News Public Health Data & Training Center

SPOTLIGHT! Center collaborator uses machine learning to improve how we look at firearm injury data

Written by Kim Furlow, communications manager, Institute for Public Health


Rachel Ancona, PhD
Assistant Professor, Department of Emergency Medicine
WashU School of Medicine in St. Louis

When you read Rachel Ancona’s bio, you’ll notice that she includes the fact that emergency departmentsor EDs as they are more commonly known by those who work there—are often the only way some people access medical care. “This makes EDs an ideal setting for many public health interventions”, the bio reads. As an assistant professor in WashU’s School of Medicine, Ancona’s research interests include making sure interventions are pragmatic. She also assesses the best ways to leverage electronic medical record data and machine learning, a technology where computers learn from data to make predictions or decisions, to identify patients with unmet medical and/or social needs and link them to care. We caught up with this dedicated professional and collaborator with the Public Health Data & Training Center (Data Center) at the Institute for Public Health.

What is the focus of your latest research?

Firearm injuries are a significant public health concern in the United States, with a high risk of recurrence among survivors.

My latest research focuses on addressing the critical challenge of accurately classifying new firearm injury encounters in hospital data. One of the current problems in this line of inquiry is differentiating between new injuries and follow-up encounters for prior injuries, which is essential for understanding the true incidence and recurrence of firearm injuries.

In our study, we developed and validated a supervised machine learning model aimed at improving the precision of identifying new firearm injury encounters in hospital data. By comparing the performance of this machine learning model with traditional methods, such as using emergency department visits, time-based cut-offs (the number of days between successive firearm injury encounters), and ICD-10 coding (a standardized system of codes used to classify diagnoses and medical procedures), we demonstrated that machine learning can provide a more accurate classification, particularly in ruling out follow-up visits.

The primary beneficiaries of this research are researchers and public health professionals who are working on firearm injury prevention and intervention strategies. By improving the accuracy of injury classification, our work can lead to better-targeted intervention programs, more reliable epidemiological studies, and ultimately, more informed policies to reduce firearm-related harm. Additionally, the findings of this study have implications for enhancing the evaluation of hospital-based violence intervention programs, which are critical in mitigating the cycle of violence in communities, such as the program highlighted in this study, Life Outside of Violence (LOV). The LOV program is a collaboration of four Level I trauma centers in the St. Louis region working with patients to prevent firearm injury recurrence.

How have you engaged with the Data Center and in what way(s) does this collaboration benefit your work?

I have collaborated closely with the Data Center to develop not only the firearm injury model but also a comprehensive classification model for all violent injuries included in the LOV program (blunt, penetrating, and firearm injuries). These models have significantly enhanced the precision of new injury estimates, contributing to several published manuscripts and others currently in development. Moving forward, we plan to continue this fruitful collaboration by exploring additional machine learning and artificial intelligence approaches to further refine and improve violent injury classification.

Collaborating with the Data Center has been an exceptionally positive experience, particularly due to the outstanding leadership of co-Director, Ben Cooper. Ben and his team maintain an administrative dataset of the highest quality, marked by meticulous attention to detail and adherence to best practices in data cleaning and management.

Researchers and faculty who engage with the Data Center can be confident in the top-notch quality of data they will use to address critical public health issues.

The team at the Data Center is not only highly knowledgeable and competent, but also exceptionally approachable and collaborative. They truly listen to and engage with their partners, elevating the quality and impact of any investigator’s work.


An abstract from the article, “Acute and Recurrent Firearm Injury Rates in an Urban Population (2010-2021): Using Machine Learning to Improve Classification” by Rachel Ancona, PhD; Ben Cooper, MPH; Kristen L Mueller, MD, et al…was recently published in The Journal of the American Medical Informatics Association. Read their full article here.