Multisite External Validation of a Clinical Screening Tool for Interpersonal Firearm Violence Risk.

BACKGROUND

Screening tools for interpersonal firearm violence (FV) are needed to facilitate prevention.

OBJECTIVE

To validate the 4-item, 10-point SaFETy (Serious fighting, Friend weapon carrying, community Environment, and firearm Threats) score.

DESIGN

Prospective longitudinal study.

SETTING

4 level 1 emergency departments (EDs) in 3 cities.

PARTICIPANTS

Adults aged 18 to 24 years seen in an ED for any reason.

MEASUREMENTS: FV (shooting someone or being shot) 12 months after baseline from self-report and medical record review, SaFETy score, and self-reported baseline covariates (demographic characteristics; baseline assault injury; violence-related ED use in the past 6 months; drug misuse; anxiety, depression, and posttraumatic stress screening; and FV in the past 6 months).

RESULTS

Among 1506 participants (61.4% female; mean age, 21.3 years; 3.8% with assault injury at baseline), 12-month FV could be ascertained in 1122 (74.5%); of those, 73 (6.5%) had 12-month FV. For baseline SaFETy scores of 0, 1 to 5, and 6 or greater, 12-month FV rates were 1.8% (12 of 654), 12.1% (49 of 406), and 25.0% (10 of 40), respectively. The area under the receiver-operating characteristic (ROC) curve (AUC) for the score was 0.78 (95% CI, 0.72 to 0.83). The optimal ROC cut point was a SaFETy score greater than 0, with a sensitivity of 83.1% and specificity of 62.4%; a SaFETy score greater than 4 optimized positive predictive value (31.6%). Logistic regression of 12-month FV, including the full covariate set, examined estimated risks for patients grouped by SaFETy score and found that model-based predictions underestimated risk among those with SaFETy scores of 0 and overestimated risk among those with SaFETy scores of 1 to 5 or 6 or greater. Adding the SaFETy score to the full covariate set improved the predictions' AUC (0.84 vs. 0.81; P = 0.025). The added contribution of the SaFETy score to predictions based only on variables typically available at triage (demographic data, ED visit reason, and recent ED use) was larger.

LIMITATION: Outcomes were primarily self-reported, and the highest-risk subsamples were more likely to have missing data.

CONCLUSION

The SaFETy score predicts FV risk in general samples of young adults in the ED. A comprehensive covariate set, involving factors that are difficult or intrusive to measure, did not reproduce the SaFETy score's risk gradient or explain its discriminatory power, suggesting that the score provides distinct predictive information.

PRIMARY FUNDING SOURCE: Centers for Disease Control and Prevention.