The study draws on baseline data from Project SPARK (Screening to Predict Young Adults at Risk for Firearm Violence), a prospective longitudinal study of 1,506 adults aged 18 to 24 recruited between 2021 and 2023 from four Level 1 trauma centers in Philadelphia, Flint, and Seattle. Among this urban emergency department sample, 14.4 percent reported past-six-month firearm assault — either as a victim, an aggressor, or both, including threats. The cohort was demographically broad: mean age 21.3, 61.4 percent female, 35.9 percent male, racially diverse, with roughly half reporting some form of past-six-month violence involvement. More than 60 percent screened positive for risky alcohol use or drug misuse, and about half screened positive for at least one mental health condition, underscoring how densely clustered these challenges are in the same population.

The methodological innovation lies in how the researchers, led by Dorothy R. Stearns and Jason E. Goldstick of the University of Michigan’s Institute for Firearm Injury Prevention, treated the sixteen risk and protective factors not as independent predictors but as nodes in a network. Using regularized partial correlation models — specifically, a graphical LASSO penalty applied within the R packages lavaan and qgraph — the team estimated the conditional association structure among factors spanning substance use (alcohol and drug use), mental health symptoms (depression, anxiety, and PTSD via the PHQ-9, GAD-7, and PCL-17), firearm carrying, peer behaviors, family conflict, adverse childhood experiences, parental support, resiliency, prosociality, retaliatory attitudes, and community violence exposure. The regularization shrinks trivially small correlations to zero, allowing the resulting network diagram to highlight only the most meaningful connections. The authors set the tuning parameter at γ = 0.5, a convention regarded as a conservative and safe choice.

When the network was visualized, distinct clusters emerged with striking clarity. One grouping contained the mental health triad of depression, anxiety, and PTSD. Another combined substance use with negative behaviors, linking drug use, alcohol use, and negative peer behaviors. A third cluster gathered the protective factors — resiliency, prosocial attitudes, peer support, positive peer behaviors, and parental support — into a tightly interconnected block of resilience-oriented influences. Firearm carrying connected to the rest of the network primarily through community violence, while community violence, family conflict, and adverse childhood experiences each maintained weak but far-reaching connections across multiple sections of the network, acting as the connective tissue between otherwise separate risk domains.

The most consequential findings came from the centrality analyses. The researchers computed four metrics — betweenness, closeness, strength, and expected influence — each capturing a different dimension of how central a variable is within the network. PTSD emerged as the single most connected node, topping the strength measure and ranking near the top of expected influence. Community violence exposure and family conflict scored highly on betweenness and closeness, meaning they frequently lie along the shortest paths connecting other factors, essentially serving as bridges between distinct risk territories. Negative peer behaviors also ranked among the most central elements across multiple metrics. Bootstrap resampling with 2,000 iterations confirmed that these centrality estimates were highly stable: strength and expected influence retained correlations above 0.9 even when only a quarter of the sample was used, well beyond accepted thresholds for reliability.

Perhaps the most striking practical result concerns predictive efficiency. A logistic regression model using all sixteen variables achieved an area under the receiver operating characteristic curve (AUROC) of 0.86 in discriminating between young adults with and without firearm assault involvement. Remarkably, a model built from just the four central network factors — PTSD, community violence exposure, negative peer behaviors, and family conflict — captured more than 96 percent of that discriminatory power, achieving an AUROC of 0.83. Community violence exposure alone outperformed every other individual factor, with an AUROC of 0.81, representing 93 percent of the full model’s performance. By contrast, the protective factor grouping collectively produced a much weaker model (AUROC 0.66), and firearm carrying frequency showed only modest standalone discrimination (AUROC 0.60).

The dominance of community violence exposure as a standalone predictor reinforces a growing body of evidence that firearm violence risk is shaped as much by environment as by individual psychology. The authors contextualize this within the framework of structural racism: residential segregation, economic disinvestment, and discriminatory policies concentrate violence exposure in neighborhoods disproportionately inhabited by minority and marginalized communities. These same structural conditions generate cumulative trauma, chronic stress, and elevated rates of anxiety and depression, feeding directly into the mental health nodes that the network identifies as centrally interconnected with community violence. In other words, the statistical architecture of the network mirrors the social architecture of inequality — risk factors do not operate in isolation but propagate through one another.

The study’s implications for clinical practice are immediate. Emergency departments and trauma centers represent critical touchpoints for young people at elevated risk, and more than one in seven young adults entering urban emergency departments report recent exposure to firearm violence. Brief screening protocols cannot feasibly assess every possible risk domain, but the network analysis suggests they do not need to: asking about community violence exposure, PTSD symptoms, negative peer behaviors, and family conflict captures nearly all of the predictive signal available from a much broader battery. The findings also point toward trauma-informed, multifaceted interventions that address social and structural determinants rather than targeting single factors — for example, combining PTSD treatment with neighborhood safety initiatives, peer norm reshaping, and family support programs.

The authors are careful to note the limitations of their approach. The cross-sectional baseline design cannot establish causality or directionality; PTSD, for instance, may be a consequence of firearm assault exposure rather than a precursor. Self-report measures may be subject to recall bias and underreporting of stigmatized behaviors, and network models estimate conditional associations that can be sensitive to variable selection and regularization choices. The study also combined victimization and aggression into a single indicator, consistent with prior literature documenting substantial overlap between the two, though an aggression-only analysis might have yielded somewhat different results. Exclusion of patients unable to consent may further limit generalizability, although the multi-site design strengthens confidence in the observed associations.

Even with these caveats, the analysis offers a compelling reframing of firearm violence prevention. Rather than a long list of independent hazards, the risk landscape appears as a network with identifiable hubs — and those hubs, particularly community violence exposure and PTSD, represent high-leverage points where intervention could ripple outward through the entire system. As the researchers conclude, individual-level interventions alone cannot address all risk factors simultaneously; reducing firearm violence among young adults will require coordinated strategies that confront personal trauma, social influence, and structural inequity in the same breath. In a field often dominated by single-factor studies, this network perspective provides both a sharper diagnostic lens and a clearer roadmap for action.

Subject of Research: Network analysis of psychosocial, familial, and community risk factors for firearm assault among young adults

Article Title: Network analysis of risk factors for firearm assault

Article References: Stearns, D. R., Carter, P. M., Stallworth, P., Delgado, M. K., Whiteside, L. K., Bonar, E. E., & Goldstick, J. E. (2026). Network analysis of risk factors for firearm assault. Journal of Behavioral Medicine. https://doi.org/10.1007/s10865-026-00698-8

Image Credits: AI Generated

DOI: 10.1007/s10865-026-00698-8

Keywords: firearm assault, network analysis, PTSD, community violence, risk factors, young adults, emergency department, substance use, family conflict, negative peer behaviors, violence prevention, Project SPARK