A Cooling Center Saves a Life Every Thousand Visits — or Every 1.6 Million. It Depends Who Walks In.

By Common Ladder · August 6, 2026 · 7 min read

Every summer the same sentence appears in Phoenix press releases: the county has opened cooling centers. It is a reassuring sentence. It sounds like a response proportionate to a problem that killed 430 people here last year.

The evidence underneath it is thinner than almost anyone assumes — and, read carefully, it does not say cooling centers don't work. It says something more useful and more uncomfortable: a cooling center is an intervention whose entire value depends on who is sitting in the chair.

What the evidence actually shows

In July 2025 a team at the UK Health Security Agency published the first systematic review of whether cooling centres reduce heat-related illness and death. They searched five databases across fourteen years of literature. Five studies met the criteria. All five were from North America, three of them from Maricopa County. Their finding, stated plainly: "No studies reported real-world health outcomes for cooling centre users."¹

That is not a finding of failure. The authors are explicit that it is not: "this does not mean that there is good quality evidence of no effect either."¹ It is a finding that a public-health intervention deployed in hundreds of American cities has never been tested the way we would test a drug.

What has been tested is the physiology. Two Canadian laboratory studies put forty healthy older adults through a simulated nine-hour heat wave; half spent two midday hours in an air-conditioned room before returning to the heat. Cooling produced real improvements in cardiovascular strain — and those improvements dissipated quickly once participants went back out.⁷ The companion paper found mood and symptoms improved and lasted longer, and then noted the trap in that: feeling better may lead people to underestimate their ongoing risk and skip the precautions they would otherwise take.⁸

The one population-level study is older and weaker: an area-level analysis of Maricopa County found heat mortality rose faster in neighborhoods with less publicly accessible cool space, independent of social vulnerability. But its definition of "cool space" included shopping malls, museums and restaurants, and the review rated it low quality.⁹

The number that changes everything

In 2022 a team modeled how many people would have to visit a cooling center on a hot day in Maricopa County to prevent one heat death. Their answer, under most assumptions, was more than a million — about 1.6 million visits per life saved. The implied conclusion was blunt: you could not build enough cool space to matter.²

Two years later, researchers at Arizona State and the county reran that model with one change. Instead of the general population, they used the population of people experiencing homelessness in the same county, whose estimated daily heat mortality rate is about 1,972 per million. Same model. Same assumptions. Same city. The number needed to visit fell to fewer than 1,000.³

A factor of roughly 1,600. Nothing about the building changed — not its air conditioning, not its hours, not its budget. The only variable was who came through the door.

Both models are crude, and the systematic review rated both as very low quality; they rest on assumptions the authors chose rather than measured.¹ Do not treat either figure as a measurement. Treat the ratio as the finding, because the ratio is driven by something we do know with confidence: baseline risk. An intervention that prevents deaths can only prevent deaths among people who were going to die.

Who is actually in the room

Which makes the next finding the most important one in this brief. In 2017 Maricopa County ran an evaluation of its own cooling centers — observations at 53 facilities, interviews with 52 managers and 658 visitors. Two-thirds of the visitors had a permanent residence. Eighty-nine percent had air conditioning at home. Seventy-seven percent stayed less than four hours.⁴ The study's authors drew a more generous conclusion than we are drawing here — they found the centers reached "some of the region's most vulnerable populations," and many visitors were unemployed or homeless. Both readings fit the same data. The fieldwork is also now more than a decade old.

Most of the capacity, in other words, was being used by people whose baseline heat-death risk was near the bottom of the distribution — the 1.6-million-visits group — while the people in the fewer-than-1,000 group were disproportionately outside. That is not a criticism of anyone who walked in. It is a description of a program that was built to be available and was never built to be aimed.

And the barriers that keep the highest-risk group out are mundane and fixable. A 2025 rapid review in the Journal of Urban Health found cooling centres are "often not well advertised, inconveniently located, and operate with restrictive rules that exclude unhoused individuals" — rules on substance use, on pets, on personal belongings, and requirements to show identification or sign in.¹⁰ People who cannot leave a dog, a cart, or a partner at the curb do not come in. The same review found public libraries were preferred precisely because they ask for nothing.

Maricopa County has since asked its visitors what would help, and the answers are logistical rather than philosophical. In a 2023 assessment published in MMWR, surveys of 944 cooling-center visitors and 1,260 members of the general public found that the most common barrier was simply getting there — described by 31% of visitors — that a majority wanted centers open until at least 7 p.m., and that street signage was rated the best way to advertise a site. The same report notes something worth sitting with: between 2014 and 2023 the county doubled its cooling centers to 112 while heat deaths rose from 61 to 645.¹¹ Those two trends are confounded by a hotter decade, a growing unsheltered population, and fentanyl and methamphetamine. But they are not consistent with a story in which capacity alone is the constraint.

There is also a dosing problem. A 2025 study of 44 older adults experiencing homelessness using a Phoenix respite center — the first real-world physiological work of its kind — recommends a minimum of about 70 minutes and a suggested 120 minutes of cooling after extended heat exposure, with structured rehydration.⁶ A site that seats someone for twenty minutes and moves them along has not delivered the intervention. It has interrupted it.

Being precise about what this does and doesn't establish

This is not proof that targeted cooling centers save lives. It is proof that the studies capable of answering that question have not been done, and an argument about where the plausible benefit is concentrated. The systematic review's own recommendation is to run experiments on the highest-risk groups precisely because that is where an effect would be large enough to detect.¹

Cooling centers are not housing, and they are not shelter. The rapid review is unambiguous that housing access is the long-term intervention and everything else is interim.¹⁰ Nothing here argues for spending on cool rooms instead of spending on units.

Heat is rarely the only cause. In Maricopa County's 2025 deaths, 55% involved substance use, 82% of those involving stimulants, and cardiopulmonary disease contributed in 47% of cases.⁵ Heat kills people who are already carrying something else. That is an argument for pairing cooling with medical outreach and treatment, not for discounting the deaths.

And the indoor deaths are a different problem with a different fix. Of the 111 indoor heat deaths in 2025, an air conditioner was present in 94% of cases — and in 72% of those, it was not working.⁵ No cooling center reaches that person. Utility assistance and AC repair do.

What this looks like in Maricopa County right now

The county identified 430 heat-related deaths in 2025 — a second consecutive annual decline from 645 in 2023 and 608 in 2024, and still a number that would be a front-page emergency almost anywhere else. 208 of those people were experiencing homelessness. Seventy-four percent of deaths occurred outdoors, and 60% of the outdoor deaths were in an urban area, not the desert. Roughly 60% of all decedents were 50 or older. Between June 13 and September 5, 2025, someone died of heat in this county every single day for 85 consecutive days.⁵

The county's own report contains the sentence that ought to govern the whole discussion: "all heat-related deaths are preventable."⁵

2026 is not tracking better. Phoenix recorded its hottest spring on record, and by early July the county had confirmed 16 heat deaths with 192 more under investigation — 13 more confirmed than at the same point in 2025.¹² Metro Phoenix set a 115°F record on August 3.¹³ In-season counts lag by two to three months while the Medical Examiner completes each investigation, so today's number is always an undercount, and the share identified as experiencing homelessness rises as investigations close.¹⁴

What follows from this

If you run or fund heat relief, the evidence does not tell you to open more sites. It tells you that the return on each site is set almost entirely by whether the people at highest risk can and will use it. That points at a short, unglamorous list: place sites where unsheltered people already are rather than where buildings happen to be free; keep them open through the hours people actually die, which in 2025 meant late afternoons across a season that ran into September; let people bring their dog, their cart, and their partner; drop the sign-in sheet and the ID; staff for a two-hour stay with water rather than a twenty-minute pass-through; and put outreach and medical care in the room, since the people who most need the cooling are also the people carrying the cardiopulmonary disease and the stimulant use that turn 112°F into a death certificate.

None of that requires new science. It requires deciding that a cooling center is a targeted clinical intervention rather than a courtesy, and building it accordingly. The difference between those two framings, in this county, is measured in the hundreds.

Have a correction or an updated figure? Email corrections.commonladder@gmail.com.