Out-of-Season Heat Tracked 3x the Youth Mental-Health Admissions

It is late in an unseasonable warm spell, and a parent is standing in a kitchen doing arithmetic that has nothing to do with the weather: three days of poor sleep, a fourteen-year-old wound tighter each evening, a fraying that started roughly when the forecast did. The week's headlines had said something about heat and children's mental health, and the claim lodged somewhere it can no longer be ignored. Whether those two things have anything to do with each other is the sort of question that used to be answered with a shrug.

The study behind those headlines is more specific than the coverage suggested, and more interesting. It is about hospital admissions rather than emergency-department visits, it runs to nearly three quarters of a million of them, and its most striking finding is one almost no headline carried: the temperature signal was strongest out of season.

Wen-Qiang He, Cybele Dey, Chenxi Li, Nan Hu, Iain Perkes, Haidong Kan, Raghu Lingam, Grant Sara, Ollie Jay and Natasha Nassar assembled every hospital admission for a mental health disorder in New South Wales, Australia, from 2001 through 2022, among children aged 12 and under, adolescents aged 13 to 17, and young adults aged 18 to 24. That came to 719,375 admissions. The exposure was daily average temperature. The analysis, published in the Journal of the American Academy of Child and Adolescent Psychiatry in May 2026, used a case-crossover time-series design with conditional logistic regression and lag-distributed non-linear models.

The case-crossover design is worth understanding, because it determines what the numbers can mean. Each admitted patient effectively serves as their own control: the temperature on the day of admission is compared against temperatures on other, nearby days for that same person. Anything fixed about a child — diagnosis, sex, family, neighbourhood, the quality of their school — is held constant by construction, because it is the same on both days. What the design cannot hold constant is anything else that moves with the temperature.

Rather than reporting one number for the whole year, the authors split the analysis by season and compared the 99th-percentile temperature within each season against that season's minimum morbidity temperature — the temperature at which admissions were lowest. That choice is what surfaced the finding underneath.

On the hottest cold-season days, the relative risk of admission was 3.03 (95% CI 2.56-3.60). On the hottest warm-season days it was 1.92 (95% CI 1.58-2.34). Both intervals sit well clear of 1, and they do not overlap each other — the cold-season effect is the larger one, and not marginally. The plainest reading is that what matters is not absolute heat but heat the body, the household and the building are unprepared for.

The pattern held across most diagnostic categories rather than resting on one. Substance use, depressive, anxiety, obsessive-compulsive, reaction and adjustment, and attention-deficit, disruptive and impulse-control disorders all showed elevated risk in both seasons, with relative risks ranging from 2.02 to 5.49 across that group. Schizophrenia, eating disorders and deliberate self-harm showed elevated risk in the cold season only, ranging from 2.22 to 5.23. Manic episode and bipolar disorders showed no association in either season — a null result in the middle of a positive paper, which is a mild reassurance about the analysis, since a method that manufactured signal would have manufactured it there too.

The subgroup findings cut in different directions by season: risks were higher among female patients and young adults in the cold season, and among male patients and the youngest children, aged 0 to 11, in the warm season. The authors also report that the effects have grown in recent years.

Then comes the number that should govern how the rest is read. Heat above the minimum morbidity temperature accounted for 8.36% of admissions in the cold season and 2.94% in the warm season. A relative risk of three sounds enormous, and it is — on the handful of days each year that reach the 99th percentile. Spread across every admission in two decades, heat explains roughly eight in a hundred cold-season admissions and three in a hundred warm-season ones. Both figures can be true at once, and only one of them describes the caseload of an average week.

The paper also models forward. By 2090-2099, it projects heat-attributable admissions rising by 6.0% under a low greenhouse-gas emissions scenario, 7.7% under a medium one, and 20.8% under a high one. That spread is the paper's argument for treating mental health as a climate-adaptation question rather than a seasonal inconvenience.

It is also the part of the paper resting on the most assumptions. A projection to the end of the century holds the measured temperature-admission relationship fixed for seventy years, across changes in air conditioning, housing, psychiatric service capacity, admission thresholds and treatment that nobody can specify. The authors' own finding that effects have strengthened in recent years cuts both ways here: it suggests the relationship is not fixed, which is precisely what the projection assumes it is.

An admission is the severe end of a distribution. For every young person who reaches a hospital bed on a hot day, there are many more having a worse week at home, and this study by design says nothing about them — which means the effect it measures is the visible tip of something wider, not the whole of it.

What survives the caveats is a shape worth knowing: the association is real, it is measured across 719,375 admissions and twenty-two years, it spans most diagnostic categories rather than one, and it is largest when the heat is unexpected. Clinicians weighing seasonal patterns in demand, and services deciding when to staff for surges, have generally organised around summer. On this evidence the shoulder seasons deserve a share of that attention — and the reason the finding is counterintuitive is exactly why it had gone unmeasured.

The projections are what will get quoted and deserve the least weight; the cold-season result is what deserves the most. A finding that the hottest days of the cool months carry a higher relative risk than the hottest days of summer is the kind of result that is hard to produce by wishful analysis, because nobody was looking for it and it embarrasses the intuitive story. It also has a plausible reading that costs nothing to hold: acclimatisation is relative, and an unseasonable 30-degree day lands on unprepared bodies, unprepared buildings and unprepared routines in a way that the twentieth such day of a summer does not. Set against that, the attributable fractions are a useful deflation — heat is a real contributor to youth mental-health admissions and a minor one in the aggregate, and both halves of that sentence are load-bearing. What this paper has genuinely established is that the seasonal shape of the problem is not the one most people assumed, measured well enough and at enough scale to be worth acting on before the mechanism is understood.