As a pediatric anesthesiologist, Dyann Daley, MD, sees her share of injured kids. But one case in 2012 changed her life.
“There was a little boy who had a ruptured liver. [We took him] to the operating room, not knowing what the injury was, and he exsanguinated, he bled to death, on the table,” she said.
The toddler’s injury had come from child abuse.
“It was really hard to imagine how someone could take a toddler’s life like that and just throw it away,” she said.
So Dr. Daley, who worked for Cook Children’s Medical Center in Fort Worth at the time, began looking into ways to prevent child abuse. At first, she focused on training other physicians to recognize signs of child abuse. Her work led her to geospatial risk analysis — crunching statistics to predict areas where crime most likely will occur.
Many big-city police departments already use this type of analysis, called “risk terrain modeling,” to predict violent crimes like shootings and robberies. Dr. Daley worked with Michael Bachmann, PhD, an associate professor of criminal justice at Texas Christian University, and Joel Caplan, the creator of Risk Terrain Modeling software, to use this tool for the first time to predict hot spots for child abuse and neglect. Their model relied on publicly available data on domestic violence arrests, aggravated assaults, as well as the locations of potentially risky establishments, such as laundromats and bars.
The map Dr. Daley helped produce in 2013 predicted Fort Worth’s child abuse and neglect cases with 98 percent accuracy. In other words, only 2 percent of the actual cases were outside the predicted areas. In a peer-reviewed article she co-wrote for the journal Child Abuse & Neglect (tma.tips/DaleyStudy), Dr. Daley says this approach has several advantages over other methods: It is more accurate, it helps prioritize responses, the data are relatively simple to obtain, and it is adaptable to changing conditions.
Dr. Daley has since left Cook Children’s to set up her own nonprofit, Predict-Align-Prevent. The organization is currently working with cities in Virginia and Washington, and Dr. Daley is in talks with other locales as well.
Carl Ayers, director of Family Services at the Virginia Department of Social Services, says the Virginia project is in its early stages. But he says being able to scientifically identify areas most prone to child abuse and neglect means the state can focus its scarce resources more efficiently. Those resources could include anything from conducting an advertising campaign to launching neighborhood programs or improving regional health care.
“I can’t emphasize enough to you: This is a new area of child welfare,” Mr. Ayers said. “This is new ground that we’re blazing.”
Dr. Daley collaborates with Ken Steif, PhD, director of the Master of Urban Spatial Analytics program at the University of Pennsylvania. He says many of the companies that help police departments with geospatial work use data modeling that is both expensive and proprietary. On the other hand, his work with Dr. Daley is designed to empower communities to learn how to find locations with higher risk for child abuse and neglect.
“The long-term goal of our work is to put these algorithms out there open source so that other communities can start developing them on their own,” Dr. Steif said.
Dr. Daley still works as a pediatric anesthesiologist, though she does it now as a locum tenens to free up time for her nonprofit work, which Texas Medicine talked to her about more in-depth.
After that toddler died from the ruptured liver, what did you do?
I learned after taking care of him that it’s very likely he had a previous injury that was either missed or not reported by another medical provider. So often children who come to the hospital with terrible or fatal maltreatment (abuse or neglect) injuries had a sentinel injury before that was missed. Recognizing that specialists receive minimal training in child maltreatment, the physiology of toxic stress, and adverse childhood experiences, it seemed like an opportunity to develop some training for medical professionals, which we did. (See “What You Need to Know About ACEs,” September 2017 Texas Medicine, www.texmed.org/WhatToKnowAboutACEs/.)
How did you move from that to using statistics to predict future neglect and abuse?
Zero- to 3-year-old children, who are the most likely to die from abuse and neglect, are easy to miss. The American Academy of Pediatrics has been focusing on screening individuals and families for exposure to adverse childhood experiences, but what if a vulnerable child never goes to primary care? So instead of screening individuals and families, I thought, what if we look geographically at where adverse childhood experiences and other social determinants of health coexist spatially and see what’s happening to the children there?
What did you find?
What we found was that we could predict where about 98 percent of substantiated child maltreatment cases would occur in the city of Fort Worth. While the areas in general were not surprising to police officers and social workers, the correlated risk factors were. Ranked in order of relative risk, the most important contributors to risk were the density of domestic violence arrests, runaways, aggravated assaults, and sexual assaults. If there is a high density of those four things in a 400 x 400 square foot area, that’s where more than half of the following year’s substantiated child maltreatment cases will occur. This was surprising because so much focus is on poverty. Many of these areas are also impoverished, but poverty casts a net very wide, and by looking at where the violent crimes occur along with poverty you can increase your predictive accuracy.
How does it help to map these areas?
What the mapping really does is help to focus where to prioritize resources, what to focus resources on, and also how to align existing programs and services in the places where they’re likely to do the most good. The maps also provide a way [to visualize how] all the different, but geographically-related initiatives that cross sectors [interrelate], and how collaboration can be cost saving while improving program effectiveness.
Do you know yet if this actually brings down rates of child maltreatment?
No. It hasn’t been quite long enough yet. It’s really hard to say [how long it will take] because we as a society don’t yet know what the components of a primary prevention bundle look like — a primary prevention bundle of services and supports that reliably and repeatedly prevent child maltreatment. As we model and align resources in multiple cities and states, what we’re seeking are areas where the alignment of resources actually reduces child maltreatments. Once we find bright spots [of positive change], we’ll figure out how those communities are different and replicate what they are doing elsewhere.
Predictive analytics also has been used by police departments to predict crime rates and distribute resources. Has that caused problems for you?
Our work is about allocating resources in the most effective way to the people who need them the most. That is an entirely different focus than using predictive analytics for [law enforcement], to punish people, which is what people are used to hearing about.
Do you run into problems collecting so much data?
We’re looking more at how to change the environment to support behaviors that are healthier and safer for children. All we use is location data. Since we are not seeking data such as names, birthdays, or social security numbers, we have an easier time accessing data.
How is your approach received among those who work with abused children?
Whenever I talk to people, my most favorite compliment is, “You mean we’re not already doing that?” — because it’s such a logical progression to do this work in this way. People find hope in this. Our work helps to allocate resources in a more efficient and effective way and adds a layer of accountability by monitoring population-level health and safety outcomes to see if the prevention efforts are working.
Tex Med. 2018;114(7):10-11
July 2018 Texas Medicine Contents
Texas Medicine Main Page