Nursing Home Compliance Intelligence Dataset

About the Dataset

The Nursing Home Compliance Intelligence Dataset is an analytics-ready intelligence layer over U.S. nursing home data from the Centers for Medicare & Medicaid Services (CMS) — a CCN-spined facility master, 7+ years of monthly facility history, effective-dated PECOS ownership with self-reported private-equity (PE) and real estate investment trust (REIT) flags, chain intelligence reconciled to CMS's own chain file, Payroll-Based Journal (PBJ) staffing compliance, conformed enforcement and deficiency history, and observed closure signals. It joins Care Compare (Nursing Home Compare / Five-Star), PECOS ownership, PBJ staffing, and enforcement feeds onto a single stable facility spine so operators, investors, lenders, and researchers can query nursing home ownership, staffing, and compliance without building the pipeline themselves.

Get Full Access | Snowflake Marketplace | Free Trial

Data coverage:

Effective-dated ownership intervals: Feb 2023–present (PECOS). Labeled owner-name observations: Jan 2019–present (Care Compare).

Quick Access

Tables: FACILITY_MASTER, FACILITY_MONTH, OWNERSHIP_HISTORY, OWNERSHIP_EVENTS, CHAIN_MEMBERSHIP_HISTORY, STAFFING_COMPLIANCE, ENFORCEMENT_EVENTS, CLOSURE_SIGNALS, CHAIN_MONTH, SNAPSHOT_CALENDAR Sources: CMS Care Compare (Nursing Home Compare / Five-Star), PECOS ownership, Payroll-Based Journal (PBJ), NH Chain Performance Measures, Provider of Services (POS) Coverage: 16K+ certified nursing homes, Jan 2019–present (7+ years of monthly snapshots) Update Frequency: Monthly — pipeline monitors CMS for new releases and loads within days of publication Geography: United States (all states and territories)

Signal, not prediction. Every flag in this dataset is a published, deterministic, seed-versioned formula computed over CMS's own data. Flags are 0/1/NULL — NULL means not evaluable, never a silent 0 — and each carries a reason code and an inputs-coverage column. There is no machine learning, no predictive scoring, and no name-based inference. See the flag methodology.

Overview

The Nursing Home Compliance Intelligence Dataset provides analytics-ready access to CMS nursing home compliance data including:

• FACILITY_MASTER — One row per nursing home (CCN): current identity, location, certification, Five-Star ratings, chain, and observed-history bookends

• FACILITY_MONTH — CCN × observed month history (1.2M+ rows) with monthly snapshot state, PBJ staffing, ID-based ownership summary, and the five computable compliance flags plus the operational-stress composite

• OWNERSHIP_HISTORY — Effective-dated PECOS ownership intervals (Feb 2023–present) with PE/REIT/holding-company typing, plus labeled Care Compare owner-name observations (Jan 2019–present)

• OWNERSHIP_EVENTS — Detected ownership-change events and official change-of-ownership (CHOW) confirmations

• CHAIN_MEMBERSHIP_HISTORY — CMS-reported facility chain-membership intervals

• STAFFING_COMPLIANCE — CCN × quarter Payroll-Based Journal staffing with case-mix expectation and Five-Star staffing passthrough

• ENFORCEMENT_EVENTS — Conformed, de-duplicated deficiency citations and monetary penalties with true-repeat occurrence indexing

• CLOSURE_SIGNALS — Observed-only closure-type signals (snapshot disappearance, POS termination, beds-to-zero)

• CHAIN_MONTH — Chain × observed month rollups reconciled to CMS's own chain performance file (within ±5%)

• SNAPSHOT_CALENDAR — Per-feed-family observed-month calendar that anchors every vintage comparison

Metadata Tables

Every Dataplex data product includes these standard metadata tables:

Table	Purpose
FEEDS	Dataset catalog — available tables, descriptions, update dates
FEEDS_FILES	Batch load history with is_latest flag for data freshness
CHANGELOG	Change log — data loads, schema changes, corrections
DATA_DICTIONARY	Column descriptions for all tables

Ask it questions — Cortex Analyst and the semantic view

The share includes the FACILITY_ANALYST semantic view — a semantic model over the facility master, the monthly history, and PBJ staffing, with relationships, synonyms, and business definitions built in. Point Cortex Analyst at it (in Snowsight: AI & ML → Cortex Analyst → select DWV.FACILITY_ANALYST) and ask questions in plain English:

• "Which states have the highest average nurse staffing hours per resident day?"

• "Compare occupancy by ownership type."

• "How many facilities are in each county in Ohio?"

New to these features? See Snowflake's documentation on Cortex Analyst and semantic views.

The same semantic model is queryable directly in SQL with the SEMANTIC_VIEW() construct:

-- Average total nurse hours per resident day, by state
SELECT *
FROM SEMANTIC_VIEW(
    DWV.FACILITY_ANALYST
    DIMENSIONS state
    METRICS avg_total_nurse_hprd, facility_count
)
ORDER BY avg_total_nurse_hprd DESC NULLS LAST;

-- Occupancy by ownership type — the semantic view resolves the
-- facility joins for you
SELECT *
FROM SEMANTIC_VIEW(
    DWV.FACILITY_ANALYST
    DIMENSIONS ownership_type
    METRICS avg_occupancy, facility_count
)
ORDER BY avg_occupancy DESC NULLS LAST;

Entity Relationship Diagram

Facility-centric model: FACILITY_MASTER is the CCN spine; facility-keyed tables join on ccn, chain rollups join on chain_id, and every data table carries feed_id / feeds_files_id lineage back to FEEDS and FEEDS_FILES

The CCN (CMS Certification Number) is the spine of the product: FACILITY_MASTER holds one row per facility, and FACILITY_MONTH, OWNERSHIP_HISTORY, STAFFING_COMPLIANCE, ENFORCEMENT_EVENTS, and CLOSURE_SIGNALS all join back to it on ccn. Chain rollups in CHAIN_MONTH join on chain_id. Every data row also carries feeds_files_id (FK to FEEDS_FILES) so you can trace it to the exact batch that loaded it.

Facility Tables

Facility Master (FACILITY_MASTER)

One row per nursing home (CCN) — current-state identity, location, certification, Five-Star ratings, chain, and observed-history bookends. There is no SCD2 history and no is_closed; longitudinal state lives in FACILITY_MONTH.

Key Features:

• 16K+ certified nursing homes on a stable CCN spine

• Current provider name, legal business name, address, county, and SSA county code

• Certified beds, average residents per day, and current Five-Star ratings

• ccn_reuse_suspect and successor_ccn_hint — observed-only heuristics for likely certification-number reuse (never name-based inference)

Column Reference

Column	Type	Description
ccn	VARCHAR	6-char CMS Certification Number — the product spine
provider_name	VARCHAR	Current provider name (latest snapshot)
legal_business_name	VARCHAR	Current legal business name
provider_address	VARCHAR	Street address
city	VARCHAR	City
state	VARCHAR	State
zip_code	VARCHAR	ZIP code
county	VARCHAR	County / parish
ssa_county_code	VARCHAR	SSA county code
ownership_type	VARCHAR	CMS ownership type (for-profit, non-profit, government)
number_of_certified_beds	NUMBER	Certified beds
average_number_of_residents_per_day	NUMBER	Average daily resident census
current_chain_id	VARCHAR	Current CMS-reported chain id
current_chain_name	VARCHAR	Current CMS-reported chain name
special_focus_status	VARCHAR	Special Focus Facility (SFF) / SFF-candidate status
abuse_icon	VARCHAR	Care Compare abuse icon flag
overall_rating	NUMBER	Five-Star overall rating
health_inspection_rating	NUMBER	Five-Star health inspection rating
staffing_rating	NUMBER	Five-Star staffing rating
qm_rating	NUMBER	Five-Star quality measure rating
first_observed_month	DATE	Earliest month this CCN appears
last_observed_month	DATE	Latest month this CCN appears
present_in_latest_snapshot	BOOLEAN	TRUE if in the most recent snapshot month
ccn_reuse_suspect	BOOLEAN	Observed-only reuse heuristic (state or name changed across history)
successor_ccn_hint	VARCHAR	Address-match successor CCN hint for a disappeared CCN
feed_id	VARCHAR	FK to FEEDS
feeds_files_id	VARCHAR	FK to FEEDS_FILES
created_at	TIMESTAMP	When the source data was loaded
updated_at	TIMESTAMP	When the table was last rebuilt

Facility Month (FACILITY_MONTH)

The gold table: one row per CCN × observed month (1.2M+ rows). Monthly snapshot state, the PBJ staffing quarter, an ID-based ownership summary, and the five computable compliance flags plus the operational-stress composite. Keyed on observed months only — there are no ghost rows for months CMS did not publish. Flag formulas and thresholds are documented in the flag methodology.

Key Features:

• 1.2M+ facility-months, Jan 2019–present

• Five 0/1/NULL compliance flags, each with a _reason code and an _inputs_coverage boolean

• operational_stress_indicator_count — a 0–5 count of fired components (NULL unless at least 4 of 5 are evaluable)

• method_version stamped on every row for reproducibility

Column Reference

Column	Type	Description
ccn	VARCHAR	CMS Certification Number
observed_month	DATE	Published snapshot month
observed_quarter	DATE	Calendar quarter (PBJ staffing join key)
provider_name	VARCHAR	Provider name in this snapshot
city, state, zip_code, county	VARCHAR	Location in this snapshot
ownership_type	VARCHAR	Ownership type in this snapshot
number_of_certified_beds	NUMBER	Certified beds
average_number_of_residents_per_day	NUMBER	Average daily census
occupancy_ratio	NUMBER	Residents per day ÷ certified beds
overall_rating	NUMBER	Five-Star overall rating
health_inspection_rating	NUMBER	Five-Star health inspection rating
staffing_rating	NUMBER	Five-Star staffing rating
qm_rating	NUMBER	Five-Star quality measure rating
special_focus_status	VARCHAR	SFF / SFF-candidate status
abuse_icon	VARCHAR	Abuse icon flag
chain_id, chain_name	VARCHAR	CMS-reported chain for the month
pbj_submission_status	VARCHAR	reported_full / partial / missing
pbj_total_nurse_hprd	NUMBER	Total nurse hours per resident day (quarter)
case_mix_expected_total_nurse_hprd	NUMBER	CMS case-mix expected total nurse HPRD
pbj_avg_daily_census	NUMBER	PBJ average daily census
in_pecos_coverage	BOOLEAN	TRUE when this month is inside PECOS ownership coverage
disclosed_owner_count	NUMBER	Distinct disclosed owners this PECOS vintage
pe_owner_count	NUMBER	Distinct PE-flagged owners
reit_owner_count	NUMBER	Distinct REIT-flagged owners
has_undisclosed_upstream_owner	BOOLEAN	Current-state dead-end disclosure measure
staffing_shortfall_flag	NUMBER	0/1/NULL — reported HPRD below case-mix expectation
staffing_shortfall_reason	VARCHAR	Reason code
staffing_shortfall_inputs_coverage	BOOLEAN	Were the inputs present
repeat_deficiency_flag	NUMBER	0/1/NULL — deficiency tag recurred across standard surveys
repeat_deficiency_reason	VARCHAR	Reason code
repeat_deficiency_inputs_coverage	BOOLEAN	Were the inputs present
ownership_disclosure_gap_flag	NUMBER	0/1/NULL — ownership disclosure gap in-coverage
ownership_disclosure_gap_reason	VARCHAR	Reason code
ownership_disclosure_gap_inputs_coverage	BOOLEAN	Were the inputs present
enforcement_escalation_flag	NUMBER	0/1/NULL — annual fine total escalated
enforcement_escalation_reason	VARCHAR	Reason code
enforcement_escalation_inputs_coverage	BOOLEAN	Were the inputs present
deactivation_signal_flag	NUMBER	0/1/NULL — observed closure signal fired
deactivation_signal_reason	VARCHAR	Reason code
deactivation_signal_inputs_coverage	BOOLEAN	Always TRUE — always evaluable
sff_or_abuse_icon	NUMBER	Composite component: SFF status or abuse icon
pbj_census_decline_gt20pct_yoy	NUMBER	Composite component: census down >20% YoY
chow_within_12m	NUMBER	Composite component: confirmed CHOW within 12 months
stress_indicators_evaluated	NUMBER	0–5 count of evaluable components (the denominator)
operational_stress_indicator_count	NUMBER	0–5 count of fired components (NULL unless ≥4 evaluable)
stress_indicator_reason	VARCHAR	Which components fired, or insufficient_evaluable_components
method_version	VARCHAR	Methodology version stamped on the row
anchor_source_file_id	VARCHAR	Upstream CMS load pointer (resolves in FEEDS_FILES)
feed_id	VARCHAR	FK to FEEDS
feeds_files_id	VARCHAR	FK to FEEDS_FILES
created_at	TIMESTAMP	When the source data was loaded
updated_at	TIMESTAMP	When the table was last rebuilt

Ownership History (OWNERSHIP_HISTORY)

Effective-dated, ID-keyed PECOS ownership intervals (Feb 2023–present) with self-reported PE/REIT/holding-company typing, plus labeled Care Compare owner-name observations (Jan 2019–present). PECOS rows (source = 'pecos') are the only branch that carries interval math, is_current, and ID-based counts; Care Compare rows (source = 'care_compare_name') are name-grain observations, never intervals.

Key Features:

• 13M+ ownership rows across both sources on a shared CCN spine

• ID identity via owner_associate_id (PECOS ASSOCIATE_ID) — names churn, IDs do not

• Dual dating: valid_from_observed (fact, used for interval math) vs valid_from_reported (self-reported claim, carried verbatim)

• Self-reported PE, REIT, holding-company, investment-firm, and chain-home-office typing

Column Reference

Column	Type	Description
ccn	VARCHAR	Facility CMS Certification Number
source	VARCHAR	pecos (ID-keyed interval) or care_compare_name (labeled name observation)
enrollment_id	VARCHAR	PECOS enrollment id
owner_associate_id	VARCHAR	PECOS ASSOCIATE_ID (identity for pecos rows)
role	VARCHAR	Ownership / control role
owner_display_name	VARCHAR	Owner display name
owner_kind	VARCHAR	Organization vs individual
name_bridge_match_tier	VARCHAR	Care Compare name → ID bridge tier (hint only)
ownership_percentage	NUMBER	Reported ownership percentage
is_owned_by_another_org_or_ind	BOOLEAN	Owner is owned by another org/individual
has_undisclosed_upstream_owner	BOOLEAN	Curated current-state dead-end measure (pecos rows)
is_private_equity	BOOLEAN	CMS self-reported PE flag
is_reit	BOOLEAN	CMS self-reported REIT flag
is_holding_company	BOOLEAN	Holding-company flag
is_investment_firm	BOOLEAN	Investment-firm flag
is_chain_home_office	BOOLEAN	Chain home-office flag
is_for_profit / is_non_profit	BOOLEAN	Profit-status flags
valid_from_observed	DATE	First vintage the interval appears (interval math + is_current)
valid_from_reported	DATE	Self-reported association date (verbatim; never effective_from)
last_observed_vintage	DATE	Latest vintage the interval reaches
prev_available_vintage_month	DATE	Previous available vintage (from SNAPSHOT_CALENDAR)
is_current	BOOLEAN	TRUE when the interval reaches the latest vintage
date_confidence	VARCHAR	observed_only / reported_precedes_coverage / reported_within_coverage / reported_anomalous
end_reason	VARCHAR	ongoing / not_observed_after / chow / label_observation
anchor_source_file_id	VARCHAR	Latest contributing upstream CMS load
source_row_ids	ARRAY	All upstream loads that contributed to the interval
feed_id / feeds_files_id	VARCHAR	Lineage FKs
created_at / updated_at	TIMESTAMP	Load and rebuild timestamps

Ownership Events (OWNERSHIP_EVENTS)

Detected ID-keyed ownership-change events plus unmatched official CHOW rows. Useful for change monitoring and diligence.

Key columns: event_id, ccn, event_type (owner_appears / owner_gone / ownership_percentage_change / facility_enrollment_appears / facility_enrollment_gone / official_chow), ccn_join_confidence (matched / unmatched), alert_tier (chow_signal / watchlist / disclosure), owner_associate_id, owner_name, old_ownership_percentage, new_ownership_percentage, is_private_equity, is_reit, is_holding_company, chow_confirmed, chow_type_text, chow_effective_date, plus lineage columns.

Chain Membership History (CHAIN_MEMBERSHIP_HISTORY)

CMS-reported facility chain-membership intervals from the provider-information snapshots. CMS-reported only — no inferred chains. no_chain_reported is an explicit interval value.

Key columns: ccn, chain_id, chain_name, no_chain_reported, valid_from_observed, last_observed_month, observed_month_count, is_current, plus lineage columns.

Staffing Compliance (STAFFING_COMPLIANCE)

CCN × quarter Payroll-Based Journal staffing with case-mix expectation and Five-Star staffing passthrough. pbj_submission_status is explicit; suppressed values stay NULL (never a fabricated zero).

Column Reference

Column	Type	Description
ccn	VARCHAR	Facility CMS Certification Number
staffing_quarter	DATE	First day of the PBJ calendar quarter
cy_qtr	VARCHAR	Calendar-year quarter label
pbj_submission_status	VARCHAR	reported_full / partial / missing
pbj_days_reported	NUMBER	Days present in PBJ for the quarter
pbj_expected_days	NUMBER	Days expected in the quarter
pbj_avg_daily_census	NUMBER	Average daily resident census
pbj_total_nurse_hprd	NUMBER	Total nurse hours per resident day
pbj_rn_hprd	NUMBER	RN hours per resident day
pbj_lpn_hprd	NUMBER	LPN hours per resident day
pbj_cna_hprd	NUMBER	CNA hours per resident day
pbj_contract_staff_share	NUMBER	Share of hours from contract staff
five_star_staffing_rating	NUMBER	Five-Star staffing rating passthrough
reported_total_nurse_hprd	NUMBER	Care Compare reported total nurse HPRD
case_mix_expected_total_nurse_hprd	NUMBER	CMS case-mix expected total nurse HPRD
feed_id / feeds_files_id	VARCHAR	Lineage FKs
created_at / updated_at	TIMESTAMP	Load and rebuild timestamps

Enforcement Events (ENFORCEMENT_EVENTS)

Conformed, de-duplicated union of deficiency citations and monetary penalties (1.9M+ events), one row per event with occurrence_index for true repeats of the same tag.

Key columns: event_id, ccn, event_type (health_deficiency / fire_safety_deficiency / penalty), event_date, tag, scope_severity_code, category, description, standard_deficiency, complaint_deficiency, corrected, correction_date, fine_amount, payment_denial_start_date, payment_denial_length_days, occurrence_index, first_reported_month, last_reported_month, source_row_ids, plus lineage columns.

Closure Signals (CLOSURE_SIGNALS)

Observed-only closure-type signals. Gap months emit nothing; a signal inside CMS's official change-of-ownership publication lag is held at pending_confirmation.

Key columns: ccn, signal_type (snapshot_disappearance / pos_termination / bed_count_zero), signal_month, signal_tier (top / corroborating), detail, explained_by_official_chow, signal_status (observed / pending_confirmation), source_row_ids, plus lineage columns.

Chain Month (CHAIN_MONTH)

Chain × observed month rollups computed from FACILITY_MONTH for each chain's member CCNs, with a coverage_ratio, an aggregates_partial flag (set when under 90% of members had the underlying measure), and CMS_REPORTED_* columns from CMS's own chain performance file for reconciliation (within ±5%).

Key columns: chain_id, chain_name, observed_month, member_count, avg_total_nurse_hprd, avg_overall_rating, avg_health_inspection_rating, avg_staffing_rating, avg_qm_rating, avg_operational_stress_indicator_count, coverage_ratio, aggregates_partial, cms_reported_facility_count, cms_reported_avg_overall_rating, cms_reported_avg_total_nurse_hprd, plus lineage columns.

Snapshot Calendar (SNAPSHOT_CALENDAR)

Per-feed-family observed-month calendar — the source of prev_available_vintage_month everywhere, so no query ever does month arithmetic across a gap.

Key columns: feed_family, observed_month, prev_available_vintage_month, months_since_prev_vintage, prev_vintage_spans_gap, vintage_seq, snapshot_row_count.

Full column-level descriptions for every table are available in the DATA_DICTIONARY table.

Data Quality

Standardization

• All columns use UPPERCASE naming consistent with Snowflake conventions

• CCN is normalized to a 6-character certification number and used as the join key across every table

• Boolean flags are native BOOLEAN; compliance flags are SMALLINT 0/1/NULL (never a silent 0)

• Every vintage comparison reads the previous available month from SNAPSHOT_CALENDAR — never a naive month-minus-one across a gap

• PBJ staffing measures are NULL when suppressed or unreported, never fabricated as zero

• Ownership identity is always the CMS-assigned owner_associate_id; owner-name strings are display data, not identity

Data Freshness

Check when data was last updated:

Snowflake

SELECT source_name, data_period, row_count, created_at, is_latest
FROM DWV.FEEDS_FILES
WHERE is_latest = TRUE;

Databricks

How to Query Nursing Home Compliance Data

Platform Schema Reference

This dataset is available on both Snowflake and Databricks. Queries use schema-only references — the database is already set by the share or catalog context:

Platform	Schema	Example
Snowflake	DWV	DWV.FACILITY_MONTH
Databricks	cms_nh_compliance_dwv	cms_nh_compliance_dwv.facility_month

Discover Available Data

Start with the FEEDS table to see what's available, and FEEDS_FILES to understand data freshness and load history.

Snowflake

-- Browse the data catalog
SELECT table_name, description, last_loaded_at, total_rows_extracted, total_batches
FROM DWV.FEEDS
ORDER BY table_name;

Databricks

Working with Data Lineage

Every data row links to FEEDS_FILES via feeds_files_id, which tells you exactly which batch loaded that data. Use this to filter to the current data version or trace any row back to its source load.

Snowflake

-- Current facility-month rows with batch lineage
SELECT fm.ccn, fm.observed_month, fm.provider_name, fm.overall_rating,
       ff.data_period, ff.created_at AS data_loaded_at
FROM DWV.FACILITY_MONTH fm
JOIN DWV.FEEDS_FILES ff ON fm.feeds_files_id = ff.id
WHERE ff.is_latest = TRUE
LIMIT 20;

Databricks

Staffing-Shortfall Screening

Screen facilities whose payroll-based reported nurse staffing fell below CMS's case-mix expectation. The flag is only set where PBJ was fully reported — NULL (not evaluable) is excluded automatically by = 1.

Snowflake

SELECT fm.ccn, fm.provider_name, fm.state, fm.observed_month,
       fm.pbj_total_nurse_hprd, fm.case_mix_expected_total_nurse_hprd,
       fm.staffing_shortfall_reason
FROM DWV.FACILITY_MONTH fm
JOIN DWV.FEEDS_FILES ff ON fm.feeds_files_id = ff.id AND ff.is_latest = TRUE
WHERE fm.staffing_shortfall_flag = 1
ORDER BY fm.observed_month DESC, fm.state
LIMIT 100;

Databricks

Effective-Dated Ownership with PE/REIT for a Facility

Trace the effective-dated PECOS ownership of a single facility, including private-equity and REIT typing and effective-dating confidence.

Snowflake

SELECT ccn, owner_display_name, role, ownership_percentage,
       is_private_equity, is_reit, is_holding_company,
       valid_from_observed, valid_from_reported, date_confidence,
       is_current, end_reason
FROM DWV.OWNERSHIP_HISTORY
WHERE ccn = '345003' AND source = 'pecos' AND is_current = TRUE
ORDER BY is_reit DESC, owner_display_name;

Databricks

Chain-Month Rollup vs CMS-Reported

Compare our chain rollups to CMS's own chain performance file. coverage_ratio and aggregates_partial tell you how much of the chain the rollup covers.

Snowflake

SELECT chain_id, chain_name, observed_month, member_count,
       avg_total_nurse_hprd, avg_overall_rating,
       coverage_ratio, aggregates_partial,
       cms_reported_facility_count, cms_reported_avg_total_nurse_hprd
FROM DWV.CHAIN_MONTH
WHERE cms_reported_facility_count IS NOT NULL
ORDER BY observed_month DESC, member_count DESC
LIMIT 100;

Databricks

Enforcement Escalation + Repeat Deficiencies for Diligence

Screen facilities whose latest snapshot shows both an escalating penalty pattern and a recurring deficiency, with the count of repeat tags and the maximum tag recurrence — a one-query diligence shortlist.

Snowflake

SELECT fm.ccn, fm.provider_name, fm.state, fm.observed_month,
       fm.enforcement_escalation_flag, fm.repeat_deficiency_flag,
       COUNT(DISTINCT e.tag)        AS distinct_repeat_tags,
       MAX(e.occurrence_index)      AS max_tag_occurrences
FROM DWV.FACILITY_MONTH fm
JOIN DWV.ENFORCEMENT_EVENTS e
  ON e.ccn = fm.ccn AND e.event_type = 'health_deficiency' AND e.occurrence_index >= 3
WHERE fm.observed_month = (SELECT MAX(observed_month) FROM DWV.FACILITY_MONTH)
  AND fm.enforcement_escalation_flag = 1
  AND fm.repeat_deficiency_flag = 1
GROUP BY ALL
ORDER BY max_tag_occurrences DESC
LIMIT 100;

Databricks

Closure Signals with Corroboration

Find facilities where a top-tier snapshot disappearance is corroborated by a second observed signal (a POS certification termination or a beds-to-zero transition). Remember: a signal is a change in the federal record, not a statement that a building's doors have closed.

Snowflake

SELECT cs.ccn, m.provider_name, m.state,
       cs.signal_type, cs.signal_tier, cs.signal_month,
       cs.signal_status, cs.explained_by_official_chow, cs.detail
FROM DWV.CLOSURE_SIGNALS cs
JOIN DWV.FACILITY_MASTER m ON m.ccn = cs.ccn
WHERE cs.ccn IN (
    SELECT ccn FROM DWV.CLOSURE_SIGNALS
    GROUP BY ccn HAVING COUNT(DISTINCT signal_type) > 1
)
ORDER BY cs.ccn, cs.signal_month
LIMIT 100;

Databricks

Tracking Data Changes Over Time

FEEDS_FILES records every batch load with row_count_delta showing what changed. Use this to monitor source data updates.

Snowflake

SELECT source_name, data_period, row_count, row_count_delta,
       created_at AS loaded_at, is_latest
FROM DWV.FEEDS_FILES
WHERE source_name = 'pbj_daily_nurse_staffing'
ORDER BY created_at;

Databricks

Methodology, Lineage & Caveats

This dataset is built to be cited in diligence, underwriting, and compliance work. Three companion pages document exactly how it behaves:

• Flag Methodology — the exact formula, thresholds, evaluability rules, and worked examples for each of the five compliance flags and the operational-stress composite.

• Lineage & Reproducibility — the lineage columns, how the monthly snapshot calendar works, and the steps to reproduce any value from the CMS source files.

• Caveats & Coverage — the data-coverage statement and the source-data properties you must reason about before drawing a conclusion.

Who Uses This Data

Common Use Cases

• Private-equity and REIT ownership diligence — Trace effective-dated, ID-keyed ownership intervals with CMS self-reported PE, REIT, and holding-company flags for investor and lender diligence on chains and facilities

• Staffing compliance monitoring — Monitor payroll-based (PBJ) nurse staffing against CMS's case-mix expectation by facility and quarter, with explicit coverage flags where PBJ is missing, partial, or suppressed

• Enforcement and survey diligence — Query conformed, de-duplicated deficiency citations and penalties with occurrence indexing to surface true repeat deficiencies across a facility's full survey history

• Chain intelligence and benchmarking — Analyze chain membership and chain-month rollups reconciled to CMS's own chain performance file, with explicit partial-coverage flags

• Longitudinal facility research — Study 7+ years of facility history on a stable CCN spine — staffing, quality, enforcement, ownership, and observed closure signals — for researchers, journalists, and analytics teams

• Change-of-ownership monitoring — Track detected ownership-change events and official CHOW confirmations for portfolio and market surveillance

Related Datasets and Research

This dataset pairs well with:

• FacilityGuard — Facility Ownership Intelligence — The interactive app built on the same curated CMS PECOS ownership model; has_undisclosed_upstream_owner here is the same figure FacilityGuard reports

• CMS Data Feeds Dataset — The broader CMS nursing home Care Compare feeds this product joins together, plus Medicare provider spending for cross-program analysis

• CMS NPPES Provider Dataset — NPI provider attributes for enriching owners and administrators with taxonomy and practice-location detail

• HRSA Healthcare Resources — County-level workforce and shortage-area data for correlating staffing patterns with local provider supply

Frequently Asked Questions

What facilities are included? All Medicare/Medicaid-certified U.S. nursing homes (skilled nursing facilities and nursing facilities) that appear in the CMS Care Compare provider-information snapshots — 16K+ facilities on a stable CCN spine, spanning Jan 2019–present.

How often is the data updated? Our pipeline automatically monitors CMS for new releases and loads updated data within days of publication — no manual intervention needed. CMS publishes the nursing home provider-information, ownership, deficiency, and penalty files monthly; PBJ staffing is published quarterly and arrives roughly two quarters behind. Coverage grows with each CMS release.

Why do the ownership tables start in Feb 2023 but the facility history goes back to 2019? Effective-dated, ID-keyed ownership intervals begin in February 2023, when CMS's PECOS ownership data becomes usable as dated intervals. Ownership before that date ships only as labeled owner-name observations from the Care Compare ownership feed, back to January 2019 — carried as name-grain rows, never as intervals, and never counted in any ID-based figure.

What does a NULL flag mean? NULL means not evaluable — the inputs needed to decide were missing, suppressed, or outside coverage. It is never a silent 0. A facility that never reported payroll staffing for a quarter gets NULL on the staffing flag, not 0, because "we could not evaluate this" is a different statement from "we evaluated this and found no shortfall." Each flag carries a _reason code and an _inputs_coverage boolean so you always know which branch produced the value.

Are these flags predictions of closure or care quality? No. Every flag is a published, deterministic formula over CMS's own data — signal, not prediction. There is no machine learning, no predictive scoring, and no name-based inference. The composite is named operational_stress_indicator_count because it is a count of fired indicators, not a distress score or a closure probability. See the flag methodology.

What's the difference between this and downloading from CMS directly? CMS publishes nursing home data across many separate monthly files — Care Compare, PECOS ownership, PBJ staffing, deficiencies, penalties, chain performance — each with its own keys, gaps, and quirks. This product joins them onto one stable CCN spine, resolves ownership identity to CMS-assigned IDs, dates ownership intervals, conforms and de-duplicates enforcement events, and adds published compliance flags with full batch lineage — analytics-ready in Snowflake and Databricks, updated automatically, with every row traceable to its CMS source file.

Ready to access Nursing Home Compliance Intelligence data?

Questions? Contact our team for a walkthrough.

Platform	Action
Snowflake	View listing on Snowflake Marketplace
Databricks	Free Trial