Imagine a hospital emergency department installs a defibrillator in the lobby. Heart attack deaths drop — but cardiac admission numbers go up. That's not a failure; it means people who would have died in the parking lot are now making it inside alive. This paper finds that drug consumption rooms do exactly the same thing, but for overdoses. Ana Armendariz exploits the staggered rollout of drug consumption rooms across Swiss cities between 1998 and 2022, using individual-level hospital discharge records and a difference-in-differences estimator. The headline: DCR openings reduce drug-related hospitalization rates by roughly 35% within 5 km of a facility. But the interesting result isn't the topline — it's where the 35% comes from. Fatal outcomes drop by 35%. Narcotics poisoning hospitalizations — the kind where someone survives — rise by a comparable amount. This is the paper's core claim, and it's mechanistically clean: DCRs are converting potentially fatal overdoses into non-fatal hospitalizations. The defibrillator-in-the-lobby analogy isn't decorative; it IS the mechanism. Supervised consumption means someone is present to administer naloxone or call an ambulance before the overdose kills. Mental and behavioral disorder hospitalizations decline by 32%, which Armendariz interprets as DCRs functioning as a gateway to broader healthcare engagement — users who show up for safe injection also get referred to psychiatric and behavioral services. Hepatitis and HIV estimates are imprecise, which the author acknowledges honestly. Those diseases have long incubation periods and complex transmission dynamics that a hospitalization-based outcome measure won't capture cleanly. The identification strategy leans on geographic proximity — the 5 km radius around each DCR — and the staggered timing of openings across cities. This is a well-understood DiD setup, and the staggered adoption across Swiss municipalities over 24 years provides real variation. The individual-level hospital records are a strength; this isn't ecological inference from city-level aggregates. Switzerland's universal coverage system means the hospital records are essentially population-level, not a biased sample of the insured. The limitation worth naming plainly: this is one country with a specific institutional context — universal healthcare, relatively high state capacity, and a political culture that tolerated DCR experimentation starting in the late 1990s. Whether the same mechanism operates in, say, US cities with fragmented insurance and hostile state legislatures is a policy question the data can't answer. The internal validity looks solid; external validity is genuinely uncertain. The paper lands squarely in the harm-reduction vs. abstinence-only debate that has defined drug policy for three decades. The finding that DCRs don't reduce total drug use (poisoning hospitalizations rise) but dramatically reduce deaths is the kind of result that forces a choice: do you measure success by whether people stop using drugs, or by whether they stop dying? This paper gives the harm-reduction side its cleanest causal estimate to date in a national-scale dataset.
