Statement of purpose Absent a single data source, researchers have characterised firearm injury epidemiology using trauma registry data. To understand bias associated with this approach, we compared trends in firearm assault incidence and mortality in Philadelphia over 10 years from two sources: the Pennsylvania Trauma Outcomes Study (PTOS) and the Philadelphia Police Department database (PPD).

Methods We included PTOS Philadelphia County firearm assaults patients and PPD firearm assault victims from 2005–2014. We compared subject demographics and calculated counts of fatal and non-fatal incidents between the two data sources. We used generalised linear models adjusting for seasonality to estimate temporal trends in firearm assault rates and fatality over the study period for both data sources.

Results A total of 6,988 PTOS and 14,172 PPD subjects met inclusion criteria. Compared to PTOS subjects, PPD subjects were younger (27±10 vs. 28±13 years; p<0.001), more likely to be female (7.8% vs. 5.9%; p<0.001), and less likely to be black (84.2% vs. 89.3%; p<0.001). In both datasets, the incidence of firearm assaults decreased significantly over the study period (PTOS: 5.19 to 3.43 person-years, change/year=−0.21; 95% CI:−0.26,–0.16; p<0.001; PPD: 10.97 to 6.71 person-years, change/year=−0.53; 95% CI:−0.62,–0.44; p<0.001). The PTOS mean case-fatality rate was 26.5% and decreased significantly (change/year:−0.41, 95% CI:-0.78%,0.04%;p=0.03); the PPD mean case-fatality rate was 18.9% and there was no significant change in mortality (p=0.41).

Conclusions Relative to PPD data, PTOS data underestimated firearm assault incidence and overestimated case fatality rate. Both data sources demonstrated decrease in the incidence of firearm assault, but the rate of decrease was lower in PTOS data.

Contributions to Injury/Violence Prevention Science Local law enforcement data is a promising source for studying city-level firearm injury epidemiology. As trauma registry data likely underestimates firearm assault incidence, future research should use inclusive or linked data sources to reduce this risk of bias.