AINPI·ainpi.dev
NDH·2026-08-04
Release update · 2026-08-04

A third of the hospitals.A seventh of the people.

US hospitals in nonmetro counties
34.4%
Of residents living nonmetro
13.8%
Critical Access hospitals
1,338

1,847 of the 5,366 hospitals CMS lists sit in nonmetro counties, which hold 13.8% of US residents. Rural facilities are about 2.5 times as numerous as population alone implies. New rural health section, with a state map and a Pennsylvania deep dive on which hospitals software can actually find.

Read the full report →
methodology v0.7.2-draftNDH 2026-08-04

2026-08-04 update: a third of American hospitals serve a seventh of the people

New section on the site: rural health, with a state-level map, a national baseline, and a hospital-by-hospital Pennsylvania breakdown of which ones software can actually find.

The national number

Joining every hospital CMS lists to the USDA county rural classification:

CountShare
Hospitals CMS lists5,366
In nonmetro counties1,84734.4%
Critical Access1,338
US residents in nonmetro counties45.8M13.8%

A third of the hospitals serve a seventh of the people. Rural facilities are about 2.5 times as numerous as population alone would imply, because distance rather than density decides where a hospital has to be.

That ratio sits under rural health funding formulas, and it is rarely stated in one sentence. A funding formula weighted by population under-serves the facilities. One weighted by facility count over-serves them relative to residents. Both numbers now sit in the same payload, per state, so a reader can check either.

Vermont leads on hospital share at 88.2%. Kansas, Texas and Iowa carry the most Critical Access hospitals. Rural share and rural population rank states differently, which is visible when you sort the table by different columns.

Pennsylvania: counting hospitals is not the same as reaching them

Knowing where hospitals are does not tell you whether the software that routes patients, records and payment can find them. We measured that for Pennsylvania's 187 hospitals against the endpoint directories that certified EHR vendors publish. That single join answers two questions at once: whether a facility is digitally reachable, and which EHR it runs.

The result inverts the usual assumption. Among acute care hospitals we located 26 of 27 rural facilities (96%) against 84 of 105 metro ones (80%). The rural denominator is small, so read the fractions rather than the percentages. Rural runs ahead of metro on every rural/metro split we ran: by hospital type, by ownership, and on the endpoint-linked measure.

The apparent Critical Access gap turned out to be a publishing pattern rather than missing technology. Six of Pennsylvania's 17 Critical Access hospitals do not appear under their own name. Four of the six carry a system brand: three Penn Highlands, one LECOM Health. Penn Highlands published 1 of its 7 Pennsylvania facilities, and that one runs MEDITECH, so the system plainly has a certified EHR. Geisinger, by contrast, published all nine of its facilities.

Dashboard: https://ainpi.dev/states/pa/rural-health

A correction

An earlier version of the Pennsylvania analysis reported that Epic cross-links 49 of 4,445 organizations and concluded that endpoint traversal would find nothing. The count was right and still ships in the payload. The conclusion was wrong, and the error was ours: we measured Organization.endpoint on the matched record and never checked whether the vendor links the other way.

Epic publishes a hierarchy. All 1,187 of its national brand-level organizations carry an endpoint, and the 83,678 facility records beneath them reach it through partOf. Other vendors publish flat, one endpoint per organization. Both are valid FHIR. We had measured only the matched record and drew a conclusion the data did not support.

The corrected finding is more useful anyway: an integration that checks only the record it matched gets a false negative for every hospital published under the hierarchical shape, which is all 86 Epic-published Pennsylvania hospitals in this set. Resolve partOf first.

A bug worth naming

Fact-checking this work turned up a county-name join failure. CMS writes MC KEAN, USDA and Census write McKean, so an uppercase string compare failed and one hospital resolved to no county at all. It dropped out of every county rollup: McKean read as a county with no hospital, while the hospital table listed a hospital in McKean.

One character of whitespace, three wrong published counts (nonmetro hospitals 42 to 43, rural hospitals in a vendor bundle 34 to 35, Critical Access in nonmetro counties 11 to 12), and a self-contradiction between two pages. Both pipelines now normalize county names to alphanumerics and warn loudly if any hospital fails to resolve.

Method

Rural means the hospital's county carries USDA ERS Rural-Urban Continuum Code 4 to 9 for 2023. Critical Access is the CMS facility-level designation. County classification means a metro-county hospital may still serve a rural population. 239 hospitals could not be matched to a county code and are reported in each state's unmatched field rather than assigned to either group. Territories are excluded because ERS publishes no continuum code for them.

Everything here is public data and costs nothing to reproduce.