“Let’s say you have two people with underlying heart disease; somebody has a heart attack versus somebody has a heart attack because it’s too hot,” Kristie Ebi, an epidemiologist at the University of Washington and an expert on heat-related mortality, explained to me. “Will the second one be recorded as a heat death or will it just be recorded as a heart attack? Frankly, when both go into the emergency department, the number one goal is to save a life — it’s not necessarily to record whether it was because of temperature.”
But if physicians don’t code the second heart attack as a heat death— a procedure designed for insurance and billing rather than getting to the root of underlying environmental conditions — then the headline number of heat wave-related deaths will almost certainly be an undercount. In Washington during the 2021 heat dome, for example, the state health department initially reported that 129 people died from the temperatures. But later analyses compared the overall number of people who died that week in the state to the average number of people who died during the same week from 2013-2019 and concluded that there were 485 “additional” deaths compared to what would have been expected during normal early summer conditions.
Those 485 deaths are called “excess deaths,” and the number offers a broader picture of who actually dies from a heat wave. The tally captures not only those heart attacks that are coded by physicians and medical examiners as caused by the heat, but also the ones the state may have overlooked or discounted. Air pollution deaths, homicides, drownings, and accidents, for example — all of which also spike in relation to heat waves — show up. As one epidemiologist explained it to the Seattle-area NPR affiliate KUOW, a boating death might count as an excess heat death, too, because while “not directly attributable to heat in the sense of heat stroke … it arguably is attributable to heat in the sense that had it not been hot, they would not have gone out.”
Excess death analyses are also methodical in what they don’t count. “After a heat wave, there’s a deficit in the number of deaths, which means that the heat wave brought forward deaths that would have occurred anyway,” Ebi told me. The analyses also take those into account to model only the “true” excess events. This, at least, is relatively simple in scope: The advantage of heat waves for mortality accounting is that they don’t have the long tails associated with hurricanes and other weather-related tragedies. “Deaths occur over a few days of a heat wave, and then it’s over,” Ebi added.
But the calculation, while relatively straightforward, has its critics, too. “The limitation of that approach is that it doesn’t actually quantitatively attribute that excess mortality to the heat,” Christopher Callahan, a climate scientist and assistant professor at Indiana University Bloomington, told me. Take the boating accident example: Maybe if it’d been a regular summer day, the enthusiast still would have taken his pontoon out and had all those beers. Maybe during the heat wave it was also smoky, and that caused some of the excess deaths. There’s also the possibility that the baseline number of deaths already includes some baked-in heat-related deaths, obscuring the cumulative total.
A third approach, favored by academics — and recently employed by Callahan to estimate the 2,700 heat-related deaths in France last month — involves using long-term data on both temperatures and mortality for a given location and then fitting a statistical model that relates the two. This method has the advantage of generating a U-shaped relationship that shows how mortality rates change once temperatures exceed a certain threshold (or drop below it, in the case of cold-weather-related deaths, hence the “U”). Like an excess death analysis, this method “has the benefit of, again, not having to rely on individual diagnoses,” Callahan said. “It has the drawback that there is no one right statistical model. Different people have different philosophies about how to fit those models.”
The other drawback is that creating such a model and subjecting it to the rigor of peer review is time-consuming — by the time you’re able to publish a death toll, the news cycle has probably moved on. Callahan got lucky: He had already created such a model for France to study the 2003 heat wave, which killed an estimated 15,000 in the country in a couple of weeks. The model relies on a historical understanding of the relationship between temperature and mortality in France — “not a crazy assumption, but an assumption,” he admitted to me — and he published the findings in Carbon Brief earlier this month. (Callahan also estimated that 20,000 people died continent-wide in Europe during the June 2026 heat wave — a number that circulated widely, but that he told me he’s now working to revise downward.)
Notably, the number Callahan arrived at for France does not represent “real” people or “real” deaths, at least as linked to death certificates. There are no biographical or even demographic numbers attached to it. (That said, you can create models of the same U-shaped relationship for anything: temperature and age, income, race.) More mind-bending, though, is that because of this, Callahan’s model can also be used to predict. Had there been a way to know the exact temperatures before the European heat wave, he could have told you how many people would likely die before they actually did.
In the case of France, the simple excess death count put the toll at 2,025, though officials say they expect the number to rise. While Callahan’s number and the official tally from France differ by what seemed to me to be a lot — 675 deaths — Callahan told me he’s actually encouraged by how close his model came to the government’s empirical count, given that the two use completely different methodologies.
After all, heat death counts can vary by orders of magnitude, including within a single government. Before 2020, the Centers for Disease Control and Prevention reported that only about 700 Americans died each year from heat, relying primarily on physician diagnostic codes. After moving in recent years to better incorporate underlying and contributing causes of death, the CDC adjusted its estimate upward to about 2,000 heat-related deaths per year in the United States. Still, the government’s numbers remain extremely conservative; independent researchers studying heat mortality say the figure is likely closer to 12,000.
But even more holistic heat-related mortality numbers have their critics. For example, models don’t work as well for many lower-income countries, where mortality may be reported monthly, thereby making day- or week-level heat attribution impossible.
Granularity presents its own set of problems. Excess deaths and modeling analyses both have to define the first “heat day” of an event. You can do that by setting a fixed threshold — say, anything above 90 degrees Fahrenheit counts as “high heat” — but Ebi told me there is little value in analyses or policies that take that approach. That’s because heat is contextual: “My standard joke is, if we had the temperatures here in Seattle that they have in Phoenix, we basically all die, because we don’t have the infrastructure and we’re not acclimatized,” she said.
A slightly better metric might be a relative threshold — say, temperatures above the 95th percentile of historical temperatures for a specific location count as “extreme heat.” The problem there, though, is that it may need to be stratified further by vulnerable populations that feel the effects much sooner, like adults over the age of 65, pregnant women, outdoor workers, or people with certain medical conditions. While that approach might seem overly complicated, parts of Asia already use nuanced thresholds to warn older adults to take precautions. “It’s going to be more challenging to communicate, I grant you that,” Ebi told me of such an approach — much less to try to model. “But it’s also going to be more useful.”
Even so, a larger problem remains: The multiple systems for calculating heat deaths are honed to address different questions, which makes them impossible to compare. The Federation of American Scientists has pushed for the CDC to upgrade and standardize its heat-mortality tracking. “We’ve thought about if it’s possible to ever set a goal of bringing heat-related deaths down by 50%, or something like that,” Grace Wickerson, the senior manager of climate and health at FAS, told me. “But we don’t even have a baseline number or a way to say, ‘This is the starting point for this goal or strategy.’”
Wickerson also suggested, though, that there might be things we lose in trying to nail down the most correct heat-related mortality number. “I’m almost a bit weary of the pursuit of large numbers,” she said. “At least to me, what feels more important is why people are suffering and dying, what types of people they are, and what stories, messages, and stakeholders we need to engage and target to actually build meaningful policy strategies.”
Despite being deeply engrossed in the calculations, Callahan stressed that he wants readers to have a similar takeaway from his own research. Improved “healthcare access and access to cooling, shade, and shelter” — or in the case of heat-related mortality from climate change, “reducing greenhouse gas emissions” — lead to fewer heat deaths, meaning the vast majority are preventable.
“The relationship between environmental conditions and a person’s mortality is not fixed or necessary,” he told me. “It can be stopped.”
Lights out in Cuba. Magdalena Chodownik/Anadolu via Getty Images
The startup team observes the test reactor at the Idaho National Lab. Clean Core Thorium Energy
