Artificial intelligence improves accuracy in estimating child abuse

Artificial intelligence improves accuracy in estimating child abuse | Image Credit: © mihakonceptcorn - stock.adobe.com.

A new study reveals that artificial intelligence (AI) can significantly improve the accuracy of estimating child physical abuse (PA) among children seen in emergency departments. The research, which will be presented at the Pediatric Academic Societies (PAS) 2025 Meeting in Honolulu from April 24-28, demonstrates that a machine-learning model offers more precise predictions of abuse rates than traditional methods relying solely on diagnostic codes.

Current coding practices and their limitations

The study emphasizes the limitations of using the International Classification of Diseases, 10th Revision, Clinical Modification (ICD-10-CM) codes for determining the prevalence of child physical abuse, particularly within emergency department settings. These codes, commonly used to track and classify diagnoses, may not provide a complete or accurate picture of PA incidence. Researchers suggest that “consideration of injury codes along with abuse-specific codes may enable more accurate PA prevalence estimates”. The reliance on abuse codes alone led to an average misdiagnosis rate of 8.5% in the study.

AI-driven approach for enhanced accuracy

To address the shortcomings of current coding practices, researchers developed a machine-learning model to estimate instances of child abuse. This model analyzed diagnostic codes for both high-risk injury and physical abuse to generate more accurate predictions. The AI-driven approach was compared against estimates derived exclusively from diagnostic codes entered by providers or administrative staff.

Study data and demographics

The study examined data from 3,317 emergency department visits related to injury and abuse across seven children’s hospitals between February 2021 and December 2022. The research focused on children under the age of 10, with nearly three-quarters of the children being under two years old. The median age of the children involved in the study was 8.4 months, with 59% being less than one year old.

Key findings and implications

The findings of the study indicate that the machine-learning model provides a more accurate estimation of PA prevalence compared to estimates based solely on abuse-specific codes. The AI approach reduced errors in prevalence estimation, demonstrating its superiority in identifying child abuse cases. In contrast, estimates relying only on abuse codes overestimated prevalence, with estimation errors ranging from 2.0% to 14.3%. The predictive models showed reduced errors, ranging from -3.0% to 2.6%.

“Our AI approach offers a clearer look at trends in child abuse, which helps providers more appropriately treat abuse and improve child safety,” said Farah Brink, MD, child abuse pediatrician at Nationwide Children's Hospital, and assistant professor at The Ohio State University.

Potential to revolutionize data interpretation

The study suggests that AI-powered tools have the potential to “revolutionize how researchers understand and work with data on sensitive issues, including child abuse,” according to Dr. Brink. By providing a more accurate understanding of child abuse prevalence, this research can inform clinical practice, improve intervention strategies, and ultimately contribute to better outcomes for children.

References:

1. Brink F, Lo CB, Rust SW, et al. A machine learning approach to improve estimation of physical abuse. Abstract. Presented at: 2025 Pediatric Academic Societies Annual Meeting. Honolulu, Hawaii.

2. Pediatric Academic Societies. Study: Artificial intelligence more accurately identifies child abuse Pediatric Academic Societies. Press release. April 25, 2025. Accessed April 25, 2025.