The same 30%, and where it lands

A new simulation out of Pennsylvania asked a plain question and got an answer that should be impossible.

Raise naloxone distribution by the same 30% in six counties. Project the overdose deaths in 2029. In Allegheny, the big urban county, deaths fall by about 70% (95% interval, 55 to 81). In Erie, mid-sized, 11% (4 to 17). In Dauphin, under 1%. In Clearfield, rural, 28% (2 to 64).

Same drug. Same proportional increase. Same five-year window. The gap between the best county and the worst is not a rounding error. It is the whole finding.

Ahmed, Rahimian, Chen and Kumar built a model of opioid use disorder and calibrated it separately to each county, urban to rural (arXiv 2609.22419, 18 Sep 2026). What predicted the spread was not the county's death rate. It was each county's own history of naloxone dispensing, how steady or how erratic the supply had been. Counties that had been handing out naloxone in fits responded differently to a smooth increase than counties that had been handing it out evenly. The authors put it plainly: a pattern invisible from mortality statistics alone.

Buprenorphine behaved differently again. Its projected reductions were consistently smaller, 10 to 23%, and in four of five counties the uncertainty intervals crossed zero. The exception is Erie, where buprenorphine beat naloxone, 20% against 11%. So even the ranking flips by county.

The model states its own limits, which is why I trust the shape of it more than the numbers. It is a simulation. It was calibrated on 2018 to 2022 and held out 2023 to 2024 for four of six counties. It underestimated deaths in Allegheny, and recalibrating through 2024 cut Allegheny's projected reduction from 70% to 59%, which the authors read as robustness. The supply figures are pharmacy dispensing data, which they say to read as an indicator of change, not as absolute supply.

Here is what stays with me. “Naloxone works” is true and useless. The same act is not the same act everywhere. An intervention is not a property of the intervention. It is a property of the ground it lands on, and the only honest number is the one that knows which ground it is standing on, and says so.