The Reversal

In 2021 the pandemic surge looked like the start of a runaway crisis. Provisional counts put 93,129 deaths in 2020, up 30% on 71,740 in 2019, the largest one-year rise in this data. Deaths kept climbing for two more years, plateaued, and then, beginning in the autumn of 2023, bent the other way.

The national 12-month total has now fallen in each of the last 33 monthly releases. Those windows overlap, sharing eleven months each, so read that as one long slide of 33 months rather than 33 independent drops. It has taken the country back to where it stood in January 2017.

Final data confirm it. NCHS recorded 105,007 deaths in 2023 and 79,384 in 2024, a 26.2% fall in the age-adjusted rate, the largest in the series (Data Brief 549). The provisional file puts 2025 at 70,735, a third straight annual decline.

The reversal is several stories at once. Deaths involving synthetic opioids, overwhelmingly illicit fentanyl, are 55.9% below their Jul 2023 peak. Heroin has all but vanished from death certificates: from 16,288 deaths a year in Jun 2017 to 1,714 now, down 89.5%, as fentanyl first displaced it and then declined itself.

The stimulants, which travel with fentanyl, peaked alongside it: meth in Aug 2023, cocaine in Aug 2023. Prescription opioids, the epidemic's first wave, topped out back in May 2017.

The reversal set off a scramble for credit: naloxone on drugstore shelves, easier buprenorphine prescribing, billions in settlement money, precursor crackdowns in China, upheaval inside the Sinaloa cartel. The figure places those events against the curve so the timing can be judged directly.

In 2021 the map was red almost everywhere: 50 of 53jurisdictions recorded more deaths in 2020 than in 2019 (switch Fig. 4 to “2020 vs. 2019”). Today the picture has turned over.

52 of 53 jurisdictions are below their own peak. The exception is New Mexico, where the latest 12-month total is the highest on record. The deepest falls are in Rhode Island† (−61%), Ohio (−59%) and West Virginia (−58%).

The newer story is that a decline is not the same as still declining. Against a year earlier, New Mexico (+32%), Colorado (+10%), Wisconsin (+2%) and Arkansas (+1%) are up, as are Nebraska† and Wyoming†, though counts there are small enough that percentages jump around.

All 53 jurisdictions NCHS publishes: the 50 states, the District of Columbia, New York City (reported separately from the rest of New York State) and Puerto Rico, which is published but not counted in the national total.

Beyond the trend, the explorer shows overdoses as a share of all deaths, as the 2021 notebook did. Nationally that share was 1.8% in the year ending Jan 2015, reached 3.5% at the Aug 2023 peak, and is 2.1% now. Below it, any years of the same place can be laid over one another month by month, the notebook's 2019-against-2020 comparison extended to every year on file.

The source is CDC/NCHS's VSRR Provisional Drug Overdose Death Counts (dataset xkb8-kh2a), a monthly file of 12-month-ending counts by jurisdiction of occurrence. A Python build downloads it, records its checksum, runs the checks below, and computes every figure on this page. A failed check stops the build rather than publishing a wrong number.

• ✓Passed: One row per jurisdiction, indicator and month0 duplicate keys in 87,543 rows
• ✓Passed: Single period typeevery row is a 12-month-ending count
• ✓Passed: Complete monthly grid for overdose totals54 jurisdictions incl. US x 137 months (2015-01 to 2026-05), no gaps
• ✓Passed: Predicted count present for every overdose total0 missing of 7,398
• ✓Passed: U.S. reported total equals the sum of its 51 jurisdictions50 states + DC + NYC, excl. Puerto Rico; max monthly difference 0 deaths
• ✓Passed: Predicted totals: jurisdictions vs. U.S. modelNCHS models each level separately; max divergence 0.56% (tolerance 1.5%)
• ✓Passed: No drug class exceeds total overdose deaths0 violations across 6 classes
• ✓Passed: Overdose deaths never exceed all-cause deaths0 violations, 0 missing all-cause counts in 7,398 rows
• ✓Passed: U.S. peak sits on the delay-adjusted seriespeak 2023-08 at 112,418 predicted deaths

Choices that change the numbers

• Delay-adjusted counts for every statistic. Deaths take months to certify, so reported counts in recent periods run low: 1.7% nationally in the latest month, but 34.8% in Oregon, 10.3% in Nevada and 5.7% in Wisconsin. Measured on reported counts, those states would appear to be falling faster than they are. Charts draw the reported series alongside so the gap stays visible.
• Each place against itself. The file has no population denominators or age adjustment, so this page compares each jurisdiction with its own past rather than ranking burden across states. D.C.'s counts also include non-residents who died there.
• Small bases are flagged, not dropped. Montana†, North Dakota†, Nebraska†, New Hampshire†, Rhode Island†, South Dakota†, Vermont† and Wyoming† record fewer than 250 deaths a year; their percent changes move on a handful of deaths and carry a dagger (†) everywhere they appear.
• Drug classes overlap and are withheld where unreliable. Most overdoses involve several drugs, so class counts cannot be summed. NCHS withholds drug-specific counts for the latest period in Louisiana and Oregon.
• Provisional is not final. These counts are by place of death and will not match final, residence-based annual totals: the provisional file gives 81,313 for 2024; the final count is 79,384.