Peer-reviewed article
Pediatric firearm risk prediction in trauma centers and after discharge: A machine learning analysis.
Craig D Newgard, Sean Babcock, Amber Lin, Susan Malveau, Jason E Goldstick et al. ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This study used machine learning to predict which injured children and adolescents treated in U.S. trauma centers had firearm injuries, and which survivors would die within a year after discharge. The researchers analyzed data from nearly 503,000 injured children and adolescents admitted to trauma centers between 2014 and 2021. Black race, neighborhood poverty, and local opportunity were the strongest predictors of firearm injury risk. Most deaths after discharge happened within the first week and were often from another firearm injury, with risk linked to neighborhood conditions.
Key points
- 1.1% of injured children (aged 0-10) and 9.2% of injured adolescents (aged 11-17) had firearm injuries
- Black race, Child Opportunity Index, and median household income were the strongest predictors of firearm injury
- Post-discharge mortality among firearm injury survivors was 0.38%, but most deaths occurred within one week
- Repeat firearm injury was a common cause of death after discharge
- Risk of post-discharge mortality was influenced by neighborhood characteristics rather than individual clinical factors
Limitations
This study only examined injured children already admitted to trauma centers and does not address non-fatal firearm injuries or prevention in the broader community. The analysis cannot determine whether neighborhood characteristics are causal factors or proxies for other unmeasured social determinants.
Abstract
OBJECTIVE: We used machine learning (ML) to develop firearm risk prediction models for injured children and adolescents admitted to U.S. trauma centers, including prediction of death after discharge among survivors of firearm injury. METHODS: We used a retrospective cohort of injured patients 0-17 years admitted to 982 trauma centers participating in the National Trauma Data Bank from 1/1/2014 to 12/31/2021, stratified by children (0-10 years) versus adolescents (11-17 years). We followed a subset of patients to one year (through 12/31/2022). We used ML to analyze 119 predictors and a primary outcome of firearm injury, and then 141 predictors to model a secondary outcome of death after discharge. RESULTS: There were 260,098 children (2785 [1.1 %] with firearm injuries) and 242,669 adolescents (22,318 [9.2 %] with firearm injuries). For children, the high specificity model had an area under the curve (AUC) 0.787 with 31.2 % sensitivity, specificity 95.3 %, and positive predictive value (PPV) 6.9 %. For adolescents, the model had an AUC 0.763, sensitivity 34.7 %, specificity 95.0 %, and PPV 41.3 %. High-yield predictors included Black race, Child Opportunity Index, and median household income. Post-discharge mortality among those surviving a firearm injury was 0.38 %, but deaths often occurred within one week of discharge, frequently from a repeat firearm injury, and were predicted by home neighborhood characteristics. CONCLUSIONS: Pediatric firearm injury risk stratification can be performed using information available during admission, which could guide injury prevention efforts. The first week after discharge following a firearm injury is high-risk for mortality, with risk influenced by neighborhood characteristics.
Findings the Hub has extracted
Each claim below is labelled by its type of design and links back to this report.
- Intervention
- Machine learning predictive models using 119 clinical and demographic predictors for firearm injury identification and 141 predictors for post-discharge mortality prediction
- Population
- Children aged 0-17 years admitted to U.S. trauma centers with injuries (260,098 children aged 0-10 and 242,669 adolescents aged 11-17)
Key finding
For children (0-10 years): AUC 0.787, sensitivity 31.2%, specificity 95.3%, PPV 6.9%
Newgard et al. · PubMed (US National Library of Medicine) · 26 Mar 2026