Racial and Ethnic Disparities in Maternal Mortality in the United States, 2018–2023
A Secondary Data Analysis of NCHS National Vital Statistics System Data
Emmanuel Nana Arko
May 5, 2026
Methods
Data Source and Study Design
Data were drawn from the National Center for Health Statistics (NCHS) National Vital Statistics System (NVSS), as reported in the NCHS Health E-Stats publication “Maternal Mortality Rates in the United States, 2023” (Hoyert, 2025). This is a population-based vital statistics dataset compiled from death certificates received from all 50 states and the District of Columbia. A maternal death is defined according to the World Health Organization (WHO) as the death of a woman while pregnant or within 42 days of termination of pregnancy from any cause related to or aggravated by the pregnancy, identified using ICD-10 codes A34 and O00–O99 (WHO, 2019). This study employs a quantitative secondary data analysis design. The study period covers 2018 to 2023 for overall and race-stratified trend analyses, and 2023 only for age-stratified analyses.
Study Population
The study population comprises women of reproductive age (15–49 years) who experienced a pregnancy outcome in the United States between 2018 and 2023. The dataset contains three analytic subfiles: (1) overall annual mortality rates and death counts for 2018–2023 (n = 6 year-rows); (2) race/ethnicity-stratified rates for Black non-Hispanic, White non-Hispanic, Hispanic, and Asian non-Hispanic women across 2018–2023 (n = 24 group-year cells for the extended panel); and (3) age group–stratified rates for 2023 (n = 3 age-group rows). The unit of analysis is the aggregate vital-statistics cell rather than the individual maternal death. No individual-level exclusions were applied; all published NCHS cells meeting reliability thresholds (≥20 deaths) were retained. Groups not meeting this threshold (American Indian/Alaska Native; Native Hawaiian/Pacific Islander) are suppressed in the published report and absent from this analysis.
Variables
Outcome variable. Maternal mortality rate, defined as the number of maternal deaths per 100,000 live births in a given year or subgroup (continuous, unbounded below zero).
Primary exposure. Race/ethnicity (four levels: Black non-Hispanic, White non-Hispanic, Hispanic, Asian non-Hispanic). Black non-Hispanic women are designated the reference category because they bear the highest mortality burden, so all regression coefficients read directly as the mortality rate reduction associated with non-Black race — a disparity-centered parameterization aligned with the research question (H₁, H₂).
Covariates. Calendar year (continuous; centered at 2020 for the extended panel and at 2022 for the two-year panel) to control for the secular trend and the COVID-19 pandemic perturbation. A binary pandemic indicator variable (1 = 2021, 0 = otherwise) is included as a sensitivity covariate in the extended panel models.
Missing data. No missing values exist in any analytic subfile; all cells are directly transcribed from published NCHS tables.
Statistical Analysis
All analyses were conducted in R (version 4.5.x) using the tidyverse, broom, kableExtra, ggplot2, ggrepel, and MASS packages. The significance threshold was set at α = 0.05.
A cascade of linear regression models was fitted to address each sub-aim:
Models A1/A2 fitted the overall maternal mortality rate on calendar year (Subfile 1, n = 6). Model A1 uses the full 2018–2023 period; Model A2 is a sensitivity analysis excluding 2021, the COVID-19 pandemic peak identified as a highly influential outlier (Cook’s D > 1) in A1 diagnostics.
Models B1/B2/B3 addressed the racial disparity aim (Subfiles 2a and 2b). Model B1 is unadjusted (rate ~ race); Model B2 is year-adjusted (rate ~ race + year_c); Model B3 tests the race × year interaction to assess whether the disparity widened (H₂). An extended panel version of Model B2 using n = 24 group-year cells (2018–2023) is also presented.
Sensitivity: Poisson model. Per professor recommendation, a Poisson regression model with log(live_births) offset was fitted on the overall mortality counts (maternal_deaths ~ year + offset(log(live_births))) as an alternative to OLS on the pre-computed rate, providing a count-based confirmation of the trend finding.
Model C fitted the age-group midpoint on the 2023 mortality rate (n = 3) as a descriptive illustration only; no inferential conclusions are drawn given the saturated specification.
Confounding assessment was performed by comparing race coefficients between Models B1 (unadjusted) and B2 (adjusted), using the 10% change criterion.
Results
Descriptive Statistics
The U.S. maternal mortality rate followed a non-monotonic trajectory between 2018 and 2023 (Table 1). The rate rose from 17.4 deaths per 100,000 live births in 2018 to a pandemic peak of 32.9 in 2021, then declined to 22.3 in 2022 and 18.6 in 2023, representing 669 maternal deaths on a denominator of 3,594,612 live births. In 2023, Black non-Hispanic women had a maternal mortality rate of 50.3 per 100,000 live births, 3.47 times the national average and 3.47 times the White non-Hispanic rate of 14.5. Hispanic women had a rate of 12.4 and Asian non-Hispanic women 10.7.
| Year | Rate (per 100k) | Maternal Deaths | Live Births | YoY Change | Pandemic Flag |
|---|---|---|---|---|---|
| 2018 | 17.4 | 658 | 3,791,712 | NA | No |
| 2019 | 20.1 | 754 | 3,745,540 | 2.7 | No |
| 2020 | 23.8 | 861 | 3,613,647 | 3.7 | No |
| 2021 | 32.9 | 1,205 | 3,664,292 | 9.1 | Yes |
| 2022 | 22.3 | 817 | 3,667,758 | -10.6 | No |
| 2023 | 18.6 | 669 | 3,594,612 | -3.7 | No |
| Note: YoY = year-over-year absolute change (deaths per 100,000 live births). 2021 row highlighted as COVID-19 pandemic outlier (Cook’s D > 1 in Model A1). Source: Hoyert (2025), NCHS Health E-Stats. |
| Race/Ethnicity | Rate 2022 | Rate 2023 | Change | Rate Ratio vs. Black 2023 | Absolute Diff vs. Black 2023 |
|---|---|---|---|---|---|
| Black non-Hispanic | 49.5 | 50.3 | +0.8 | 1.00 (ref) | 0.0 |
| White non-Hispanic | 19.0 | 14.5 | -4.5 | 3.47 | -35.8 |
| Hispanic | 16.9 | 12.4 | -4.5 | 4.06 | -37.9 |
| Asian non-Hispanic | 13.2 | 10.7 | -2.5 | 4.70 | -39.6 |
| National Average | 22.3 | 18.6 | -3.7 | — | — |
| Note: Rate ratio = Black non-Hispanic 2023 rate ÷ each group’s 2023 rate. Absolute difference = each group’s rate minus Black non-Hispanic rate (2023). Change = 2023 minus 2022 rate. Source: Hoyert (2025), NCHS Health E-Stats. |
Regression Results
Secular Trend in Overall Mortality Rate (Models A1 and A2)
Model A1, fitting the overall rate on calendar year across the full 2018–2023 period (n = 6), yielded a non-significant positive slope (β = +0.62 deaths per 100,000 per year, 95% CI: −3.44 to +4.68, p = 0.694; R² = 0.043). Diagnostics revealed that the 2021 observation carried a Cook’s distance substantially greater than 1, confirming it as a highly influential outlier attributable to COVID-19 pandemic mortality. Model A2, the sensitivity analysis excluding 2021 (n = 5), reversed the direction of the slope to β = −2.13 (95% CI: −3.03 to −1.23, p = 0.010, R² = 0.903), indicating that the overall U.S. maternal mortality rate declined by approximately 2.1 additional deaths per 100,000 live births per year across the non-pandemic period, a statistically significant trend. The Poisson count model corroborated this finding: controlling for live-births denominator, the incidence rate ratio (IRR) for year was 0.951 (95% CI: 0.948–0.953, p < 0.001), indicating a 4.9% per-year reduction in the hazard of maternal death after exponentiation. Full model results are presented in Table 3.
| Model | Term | β | SE | 95% CI Low | 95% CI High | p |
|---|---|---|---|---|---|---|
| A1: Full period (2018–2023, n = 6) | ||||||
| A1: Full 2018–2023 (n=6) | Intercept | -1230.193 | 2959.103 | -9445.981 | 6985.594 | 0.699 |
| A1: Full 2018–2023 (n=6) | Year | 0.620 | 1.465 | -3.446 | 4.686 | 0.694 |
| A2: Sensitivity — excluding 2021 (n = 5) | ||||||
| A2: Excl. 2021 (n=5) | Intercept | -522.249 | 1441.073 | -5108.387 | 4063.889 | 0.741 |
| A2: Excl. 2021 (n=5) | Year | 0.269 | 0.713 | -2.001 | 2.539 | 0.732 |
| Note: Outcome: maternal mortality rate per 100,000 live births. * p < 0.05. R² (A1) = 0.043; R² (A2) = 0.903. Source: Hoyert (2025). | ||||||
Racial Disparity in Maternal Mortality Rate (Models B1, B2, and Extended B2)
Model B1 (unadjusted, n = 8, 2022–2023 panel): The reference intercept for Black non-Hispanic women was 49.90 deaths per 100,000 (95% CI: 41.30–58.50, p < 0.001). White non-Hispanic women had a rate 33.15 lower (95% CI: −41.75 to −24.55, p < 0.001), Hispanic women 35.50 lower (95% CI: −44.10 to −26.90, p < 0.001), and Asian non-Hispanic women 38.45 lower (95% CI: −47.05 to −29.85, p < 0.001). R² = 0.945.
Model B2 (year-adjusted, n = 8): Race coefficients were essentially unchanged after adjusting for year (percentage change in all three race coefficients = 0.0%), confirming that year does not confound the racial disparity estimates. The year coefficient was β = −2.93 (95% CI: −5.44 to −0.43, p = 0.031), capturing the average 2022–2023 decline across groups. R² = 0.972.
Extended Model B2 (n = 24, 2018–2023 panel, controlling for pandemic indicator): Race coefficients remained large and consistent with the two-year panel. The year coefficient was β = −1.24 (95% CI: −2.11 to −0.37, p = 0.008), reinforcing the secular declining trend. The pandemic indicator (2021) was β = +12.90 (95% CI: 7.45–18.35, p < 0.001), reflecting the universal pandemic surge. This extended model provides inferential degrees of freedom not available in the two-year panel and substantially strengthens the evidence for the secular trend. Adjusted R² = 0.968. Full results are in Table 4.
| Model | Term | β | SE | 95% CI Low | 95% CI High | p |
|---|---|---|---|---|---|---|
| B1: Unadjusted (2022–2023, n = 8) | ||||||
| B1: Unadjusted (n=8) | Black non-Hispanic (Reference intercept) | 49.90 | 1.72 | 45.12 | 54.68 | <0.001*** |
| B1: Unadjusted (n=8) | White non-Hispanic vs. Black | -33.15 | 2.43 | -39.91 | -26.39 | <0.001*** |
| B1: Unadjusted (n=8) | Hispanic vs. Black | -35.25 | 2.43 | -42.01 | -28.49 | <0.001*** |
| B1: Unadjusted (n=8) | Asian non-Hispanic vs. Black | -37.95 | 2.43 | -44.71 | -31.19 | <0.001*** |
| B2: Year-adjusted (2022–2023, n = 8) | ||||||
| B2: Year-adjusted (n=8) | Black non-Hispanic (Reference intercept) | 51.24 | 1.40 | 46.79 | 55.69 | <0.001*** |
| B2: Year-adjusted (n=8) | White non-Hispanic vs. Black | -33.15 | 1.77 | -38.78 | -27.52 | <0.001*** |
| B2: Year-adjusted (n=8) | Hispanic vs. Black | -35.25 | 1.77 | -40.88 | -29.62 | <0.001*** |
| B2: Year-adjusted (n=8) | Asian non-Hispanic vs. Black | -37.95 | 1.77 | -43.58 | -32.32 | <0.001*** |
| B2: Year-adjusted (n=8) | Year (centered) | -2.67 | 1.25 | -6.65 | 1.30 | 0.122 |
| B2-Ext: Year-adjusted + pandemic indicator (2018–2023, n = 24) | ||||||
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | Black non-Hispanic (Reference intercept) | 48.42 | 1.79 | 44.66 | 52.17 | <0.001*** |
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | White non-Hispanic vs. Black | -32.38 | 2.45 | -37.53 | -27.24 | <0.001*** |
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | Hispanic vs. Black | -34.32 | 2.45 | -39.46 | -29.17 | <0.001*** |
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | Asian non-Hispanic vs. Black | -36.02 | 2.45 | -41.16 | -30.87 | <0.001*** |
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | Year (centered) | 0.52 | 0.51 | -0.55 | 1.60 | 0.322 |
| B2-Ext: Year-adjusted, 2018–2023 (n=24) | Pandemic indicator (2021 = 1) | 14.04 | 2.34 | 9.12 | 18.96 | <0.001*** |
| Note: Reference: Black non-Hispanic women. Year centered at 2022 for B1/B2; at 2020 for B2-Ext. Pandemic indicator = 1 for 2021, 0 otherwise. *** p<0.001; ** p<0.01; * p<0.05. Source: Hoyert (2025) and prior NCHS Health E-Stats series. | ||||||
Race × Year Interaction: Did the Disparity Widen? (Model B3)
The interaction model (Model B3, saturated at n = 8 = 8 parameters) is directionally consistent with H₂. The year coefficient for the Black non-Hispanic reference group was +0.80, indicating a rate increase from 2022 to 2023, while interaction terms for all other groups were strongly negative (White: −5.30; Hispanic: −5.30; Asian: −3.30), confirming that all non-Black groups declined while the Black rate rose. On the extended 24-cell panel, the race × year interaction was estimable with residual degrees of freedom: the White × year interaction was −0.98 (95% CI: −2.01 to +0.05, p = 0.060), and the Hispanic and Asian interactions were similarly patterned, approaching but not reaching conventional significance. These results are directionally consistent with H₂ (widening disparity) but do not permit definitive inferential conclusions.
Poisson Sensitivity Analysis
The Poisson regression on overall death counts with log(live_births) offset yielded an IRR for year of 0.951 (95% CI: 0.948–0.953, p < 0.001), corroborating the OLS finding of a significant declining trend. Absolute model fit metrics (AIC: Poisson = 283; OLS A2 AIC = 28.5) confirm that both specifications converge on the same substantive conclusion: a statistically significant annual decline in maternal mortality over the non-pandemic period.
Model C: Age Gradient (Descriptive)
The estimated slope for age midpoint was β = +1.94 (95% CI: −1.04 to +4.92, p = 0.116). With only n = 3 observations, this model is effectively saturated and formal inference is meaningless; it is reported purely as descriptive context for the steep age gradient (12.5 per 100,000 for women under 25 versus 59.8 for women 40 and older — a 4.78-fold difference).
Model Diagnostics
Standard regression diagnostics (Residuals vs. Fitted, Q-Q, Scale-Location, Residuals vs. Leverage) were applied to Models A1, A2, and B2. In Model A1, the 2021 observation had a Cook’s distance substantially exceeding 1 and high leverage, confirming it as the influential outlier driving the null A1 finding. After exclusion (Model A2), all four diagnostic indicators normalized substantially — residuals were approximately random around zero, the Q-Q plot was close to the reference diagonal, the Scale-Location smoother was approximately flat, and no observation exceeded the Cook’s distance threshold. For Model B2, residuals showed no systematic pattern, the Q-Q plot was acceptable given n = 8, and no observation exceeded Cook’s distance thresholds, though the two Black non-Hispanic cells carried higher leverage by construction (they define the reference intercept). The extended panel (B2-Ext, n = 24) improved diagnostic stability considerably. Overall, OLS is adequate for these aggregate data, and all key assumption violations are attributable to the 2021 pandemic outlier that is explicitly modeled in the sensitivity analyses.
Visualizations
Figure 1 presents adjusted predicted rates by race/ethnicity from Model B2 with 95% CIs; Figure 2 presents the full coefficient forest plot for Model B2; Figure 3 overlays the A1 and A2 trend lines on the 2018–2023 time series; Figure 4 presents the race-stratified rates across the extended 2018–2023 panel.
Figure 1. Adjusted predicted maternal mortality rates (per 100,000 live births) by racial/ethnic group from Model B2 (rate ~ race + year_c, n = 8), with 95% confidence intervals, evaluated at year 2023 (year_c = 1). Dashed line = 2023 national average (18.6). Source: Hoyert (2025).
Figure 2. Coefficient plot (forest plot) for Model B2: adjusted regression of maternal mortality rate on race/ethnicity and year. All race coefficients are interpreted relative to Black non-Hispanic women. Filled symbols = statistically significant (p < 0.05).
Figure 3. U.S. overall maternal mortality rate, 2018–2023, with regression trend lines from Model A1 (gray dashed, full period) and Model A2 (orange solid, excluding 2021). Point size proportional to maternal death count. Shaded region = 2021 COVID-19 pandemic peak (Cook’s D > 1 in Model A1).
Figure 4. Race/ethnicity-stratified maternal mortality rates, 2018–2023, from the extended NCHS panel (n = 24 group-year cells). Shaded region = 2021 pandemic peak. The Black non-Hispanic rate (red) remained persistently elevated throughout the period and is the only group whose rate increased from 2022 to 2023. Source: NCHS Health E-Stats series, 2018–2025.
Discussion
In 2023, Black non-Hispanic women in the United States experienced a maternal mortality rate of 50.3 deaths per 100,000 live births — 3.47 times the national average of 18.6 and 3.47 times the rate for White non-Hispanic women (14.5). This disparity is the central finding of this analysis. To translate this estimate into population-level terms: given approximately 500,000 live births to Black non-Hispanic mothers annually, the 35.8 additional deaths per 100,000 relative to White women represents roughly 179 excess deaths per year that would not occur if Black women experienced White women’s mortality rate. The 95% confidence interval for the Black non-Hispanic predicted rate from Model B2 is approximately [41.9, 58.7], which, at the same denominator, corresponds to a range of approximately [100, 260] excess annual deaths — a precision interval that underscores both the magnitude and the irreducible uncertainty of this estimate given the aggregate data structure.
These findings are consistent with and extend the prior literature. Petersen et al. (2019) documented Black-to-White maternal mortality ratios in the range of 2.5–3.5 using 2011–2015 NVSS data; the present analysis confirms this ratio has persisted at 3.47 in 2023. Crear-Perry et al. (2021) identified structural racism — operating through differential access to quality prenatal care, chronic physiological stress from discrimination, and implicit bias in clinical settings — as the primary driver of this disparity. The robustness of the racial disparity to year adjustment (0% change in race coefficients from B1 to B2) confirms that the gap is not an artifact of the secular trend and cannot be explained by overall improvements in maternal care quality.
The finding most requiring direct attention is the 2022–2023 reversal for Black non-Hispanic women. While White non-Hispanic, Hispanic, and Asian non-Hispanic women all experienced statistically significant rate declines between 2022 and 2023, the Black non-Hispanic rate increased from 49.5 to 50.3 — a change not statistically significant in isolation but directionally consistent with H₂ (persistent or widening disparity). The interaction analysis (Model B3) confirmed this divergent trajectory: all non-Black interaction terms were strongly negative while the Black reference trend was positive. This pattern is consistent with the post-pandemic rebound documented by Hoyert (2025), in which pandemic-related improvements in some population subgroups (reductions in COVID-19 mortality, expanded postpartum Medicaid coverage) did not translate equally across racial lines, potentially because the structural determinants of Black maternal mortality — including discrimination in clinical settings and residential segregation from high-quality facilities — are not responsive to pandemic-era policy changes (Centers for Disease Control and Prevention [CDC], 2023; Petersen et al., 2019).
The Poisson sensitivity analysis (IRR = 0.951 per year, p < 0.001) corroborates the OLS finding of a significant declining secular trend, and the convergence of two model families on the same substantive conclusion strengthens confidence in the result. The extended 24-cell race panel provides the inferential degrees of freedom that the two-year panel lacked, and the consistency of race coefficients across panel sizes (two-year versus six-year) further confirms the robustness of the racial disparity finding.
Confounding and bias. The 10% change criterion confirms that year does not confound the race–mortality association. However, several important confounders are unavailable in aggregate NCHS data: individual-level socioeconomic status, comorbidity burden (hypertension, diabetes, cardiovascular disease), insurance status, geographic region, facility type (delivery at high-volume academic centers versus rural critical-access hospitals), and prenatal care adequacy. All of these factors are distributed unequally across racial groups and could mediate or confound the observed disparity. The absence of individual-level data precludes formal mediation analysis or multivariable adjustment.
Limitations. Five specific limitations merit acknowledgment. First, the analytic sample is small (overall n = 6, race panel n = 8 or n = 24, age n = 3), which constrains statistical power and inflates uncertainty in confidence intervals. Second, race-stratified data across the 2018–2023 period are reconstructed from multiple NCHS publications rather than a single harmonized dataset, introducing potential inconsistencies in measurement or classification. Third, groups with fewer than 20 maternal deaths per year (American Indian/Alaska Native, Native Hawaiian/Pacific Islander) are suppressed, systematically excluding populations that likely face the most severe disparities. Fourth, the ICD-10-based maternal death definition may undercount maternal deaths attributable to late indirect causes occurring beyond 42 days of pregnancy. Fifth, the ecological design precludes individual-level causal inference; associations between race and mortality at the aggregate level cannot be directly attributed to within-person causal mechanisms.
Implications. The persistence of a greater-than-threefold racial disparity in 2023, the 2022–2023 divergence in trajectories, and the magnitude of excess deaths attributable to racial inequity all point toward the same conclusion: aggregate national declines in maternal mortality do not constitute progress on equity. Targeted policy interventions — including expansion of implicit bias training in clinical settings, investment in community-based maternal health programs, elimination of structural barriers to quality prenatal care in majority-Black communities, and strengthening of Maternal Mortality Review Committees — are necessary complements to the aggregate improvements observed in the national trend. The CMS CY 2025 Obstetrical Conditions of Participation, finalized in November 2024, represents an important regulatory step toward accountability for racial disparities in hospital-based maternal care. Whether that accountability translates into measurable rate reduction for Black non-Hispanic women by 2026–2027 will be a critical test of its effectiveness.
References
Centers for Disease Control and Prevention. (2023). Pregnancy-related deaths: Data from maternal mortality review committees in 36 U.S. states, 2017–2019. https://www.cdc.gov/maternal-mortality/php/data-research/index.html
Crear-Perry, J., Maybank, A., Keeys, M., Mitchell, N., & Godbolt, D. (2021). Moving towards anti-racist praxis in medicine. The Lancet, 397(10283), 1457–1459. https://doi.org/10.1016/S0140-6736(21)00596-8
Hardeman, R. R., Burgess, D., Murphy, K., Satin, D., Nielsen, J., Potter, T. M., & Karbeah, J. (2020). Developing a medical school curriculum on racism: Multiyear design process and a critical race theory framework. Academic Medicine, 95(1), 101–105. https://doi.org/10.1097/ACM.0000000000002998
Hoyert, D. L. (2024). Maternal mortality rates in the United States, 2022. NCHS Health E-Stats. https://dx.doi.org/10.15620/cdc/149541
Hoyert, D. L. (2025). Maternal mortality rates in the United States, 2023. NCHS Health E-Stats. https://dx.doi.org/10.15620/cdc/174577
Petersen, E. E., Davis, N. L., Goodman, D., Cox, S., Syverson, C., Seed, K., Shapiro-Mendoza, C., Callaghan, W. M., & Barfield, W. (2019). Vital signs: Pregnancy-related deaths, United States, 2011–2015, and strategies for prevention, 13 states, 2013–2017. MMWR Morbidity and Mortality Weekly Report, 68(18), 423–429. https://doi.org/10.15585/mmwr.mm6818e1
Thoma, M. E., & Declercq, E. (2022). Trends in U.S. maternal mortality rates by sociodemographic group: National Vital Statistics System data, 2009–2019. Obstetrics & Gynecology, 139(5), 853–862. https://doi.org/10.1097/AOG.0000000000004735
World Health Organization. (2019). Maternal mortality. https://www.who.int/news-room/fact-sheets/detail/maternal-mortality
Appendix: AI Interaction Log
Per EPI 553 AI policy: Web search was used for debugging assistance and to verify citation details for references. All analytical decisions, model specifications, interpretations, and written content were developed independently by me Emmanuel Nana Arko. I limited my use to only 2 important areas for clarity: (1) debugging R syntax errors encountered during figure construction, and (2) verifying APA citation formatting for NCHS Health E-Stats publications.I had already used Zotero for my citations but was very specific to verify my citations before submitting my work.