Rural Americans have died at meaningfully higher rates than urban Americans for 20+ years — and the gap is mostly a within-state, not a between-state, story

Summary

Rural (nonmetro) America has had meaningfully worse health outcomes than urban (metro) America for at least the last 20-25 years, and the gap has been widening, not narrowing. CDC/NCHS data show the age-adjusted all-cause death rate in rural counties was 7% higher than urban counties in 1999; by 2019 it was 20% higher, because urban rates kept falling while rural rates stalled after about 2010. For working-age adults (25-54), the gap is far starker: rural natural-cause mortality went from roughly equal to urban in the mid-1990s to 43% higher by 2019. A current cross-section (County Health Rankings, 2025 release) shows nonmetro counties still running an 18% higher premature-death rate than metro counties (11,051 vs 9,359 years of life lost before age 75, per 100,000 people).

On the second half of the question: most of the health gap is a within-state story, not a between-state one. Decomposing the county-to-county variance in premature-death rates nationally, only about 29% of the total variance is explained by which state a county is in. The rural/urban split itself, measured within states, explains a much smaller share (about 6% of total variance). The largest share by far — about 65% — is variance between individual places within the same state that share the same metro/nonmetro status: two nonmetro counties in the same state can differ from each other as much as a nonmetro county differs from a metro one. In other words, rural disadvantage is not simply "which state you live in" or even "rural vs urban" as a category — it is highly local, driven by county-specific conditions (economic base, provider access, demographics) that a state-level or even a simple rural/urban label does not capture.

How much worse, over the last 20 years

Source: CDC/NCHS Data Brief 417, "Trends in Death Rates in Urban and Rural Areas: United States, 1999-2019" (NVSS age-adjusted death rates).

• 1999: rural 923.8 deaths per 100,000 vs. urban 865.1 — rural 7% higher.
• 2010: rural rate had fallen to 837.6; urban continued falling.
• 2019: rural 834.0 vs. urban 693.4 — rural now 20% higher. Urban rates fell ~20% over the two decades; rural rates essentially flatlined after 2010.
• For the 10 leading causes of death, rural rates exceeded urban in every category by 2019, led by heart disease (189.1 vs. 156.3 per 100k), cancer (164.1 vs. 142.8), and chronic lower respiratory disease (52.5 vs. 35.4) — and these particular gaps widened over the two decades.
• The disparity is sharpest at working ages. USDA ERS's analysis of the same NVSS data found the natural-cause mortality rate for adults 25-54 was roughly equal between rural and urban areas in 1999 (rural 6% higher) but 43% higher in rural areas by 2019 — driven by urban rates falling 37% for this age group while rural rates rose 14%.

A current cross-section confirms the disparity persists past 2019: County Health Rankings' 2025 release (built from recent NCHS multi-year mortality files) shows the average nonmetro county's premature-death rate (11,051 YPLL/100k) is 18.1% above the average metro county's (9,359 YPLL/100k), computed here directly from health.chr_premature_death joined to geo.rural_urban_continuum across 3,080 U.S. counties.

Between states, or between places in the same state? — a variance decomposition

To answer this directly rather than assume a narrative, we ran an ANOVA-style decomposition of the total county-to-county variance in the current premature-death rate (n=3,080 counties with both a CHR rate and a rural-urban classification):

Reading this: if you wanted to predict a county's premature-death rate, knowing its state gets you about 29% of the explainable variance nationally. Knowing whether it is metro or nonmetro within that state gets you only another 6 points. The remaining two-thirds of the variation is neither — it is the difference between one nonmetro county and its neighboring nonmetro county in the same state, or one metro county and another metro county in the same state. This matches the qualitative finding in USDA ERS's own report on the nature of the gap ("The Nature of the Rural-Urban Mortality Gap," EIB-265): within rural areas, the more rural the county, the worse the trend, and effects vary substantially by county, sex, race/ethnicity, region, and state simultaneously — no single grouping (state or rural/urban alone) accounts for most of the spread.

Bottom line for the second half of the question: the rural-urban health gap is not mainly a matter of which states are worse off. It is overwhelmingly a matter of which specific places are worse off, even among places with the same rural/urban label inside the same state.

Methodology notes and what this corpus could and could not directly verify

• The 20-year TREND (1999-2019, all-cause and cause-specific) is drawn from CDC/NCHS Data Brief 417 and USDA ERS's EIB-265, both fetched and read directly for this report — this warehouse does not carry a county- or state-level mortality time series long enough to reproduce that trend itself (health.cdc_mortality is state-level annual 1999-2017/weekly 2014-2023 but is not rural/urban-classified; health.chr_premature_death, the one table here with both county grain and a rural/urban join key via geo.rural_urban_continuum, is loaded as a single current-year (2025) release, not a multi-year series).
• The variance decomposition (between-state vs. rural/urban vs. residual) was computed directly against this corpus's current cross-section, not against a 20-year window — it answers "where does the current gap sit geographically," a separate but directly relevant question from the trend, and the ERS EIB-265 report's qualitative finding (heavy county-level heterogeneity, not explained by state or rural/urban categories alone) corroborates the same conclusion over the full 1999-2019 window.
• geo.rural_urban_continuum is the 2023 USDA vintage (RUCC updated decennially), joined by county FIPS to health.chr_premature_death's 2025-release county rows; 3,080 of 3,152 CHR county rows matched (72 counties dropped for missing rate or unmatched RUCC, mostly suppressed-count counties) — dropping them is unlikely to change the 29%/6%/65% split materially since they are a small, geographically scattered share of counties, but this was not separately re-tested with the exclusion reversed.
• County counts, not populations, were used as decomposition weights (standard for a place-to-place/county-level question); a population-weighted version would shift some weight toward large metro counties but would not change which bucket (state vs. rural/urban vs. residual) dominates.

Every query behind this report

3 warehouse calls ran in this session, in order. Each is reproducible against the same snapshot.

SELECT release_year, count(*) FROM health.chr_premature_death WHERE geo_level='county' GROUP BY release_year ORDER BY release_year

SELECT r.metro_nonmetro, count(*) n, avg(c.premature_death_rate) avg_rate, stddev_samp(c.premature_death_rate) sd
FROM health.chr_premature_death c
JOIN geo.rural_urban_continuum r ON c.fips_code = r.county_fips
WHERE c.geo_level='county' AND c.premature_death_rate IS NOT NULL
GROUP BY r.metro_nonmetro

WITH base AS (
  SELECT c.premature_death_rate AS y, r.state_fips AS st, r.metro_nonmetro AS mn
  FROM health.chr_premature_death c
  JOIN geo.rural_urban_continuum r ON c.fips_code = r.county_fips
  WHERE c.geo_level='county' AND c.premature_death_rate IS NOT NULL
),
grand AS (SELECT avg(y) gm, count(*) tn FROM base),
state_means AS (
  SELECT st, avg(y) sm, count(*) scnt FROM base GROUP BY st
),
cell_means AS (
  SELECT st, mn, avg(y) cm, count(*) ccnt FROM base GROUP BY st, mn
)
SELECT
 (SELECT gm FROM grand) AS grand_mean,
 (SELECT tn FROM grand) AS total_n,
 (SELECT sum((y-(SELECT gm FROM grand))*(y-(SELECT gm FROM grand))) FROM base) AS SST,
 (SELECT sum(scnt*(sm-(SELECT gm FROM grand))*(sm-(SELECT gm FROM grand))) FROM state_means) AS SSB_state,
 (SELECT sum(ccnt*(cm-sm)*(cm-sm)) FROM cell_means cmx JOIN state_means smx ON cmx.st=smx.st) AS SSB_metro_within_state