We conducted a serial cross-sectional, population-based descriptive analysis of firearm deaths among US children, adolescents, and young adults aged 0 to 24 years between 1999 and 2024, using the CDC WONDER Underlying Cause of Death database. WONDER compiles death certificate records through the National Vital Statistics System and provides annual counts, population denominators, and crude rates by demographic and cause-of-death categories [18]. Population denominators were drawn from CDC WONDER bridged-race population estimates for period 1 (1999 to 2020) and single-race population estimates for period 2 (2018 to 2024). The analytic population was restricted to US residents of the 50 states and the District of Columbia.

The current study population was defined as individuals aged 0 to 24 years, comprising early childhood (0 to 12 y), adolescence (13 to 17 y), and young adulthood (18 to 24 y), consistent with established developmental and public health frameworks [19,20]. This upper boundary extends beyond the 0-to-17-year range commonly used in pediatric firearm surveillance reports [2,7,21], which exclude young adults aged 18 to 24 years, a developmental demographic that bears a disproportionate share of the national firearm homicide burden, particularly among Black and Hispanic male youth [3,21].

The 25-year analytic window was bounded by historical and administrative considerations of data continuity. The lower bound of 1999 represents the first full calendar year in which the United States implemented the ICD-10 (International Statistical Classification of Diseases, Tenth Revision) for underlying cause-of-death coding, and the first year for which CDC WONDER bridged-race population denominators are available [18,22]. The upper bound of 2024 incorporates the most recently finalized national mortality file released by the NCHS in January 2026 [18]. The window spans 25 years, from 1999 through 2024, and comprises 26 annual observations. This window enables the contextualization of contemporary, postpandemic shifts within a quarter-century historical trajectory.

This analysis used deidentified, publicly available, aggregated vital statistics data and was therefore determined not to constitute human participants research as defined under 45 CFR (Code of Federal Regulations) §46.102. Institutional review board review was not required under institutional policy governing the secondary analysis of public-use, fully anonymized federal data, and no application was submitted. The CDC WONDER public-use data files contain no direct or indirect personal identifiers. NCHS suppresses cells with fewer than 10 deaths, including finely stratified national cells, to protect confidentiality. The present analysis used visible cells and explicitly identified suppressed and statistically unreliable estimates that satisfied federal analytic reliability thresholds, with unstable rates explicitly flagged. As the underlying records are fully anonymous and collected via routine statutory vital statistics registration, informed consent was not applicable.

Firearm fatalities were identified using ICD-10 underlying cause-of-death codes. In this analysis, the term “intent” refers to the manner of death recorded on the death certificate, representing the standard NCHS classification of the circumstances surrounding the fatal injury. Five mutually exclusive intent categories were retained: firearm homicide (X93-X95); firearm suicide (X72-X74); unintentional firearm injury, defined as a fatal accidental discharge of a firearm rather than a category of homicide or suicide (W32-W34); firearm injury of undetermined intent (Y22-Y24); and legal intervention involving a firearm (Y35.0).

Firearm homicide and suicide served as the primary intent categories for all intersectional analyses. Unintentional, undetermined, and legal intervention deaths were retained in cumulative all-cause totals but were excluded from intent-specific trend estimation, given high suppression rates across multiway demographic and intersectional strata. Cumulative 25-year death totals were calculated by aggregating all period 1 deaths (1999‐2020) and adding unique deaths from the nonoverlapping years of period 2 (2021‐2024) to avoid duplication. All 2024 data reflect finalized counts released by CDC WONDER in January 2026 [18].

In data year 2021, the CDC fully transitioned its multiple-cause-of-death files from a bridged 4-category race classification scheme (White, Black, AI/AN, and Asian or Pacific Islander, with multiple-race decedents algorithmically assigned to a single category) to a 6-category single-race scheme that separately identifies Asian, NH/OPI, and multi-racial decedents [22]. The legacy bridged-race data preserves long-term trend comparability back to 1999 but obscures contemporary diversity, whereas the single-race file allowed for greater contemporary sociodemographic comparability (back to 2018) but cannot be extended to the prebridging era. As these 2 population denominator systems introduce a structural trend break, a single continuous analytic framework spanning 1999 to 2024 was computationally untenable.

Accordingly, race and ethnicity were combined into a single mutually exclusive variable for all analyses (eg, non-Hispanic Black and Hispanic). The overlapping years (2018 to 2020) were used as an internal validation window to verify rate consistency and mathematical stability across both classification methods. The descriptive simultaneous stratification by race or ethnicity, sex, age, and injury intent used in this study constitutes a surveillance application of intersectional theory to isolate overlooked risk concentrations, rather than a formal quantitative intersectional decomposition [16].

Crude firearm death rates per 100,000 population were calculated for each demographic subgroup. Crude rather than age-adjusted rates were prioritized because simultaneous multiway stratification renders within-stratum age adjustment statistically volatile and analytically untenable for low-denominator populations, such as AI/AN, NH/OPI, and multiracial youth [9,21,23]. Disparity ratios were computed using non-Hispanic White males as the reference group for male comparisons and non-Hispanic White females for female comparisons; absolute crude rates were also compared directly across sex lines where female burdens eclipsed male baselines. Further, 95% CIs for rates and ratios were derived using the Byar approximation, which provides robust coverage for the small expected cell counts characteristic of fine-grained intersectional strata [24].

Cells based on fewer than 20 deaths were flagged as analytically unreliable per NCHS standards. Within the primary pooled-age contemporary intersectional matrix for period 2 (6 racial groups x 2 Hispanic-origin categories x 2 sexes x 3 primary intent categories [homicide, suicide, and unintentional injury]), there were 72 distinct cells. Of these, 19 of 72 (26.4%) cells had counts fewer than 10 and were omitted from detailed stratification, and 11 of 72 (15.3%) cells were flagged as unreliable, yielding fully stable estimates for 42 of 72 (58.3%) baseline cells. Rate instability was heavily concentrated among NH/OPI youth (10/12, 83.3% cells) and within specific unintentional injury strata.

To characterize directional trends within period 2 (2018-2024), average annual percentage changes (AAPCs) were estimated by log-linear regression of the natural logarithm of the annual crude rate on calendar year. These were calculated separately for firearm homicide and suicide across non-Hispanic race-sex strata that demonstrated 3 or more reliable annual observations. The AAPC was calculated as (exp[β] − 1)×100, where β is the estimated slope parameter. Further, 95% CIs were derived from the t-distribution of β with n − 2 degrees of freedom. Statistical significance was assessed using 2-sided t tests with an a priori alpha =.05. As this study’s period encompasses a nonlinear post-2020 mortality surge and subsequent decline, the log-linear AAPC is explicitly interpreted as the net annualized trend velocity across the interval rather than a monotonic trajectory. All analyses were executed in Python (version 3.11; Python Software Foundation) using the Pandas, NumPy, SciPy, Matplotlib, and Seaborn libraries. This study adhered strictly to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting guideline; see Checklist 1.

From 1999 through 2024, a total of 200,704 firearm deaths occurred among US youth and young adults aged 0 to 24 years (period 1: 161,313; period 2, nonoverlapping years 2021‐2024: 39,391). Firearm mortality remained concentrated among male youth throughout the 25-year period, though the female share of the burden rose modestly, from 11.4% (18,459/161,313) of deaths in period 1 to 13% (8543/65,472) in period 2, even as the sex-specific trajectories diverged by intent (Figure 1). Across the full 25-year window, overall crude firearm mortality declined steadily from 1999 through 2013, reversed direction in 2014, and accelerated sharply during the pandemic period, rising 28.9% between 2019 and 2020 (from 7.69 to 9.91 per 100,000). Rates remained elevated through 2022 before partially stabilizing in 2023 and declining in 2024 (Figure 1). The recent 2024 overall rate of 8.24 per 100,000 represented a 12% decrease from 2023, the largest single-year decline observed during this 25-year period, though overall youth firearm mortality remained above the 2019 prepandemic baseline.

When examined by intent, firearm homicide and firearm suicide followed divergent trajectories within this aggregate trend. Among Black non-Hispanic male youth, homicide rates rose from 33.44 per 100,000 in period 1 to 44.03 per 100,000 in period 2, and Black non-Hispanic female homicide rates nearly doubled from 3.46 to 6.03 per 100,000 (Tables 1 and 2). Firearm suicide rates among AI/AN non-Hispanic male youth rose from 8.47 to 11.74 per 100,000 across the 2 periods, and Black non-Hispanic male suicide rates rose from 3.51 to 6.33 per 100,000. The 2 periods nearly coincide across the 2018 to 2020 overlap in Figure 1. The small residual differences reflect changes in population denominators between the bridged-race and single-race files rather than true changes in mortality; therefore, rates were not directly compared across periods.

Within period 2, age-specific analysis revealed steep developmental gradients concentrated in 2 intersectional locations (Multimedia Appendix 1). Among Black non-Hispanic male youth, firearm homicide rates increased from 6.63 per 100,000 at ages 10 to 14 years to 86.06 at ages 15 to 19 years and 114.44 at ages 20 to 24 years. The Black-to-White non-Hispanic male homicide rate ratio widened across consecutive age bands, from 26.1 at ages 15 to 19 years to 27 at ages 20 to 24 years, indicating that the greatest absolute firearm homicide burden occurred during late adolescence and young adulthood. This pattern challenges prevention frameworks that define youth as ending at age 17 or 18 because they may exclude young adults aged 18 to 24, including the age group with the highest observed firearm homicide rate.

Among AI/AN non-Hispanic male youth, firearm suicide rates increased from 2.92 per 100,000 at ages 10 to 14 years to 20.16 at ages 15 to 19 years and 31.71 at ages 20 to 24 years. In the oldest age band, the AI/AN non-Hispanic male firearm suicide rate was 1.6 times the corresponding White non-Hispanic male rate (19.61 per 100,000). The parallel adolescent-to-young-adult escalations observed for AI/AN male firearm suicide and Black non-Hispanic male firearm homicide underscore the need for prevention frameworks that extend beyond adolescence into the 18-to-24-year developmental window.

Among female youth, age gradients were less pronounced for homicide but evident for suicide. White non-Hispanic female firearm suicide rates increased from 0.47 per 100,000 at ages 10 to 14 years to 2.40 at ages 20 to 24 years. Black non-Hispanic female firearm suicide rates showed a similar but shallower gradient, increasing from 0.52 to 2.44 per 100,000 across the same age bands. In contrast to the pattern among males, Black and White non-Hispanic female firearm suicide rates across age groups were statistically indistinguishable, with overlapping CIs at ages 10 to 14 and at ages 20 to 24 years, warranting continued monitoring in future surveillance cycles as an indicator of the shifting racial distribution of firearm suicide risk.

This 25-year national analysis advances 3 critical findings that aggregate surveillance routinely conceals. Primarily, firearm homicide remains concentrated with extreme disproportionality among Black non-Hispanic male youth and, to a degree not captured by race-only or sex-aggregated reporting, among Black non-Hispanic girls. Annual firearm homicide rates among Black non-Hispanic male youth rose sharply during 2020 and 2021 and substantially reverted by 2024, with the 2024 rate approaching the 2018 level. However, the disparity ratio relative to White non-Hispanic male youth did not narrow across this period. The Black-to-White non-Hispanic male homicide rate ratio increased from 19.3 in 2018 to 24.4 in 2024, indicating that improvement in the absolute homicide rate among Black non-Hispanic males was not accompanied by a reduction in relative inequality.

In addition, firearm suicide burden was highest among AI/AN non-Hispanic male youth and intensified with age, peaking at ages 20 to 24 years. Furthermore, firearm suicide rates among Black non-Hispanic male youth increased at a statistically significant rate of +9.2% per year, with a comparable-magnitude rise among Black non-Hispanic female youth and stable rates among White non-Hispanic male and female youth. Consequently, the Black-to-White non-Hispanic male suicide rate ratio shifted from 0.67 in 2018 to 1.07 in 2024, marking the first year in recent surveillance history that Black non-Hispanic male rates exceeded those of White non-Hispanic male youth. Our findings show that firearm mortality among US youth represents 2 epidemiologically distinct phenomena that require intent-specific, intersectionally disaggregated surveillance and prevention paradigms.

Three findings illustrate why an intersectional multi-year lens, rather than race-only or sex-only stratification, is necessary to characterize the firearm mortality burden in this age group. Primarily, the firearm homicide rate among Black non-Hispanic female youth (6.03 per 100,000) exceeded male homicide rates in 7 of 12 other racial and ethnic male subgroups examined, a result that is invisible in sex-aggregated reporting and in race-only reporting that pools Black non-Hispanic girls with girls of all other groups. This is the empirical pattern that intersectional theory predicts, noting that risk concentrates at specific social locations defined by the simultaneous operation of multiple structural axes [14-16]. Moreover, the elevated firearm suicide rate among AI/AN non-Hispanic young adult males (31.71 per 100,000 at ages 20 to 24 y) similarly emerges only when race or ethnicity, sex, age, and intent are considered together, reflecting the joint operation of historical, geographic, and behavioral health system exposures rather than any single attribute [29].

In addition, the convergence of Black and White non-Hispanic male firearm suicide rates between 2018 and 2024 demonstrates that intersectional locations are not static. The rate ratio shifted from 0.67 to 1.07 over 6 years, indicating that the demographic distribution of firearm suicide risk is itself in motion. This shift is reinforced by a parallel rise among Black non-Hispanic female youth (AAPC +11.7%; 95% CI −1.7 to +27.0), which approaches but does not reach the conventional threshold for statistical significance and which is similarly invisible in sex-aggregated or race-only reporting. The sex-symmetric character of the Black non-Hispanic firearm suicide rise is precisely the kind of dynamic that intersectional surveillance can detect and that aggregated reporting cannot. For prevention planning, these findings imply that resources designed around the historical White male suicide profile or the historical Black male homicide profile cannot be assumed to reach the youth populations now bearing the greatest and most rapidly changing burden. Intersectional surveillance must be matched with intersectional intervention design, developed with the affected communities [15,16].

Several populations with elevated burden in this analysis, including Black non-Hispanic girls, AI/AN youth, and NH/OPI adolescents, remain difficult to characterize reliably in national datasets because of small population denominators, historical racial category aggregation, misclassification of injury intent, and suppression thresholds that remove cells with fewer than 10 deaths [13,30]. Of the 72 possible intersectional cells in period 2, only 42 (58.3%) yielded stable estimates. NH/OPI youth were the most severely affected by suppression, with 10 of 12 (83.3%) cells either suppressed or unreliable, meaning that a population group with documented elevated homicide burden is effectively invisible in standard surveillance outputs. This gap reflects a systemic shortcoming in national mortality surveillance, particularly for populations facing the highest levels of risk.

Addressing these gaps requires action at multiple levels. Vital statistics systems should improve the completeness and accuracy of race and ethnicity classification, particularly for AI/AN, Hispanic, Asian, and multiracial decedents. In addition, supporting tribally led and tribally governed data-linkage initiatives, developed through sovereign-to-sovereign partnerships among Tribal Nations, Tribal Epidemiology Centers, Urban Indian Organizations, the Indian Health Service, and state vital statistics agencies, may help reduce racial misclassification in mortality records and improve identification of AI/AN deaths. Such initiatives must proceed under tribal authority, community-defined governance, formal data-use agreements, and appropriate tribal review, consistent with Indigenous Data Sovereignty and CARE principles (Collective benefit, Authority to control, Responsibility, Ethics). Because linkage primarily improves numerator accuracy rather than the size or stability of the underlying population denominator, it should be paired with multiyear estimates, transparent uncertainty reporting, and improved population-estimation methods for small and geographically diverse AI/AN populations. Linkage cannot, however, resolve the small-denominator problem itself, as the instability of rates in numerically small populations is a function of population size rather than of counting error.

Tribal data sovereignty frameworks that enable AI/AN communities to generate, govern, and analyze their own public health data could also improve the capacity to detect and respond to concentrated burden [30]. Finally, the 2018 race-classification transition, while methodologically necessary, created a discontinuity that has constrained intersectional analyses for nearly a decade. Prospective planning for future classification updates should include explicit transition protocols that preserve analytic continuity (e.g., bridging algorithms), rather than requiring investigators to construct post hoc bridging designs.

National vital statistics surveillance is not the only level at which firearm injury can be counted or addressed. Community-based surveillance systems, including hospital trauma registries, school-based injury reporting, faith-based and neighborhood-level violence interruption networks, and tribal injury surveillance partnerships, may generate higher-resolution and more contextually grounded data than national mortality files in populations where suppression and small denominators limit the utility of vital statistics; the White Mountain Apache surveillance and prevention system offers one such model [29]. Pairing these systems with community-based interventions [27], such as hospital-linked violence interruption programs, school-based behavioral health, and lethal-means counseling [28], could allow detection and response to be matched at the scale where families and providers actually encounter risk, moving prevention science well beyond universal screening and counseling [27,29]. These findings indicate which populations bear concentrated burden, and do not posit that descriptive mortality data can adjudicate these issues; that work requires evaluation designs this analysis does not provide.

This analysis has several limitations. CDC WONDER suppresses strata with fewer than 10 deaths, which limits estimation for some intersectional subgroups. Small denominators among AI/AN, NH/OPI, and multiracial youth increase rate instability. The 2018 race-classification transition prevents direct cross-period comparisons. Additionally, crude rather than age-adjusted rates were used to preserve interpretability within fine-grained strata but may limit comparability with age-adjusted estimates reported elsewhere. AAPC estimates rest on a 7-year annual series (2018‐2024); the short window precludes reliable detection of trend inflection points, and log-linear regression may understate nonmonotonic dynamics such as the pandemic-era homicide surge and subsequent partial reversion. Our use of AAPC assumes a linear trajectory across the 2018‐2024 period. However, firearm homicides in heavily burdened strata followed a distinctly nonmonotonic path (a sharp pandemic-era spike followed by a post-2022 decline). The resulting nonsignificant P values reflect this U-shaped deviation rather than a lack of meaningful year-to-year shifts. Future analyses with longer longitudinal windows should use nonlinear modeling to capture these specific inflection points. AAPCs should be interpreted as directional rather than definitive. Our descriptive design does not account for socioeconomic conditions, urbanicity, or state-level firearm violence policies and precludes causal inference. Sex was determined from death certificate records, which did not include gender identity data. Consequently, the analysis could not examine firearm mortality among transgender, nonbinary, and other gender-diverse populations at the intersections of race and ethnicity, age, and gender. Intent and race or ethnicity classification on death certificates are similarly imperfect, particularly for AI/AN, Hispanic, and Asian American or Pacific Islander decedents [13,30]. Geographic variation was not examined, though prior work demonstrates substantial heterogeneity [2,10].

Twenty-five years of national mortality data show that firearm mortality among US youth follows distinct, intent-specific patterns across race or ethnicity, sex, and age that aggregate surveillance routinely obscures. The post-2020 surge in firearm homicide among Black non-Hispanic male youth has reverted by 2024, with the 2024 rate approaching the 2018 baseline. However, the Black-to-White non-Hispanic male homicide rate ratio widened modestly across the period (from 19.3 in 2018 to 24.4 in 2024), indicating that the partial reversion was concentrated in the reference group rather than in the burdened group, and that the structural drivers producing this disparity did not narrow. Frameworks that interpret the 2024 decline as evidence of progress should attend to this distinction. Concurrently, firearm suicide rates among Black non-Hispanic male and female youth rose at a comparable magnitude while White non-Hispanic rates remained stable, producing the first year in this dataset in which Black non-Hispanic male suicide rates exceeded those of White non-Hispanic males. This study advances 3 contributions to the surveillance literature: (1) a coherent bridging design that links bridged-race and single-race surveillance eras within a single intersectional framework, enabling continued long-term monitoring across the 2018 classification transition; (2) finalized 2024 estimates with explicit attention to nonmonotonic postpandemic dynamics, including parallel computation of homicide and suicide AAPCs that prior partially stratified analyses have not provided; and (3) an empirical case for intersectional surveillance as the unit of analysis appropriate to populations whose burden and trajectory cannot be characterized by race alone or sex alone. Intent-specific, intersectionally disaggregated surveillance and prevention strategies, developed with the affected communities, continue to be the appropriate response [27,29].

For firearm homicide, the log-linear AAPC understates nonmonotonic dynamics. Annual homicide rates among Black non-Hispanic male youth illustrate this pattern: the rate rose from 33.28 in 2018 to a peak of 53.75 in 2021 (a 61.5% relative increase over 3 years), then declined progressively to 37.06 in 2024 (a 31% relative decline over 3 years). Consequently, the log-linear fit across the overall 7-year window did not reach statistical significance because the pandemic-era increase was partially offset by subsequent declines. The pattern is interpreted as pandemic-era elevation with partial reversal rather than as a sustained directional trend.

For firearm suicide, 2 strata show sustained directional trends consistent with the intersectional pattern described in the main text: Black non-Hispanic male youth (AAPC +9.2%, P=.01) and Black non-Hispanic female youth (AAPC +11.7%, P=.08, approaching but not reaching the conventional threshold). White non-Hispanic male and female rates are stable, producing the convergence in the Black-to-White non-Hispanic male suicide rate ratio reported in the main text.