I analyzed 31 GB of CMS data. Here’s what I found.
I wanted to know how much a nursing home’s star rating tells us about its staffing and care outcomes. Three patterns stood out: high overall stars can coexist with low staffing stars; quality-measure stars are more closely related to later emergency-department use; and higher nurse turnover accompanies more emergency visits.
The download archive totaled 31.75 GB across 311 completed files. I checked the whole collection for integrity, then rebuilt the ratings, claims, and staffing files relevant to these questions. The results below come from those selected files, not a model of every byte in the archive.
High overall stars can coexist with low staffing stars
A high overall rating does not guarantee a high staffing rating. In the October 28, 2026 CMS release, 1,273 of 5,939 homes with four or five overall stars and an observed staffing rating had only one or two staffing stars: 21.4%.
Among four-star homes, 809 of 2,813 had low staffing stars (28.8%). Among five-star homes, it was 464 of 3,126 (14.8%). Excluding homes with staffing-related footnotes reduced the combined share to 1,196 of 5,862, or 20.4%.
CMS does not calculate overall stars by averaging the three domain ratings. It starts from the health-inspection rating and applies specific staffing and quality-measure adjustments. A strong overall rating can therefore coexist with a weak staffing rating.
A low staffing rating can also reflect reporting-related penalties. The mismatch is a reason to inspect the component ratings and footnotes, not evidence of fraud or a direct measurement of a resident’s experience.
I preserved CMS certification numbers as identifiers and excluded missing ratings from the denominator. The release contained 14,686 homes; 14,483 had both overall and staffing ratings.
Which stars track later emergency care?
Quality-measure stars were more closely related to next-year long-stay emergency-department use than overall stars. Across the same 9,598 homes, the correlations with 2025 ED use were −0.388 for QM stars and −0.220 for overall stars.
I compared all four rating domains on the same homes. December 20, 2024 ratings were paired with outcomes for care delivered in 2025. A separate comparison paired December 21, 2023 ratings with 2024 care.
The earlier comparison showed the same ordering: QM stars had a correlation of −0.361 with long-stay ED use, versus −0.224 for overall stars, across 9,712 homes. Negative values mean higher ratings accompanied lower ED-use rates. Spearman correlations measure a rank relationship, not a change in visits caused by one additional star.
The common samples required complete data for four rating domains and four claims outcomes within each year. This keeps the rating comparison consistent, but excludes homes with missing measures. CMS certification numbers linked the files, and repeated publications of the same care period were deduplicated.
Earlier care outcomes contribute to quality-measure ratings. The association may partly reflect persistence in those outcomes. It is not evidence that the QM formula independently predicts quality, and many homes appear in both years.
Overall stars still contain information. In a separate 2025 hospitalization comparison, homes with one prior overall star averaged 2.06 hospitalizations per 1,000 resident-days, versus 1.68 for five-star homes (11,114 homes; rank correlation −0.164). There was substantial overlap between rating groups.
These are exploratory facility-level associations. ED visits can be medically necessary; a lower rate alone is not a complete definition of better care.
Higher turnover accompanies more emergency visits
Homes with higher nursing staff turnover also had higher long-stay emergency-department use. In an adjusted 2025 model of 9,218 homes, ten percentage points more turnover was associated with 0.122 more ED visits per 1,000 resident-days. The 95% state-clustered confidence interval was 0.096–0.149.
The raw rank correlation was 0.239 across 10,462 homes in 2025 and 0.216 across 10,722 homes in 2024. The chart groups homes into fixed turnover bands; it describes the data and does not estimate a causal effect.
For 2025, average ED-use rates rose from 1.389 visits per 1,000 resident-days among homes below 30% turnover to 2.250 among homes at 60% or more. The five bands contained 1,483; 2,373; 2,853; 2,153; and 1,600 homes, respectively. Each home receives equal weight.
The adjusted model controlled for state, facility size, occupancy, case mix, Medicaid share, urban location, ownership, chain status, and total staffing hours from Payroll Based Journal data.
This does not show that reducing turnover would prevent that many visits. Turnover and ED use were measured during the same year. Resident needs, staffing difficulties, local labor markets, and financial resources could influence both. State-clustered uncertainty also does not capture every dependence within chains operating across states.
Emergency visits can be necessary and appropriate. The association is a reason to investigate staffing continuity alongside outcomes, not a basis for judging an individual home or patient from turnover alone.
What went into the analysis
The download archive totaled 31.75 GB across 311 completed files. I checked the collection for integrity, then rebuilt the files relevant to these questions. The rebuild reopened 35 monthly CMS archives and reprocessed 10,621,987 daily nursing-staffing records from eight quarters in 2024–2025. The annual panel contained 29,449 facility-year records across 14,793 homes.
Repeated monthly publications were deduplicated by care period, and missing outcomes were excluded rather than treated as zero. Previously derived annual case-mix and payer controls were reused with recorded hashes. Twenty-two planned downloads remained unresolved.
A plan was saved before calculating 72 correlations and 36 adjusted models. It was not externally preregistered, and the archive had been explored previously. The three findings above were selected from that exploratory screen. All results were retained, including weak and adjustment-sensitive findings.
What I would not conclude
These results do not show that raising a rating or reducing turnover would cause better outcomes. Resident needs, resources, local labor markets, and reporting can influence both. Many homes appear in both years, so the second year is not an independent validation sample.
Some relationships were weak: weekend staffing reductions had little association with hospitalization. Contract staffing had a near-zero raw association that changed after adjustment. The adjusted RN-share interval crossed zero. These results matter when judging a search across many possible correlations.
Sources
Analysis completed November 5, 2026. Page design and consolidation updated November 6, 2026. AI-assisted coding and writing. These are exploratory facility-level comparisons, not a peer-reviewed study or predictions of an individual resident’s care.
↑Back to Top