Each year, about 125,000 children in the United States experience physical abuse – and those are only the substantiated cases. Children under age 1 are the most vulnerable; they’re twice as likely to be physically abused as other children and three times more likely to die from the abuse.
Tragically, in about one in three cases, children who sustain serious abusive injuries have had previous, usually lesser injuries – certain bruises, broken bones, mouth injuries, and burns – that could have raised red flags that the children were being abused, but those warning signs were missed and the children weren’t protected.
In many cases, these children were seen at emergency departments and urgent-care clinics, yet the young patients left without an abuse evaluation, says Daniel Lindberg, MD, a professor in the University of Colorado Anschutz Department of Emergency Medicine.
“When you look at kids who have the most severe, devastating forms of abuse, and you look backwards, often you see that they've come to emergency departments, and sometimes they have things that could have been an opportunity to pick up what’s happening,” says Lindberg, a member of CU Emergency Medicine’s Family Violence Recognition Lab and a clinical researcher into improving care for children experiencing abuse.
“We see this as one of the biggest opportunities to prevent lifelong disability or death in the youngest kids. If you make a mistake in a 9-month-old, and they spend decades requiring special equipment or 24/7 care, it's a major loss. It’s a major win when you can find this early.”
That’s why Lindberg is leading an ongoing research project into using an artificial intelligence tool to improve early recognition of child physical abuse in emergency and urgent care settings. The project is supported by a major grant from the Eunice Kennedy Shriver National Institute of Child Health and Human Development at the National Institutes of Health.
→ Daniel Lindberg, MD, Aims to Reduce Elder Abuse through Research on the ‘Whole Family’ Approach
Warning signs of physical abuse can be missed in the ED for a number of reasons, Lindberg says.
For one thing, cases involving abuse are relatively rare in any single emergency department, especially those that don’t specialize in pediatrics – about 16 cases per 10,000 visits. Also, abusive parents or other caregivers often mislead providers about the causes of injuries, or may not know if someone else is abusing the child. The guidelines for spotting suspicious injuries are complex – a bruise on the forehead of a 3-month-old means something different than the same bruise on a 9-month-old. Emergency clinicians are busy, interventions to detect abuse can be resource intensive, and racial and cultural biases can come into play.
There have been many efforts to improve rates of detection by clinicians of abuse in child patients, including enhanced training, use of tools known as clinical decision rules to calculate the risk of abuse, and guidelines by professional societies, but the problem remains.
At the same time, Lindberg says, there is concern about the need for high accuracy in detecting abuse so that families aren’t erroneously reported to a child protective services agency when abuse didn’t occur.
“Abuse can cause a skull fracture, but there are lots of infants who have skull fractures and it's not abuse,” he says. “If we end up reporting to child protective services every kid with a skull fracture, it's not always going to be the right thing for kids and for families. We could end up causing more harm.”
Lindberg’s project uses natural language processing to analyze clinical records as a way to improve early recognition of child physical abuse. The screening tool, developed in collaboration with colleagues at Yale University, searches for “sentinel injuries” – a combination of specific injuries and patient age groups – that evidence shows are highly indicative of abuse.
Natural language processing is a branch of artificial intelligence that enables computers to read and interpret human language and text – in this case, the notes that providers enter into patients’ electronic health records. The screening tool looks for specific injury descriptions that research has shown are associated with abuse in infants at specific ages. It does this in the background, without burdening clinicians.
“A computer in the background is watching you type your note on a patient, and when you type ‘bruise on the soft portion of the cheek,’ it knows if you're typing that for a 9-month-old or a 9-year-old, and if it’s a 9-month-old, it can prompt you: ‘Hey, do you want to think about abuse?’” Lindberg says.
In a preliminary test at EDs at a single health care system in Connecticut, the natural language screening tool was 93% accurate in flagging injuries indicating physical abuse. Now, Lindberg and his collaborators will test the tool in Colorado. Researchers want to see if the tool works as well in another state because of regional variations in language and terminology used in records.
The project will evaluate records on more than 100,000 emergency and urgent care visits from 2017 to 2023 at 49 UCHealth sites and eight Children’s Hospital Colorado sites for children under 12 months old to identify those with high-risk injuries. Several thousand of the flagged medical charts will be reviewed by humans to confirm that the screening tool is at least 90% accurate.
The study will also examine how many of the children flagged by the screening tool as having suspicious injuries were missed by providers, and how many of the flagged children were later reported to the state’s child protective services agency as having suffered abuse.
If the project is successful, Lindberg and his colleagues hope to distribute the screening tool widely, with the potential of preventing escalating abuse, permanent injury, or death for hundreds of children each year.
Lindberg notes the pioneering role of the CU Anschutz School of Medicine in child prevention and awareness. In 1962, C. Henry Kempe, MD – then chair of the school’s Department of Pediatrics, published the paper “The Battered-Child Syndrome,” a groundbreaking study on what had been a seldom recognized issue.
Kempe’s work led to child abuse reporting laws in all 50 states, and the Kempe Foundation was established in 1976 to lead fundraising, awareness, and advocacy efforts for abused children. Lindberg is a member of the Child Protection Team at the Kempe Center for the Prevention and Treatment of Child Abuse and Neglect on the CU Anschutz campus.
In addition to helping children, Lindberg hopes his study helps address a nagging concern among emergency providers.
“As an emergency physician, one of the scariest things is that a child you saw a few days ago had something you missed that was a bad sign – certain subtle findings, like bruises, oral injuries, or broken bones; or things that sometimes you don't even see when the child is getting an x ray for pneumonia, for example,” he says. “When we see kids with those injuries, and if we can recognize those that are from abuse, we can protect those kids.”
It’s a field that Lindberg has been interested in since midway through medical school. “I wanted to do emergency medicine and research, but I didn't have a specific area of focus. Then one of my professors gave a talk on child abuse pediatrics. It was a really good talk, and I asked myself, ‘Do you think this is something you could do for the rest of your career?’ And the answer was yes, and ever since it’s always been yes. It’s something I really enjoy focusing on, and I feel like hopefully we can make a difference.”