Adjusting US homicide rates for improvements in trauma care, 1900–2020.
Abdominal GSW mortality anchors: Cited anchor points with linear interpolation between them: ~72% (1900, interpolation -- Civil War ~87% per Barr et al. 2023; Spanish-American War abdominal mortality comparable to Civil War; Boer War laparotomy mortality 69% (18/26); no single civilian ~1900 case series); ~30% (1960, Wilson & Sherman 1961, 494 civilian cases); ~12% (late 1980s, Demetriades et al. 1988, 300 cases at urban Level I center); ~9.5% (1990s consensus, multiple series); ~11% (2010, ACS TQIP national database, 16,866 patients -- note: higher than 1990s due to worsening injury severity); ~8% (2020, modern Level I standard). WWI military data: forward hospitals reduced abdominal mortality from ~65% to ~45%. Giacopassi et al. 1992 (Memphis, 25-year intervals) corroborates mid-century improvement: overall assault lethality 11.4% (1935) → 5.5% (1960) → 3.2% (1985).
Adjustment formula: adjusted(t) = observed(t) × abd_mort(1900) / abd_mort(t). This asks: "if gunshot victims at time t died at the same rate as in 1900, how high would the homicide rate be?" This is an upper bound on the true adjustment because not all homicide victims are medically saveable -- head shots (~11% of assault GSWs, ~90% fatal regardless of era) and DOA cases are essentially invariant. The true adjustment is probably 60–80% of the displayed values for recent decades.
Other limitations: (1) Pre-1960 anchor points are from scattered clinical series and military data, not systematic civilian surveillance. (2) 1900–1910 homicide rates are from "registration states" only and undercount the violent South -- the early observed rates are too low. (3) Weapon characteristics (caliber, velocity) changed over time. (4) Linear interpolation between sparse anchors is crude. This chart is illustrative, not definitive.
Harris et al. (2002), "Murder and Medicine": The foundational study. Firearm-specific lethality dropped 65% from 1964 (15.5%) to 1999 (5.4%). Only 1.2% of the overall lethality decline was attributable to weapon mix shifts; 98.8% was genuine within-weapon improvement. Motor vehicle crash lethality dropped 67% over the same period -- nearly identical, supporting the medical explanation. Counties with hospitals had 11–24% lower lethality; regionalized trauma systems added 16% further reduction.
Giacopassi et al. (1992), Memphis study: The only study reaching back before 1960. Memphis police homicide files at 25-year intervals: 1935 lethality 11.4%, 1960 lethality 5.5%, 1985 lethality 3.2%. The 1935→1960 drop (~2×) is driven by penicillin, blood banking, and hospital proliferation.
Why gunshot hospitalizations don't solve it cleanly: CDC explicitly warns against using NEISS-AIP for firearm injury prevalence due to small, geographically unrepresentative hospital samples. Nonfatal data starts only 2001 (partially 1993). The best available proxy remains weapon-specific lethality ratios supplemented by clinical case fatality studies.
Cook (2017) / Lauritsen (2015) critique: Part of the aggravated assault increase may reflect expanded reporting, not actual violence. This would inflate assault-based denominators and overstate lethality decline. The abdominal GSW survival approach in Chart 3 sidesteps this entirely -- it uses clinical mortality, not crime reporting.