The case for spending money to stop the next pandemic before it starts leans heavily on a simple story about the environment. The argument goes that clearing a forest and disturbing the wildlife inside it invites the next virus to cross into people . A paper published today in Nature puts that story to the largest empirical test it has faced, assembling 58,319 infection records for 32 diseases across 169 countries. In places the story holds. But the strongest and most consistent pattern in the data is not about forests or climate at all. It is about detection, and how much of the world we are able to see.

It turns out that where diseases show up depends most on whether the sick live near a clinic that can diagnose them. To a first approximation, the geography of emerging diseases is the geography of health care.

The work comes from Rory Gibb at University College London, Sadie Ryan at the University of Florida, Colin Carlson at Georgetown, and 28 other authors, most affiliated with the Verena Institute , a National Science Foundation effort to put the study of viral emergence on a firmer quantitative footing.

The obstacle any such study has to overcome is that spillover—the transmission of an animal pathogen to a human—cannot be watched directly. What exists instead is a record of outbreaks that someone noticed, diagnosed and reported, gathered unevenly around the world. A febrile patient near a well-equipped hospital becomes a data point. The same infection two valleys away, in a village a day from the nearest clinic, often does not. A map drawn from such records blends where diseases actually occur and where they merely happen to be seen.

To pull those apart, the authors matched outbreak locations against population-weighted background points, then modeled the footprint of surveillance directly—travel time to the nearest health facility and the density of built-up land—so it could be subtracted before any ecological signal was measured. With hundreds of possible driver-disease combinations available, 25 of the 31 coauthors recorded their predictions before the analysis ran, guarding against the data dredging that turns noise into narrative.

Outbreak Maps Mostly Show Where We Look

Without the detection correction, the models reproduce the familiar hotspots, the same broad geography that has organized this field since Kate Jones and colleagues mapped it in 2008 . With the correction, much of that pattern dissolves. The features that best explained where outbreaks had been recorded were not ecological but infrastructural: nearness to a hospital and the density of the built environment.

For the median disease, the odds of an outbreak being recorded fell by about a third for every additional hour of travel time to the nearest health facility—for MERS and Argentine hemorrhagic fever, by 90 percent or more. Two identical spillovers, one beside a clinic and one a few hours’ drive away, have very different chances of ever becoming a known case. Roughly a quarter to more than a third of the people most at risk live more than two hours from care. Earlier work estimated that up to half of all Ebola spillovers are never identified. The new analysis suggests that scale of blindness is closer to the rule than the exception. This summer I wrote about authorities in Congo tallying fewer than half the cases in an Ebola outbreak already underway.

“It’s clear we really are noticing only the very tip of the iceberg when it comes to zoonotic spillover,” Gibb told me.

For Gibb the implication runs deeper than a gap in the data. “I think we are now reaching the stage where a paradigm shift is needed in the conventional scientific view of spillovers, away from this idea of unusual events under exceptional circumstances, and towards instead seeing spillover as ubiquitous—a part of the fabric of life as an animal, sharing landscapes and microorganisms with other animals.”

Where the Fingerprint Is Real

Subtracting detection does not erase the environment. It sharpens it.

Outbreak risk was highest in mosaic landscapes, the patchwork of forest, farmland and settlement where people and livestock live alongside fragmented habitat. That signal was strong and broadly shared among the 17 vector-borne diseases in the set—Chagas disease, spread by blood-feeding kissing bugs that shelter in the walls of rural houses; yellow fever, a mosquito-borne virus that circulates between primates and people at the forest edge.

The models also reproduced relationships that specialists already accept—intact biodiversity lowering Lyme disease risk via the dilution effect, pig density predicting Japanese encephalitis—which is a reason to take the more surprising results seriously.

The climate result was one the authors did not expect. The signal they found was long-term drying, not warming, a decades-scale decline in rainfall, and it recurred for dengue across the Americas, Africa and Asia. That complicates the tidier picture of heat and dengue that many scientists have described . “For me the strength and consistency of the effect of climate drying on outbreak frequency was probably the most unexpected finding of the study,” Gibb said. For dengue, he pointed to “urban water storage and climate whiplash (dry-then-wet) events” that concentrate mosquito habitat during a shortage and then ignite transmission when rains return. Warming still matters, he stressed, but its effect is “extremely nonlinear” and depends on geography. “So it may be that, in this study, we weren’t quite asking the right question to detect these effects,” he said.

The Drivers That Didn’t Show Up

The nulls are where the paper cuts against received wisdom. Recent forest loss, the centerpiece of the forest-first prevention argument, was a significant predictor in only 6 of the 29 disease systems where an expert had judged it plausible, and its effect ran in both directions. Long-term warming reached significance in 5 of 28. No single human pressure behaved as a general driver of emergence.

The starkest result concerns the pathogens that most worry pandemic planners: the directly transmitted zoonoses—Ebola, Marburg, Nipah, MERS and mpox—which pass from animals to people and then spread person to person without an insect in between. These shared almost no environmental drivers with one another—their reservoirs and exposure routes have little in common. Marburg strikes miners who disturb bat colonies in caves, Nipah emerges from date-palm sap contaminated by fruit bats in Bangladesh, MERS passes from camels. No single environmental lever should be expected to move all of them together.

Gibb resists reading the deforestation result as a debunking. “This was a striking result given the widespread narrative around the importance of deforestation for infectious disease emergence,” he told me. “But it is also not especially surprising, given that zoonotic diseases have a really wide variety of host communities and ecological characteristics, some of which are not especially connected to deforestation.” The lesson he draws is that interventions have to be matched to particular systems. “I’m thinking of examples like Lyme disease, or Hendra virus in Australia,” he said.

Not everyone accepts that the data can bear the weight of the authors’ conclusions. Jason Rohr, an ecologist at Notre Dame whose meta-analysis with Michael Mahon found broad and consistent effects of global change on disease, was blunt: “the conclusions they draw extend far beyond what is supported by the data and analyses.”

His first objection is definitional, and sharper than it sounds. The study counts an “outbreak” as one or more reported cases of a disease in a place in a given year. That, Rohr argues, makes the analysis one of where pathogens are detected, not of outbreaks, spillovers or emergence—terms that carry distinct meanings in epidemiology. A single imported case is none of these, yet the authors tally them the same way.

His second objection is about time. The analysis compares places, but the relevant claim is about what happens in a single place as its forests come down; averaging two decades of forest loss can erase a real but short-lived effect. “Just because you get a null result doesn’t necessarily mean that it’s not a causal driver,” he said. He noted an irony: Carlson, one of the senior authors, published a paper in Nature earlier this year arguing that establishing climate’s effect on disease demands exactly this kind of change-over-time attribution—the approach I have written about —which the fingerprint study does not attempt.

Raina Plowright, the Cornell disease ecologist behind the case for ecological countermeasures , pressed a complementary point. Every record on the map ends with a person who had to be exposed, fall ill, reach care and be diagnosed—so it was almost inevitable that access to health care would surface as the common thread. Her own system is the cautionary example: Hendra virus spills from Australian flying foxes when the trees that feed them in winter are cleared, a driver invisible to any global forest metric. “The database could never have captured that nuance,” she said. Real understanding comes “from the bottom up rather than top down,” from reconstructing the whole causal chain one system at a time.

The forests-and-pandemics story is not wrong. It was too simple. My own view, having spent a career on the ecology of disease and the problem of measuring it: much of what the field had taken for the geography of emergence was the geography of observation, and this paper is the most systematic evidence yet for that.

Two things bound how far the conclusions reach. For the pandemic-grade pathogens we most want to understand, the data are painfully thin. Hendra virus enters the analysis with 11 recorded events, and the per-disease uncertainty ranges for Marburg, Ebola, Nipah and MERS are so wide they cover nearly the whole plausible span. Finding no shared drivers among such sparse systems is at least as consistent with too little data as with a true absence of common causes. It is the question I would most like the field to settle next.

The second concerns the headline finding itself. Distance to health care shapes whether an outbreak is recorded, which is the effect the authors model. But it also shapes whether the sick are treated and whether onward transmission is contained—genuine epidemiological effects rather than artifacts of observation. The result is better understood as evidence that clinics matter, for more reasons than one, than as evidence that clinics merely explain the map.

Barbara Han, a disease ecologist at the Cary Institute and a co-author, told me she worried the paper would invite exactly the wrong simplification. “I’ve been concerned that a less careful reader or less careful press coverage of this paper will conclude ‘everything is biased so forget prediction and just invest in healthcare,’” she said, “which would be an unfortunate oversimplification.”

Even Neil Vora, a physician among the most prominent advocates for stopping spillover at its source, welcomed it. He told me the study “adds to the evidence” for a layered defense: advocates like him have long called for health-systems investment in communities where spillovers happen. “This paper provides data on why that matters,” he said. “We know enough to act on spillover prevention already.”

Ecological interventions deserve support, but aimed by evidence, system by system, not as a single global program against deforestation. Health care is the rare investment that pays off regardless of which pathogen emerges next, improving detection, treatment and prevention together.

Gibb framed the stakes better than any coefficient can. “It is also critically a question about inequality, about who is most likely to get sick, and why, and who can access healthcare, diagnostics and treatment at the point when they’re most needed.”