Surveillance of, and the impact of community policing on arrest-related deaths (ARDS): Exploring the surveillance of ARDs and the opportunities to prevent them in the United States

Background: In 2015, 1,356 deaths resulted from arrest-related interactions with US law enforcement -- a near two-fold increase in known deaths in the previous ten years, according to open-sourced records. Strong public interest and controversy surrounding recent arrest-related deaths (ARDs) in the United States highlights the importance of a better understanding of the public health burden, the trends, and the factors associated with these ARDs. Indeed, little is known about these deaths or their relationship to policing practices. Our aim was to evaluate the official surveillance system for identifying, reporting, and monitoring ARDs. Then, using open-sourced death records, we sought to determine the association between ARDs and policing practices, specifically community-based policing.;Methods: We first evaluated the current surveillance system, the Deaths in Custody Reporting Program ARD collection (hereafter DCRP), assessing the system attributes and its level of usefulness to public health. Next, to investigate the impact of Community Policing we constructed a cross-sectional dataset using counts of ARDs from the open-sourced database, Fatal Encounters, a public-driven database documenting deaths through police interactions. We first compiled a panel data set from three periods of the Law Enforcement Management and Administrative Statistics (LEMAS) survey on ARDs. However, due to poor overlap between panels, changes in survey design, and diminishing data quality in earlier years, we focused only on the most recent survey. We case-matched records of ARDs occurring between January 2012 and April 2016 from Fatal Encounters to the LEMAS responses, collected in 2012 from 2,826 law enforcement agencies. We used arrest counts from the 2012 Uniform Crime Reporting Program (UCRP) with imputed estimates for observations with incomplete reports. We included Census-derived community level characteristics for education, race, and poverty. We modeled the rate of ARDs as our outcome of interest, investigating the effect of Community Policing and other covariates.;Findings: The DCRP falls short of the level of completeness and quality necessary for any practical application as a surveillance system. State counts are reported on a quarterly basis but greatly underestimate the true burden. When restricted to homicides by law enforcement -- a subset of ARDs -- the DCRP was found to capture fewer than half of all deaths in the United States between 2003 and 2011. No studies have been identified that employ DCRP data to assess any prevention or control efforts, and no changes in policy or agency level practices have been attributed to data collected through the DCRP. 2,826 agencies completed the LEMAS in 2012. Of these, 813 (29%) were linked to one or more ARDs over the observed period. These agencies accounted for 3,380 (60.4%) of the reported ARDs in following years, with the remaining 2,219 unmatched to the law enforcement agencies sampled. Using a negative binomial model to account for the over-dispersion in our outcome, we found that greater adherence to the community policing model is associated with a reduced rate of ARDs (by a factor of 0.61; 95% CI: 0.42 to 0.87). Larger policing agencies and greater county-level poverty were both associated with higher rates of ARDs. For example, agencies with 100 additional officers have an estimated 1.65 times higher rate of ARDs than their smaller counterparts. Similarly, communities with 10% greater poverty have a 1.90 times higher rate. No significant association was observed between higher county rates of African-Americans and ARD rates.