Heat-related hospital admissions in Brazil could rise 54% to 78% by the 2050s

Most research on heat and health counts deaths. Deaths are the endpoint that is hardest to argue with, but they are not the endpoint that health ministries budget against. Two studies published in Environment International this month count something else — hospital beds in Brazil, and school readiness in Western Australia — and between them they sketch what heat costs a system that is not dying of it.

Brazil: the hospital bill

The first study, published on 20 June, assembled daily non-elective hospital admissions from the Brazil Hospital Information System for 5,459 municipalities — 98% of the country's total — covering 2008 to 2019 [s1].

Gridded daily maximum temperatures came from the European Centre for Medium-Range Weather Forecasts ERA5 reanalysis for the historical period 2008-2023, and were projected to 2060 under three Shared Socioeconomic Pathway emissions scenarios; population projections came from WorldPop [s1]. The analysis used a case time-series design with distributed lag non-linear models, fitted per state, and estimated heat-attributable hospitalisations from 2008 to 2060 alongside economic costs using a cost-of-illness approach covering direct and indirect costs [s1].

Without adaptation, annual high-temperature-related hospitalisations are projected to reach 51 per 100,000 population (95% CI 19-103) in the 2050s under SSP1-2.6, 54 (21-106) under SSP2-4.5 and 59 (25-112) under SSP5-8.5 [s1]. Against a 2010s baseline of 33 per 100,000 (9-67), those are increases of 54%, 62% and 78% respectively [s1].

Annual economic costs are projected to reach $228-264 million in the 2050s, with higher absolute costs in the South and faster relative increases in the North [s1].

The regional split is the substantive finding. A national average would obscure it entirely: the states carrying the largest absolute burden and the states experiencing the fastest growth in burden are not the same states, which means a nationally uniform adaptation policy would be misallocated by construction.

The confidence intervals are wide — the 2050s SSP5-8.5 estimate ranges from 25 to 112 per 100,000 [s1] — and the phrase "without adaptation" is load-bearing. The projection assumes the temperature-hospitalisation relationship estimated from 2008-2019 data holds unchanged for four more decades, with no acclimatisation, no air conditioning uptake and no change in clinical practice. That is a deliberate baseline, not a forecast of the most likely future.

Western Australia: the developmental bill

The second study, published on 25 June, measures a different kind of cost, and one that does not appear in any health budget [s2].

It is a population-based retrospective cohort of 59,771 children from the 2009-2015 Australian Early Development Census [s2]. Exposure was the Universal Thermal Climate Index, a biothermal measure that combines temperature with humidity, wind and radiation, spatiotemporally linked to maternal residential addresses at birth at the Statistical Area 1 level [s2]. Average UTCI exposure was calculated from conception to the census assessment, with separate analyses for the prenatal period (conception to birth) and the postnatal period (birth to assessment) [s2].

Mean UTCI exposure across the whole window was 14.4 ± 2.5 °C [s2].

After adjustment for confounders, children exposed to heat stress across the full period had 29% higher odds of being developmentally vulnerable in one or more census domains than children exposed to optimal thermal conditions, defined as 9-26 °C (OR 1.29, 95% CI 1.05-1.58) [s2]. Prenatal exposure to heat stress was associated with 34% higher odds (OR 1.34, 95% CI 1.10-1.65) [s2].

The prenatal association being stronger than the whole-period one is the result the authors highlight [s2]. It is also the result that should be held most loosely: prenatal and postnatal exposures for the same child are correlated by geography, and separating them statistically does not fully separate them causally. Families do not move randomly.

Both confidence intervals have lower bounds close to 1.0 — 1.05 and 1.10 — on a binary outcome from a mixed-effects logistic model with a large sample [s2]. Statistically significant, but not far from the null.

Two currencies, one exposure

Brazil's study prices heat in hospital admissions and dollars over four decades; Western Australia's prices it in the odds that a five-year-old starts school behind [s1][s2]. Neither can be converted into the other, and neither is a mortality figure.

Both are observational. The Brazilian analysis links administrative admissions to gridded weather without individual-level exposure; the Australian cohort assigns exposure from a residential address at birth, which is a proxy for where a child actually spent their days [s1][s2]. Neither establishes causation, and both say so.

What they establish jointly is narrower and worth keeping: the health consequences of heat are already being recorded in datasets that were built for other purposes — hospital billing systems and school-entry censuses — and those records show effects at exposure levels that are not extreme.

What to watch

Whether adaptation assumptions get built into the Brazilian projections, since the current figures are explicitly the no-adaptation case [s1]. And whether the Western Australian finding replicates in cohorts where residential mobility can be tracked rather than assumed [s2].

Sources

• [s1] Regional variations in the impacts of high temperature on hospital admissions in Brazil. Environment International, 20 June 2025. https://doi.org/10.1016/j.envint.2025.109620
• [s2] Prenatal and early childhood exposure to biothermal stress and developmental vulnerability at school entry in Western Australia: a population-based cohort study. Environment International, 25 June 2025. https://doi.org/10.1016/j.envint.2025.109642