The Centers for Disease Control's provisional figures predict about 67,800 drug overdose deaths in the United States in the twelve months to March 2026 — down 12.3 percent on the year before. In New York, predicted deaths fell from more than 6,630 in 2022 to 3,775 in 2025, a decline of roughly 45 percent, or from about eighteen deaths a day to about ten.
These are very large numbers moving in the right direction, faster than almost anyone forecast. Tens of thousands of people are alive who, on the trend line of four years ago, would not be.
The question nobody has answered is why.
Three explanations, and they are not the same
The first is policy: sustained investment in treatment capacity, naloxone distribution and harm reduction. New York's governor credited it directly, and states that spent heavily are seeing declines.
The second is supply. There is documented change in the illicit fentanyl supply, including reduced potency, linked to Chinese controls on precursor chemical exports to manufacturing operations in Mexico. If the drug itself became less lethal per dose, deaths fall without anything changing in American clinics.
The third is the population. Published research points to a cohort effect — that the people most at risk have already died, that fentanyl has saturated the market, and that drug use among younger cohorts has fallen. On this reading the decline is partly demographic exhaustion, which is the least comfortable explanation and cannot be dismissed.
Why the attribution problem is a policy problem
These explanations are not mutually exclusive and probably all contribute. But they imply completely different decisions.
If the decline is policy, funding treatment and distribution is the highest-return spending in public health and should be expanded. If it is supply-side, the decline is contingent on the behaviour of foreign regulators and criminal manufacturers, and could reverse without warning when the chemistry changes again. If it is cohort exhaustion, the improvement will continue for a period and then flatten at a level well above where it started, and any programme judged on the trend will look effective regardless of whether it works.
The trouble is that everyone will interpret the same figures in favour of whatever they already fund. A win with three fathers gets claimed three times and defended by none of them, because the defence requires showing which input produced the output, and nobody can.
Success is the dangerous phase
Public health programmes are most vulnerable when the numbers improve. The crisis framing that justified emergency appropriations does not survive a 45 percent decline, and a legislature looking for savings will reasonably ask why a programme still needs its funding when the problem is receding.
If the funding was causal, cutting it removes the thing producing the improvement — and the reversal will appear a year or two later, by which point the cut will be hard to connect to it. If the funding was not causal, maintaining it is expensive and the money would do more elsewhere. Nobody can currently distinguish these cases, and the decision will nevertheless be made this budget cycle.
What would settle it
The evidence to separate the explanations is largely obtainable. Potency data from seized samples shows whether the supply changed and when. Treatment admission and retention figures show whether capacity was actually used. Age-specific mortality shows whether the decline is concentrated in the long-exposed cohort or spread across new users. Cross-state comparison between heavy and light spenders, against a common supply shock, isolates the policy contribution better than any single state's numbers can.
Most of this exists in fragments held by different agencies, on different timetables, and provisional national data runs months behind. The delivery problems are the ordinary ones this desk keeps finding in the health system — capacity that cannot be measured without confronting its labour arithmetic, services whose reach depends on unglamorous infrastructure like interpretation in the exam room, and access that improved when remote delivery found the specialties it actually suits.
The decline is real and it is worth understanding while it is happening, rather than reconstructing after it stops. Right now the country is running the largest natural experiment in overdose prevention in its history, and is not set up to read the result.
The national provisional figures are from the CDC's National Vital Statistics System for the twelve months ending March 2026. The New York figures and the governor's remarks date from International Overdose Awareness Day, 31 August 2026. The account of reduced fentanyl potency following Chinese precursor controls, and the research attributing part of the decline to changes in drug use, are as reported in national coverage and published research. The analysis is our own.
Topics nationalpublic healthpolicydata
