How fast do nursing homes fix what surveyors cite? 28.5 days for the harmful ones

When a federal surveyor walks into a nursing home and writes a deficiency, a clock starts. The facility has to submit a plan of correction, fix the underlying problem, and document the date the correction was complete. That correction date is recorded in the same public file that lists the citation itself. Lay the two dates side by side across every citation in the country and you get something the star rating never shows you: not whether a facility was cited, but how long it took to make the problem go away.

We measured that lead time for every corrected health deficiency in the CMS Care Compare Nursing Home Health Deficiencies file — 415,849 citations that carry both a survey date and a documented correction date, drawn from a snapshot frozen on 25 May 2026. The headline is simple. Across all severities, the mean correction time is 32 days. For the citations that matter most — the harm-level findings at CMS Severity G and above — it is 28.5 days, with a median of 26. And the single most dangerous category, immediate jeopardy, is corrected fastest of all, in 26 days on average.

The 55-word version. Across 415,849 corrected CMS nursing home health deficiencies (snapshot May 2026), the mean correction time is 32 days. Harm-level violations — CMS Severity G and above — close faster, in 28.5 days on average, with a median of 26. The most severe immediate-jeopardy citations close fastest, at 26 days. Texas and Illinois correct quickest; Washington, D.C. slowest.

The severity inversion

The intuitive story is that harder problems take longer to fix. A leaking roof takes longer than a missing sign; a systemic infection-control failure takes longer than a single mislabeled medication cart. So you would expect the most severe deficiencies — the ones CMS scores as causing actual harm or immediate jeopardy — to sit open the longest. The data says the opposite.

Sort the four CMS harm bands by mean correction time and they line up almost perfectly in reverse order of severity. No-harm citations (Severity A–C) take 32.4 days on average. Minimal-harm citations (D–F), the overwhelming bulk of the file at 380,846 corrected rows, take 32.6 days. Actual-harm citations (G–I) drop to 30.0 days. And immediate-jeopardy citations (J–L) — the findings CMS defines as posing an immediate threat to resident life or safety — close in 26.2 days, more than six days faster than the least serious band.

This is not a quirk of a few outliers. The pattern holds at the median (30 days for the two least-severe bands, 27 for actual harm, 25 for immediate jeopardy) and it holds across more than twenty thousand harm-level citations. The inversion is real, and the explanation is regulatory, not clinical.

Under the federal enforcement framework at 42 CFR Part 488, a facility cited for immediate jeopardy is given almost no slack. Survey teams flag immediate jeopardy in real time, and the facility must remove the jeopardy — bring the situation below the immediate-threat threshold — on a compressed timeline, typically measured in days, or face the most severe remedies CMS can impose: denial of payment for new admissions, civil money penalties that accrue daily, and ultimately termination from Medicare and Medicaid. The clock on a J, K, or L citation is short by design. A facility that lets one sit open is not risking a slow administrative letter; it is risking its participation in the programs that pay most of its bills.

The least-severe citations carry no such pressure. A Severity D finding — the most common single category in the entire file — is an isolated deficiency with the potential for, but no actual, more-than-minimal harm. The plan-of-correction process still applies, but the consequence of a longer correction window is mild. So the least urgent findings drift, and the most urgent ones are closed under the threat of the heaviest penalties in the federal toolkit. The result is the inversion you see in the table below.

The pattern runs against intuition. The deficiencies that put residents in immediate jeopardy are corrected fastest of all — not because they are easier to fix, but because federal rules give the facility almost no time to leave them open.

Correction time by harm band

The figures below are computed from the corrected-citation universe: every row with both a survey date and a documented correction date, with negative spans (a CMS data artifact discussed in the limitations) removed. p90 is the 90th percentile — the value below which nine in ten corrections land.

A few things are worth drawing out. First, the harm-level row — actual harm plus immediate jeopardy combined — is the study's anchor: 20,297 corrected citations where a surveyor documented real harm to a resident, closed in a mean of 28.5 days. That is the number to remember. Second, the p90 column tells a quieter story than the means. Even though immediate-jeopardy citations have the lowest mean, their p90 (51 days) is close to every other band. The fast average is driven by the bulk of cases being closed quickly under regulatory pressure; the slow tail still exists. Roughly one in ten harm-level citations stays open for more than 53 days. Third, the actual-harm band (G–I) has the highest p90 of any band at 54 days — these are serious findings that did not rise to immediate jeopardy, and without the compressed jeopardy clock, their long tail runs the longest of all.

The all-corrected mean of 32.4 days and median of 30 days are the right summary statistics for the file as a whole, and they are dominated by the minimal-harm band simply because that band is 93 percent of the corrected universe. When a facility, a journalist, or a regulator talks about how long it takes nursing homes to fix problems, the honest single number depends on which problems they mean. For everything, it is about a month. For the harmful findings specifically, it is closer to four weeks — and falling as severity rises.

State by state: a two-to-one spread

Correction time is not evenly distributed across the country. Limiting the ranking to states with at least 500 corrected citations (to keep small-sample territories from dominating the extremes), the mean correction time ranges from 16.5 days in Texas to 64.5 days in Washington, D.C. — a near four-fold spread, and a two-to-one gap even between the second-fastest and second-slowest jurisdictions.

The ten fastest correctors (mean days, all corrected citations):

The ten slowest correctors:

Texas and Illinois are in a category of their own. Both close more than 24,000 citations apiece, so this is not a small-sample mirage — these are two of the largest long-term-care markets in the country, each resolving the typical deficiency in about two and a half weeks. California, the single largest market in the file at 57,629 corrected citations, sits sixth at 27.1 days. At the other end, Washington, D.C. takes nine weeks on average and New York close to eight.

It is tempting to read fast correction as good and slow correction as bad, but the metric does not support that reading on its own. A correction date is the date the facility documents that it fixed the problem; it is a measure of administrative throughput and regulatory cadence, not of care quality. State survey agencies differ in how they schedule revisits, how they accept plans of correction, and how they record completion. A state that requires an on-site revisit before recording a correction will mechanically show longer lead times than one that accepts documented self-attestation. The spread is real and worth attention, but it measures the speed of the correction process, not the goodness of the underlying care.

The Illinois paradox

The state-level results become genuinely interesting when read alongside our earlier work on the same dataset. In our companion analysis of nursing-home harm rates, Illinois had the highest rate of harm-level (Severity G+) citations per facility cited of any state — 4.57, against a national pattern far below it. Yet here, Illinois corrects faster than every state but Texas, closing the typical deficiency in 16.8 days.

High citation intensity and fast correction are not a contradiction; together they describe a particular enforcement posture. Illinois runs an aggressive survey regime that documents a large volume of serious findings, and it pairs that with a correction process that closes them quickly. New York shows the mirror image on the correction axis — a large market that takes nearly eight weeks to close the average citation. Neither number, on its own, tells you whether a resident in Chicago or Buffalo is safer. What the two studies together demonstrate is that the federal data, frozen and joined, can separate enforcement intensity from enforcement speed — two things that the public star rating blends into a single, lagging summary.

Why this is not in the star rating

The Five-Star Quality Rating that consumers see on Care Compare folds health inspections, staffing, and clinical quality measures into one number. The health inspection component is built from the weighted count and severity of deficiencies over the most recent survey cycles. It is, by construction, a backward-looking measure of how many problems a facility had and how bad they were. It says nothing about how quickly the facility responded.

Correction speed is a different signal entirely. It is the operational reflex of a facility under regulatory pressure — the gap between being told something is wrong and documenting that it is fixed. Two facilities with identical citation records can behave completely differently here: one closes its harm-level findings in two weeks, the other lets them sit for two months. The star rating treats them the same. The correction-date field, sitting unused in the same public file, distinguishes them cleanly.

This is the recurring theme of our care-quality desk. The federal data carries operational signals that the consumer-facing ratings discard. We found the same thing when we partitioned the SNF quality decline by ownership change in our analysis of the Q1 2026 quality drop, and again in the timing of provider deactivations. The ratings are a trailing summary; the underlying files are a daily record. When the two disagree, the underlying file leads.

Methodology

The analysis aggregates the CMS Care Compare Nursing Home Health Deficiencies file (table nh_health_deficiencies), a frozen snapshot taken on 25 May 2026. The file contains 418,148 citations across 14,635 nursing homes, with survey dates ranging from March 2017 to March 2026.

The correction lead time for a citation is correction_date − survey_date, expressed in whole days. Of the 418,148 citations, 415,849 carry both a survey date and a correction date; the remaining 2,299 have no correction date recorded and are treated as open and excluded. A further 5,126 citations record a correction date earlier than the survey date — a data-entry artifact in the source file — and are excluded outright rather than clamped to zero, leaving 410,723 citations with a valid non-negative span.

Severity bands follow the CMS scope-and-severity scale: no harm (A–C), minimal harm (D–F), actual harm (G–I), and immediate jeopardy (J–L). "Harm-level" throughout this study means Severity G and above (G–L). For each band we report the count, the arithmetic mean, the median (50th percentile), and the 90th percentile, computed with continuous-percentile interpolation so the figures reconcile with the percentile_cont aggregate in the published query. State rankings use the same correction-time definition, restricted to states with at least 500 corrected citations.

The exact query is reproduced below and embedded in the reproducibility block at the foot of this page. It runs unmodified against the nh_health_deficiencies table and returns the three result sets behind every figure in this study — the harm-band table, the harm-level headline, and the state ranking.

-- Correction time by CMS harm band (no harm A-C, minimal D-F,
-- actual harm G-I, immediate jeopardy J-L). Open citations and
-- negative spans excluded.
with corrected as (
  select
    (correction_date - survey_date) as days,
    upper(left(scope_severity_code, 1)) as sev
  from nh_health_deficiencies
  where survey_date is not null
    and correction_date is not null
    and (correction_date - survey_date) >= 0
)
select
  case
    when sev in ('A','B','C') then '1 no harm (A-C)'
    when sev in ('D','E','F') then '2 minimal harm (D-F)'
    when sev in ('G','H','I') then '3 actual harm (G-I)'
    when sev in ('J','K','L') then '4 immediate jeopardy (J-L)'
  end as harm_band,
  count(*) as n,
  round(avg(days)::numeric, 2) as mean_days,
  round(percentile_cont(0.5) within group (order by days)::numeric, 1) as median_days,
  round(percentile_cont(0.9) within group (order by days)::numeric, 1) as p90_days
from corrected
where sev in ('A','B','C','D','E','F','G','H','I','J','K','L')
group by harm_band
order by harm_band;

The computation is also available as a typed, unit-tested helper in the Fonteum codebase (src/lib/research/time-to-correction.ts), which classifies severity, excludes invalid and negative spans, and computes the same mean/median/p90 from plain records — so the published figures can be reproduced from the raw rows without re-deriving the SQL.

Limitations

• A correction date is an administrative event, not a clinical outcome. It records when a facility documented that a deficiency was corrected. It does not measure whether the correction held, whether residents were harmed in the interim, or whether the underlying problem recurred.
• Open citations are excluded. The 2,299 citations with no recorded correction date — the still-open findings — are not in any average here. If the slowest corrections are disproportionately still open at snapshot time, the true correction times are slightly understated.
• Negative spans are excluded, not corrected. 5,126 rows record a correction date before the survey date. These are source data-entry artifacts. We drop them rather than guess at the intended dates.
• State differences reflect process, not only performance. Survey agencies differ in revisit scheduling and in how completion is recorded. A longer mean in one state can reflect a stricter revisit requirement rather than slower remediation.
• Severity is CMS's classification, not ours. Scope-and-severity codes are assigned by CMS surveyors. We group and count them; we do not re-adjudicate any citation's severity.
• No facility-level claims. This study reports national and state aggregates only. It does not rank, rate, endorse, or warn against any individual facility, and it attaches no citation to any provider profile.
• Snapshot, not live. Figures reflect the 25 May 2026 snapshot. CMS updates the source file on a recurring cadence, and the numbers will move with each release.

Sources and provenance

The underlying data is a U.S. Government Work in the public domain. Every figure in this study traces to the frozen snapshot recorded in the reproducibility block below, whose SHA-256 digest is published so the result set can be checked byte-for-byte. Fonteum reports CMS-published enforcement records; it does not independently inspect or rate any facility.

• CMS Provider Data Catalog — Nursing Homes (source file) ↗
• CMS — Certification & Compliance for Long-Term Care ↗
• Medicare Care Compare (consumer surface) ↗
• U.S. Government Works — public-domain license ↗

Related Fonteum research

• Why 14% of skilled nursing facilities had a quality drop in Q1
• The Medicare provider deactivation spike
• LEIE exclusion trends, 2026
• Care Compare: nursing homes
• Browse the federal dataset catalog
• Source registry and Tier classification
• How every figure is signed and chained

Reviewed by Jennifer Montecillo, MD, medical reviewer. Non-practicing medical reviewer.