Rural-urban overdose gap narrowed to near zero by 2015, then reversed and widened to +2.4 deaths/100k (urban over rural) by 2020 — no official split exists past 2020

Summary

Since 2011 (the earliest year inside a true 'last fifteen years' window), the rural–urban overdose gap has not moved in one direction — it narrowed to near zero, flipped sign, and then widened again, ending up about the same size it started but on the opposite side. In 2011 rural counties had the higher age-adjusted overdose death rate (14.7 vs 12.9 per 100,000, rural minus urban = 1.8). That rural excess shrank steadily through 2015 (17.0 vs 16.2, a gap of just 0.8), then reversed: urban counties overtook rural in 2016 and the urban excess grew to 2.4 per 100,000 by 2019 and again in 2020 (28.6 urban vs 26.2 rural) — the widest gap of the whole 1999-2020 series, and now pointing the opposite way from where it started.

The honest limit of this answer: CDC/NCHS has not published an official age-adjusted urban/rural breakdown for any year after 2020 — the last such report (Data Brief No. 440) came out in July 2022 covering 2020 data, and no successor brief exists as of September 2026. So roughly the last 5-6 years of the 15-year window asked about (2021-2026) cannot be answered with the same authoritative, age-adjusted measure. I attempted to fill that gap myself from this connector's own county-level provisional overdose data joined to USDA rural-urban codes, but that computation diverged sharply from the official figures — see the data-quality section below for exactly what that connector query excluded and why. I am reporting that as a genuine data-coverage limitation rather than presenting an unreliable number as the answer.

The full 1999-2020 pattern (why 'widening or narrowing' has no single answer)

NCHS's own two data briefs on this exact question (No. 403 for 1999-2019, No. 440 for 2020) describe the pattern as oscillating, not monotonic:

• 1999-2003: urban rates higher than rural (gap starts at 2.4 per 100,000 in 1999).
• 2004-2006: rates statistically similar — the gap crosses zero for the first time.
• 2007-2015: rural rates higher than urban — the gap flips sign and widens to its most negative point, -1.8, in 2011, then narrows back toward zero by 2015 (-0.8).
• 2016-2020: urban rates higher than rural again — the gap flips sign a second time and grows to +2.4 by 2019-2020.

A companion peer-reviewed analysis (Post et al., JAMA Network Open, 2022, using the same CDC WONDER mortality data at a finer 6-category urbanicity scale) independently corroborates this shape: 'Differences in OOD [opioid overdose death] rates by urbanicity were largest at the start and end of the study period... The initial rank order, with urban counties having the highest rates and rural the lowest, reemerged by 2020.' That paper also found rural counties had the faster acceleration in rates during the fentanyl-driven wave 3 (2013-2019) even as urban counties kept the higher absolute rate — i.e., rural areas were catching up in relative terms even while the raw gap widened in urban's favor.

What's driving the reversal: the drug supply changed, not just demographics

The mid-2010s flip from 'rural higher' to 'urban higher' tracks a change in which drugs were killing people. NCHS's drug-type breakdown (also from Data Brief No. 403) shows: rates of overdoses involving natural/semisynthetic opioids (oxycodone, hydrocodone) — the first-wave prescription-opioid crisis — were higher in rural counties from 2004 through 2017, consistent with rural areas' heavier initial burden. But synthetic opioids other than methadone (illicit fentanyl) flipped from rural-higher (2001-2014) to urban-higher (2015-2019) — by 2019 the urban fentanyl-death rate (11.9) was 1.4 times the rural rate (8.3) — and fentanyl became the dominant driver of overdose deaths overall. Heroin and cocaine deaths were consistently higher in urban counties throughout 1999-2019. Meanwhile methamphetamine ('psychostimulants with abuse potential') deaths have been higher in rural counties every year since 2012, reaching 1.4x the urban rate by 2019 (6.7 vs 4.8 per 100,000) — a genuinely rural-favoring trend running in the opposite direction of the overall gap.

What happened after 2020 — the parts that can and can't be answered

National (not rural/urban-split) data confirms overdose deaths kept rising through the pandemic and peaked around August 2023 (national rate ~33.2 per 100,000, per Post et al., JAMA Network Open, 2025), then fell about 24-27% by late 2024 — the sharpest one-year national decline on record, driven mainly by falling fentanyl-involved deaths (Friedman et al., medRxiv preprint, posted 2026). None of that recent literature, however, breaks the decline out by urban vs. rural residence. State-level snapshots that do exist for this period point in different directions and don't resolve the national picture: Pennsylvania's rural counties had a materially higher overdose rate than its urban counties in 2022 (42.3 vs 34.0 per 100,000, PA Center for Rural Pennsylvania), while my own connector-derived, uncorrected 2021-2025 county data (see caveat below) shows both metro and non-metro segments falling together since roughly 2022-2023, consistent with the national decline, but cannot be trusted for the exact gap size.

Data-quality note on the connector's own attempt to extend the series

I queried health.cdc_county_overdose_deaths (CDC's provisional, 12-month-ending county feed) joined to geo.rural_urban_continuum (USDA metro/nonmetro codes) for December of each year 2020-2025. The resulting crude, unadjusted rates: Metro 28.5/100k vs Nonmetro 15.7/100k in 2020 (a 12.8-point gap) — compare that to NCHS's own official, age-adjusted 2020 figure of 28.6 urban vs 26.2 rural (a 2.4-point gap) for effectively the same year. The two disagree by roughly a factor of five on gap size.

Investigating why, I ran SELECT o.year, ... SUM(CASE WHEN o.footnote IS NOT NULL AND o.footnote <> '' THEN 1 ELSE 0 END) AS suppressed_rows, COUNT(*) AS total_rows FROM health.cdc_county_overdose_deaths o WHERE o.month='12' GROUP BY o.year. The predicate o.footnote IS NOT NULL AND o.footnote <> '' identifies exactly the county-months CDC itself excluded from any usable death count: between 1,399 and 1,630 of the 3,144 county-month rows per year (44-52%) were excluded/suppressed because the true death count was 1-9 and CDC withholds it for privacy. Those excluded county-months are not a random sample: suppression applies only to small counties, which are disproportionately rural/nonmetro — so the SUM(deaths) figure used above for Nonmetro is built from a systematically rural-skewed subset with roughly half of all rural counties' deaths dropped from the total entirely. There is no 'headline number without this exclusion' to report, because CDC itself never releases the true underlying counts for the excluded/suppressed rows — the exclusion is baked into the source feed, not a filtering choice made in this analysis. On top of that undercount, this feed's rate is crude (not age-adjusted), and rural populations skew older, which further inflates the true age-adjusted rural rate relative to what a simple crude count would show. Net effect: the connector's own 2020 nonmetro rate of 15.7/100k is almost certainly a substantial underestimate of the true rate, which is why it diverges so sharply from NCHS's official, non-suppressed 26.2 figure for the same population and year. I am flagging this as a genuine, currently unresolved coverage gap rather than presenting the connector's own crude figure as if it were comparable to the official NCHS series.

What This Report Does Not Answer

• Cover the full 'last fifteen years' (approximately 2011-2026): No official NCHS age-adjusted urban/rural breakdown exists past 2020 (the most recent such report, July 2022, covers 2020 data only). I attempted to extend the series myself using this connector's provisional county-level data but found it diverges from the official figures by roughly 5x due to heavy privacy suppression of small (mostly rural) counties, and reported that as a data-quality limitation rather than presenting an unreliable extension as fact. Approximately the final 5-6 years of the asked-about window (2021-2026) are therefore not answerable with an authoritative rural/urban rate gap.

Every query behind this report

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

SELECT MIN(year), MAX(year), COUNT(DISTINCT year) FROM health.cdc_county_overdose_deaths

SELECT DISTINCT month FROM health.cdc_county_overdose_deaths WHERE year='2023' ORDER BY month

SELECT o.year, r.metro_nonmetro,
  SUM(CAST(o.provisional_drug_overdose AS BIGINT)) AS deaths,
  SUM(r.population) AS pop
FROM health.cdc_county_overdose_deaths o
JOIN geo.rural_urban_continuum r ON o.fips = r.county_fips
WHERE o.month = '12' AND o.provisional_drug_overdose NOT IN ('Suppressed') 
GROUP BY o.year, r.metro_nonmetro

SELECT o.year, r.metro_nonmetro,
  SUM(TRY_CAST(o.provisional_drug_overdose AS BIGINT)) AS deaths,
  SUM(r.population) AS pop,
  COUNT(*) AS n_counties
FROM health.cdc_county_overdose_deaths o
JOIN geo.rural_urban_continuum r ON o.fips = r.county_fips
WHERE o.month = '12'
GROUP BY o.year, r.metro_nonmetro
ORDER BY o.year, r.metro_nonmetro

SELECT o.year, AVG(TRY_CAST(o.percentage_of_records_pending AS DOUBLE)) AS avg_pending,
  SUM(CASE WHEN o.footnote IS NOT NULL AND o.footnote <> '' THEN 1 ELSE 0 END) AS suppressed_rows,
  COUNT(*) AS total_rows
FROM health.cdc_county_overdose_deaths o
WHERE o.month='12'
GROUP BY o.year ORDER BY o.year