That Alarming Spike in Drug Overdose Deaths Last Year? It Was A Statistical Illusion

After more than a year of cautiously hopeful news about drug overdose deaths finally dropping in the United States, a government data release in June 2025 seemed to yank the rug out from under public health officials. Provisional estimates for January 2025 suggested overdose deaths had suddenly surged, reversing a long and hard-won decline. NPR ran a segment. Experts expressed alarm. Some observers even raised concerns about possible political interference with federal data.

But according to a new analysis published in the American Journal of Public Health, that spike never happened. It was a ghost in the data, produced by a statistical model trained on years of rising deaths that simply could not adjust when the trend reversed. A later government correction confirmed the decline was still intact.

How Drug Overdose Deaths Get Counted and Predicted

Tracking overdose deaths in real time is harder than most people realize. When someone dies of an overdose, the official cause of death often depends on toxicology reports and death investigations that can take months to complete. At any given moment, the most recent national figures are incomplete. To fill that gap, the National Center for Health Statistics developed a prediction tool called “UCD Predict,” which uses statistical models to estimate what the final death count will look like once all the paperwork catches up.

For roughly a decade, this tool worked well. But fentanyl, the powerful synthetic opioid that at its peak in August 2023 accounted for roughly 70% of all drug overdose deaths, did not spread evenly across the country. It displaced heroin starting around 2013 in the Midwest, Northeast, and South, but did not significantly reach the West until around 2019. That lurching, regionally uneven pattern destabilized models trained on slower, more predictable trends.

Real trouble arrived when models built on explosive 2022 growth were applied to 2023 and beyond, when deaths had started to drop. Overdose deaths had entered what the authors call “the great deceleration,” the longest decline in more than 40 years, beginning in August 2023. Models expecting continued surges instead produced inflated estimates.

Three Rounds of Data Reveal the Drug Overdose Death Mirage

Researchers examined national death data from the National Vital Statistics System spanning January 2015 through March 2025. They compared actual observed death counts against three versions of UCD Predict downloaded at three different points in time: the original series from July 2024, a first revision from February 2025, and a second revision from August 2025.

In winter 2025, the government retroactively revised its prediction series all the way back to 2015, using data from 2022 to 2023, when deaths were still climbing in large states. This first revision overestimated deaths from early 2024 and early 2025, creating the illusion of a spike and an apparent reversal of a 17-month decline.

A few numbers make the scale of the error plain. In January 2024, observed deaths totaled 105,521. The first revision estimated 110,909, an overcount of 5,388. For January 2025, observed deaths stood at 78,883, while the first revision put the figure at 82,138. It even underestimated December 2024 deaths by 137, what the authors call “an implausible negative deviation,” meaning the model suddenly swung in the wrong direction entirely. A second government revision corrected the picture, confirming the January 2025 spike was a modeling artifact, not a real resurgence.

Not a Conspiracy, but Still a Problem

Some observers speculated the misleading estimates reflected political interference, particularly since the first revision contradicted evidence of continuing declines and coincided with restrictions in other federal data sources. Researchers push back firmly on that reading. Government epidemiologists said the anomaly resulted from applying growth-era formulas to a period of decline, not from manipulation by either the Biden or Trump administrations.

Still, the episode revealed a real transparency problem. When behind-the-scenes model changes affect the provisional estimates that journalists, policymakers, and public health officials rely on in near-real time, those changes should be announced in advance. Researchers recommend the National Center for Health Statistics publish a brief change log describing what changed, when, and why, and distribute it with each affected data release.

Despite the stumble, UCD Predict remains the best near-real-time source for tracking overdose deaths, and the authors are not calling for it to be scrapped. Death certificate data underpin the entire overdose monitoring and response system, and without reliable provisional estimates, public health officials lose their earliest signal for detecting surges and directing resources. Confidence in that system is not optional.

Drug Overdose Deaths Are Declining, but the Crisis Isn’t Over

Overdose deaths peaked around August 2023 with a 12-month rolling total of more than 114,000, a staggering toll. What has followed is genuine, if fragile, progress. Beginning in August 2023, the decline represents the longest drop in more than 40 years, though the authors caution the pace of improvement is slowing.

A single alarming data release can reshape public perception overnight, and in a public health crisis of this scale, perception shapes policy. Getting the numbers right, and being upfront when models stumble, is not just a technical concern. With tens of thousands of lives still at stake each year, the overdose epidemic may finally be bending in the right direction, but it is far from over.

---

Disclaimer: This article is based on a peer-reviewed editorial. It is intended for general informational purposes only and does not constitute medical or public health advice. Findings reflect the authors’ analysis and interpretations and should not be taken as official guidance from any government agency or health authority.

---