Payer mix is one of the strongest, most consistently documented predictors of hospital closure risk

Summary

Payer mix tracks hospital closure risk closely and is one of the most consistently identified drivers in the research literature -- more so than almost any other single hospital characteristic besides raw financial performance (which payer mix itself feeds). Hospitals that depend heavily on Medicaid, Medicare, and uncompensated/self-pay care -- and have thin or negative operating margins as a result -- close at multiples of the rate of hospitals with a more commercially-insured patient base. The clearest natural experiment is Medicaid expansion: hospitals in expansion states were found to be roughly 62-84% less likely to close than otherwise-similar hospitals in non-expansion states, entirely through the payer-mix/uncompensated-care channel. A 2026 machine-learning study of 2,683 rural hospitals (139 closures, 2011-2022) confirms payer-mix-adjacent variables -- Medicaid expansion timing, uninsured share, financial margin volatility -- rank among the top predictors, alongside low patient volume.

An AskAmerica cross-check using CMS hospital quality ratings against county-level uninsured rates found only a weak, not-quite-significant association (p=0.069) -- expected, because CMS star ratings measure clinical quality, not financial distress, and are therefore a poor proxy for closure risk. The load-bearing evidence here is the published financial-statement and closure-registry literature, not this warehouse correlation, which is included only as a triangulation check.

What the literature shows directly

• Financial-vulnerability studies (rapid systematic review, U.S. studies Jan 2024-Jun 2025): closed hospitals had a higher Medicare discharge share, higher uncompensated care burden, and greater variability in Medicaid payer mix than hospitals that stayed open, plus fewer beds, lower occupancy, and more negative operating margins.
• 2026 explainable-ML study (BMC Health Services Research): among 2,683 rural U.S. hospitals tracked 2011-2022, 139 (5.2%) closed. The model (86% recall, 89% AUC) found closure risk driven by sustained financial weakness/volatility (total margin, ROE variability, current ratio), low/volatile patient volume, years since Medicaid expansion, and community uninsured/underserved share -- i.e., payer-mix-adjacent variables sit among the top predictor categories, alongside financial and operational instability.
• Medicaid expansion as a natural experiment: a Health Affairs analysis found hospitals in Medicaid-expansion states were about 84% less likely to close than those in non-expansion states; a separate rural-focused analysis put the gap at roughly 62% for rural hospitals specifically. Non-expansion/newly-expansion states accounted for about 74% of all rural closures in one accounting. Mechanism: expansion cuts uncompensated care costs and improves operating margins, directly altering the payer mix that drives closure risk.
• Sheps Center (UNC) rural hospital closure tracker: 197 rural hospital closures/conversions since January 2005 (109 complete closures + 88 conversions to reduced-service facilities), 154 since 2010. Separately, Sheps Center identifies at least 338 rural hospitals nationally with either consistent negative margins or a high Medicaid payer mix -- the pool structurally exposed to closure risk today.
• Public Citizen \"at-risk hospital\" screen (2025): used a payer-mix threshold (>=20% Medicaid and other low-income government payers) combined with negative average net margins 2022-2024 to flag 446 hospitals nationally as most exposed to Medicaid-funding cuts -- payer mix was the primary screening variable, not a secondary one.

AskAmerica cross-check (triangulation, not the primary evidence)

The AskAmerica corpus has no hospital-closure event table, so a direct replication of the closure studies above was not possible here. As a partial, weaker check, CMS overall star ratings (health.cms_hospital_quality via health.hospital_with_geography) were joined to county-level uninsured rates (census.sahie_insurance, 2023, county grain) via county FIPS. The SQL filter used was h.overall_rating <> 'Not Available' AND h.overall_rating IS NOT NULL (and, in the metro/nonmetro breakdown, the unaliased equivalent overall_rating <> 'Not Available' AND overall_rating IS NOT NULL) -- this excluded every CMS-unrated hospital, a group disproportionately made up of small, critical-access, psychiatric, or newly-opened facilities. Excluding them left 3,130 rated facilities in the chart above. No with/without comparison on the star-rating outcome itself is possible for the excluded group (they carry no rating to average), but the direction of the bias is worth stating explicitly: the excluded, unrated hospitals are exactly the small/rural facilities the literature flags as highest closure risk, so if anything this warehouse check understates the true payer-mix/risk relationship by dropping the most exposed facilities from the sample entirely.

Binning counties into uninsured-rate quintiles and testing the trend in average hospital star rating (among the 3,130 rated facilities) gave a slope of -0.073 stars per quintile step (p=0.069, not significant at 0.05, and not strictly monotonic). Nonmetro hospitals average a slightly lower star rating (3.03 vs. 3.28 for metro) and sit in slightly higher-uninsured counties (9.8% vs. 9.2%) on average (n=905 nonmetro / 2,231 metro rated facilities). These numbers are directionally consistent with the literature (worse outcomes and higher uninsured exposure co-locate) but far too weak, on their own, to establish the relationship -- because star rating is a clinical-quality measure, not a financial-solvency measure, and is therefore several steps removed from the actual closure mechanism (margin, liquidity, uncompensated care as a share of revenue). This gap is itself informative: it explains why the load-bearing evidence for this question has to come from studies that directly measured hospital financial statements and payer-mix composition, which this warehouse does not carry at hospital grain.

Bottom line and confounds not held fixed

Payer mix -- specifically the share of Medicaid, Medicare, and uninsured/self-pay patients relative to better-reimbursing commercial insurance -- is one of the most robust predictors of hospital closure identified across multiple independent research designs (retrospective financial-statement studies, ML classifiers, and the Medicaid-expansion quasi-experiment). The size of the effect in the expansion literature (a 60-85% relative reduction in closure odds) is large by the standards of health-policy research. What none of these studies (or this check) isolate is payer mix in complete isolation from hospital size, rurality, ownership, and local market competition -- all of which are correlated with payer mix and also independently affect closure risk, so the reported effect sizes are associations/quasi-experimental estimates, not a pure payer-mix-only causal effect net of every other factor.

Every query behind this report

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

SELECT year, COUNT(*) FROM census.sahie_insurance WHERE geography='county' GROUP BY year ORDER BY year DESC LIMIT 5

SELECT DISTINCT overall_rating FROM health.hospital_with_geography LIMIT 20

SELECT h.metro_nonmetro, CAST(h.overall_rating AS DOUBLE) rating, s.uninsured_rate
FROM health.hospital_with_geography h
JOIN census.sahie_insurance s ON h.county_fips = s.county_fips AND s.geography='county' AND s."year"='2023'
WHERE h.overall_rating <> 'Not Available' AND h.overall_rating IS NOT NULL

SELECT CAST(h.overall_rating AS DOUBLE) rating, s.uninsured_rate
FROM health.hospital_with_geography h
JOIN census.sahie_insurance s ON h.county_fips = s.county_fips AND s.geography='county' AND s."year"='2023'
WHERE h.overall_rating <> 'Not Available' AND h.overall_rating IS NOT NULL

SELECT h.metro_nonmetro, COUNT(*) n, AVG(CAST(h.overall_rating AS DOUBLE)) avg_rating, AVG(s.uninsured_rate) avg_uninsured
FROM health.hospital_with_geography h
JOIN census.sahie_insurance s ON h.county_fips = s.county_fips AND s.geography='county' AND s."year"='2023'
WHERE h.overall_rating <> 'Not Available' AND h.overall_rating IS NOT NULL
GROUP BY h.metro_nonmetro

SELECT type, COUNT(*) FROM census.sahie_insurance WHERE geography='county' AND "year"='2023' GROUP BY type

SELECT COUNT(*) total, COUNT(DISTINCT facility_id) distinct_fac FROM health.hospital_with_geography WHERE overall_rating <> 'Not Available' AND overall_rating IS NOT NULL

SELECT DISTINCT h.facility_id, CAST(h.overall_rating AS DOUBLE) rating, s.uninsured_rate
FROM health.hospital_with_geography h
JOIN census.sahie_insurance s ON h.county_fips = s.county_fips AND s.geography='county' AND s."year"='2023'
WHERE h.overall_rating <> 'Not Available' AND h.overall_rating IS NOT NULL

SELECT h.metro_nonmetro, COUNT(DISTINCT h.facility_id) n, AVG(CAST(h.overall_rating AS DOUBLE)) avg_rating, AVG(s.uninsured_rate) avg_uninsured
FROM (SELECT DISTINCT facility_id, county_fips, metro_nonmetro, overall_rating FROM health.hospital_with_geography WHERE overall_rating <> 'Not Available' AND overall_rating IS NOT NULL) h
JOIN census.sahie_insurance s ON h.county_fips = s.county_fips AND s.geography='county' AND s."year"='2023'
GROUP BY h.metro_nonmetro