How to Read Your County's Overdose Data — and What the Numbers Can't Tell You

A local news station runs a headline: "Overdose deaths in our county up 40 percent." A school district sends a letter home saying overdoses among teenagers have doubled. A neighbor says the county is actually doing better than ever.

All three may be citing real numbers — and all three may be quietly wrong. Not because anyone lied, but because overdose data is published with caveats attached, and the caveats almost never make it into the headline.

The good news is that the underlying data is public. Anyone with a browser can open the same federal tables that county health departments and state epidemiologists use. This guide is about reading them honestly: what they measure, where the lag comes from, when small numbers lie, and what to do with what you find.

Two numbers come with every figure — and you usually only see one

The first thing to understand is that a provisional overdose count is not a fact. It is an estimate with an error bar that the government publishes separately.

On the CDC's Provisional Drug Overdose Death Counts page, every row has two values side by side: a reported value and a predicted value. For the twelve months ending in May 2026, the national reported figure was 65,351 deaths; the predicted figure, adjusted for deaths whose records had not finished processing, was 66,478.

The difference is not a rounding error. It is the gap between "what has arrived in the database" and "what probably happened."

The percentage that changes how you read everything

Alongside those two columns sits a third: percent pending investigation.

Drug overdose deaths are often certified slowly. A medical examiner may issue a death certificate marked "pending investigation" while toxicology is outstanding, and the cause of death gets coded months later. CDC's technical notes are unusually blunt about what that does to the numbers:

• Where fewer than 1 percent of records are pending, the provisional count for a period was about 5 percent lower than the final count for that same period.
• Where more than 1 percent of records are pending, provisional counts "may underestimate the final count of drug overdose deaths by as much as 30 percent."

Read that again, because it is the single most useful sentence in local overdose reporting. A county or state with 2 percent of its records pending can be missing up to three in ten of its deaths — and the missing ones are the most recent. Which is exactly the direction a journalist, an advocate, or a worried parent is most likely to misinterpret.

Why the newest number is always the least complete

You might reasonably ask why anyone publishes incomplete data at all. The answer is that the alternative is worse: waiting for perfection means waiting years while the overdose supply changes month to month.

CDC's approach is to publish on a four-month lag — shortened from a six-month lag in February 2022 as reporting timeliness improved. So the newest information available in mid-October 2026 describes deaths through May 2026, and the current release is based on data available for analysis as of October 4, 2026.

Two more conventions matter:

• Counts are published as "12-month ending" periods — the number of deaths in the twelve months ending in the month shown. The period ending in May 2026 runs from June 2025 through May 2026. This evens out seasonal swings, but it also means a genuine change that began in April will not show up clearly until late in the year, because eleven months of older data are still averaged in.
• Provisional counts are counted by the jurisdiction where the death occurred and include deaths of non-US residents — roughly 400 to 500 extra records a year. Final annual data, published later through CDC WONDER and NCHS reports, is tabulated by state of residence. So provisional and final numbers for "your state" are not measuring quite the same thing, and a small difference between them is not a scandal.

Practical rule: if someone quotes you a single recent month, ask what the pending percentage was. If it is over 1 percent, the number is a floor, not a measurement.

The four places the real data lives

You do not need a research account. There are four doors, and each answers a different question.

1. Provisional deaths — national and state (VSRR)

The Provisional Drug Overdose Death Counts page is the fastest national picture. It shows 12-month-ending counts for the United States and every state, reported and predicted values, the percent pending investigation, and a map of percent change by jurisdiction. For May 2026 the national predicted count was 11.5 percent lower than the same period a year earlier.

Use it for: state-level trend direction, and for the pending percentage that tells you how much to trust it.

2. Provisional deaths — by county

The VSRR Provisional County-Level Drug Overdose Death Counts dataset on data.cdc.gov breaks the same twelve-month-ending counts down to individual counties, and it is downloadable as CSV, JSON or Excel. It also carries the honest small print: each county row includes a percentage of records pending field and an urban–rural classification code, so you can see at a glance whether you are looking at a large metro county or a rural county where a handful of deaths moves the rate enormously.

Use it for: your own county's direction of travel — with the caveats below.

3. SUDORS and DOSE — the detail behind the totals

Counts tell you how many. Two CDC systems tell you what and how:

• SUDORS (State Unintentional Drug Overdose Reporting System) publishes comprehensive data on unintentional and undetermined-intent overdose deaths — richer than counts alone, which is where information about what was actually in the supply comes from.
• DOSE (Drug Overdose Surveillance and Epidemiology) covers nonfatal overdoses. DOSE-DIS draws on emergency department and inpatient discharge data; DOSE-SYS draws on electronic health record data in syndromic surveillance systems.

Nonfatal data matters because it moves first. Emergency department visits can spike weeks before deaths are certified, which is why a county health department can issue a warning in the middle of a week that looks unremarkable in the annual death tables.

CDC's data resources page gathers all of these, plus the Injury, Overdose, and Violence Dashboard — which goes down to census tract level — and the pharmacy dispensing rate maps for opioids, buprenorphine, and naloxone.

4. WONDER and your own local dashboard

Finalized, closed-out annual mortality data lives in CDC WONDER, and is what annual NCHS reports are built on. If you need a number to quote in a public comment letter that will outlive the news cycle, that is the source to use.

At the local level, many county health departments publish their own dashboards — often with closer-to-real-time fatal and nonfatal figures, broken out by drug type and geography. Public Health – Seattle & King County, for example, publishes an overdose dashboard built for exactly this purpose. Your county may already have one; searching "[county name] overdose dashboard" is usually enough to find it.

Counts versus rates: why your county's total can fool you

Now the arithmetic that trips up the most well-meaning coverage.

Here are two real Pacific Northwest counties, from the same federal county dataset, for the twelve months ending March 2026:

King County's raw count is more than twice Multnomah's. But once you divide by population — using the Census Bureau's county population estimates, which are free to download — Multnomah's rate is roughly a quarter higher.

This is the difference between a count and a rate, and it is why "our county has fewer overdose deaths than theirs" is not the reassuring sentence it sounds like. Small counties also swing hard: in a county of 20,000 people, three deaths one year and six the next is a "100 percent increase" that may mean nothing at all.

Rule of thumb: counts for trends over time in the same place. Rates for any comparison between two places.

What "suppressed" means, and why your county may be missing

Open the county dataset and you will find that a large share of rows do not have a number at all. Instead there is a footnote: "One or more data cells have counts between 1-9 and have been suppressed in accordance with NCHS confidentiality standards."

In the release covering the twelve months ending March 2026, 1,629 county rows carry that footnote. It is not censorship and it is not an error. It is the same convention used across vital statistics: very small cells are withheld so that a person in a small community cannot be identified from public data.

The consequence for readers is concrete. If you live in a rural county and the number is missing, you are not going to get a reliable local death trend this month, and anyone who claims one is working from something other than the federal tables. In that case, switch to signals that are less sensitive to tiny numbers: state-level trends, emergency department data, or your local health department's own counts, which may have access to unsuppressed figures.

Same region, same direction, two very different levels of confidence

The clearest way to see all of this working at once is to compare two neighbouring states in the current release. Both are down. They are not equally knowable.

Oregon — 12-month-ending all-drug overdose deaths, with percent pending investigation:

• January 2026: 1,026 (1.12 percent pending)
• February 2026: 932 (1.33 percent)
• March 2026: 827 (1.63 percent)
• May 2026: 629 (2.36 percent)

Washington — same measure, same months:

• January 2026: 2,822 (0.05 percent pending)
• February 2026: 2,754 (0.06 percent)
• March 2026: 2,628 (0.07 percent)
• May 2026: 2,512 (0.15 percent)

Washington's decline is reported with almost no unresolved records — its pending rate stays far under 1 percent, the threshold above which CDC says counts become unreliable. Oregon's pending rate is above 1 percent and rising, and CDC's own page carries a state-specific flag: "Oregon is experiencing substantial delays in the resolution of pending records and recent trends may underestimate the death count."

Both states are almost certainly improving. But only one of those two numbers comes with a warning label printed on it, and a reader who does not know to look for the pending column will never see the difference.

This is also how to read a local spike responsibly. A rising count with a low pending rate is a real signal. A falling count with a high pending rate is a maybe. Neither is a reason to panic, and both are reasons to check the supply warnings and naloxone access in your area rather than the graph.

Five ways a local headline goes wrong

Keep this list. It covers most of what you will read this year.

1. Comparing a single month to a single month. Overdose deaths are volatile month to month. Use 12-month-ending figures, which is why CDC publishes them that way.
2. Mixing provisional and final numbers. A provisional state figure from this year compared against a final figure from CDC WONDER two years ago is not a comparison. It is two different measurement systems.
3. Using "suspected" and "confirmed" interchangeably. A suspected overdose that later proves to be something else never appears in the death count. Suspected counts are useful early warning; they are not mortality statistics.
4. Conflating all-drug and opioid-involved deaths. They are separate indicators in the same tables and they trend differently — deaths involving stimulants have risen even as opioid-involved deaths fell.
5. Quoting a count without the population, or a rate without saying which one. "Deaths up 12 percent" and "the rate is down" can both be true in the same county in the same year.

If you want one habit from this guide, make it this: when you see an overdose number, ask when it was pulled and how much was pending. Those two questions separate reporting from guessing.

How to pull your own county's numbers in about ten minutes

1. Open the state table first. Go to the CDC Provisional Drug Overdose Death Counts page, select your state or the United States, and read the table under Figure 1a. Write down three things: the reported value, the predicted value, and the percent pending investigation, for the most recent month and the same month a year earlier.
2. Then the county. Open the county-level dataset on data.cdc.gov and filter to your state. Sort by the month-ending date so the newest rows are at the top. Note the count, the pending percentage, and whether the row is suppressed.
3. Get the denominator. Download the Census Bureau's county population estimates file (co-est2024-alldata.csv), find your county's row, and divide: deaths ÷ population × 100,000. That is your rate.
4. Check the direction, not the dot. Compare the latest 12-month-ending figure with the same month one year earlier, and with three months earlier. One number is noise; a consistent line is a trend.
5. Add the nonfatal picture. Check your state or county health department's dashboard, and CDC's DOSE-based dashboards, for emergency department signals — they move before deaths do.
6. Write down the date you pulled it. These tables update monthly, and the numbers you cite will change. A footnote saying "data as of October 4, 2026" is the difference between a citation and a stale claim.

What to do with the numbers once you have them

Data is only worth the action it produces. In practice, local overdose numbers are used for five things, and families can influence all five:

• Where naloxone goes. Distribution points, vending machines, libraries, and pharmacy standing orders are usually placed using the same data you just read. If your county's numbers are rising in a specific area, that is a legitimate, specific request to make at a public meeting. Our guide to getting naloxone in your community walks through the practical routes.
• Opioid settlement spending. Counties and states receiving settlement funds are required, in most cases, to report how the money is used. Numbers with a place name attached are the most persuasive comment you can submit.
• Testing supplies. If your local supply is shifting — fentanyl in stimulants, xylazine alongside opioids — the response is test strips and honest information, not abstinence messaging. See fentanyl test strips and drug checking.
• Treatment capacity. Waiting times, mobile units, and medication availability are budget decisions. A rate that is worse than the neighbouring county's is an argument.
• How you talk to your own family. Knowing that overdoses in your area are falling does not make your household safer. Knowing that the local supply is unpredictable does. If someone you love is using, the response is the same either way: never use alone, keep naloxone within reach, and know that medication for opioid use disorder is the treatment with the strongest evidence behind it.

Where to get help right now

• SAMHSA National Helpline — 1-800-662-4357, free, confidential, 24/7, English and Spanish.
• 988 Suicide & Crisis Lifeline — call or text 988.
• FindTreatment.gov — search treatment by ZIP code and payment type.
• Poison Help — 1-800-222-1222 for suspected poisoning or a bad reaction.

Bottom line

Overdose data is not secret, and it is not complicated. It is just published with footnotes, and the footnotes are the story: a four-month lag, a pending percentage that can hide up to three deaths in ten, and small counties whose numbers are withheld entirely. Read the reported value, the predicted value, and the pending percentage together, divide by population before comparing anywhere to anywhere else, and treat a single month as a rumor.

The national picture — 65,351 reported deaths in the twelve months ending May 2026, down 11.5 percent from a year earlier — is genuinely encouraging. Your county's picture may be better, worse, or simply unknown. All three are answers. Only one of them is a reason to stop paying attention.