SourcesThe data family
healthdata.dev
A free, keyless copy of CDC's provisional drug overdose death counts and the Census Bureau's county health insurance estimates, in one format with versions you can cite.
What it is
healthdata.dev republishes public health statistics from two federal agencies: the CDC’s National Center for Health Statistics (NCHS) and the Census Bureau. It serves them as plain JSON, CSV and Parquet files behind one API, with no key or signup. Like this site, it’s run by Eric Ness, so treat this page as a description of a sibling, and check its claims the way you would any other source’s.
The site doesn’t produce any numbers itself. Every figure comes from the agencies, and each dataset names its source, its license and an exact credit line.
What it measures
The site lists four datasets. Two of them carry most of the weight:
- Drug overdose deaths by state and drug (
overdose-deaths). The NCHS Vital Statistics Rapid Release (VSRR) provisional counts for every state and the US, monthly since January 2015. Each value is the number of deaths in the 12 months ending that month. Series cover all drugs, opioids, synthetic opioids such as fentanyl, heroin, cocaine, psychostimulants such as methamphetamine and more. Columns include the reported count, CDC’s predicted final count, the percent of records still pending investigation, and flags for suppressed and underreported counts. - Drug overdose deaths by county (
overdose-deaths-county). The NCHS provisional county-level counts, 12-month periods ending each month since January 2020, for all drugs combined. - Health insurance coverage by county and state (
uninsured). The Census Bureau’s Small Area Health Insurance Estimates (SAHIE): the share and number of people without health insurance in every county and state, each year from 2008 to 2024. Series cover people under 65, adults 18 to 64 and children under 19, each with a 90 percent margin of error.
The fourth dataset is CDC’s kindergarten vaccination and exemption rates by state since the 2009-10 school year.
What the .dev layer adds is mostly packaging. Every dataset uses the same file layout, place IDs (us-ca for a state, us-county- plus the five-digit FIPS code for a county) and column conventions as the rest of the family. Each update gets a version ID, and older versions stay reachable at fixed URLs. The site checks for new state overdose data every two days and new SAHIE data every seven days, and runs validation checks before publishing. If a check fails, the last good version stays live.
What it misses
The overdose counts are provisional, and they change. CDC builds them from death records received by a monthly cutoff. Overdose deaths often need long investigations, so many records start out as “pending.” CDC warns that when more than 1 percent of records are pending, the reported count may fall short of the final count by as much as 30 percent. CDC’s predicted counts adjust for this, but CDC says the adjustment may not fully cover long delays in some states. You can see revisions happen on healthdata.dev itself: the US count for the 12 months ending May 2026 was 65,351 in the version published November 6, 2026, and 65,720 in the version published November 12.
The most recent months are the least complete. CDC says a flat or falling trend in recent months could mean incomplete data, a real change, or both, and that true trends can’t be confirmed until final data arrive. The state data run about four months behind.
State and county figures count deaths differently. The state-level provisional counts are by where the death occurred and include foreign residents. Final counts, and the county-level provisional counts, are by place of residence. Don’t line them up as if they were the same measure.
Drug categories overlap and aren’t reported evenly. One death can involve several drugs and appear in several series. CDC says reporting of specific drugs varies by jurisdiction and warns against comparing some jurisdictions by drug.
Small county counts are withheld. CDC suppresses county counts from 1 to 9. In the data these show as empty values with suppressed set to true, so a rural county may have no number at all.
SAHIE numbers are model estimates, not counts. The Census Bureau builds them from the American Community Survey combined with administrative records such as federal tax returns, SNAP participation and Medicaid and CHIP enrollment. Every value has a margin of error, and for small counties it can be wide. Two counties whose estimates differ by less than their margins may not really differ.
The .dev layer has its own limits. It covers only the datasets and series listed above, not everything NCHS or SAHIE publishes. SAHIE also breaks results down by income, sex and race, and this site carries only the three age groups. It is a copy. For anything high-stakes, such as a published paper or a policy brief, check the figure against CDC’s or the Census Bureau’s own release.
How to access it
1Browse the catalog
Open healthdata.dev/datasets for the list, or ask the API for it:
curl -s "https://api.healthdata.dev/v1/catalog"
The catalog gives each dataset’s ID, current version and date coverage. On November 14, 2026 it listed the state overdose data through June 2026, the county data through March 2026 and SAHIE through 2024.
2Read a dataset’s metadata
https://api.healthdata.dev/v1/{dataset} returns the version, coverage, source, license, credit line, series list and download links. For example, https://api.healthdata.dev/v1/overdose-deaths.
3Get one series for one place
The pattern is /v1/{dataset}/{series}/{location}.csv (or .json). This call returns the national count for all drugs, one row per month:
curl -s "https://api.healthdata.dev/v1/overdose-deaths/all-drugs/usa.csv"
Tested on November 14, 2026, it returned rows from 2015-01 to 2026-06, with columns date, deaths, predicted_deaths, percent_complete, percent_pending, pct_drugs_specified, suppressed and underreported. For uninsured rates in California, use /v1/uninsured/under-65/us-ca.csv.
4Download a whole dataset
/v1/{dataset}/download/{dataset}.csv (or .parquet) gives every series and place in one file, such as https://api.healthdata.dev/v1/uninsured/download/uninsured.csv.
5Pin a version
Add @ and a version ID after the dataset ID, and the URL will always return the same data. /v1/{dataset}/versions lists the current version and the monthly snapshots. The docs and the API reference cover caching, errors and every endpoint.
How to cite it
Cite two things: the agency that produced the numbers and the exact version you used.
The site gives a credit line for each source on its Sources page, and the docs say to cite the monthly snapshots, which are kept permanently. Other versions may be removed after 30 days. Find the snapshot ID with /v1/{dataset}/versions, then put it in the URL.
A worked example for the national overdose series:
Ahmad FB, Cisewski JA, Rossen LM, Sutton P. Provisional drug overdose death counts. National Center for Health Statistics. https://www.cdc.gov/nchs/nvss/vsrr/drug-overdose-data.htm. Accessed via healthdata.dev, dataset overdose-deaths, version 2026-11-06.8de2d201af: https://api.healthdata.dev/v1/overdose-deaths@2026-11-06.8de2d201af/all-drugs/usa.csv. Retrieved November 14, 2026.
For the insurance data, the credit line is “U.S. Census Bureau, Small Area Health Insurance Estimates (SAHIE) Program,” followed by the program’s URL. Because overdose counts are provisional, say so in your text and give the date you pulled them.
Sources & further reading
- healthdata.dev, its docs and its sources and licenses page
- NCHS, Provisional Drug Overdose Death Counts (VSRR), with technical notes
- NCHS, Provisional County-Level Drug Overdose Death Counts
- Census Bureau, Small Area Health Insurance Estimates (SAHIE)
- Census Bureau, About SAHIE: method and inputs
- Census Bureau, SAHIE technical documentation
End of page · Back to top