia Tech recently performed simulations that showed SMART scatter was better at estimating risk factors at a much more geographic-specific level than has been possible previously, enabling more accurate predictions of domestic violence events. This capability will allow for better targeting of interventions and understanding of where limited public health resources may best be allocated.

“SMART Scatter augments the tried and true method of multiple imputation with additional data sources and geography to give us additional resolution at the neighborhood level,” Professor David Higdon says.

Looking ahead, Heisey-Grove says she would like to offer communities these tools at a "plug and play" level along with recommended interventions.

"It doesn't mean you must go in and announce, 'Our data says your neighborhood is at risk of domestic violence.' Rather, community leaders might recognize the potential and choose to invest their limited funds on job counseling, substance abuse treatment, and domestic violence prevention and awareness programs."

Additional Applications of Social Determinants of Health

Heisey-Grove says there are many other possible applications for the work MITRE and Virginia Tech are doing. For instance, residents close to wooded areas may be at greater risk of tick-borne illnesses, or those who live in "food deserts" may be more susceptible to obesity, and so on. But she feels that domestic violence is the right place to start.

"Given how domestic violence tends to escalate over time and in silence, identifying the risks early has the potential to reduce or prevent a great deal of suffering."

On Wednesday, February 13, Dawn Heisey-Grove and Kevin Gormley will discuss "Leveraging Local Data for Public Health with SMART Scatter" at HIMSS19, the world's largest conference for health information and technology professionals.

—by Bill Eidson 

Explore more at MITRE Focal Point: Health.


Publication Search