Is HealthData.gov data free to use?

Two depths share one front door, and they solve different jobs.

The catalog layer. HealthData.gov Catalog spans hospital capacity and utilization, COVID-19 patient impact, community health statistics, Medicare and Medicaid program files, substance use and workforce topics, contributed by HHS agencies alongside state and local health departments. Records carry official-source provenance flags, every tabular asset rides the same publishing chassis, and six adjacent industries draw on the same spine - distributors, supplies, technology, life and health insurance, managed care, pharmaceuticals.

The facility-week panel beneath it. COVID-19 Hospital Capacity by Facility (Historical Time Series) tracks roughly 6,000 hospitals week by week from January 1, 2020 through May 3, 2024 - about 230 collection weeks, several million facility-week rows, 127 measure columns apiece: total, staffed and pediatric bed counts, ICU occupancy, confirmed and suspected inpatient census, prior-day admissions split by age band, and critical-staffing-shortage flags. A facility key matching the CMS Certification Number ties every week of a hospital's history together.

The specialists complete the shelf: Medicare Care Compare resolves roughly 40,000 certified facilities into one typed table; Open Payments holds 15,498,687 rows across 91 columns in a single program year's general payments file; HRSA shortage designations grade every area 0 to 26 down to census tract. The catalog finds the records; these stock it.

One verified state-day observation ships with the catalog record, laid out exactly as documented:

# one row = one state-day, flagship hospital capacity file
state                                        AK
inpatient_beds                               1093
inpatient_beds_used                          869
inpatient_beds_used_covid                    3
total_adult_patients_hospitalized_confirmed_covid    6
staffed_adult_icu_bed_occupancy              65
critical_staffing_shortage_today_yes         1

Read it as a system rather than a snapshot. 869 of 1,093 staffed beds were occupied - just under eighty percent utilization - on a day when three beds held confirmed-or-suspected COVID patients and exactly one hospital in the state flagged a critical staffing shortage. The measure names do the joining: inpatient_beds against inpatient_beds_used yields utilization by simple division, and the companion *_coverage columns record how many hospitals reported each metric, so thin-reporting days get down-weighted instead of silently averaged in.

At facility grain the row family widens: identifiers - facility key, name, address, county FIPS code, subtype, metro flag - sit beneath the measures, with a documented sentinel marking suppressed values so masking never masquerades as a quiet week.

Geography - the United States at every altitude: national rollups, state-day summaries, county indicators and individual facility points with FIPS codes attached. A single extract serves a one-hospital teardown and a fifty-state benchmark without re-keying anything.

Granularity - one row per hospital per collection week at the deep cut, one row per state per day on the flagship summary, one row per certified facility in the quality universe. No sampling layers anywhere, so counts recomputed from raw rows reconcile against published totals.

One interpretation rule travels with the archive: no adjustment was made for non-reporting facilities, so values reflect reports received. Filter the sentinel, weight by the coverage columns, and the series behaves; skip either step and every rate computed inherits the reporting system's gaps. Both mechanisms ship as ordinary fields in the delivery rather than footnotes in a readme.

The flagship state file documents 136 columns; eleven resolve most jobs. state keys the geography. inpatient_beds and inpatient_beds_used pair into utilization. inpatient_beds_used_covid and total_adult_patients_hospitalized_confirmed_covid split the COVID load from the baseline census. staffed_adult_icu_bed_occupancy reads as a percentage straight off the row. critical_staffing_shortage_today_yes turns staffing stress into a countable event rather than an anecdote. previous_day_admission_adult_covid_confirmed and hospital_onset_covid carry the epidemiological curve. And the *_coverage family records how many hospitals reported each corresponding metric - the denominators that turn missingness into a measurable.

Definitions travel beside every field, each with its stated caveat - an unusually honest documentation standard that names its own limits. The remaining columns fold under additional-fields-on-request rather than padding every delivery, and anything pinned down during scoping arrives verified against the dictionary rather than guessed from headers.

Four workflows pay for themselves fastest.

Utilization benchmarks. Staffed-bed counts, ICU occupancy and shortage tallies measured one way across every state give consultants and planners defensible baselines - a market's occupancy computed from the same definitions as every other market's.

Surge modeling and pandemic retrospectives. A labeled facility-week panel spanning four years, with built-in coverage counts for weighting, is scarce supervised material for epidemic models and after-action studies that need an archive which cannot drift underneath them.

Program research and exogenous features. Medicare and Medicaid program files cited straight from the publisher, and capacity, census and shortage series dropping into forecasting features alongside claims or utilization data. Newsrooms quote the federal figures with provenance intact; quant desks use them as regressors.

Ranked by how directly the records answer the day job:

1. Investors and quant researchers stress an operator thesis against what its hospitals actually lived through, then extend screens into trajectories using editions back to 2019. See investors & quants use cases.
2. Market researchers and consultants size markets from facility denominators - beds, ownership, geography - that survive scrutiny. See market researchers use cases.
3. Data scientists and ML engineers get a longitudinal, labeled panel with quality machinery in the schema itself. See data scientists use cases.
4. Journalists, academics and students cite facility-specific history - which hospitals ran short of ICU beds, when, and how shortages spread metro by metro - with every claim tracing to a named hospital in a named week. See journalists & academics use cases.
5. Developers and data-product builders learn one publishing chassis and extend across ten thousand assets. See developers & builders use cases.

Because the hard part was never the first extract - it is the tenth. Schemas are set per dataset, so depth ranges from fully documented dictionaries to little more than a title. Coverage columns decide whether a rate is real or an artifact of thin reporting. A sentinel value means one thing in the facility panel and would sail into a naive parser as data. And a deliberately frozen archive looks identical to a broken feed unless someone tells you which one you have. Each is survivable once; none is fun to re-solve in every new notebook.

Datadory normalizes before delivery: field dictionaries verified against the publishers' own documentation during the August 2026 research pass, coverage counts shipped as ordinary columns, sentinels documented and handled at load, era boundaries labeled explicitly, and provenance flags carried through so official-source rows separate from contributions without a manual pass.

This page covers one estate inside the twenty-three-record health care services stack - sixteen primaries plus seven pooling in from neighboring industries. Start with the health care services data hub for the full pooled view, or the ranked shortlist of the best health care services datasets.

Go deeper three ways: the HealthData.gov Catalog product page documents the field dictionary above and takes sample requests scoped to your states and topics; the facility-week capacity panel carries the four-year archive measure by measure; and the breadth-versus-precision argument continues on the HealthData.gov Catalog vs HRSA Data Warehouse comparison. The publishing house behind it maps onto the HHS source profile, and the operational counterpart across the Atlantic is NHS England Statistics & Data Collections.

When you want real rows instead of descriptions, request a sample - the field dictionary travels with it.