By Justin M. Feldman, Tracey Lloyd , Phillip Atiba Solomon
Background: Mortality classification for deaths in US police custody has important consequences for epidemiologic monitoring and legal outcomes. Prior literature suggests in custody death classification is inconsistent and may not reflect non-firearm force that preceded death. Methods: We analyzed the Associated Press “Lethal Restraint” national dataset (United States, 2012-2021; N = 1,036), which included deaths following police use of non-firearm force. Our primary outcomes included whether the death investigator: (1) classified manner of death as a homicide, (2) mentioned a force-related injury/condition in the cause-of-death statement, and (3) mentioned any force. Inverse-probability-weighted logistic models estimated the association of these outcomes with death-investigator jurisdiction type, local political composition (quartile of Republican Party vote %), decedent race/ethnicity, and each agency’s prior classifications. Findings: We removed 96 deaths based on exclusion criteria. Of the remaining 940 deaths, 28.5% were classified as homicide, 16.5% had cause-of-death statements mentioning a force-related injury/condition, and 42.6% mentioned any force. In contrast, 73.9% of statements mentioned drugs. Unadjusted results showed homicide classification increased from 25.0% in 2012-2014 to 32.2% in 2018-2021. Models estimating adjusted prevalence differences (aPD) showed that, compared to medical examiner jurisdictions, coroners (aPD: -0.19; 95% CI: -0.31, - 0.06) and sheriff-coroners (a PD: -0.17; 95% CI: -0.28, -0.05) were less likely to classify deaths as homicides. Model results also showed that classifications for incidents occurring in the leastRepublican counties were most likely to reflect force across all three manner and cause outcomes. Interpretation: Non-homicide classifications and cause-of-death statements making no mention of force were widespread for US in-custody deaths. We identified novel evidence suggesting coroner and sheriff-coroner jurisdictions were especially unlikely to categorize in-custody deaths as homicides, and that incidents occurring in highly Republican counties were least likely to reflect force in the cause or manner of death.