Temperature costs the NHS 3% of everything it spends — and cold is two thirds of it

Climate and health research has spent two decades counting deaths. It has spent much less effort counting what temperature does to a health system's ordinary work — the consultations booked, the prescriptions issued, the admissions that arrive whether or not anyone dies.

A study in The Lancet Planetary Health on 18 December is an attempt at that second accounting, and its authors describe it as the first comprehensive analysis of daily temperature variability across a national health-care system [s1].

The method

The team analysed 4,366,981 patient records from the NHS in England, drawn from the Clinical Practice Research Datalink GOLD, covering 1 April 2007 to 1 June 2019 [s1]. Weather came from Met Office HadUK-Gridded climate observations [s1].

They used a fixed-effects regression model to estimate the temperature–health-care relationship while controlling for seasonality and practice-specific characteristics, with outcomes measured as daily counts of health-care events and their associated costs per 1,000 GP-registered individuals, stratified by age, sex and care domain [s1]. Robustness checks used alternative lag structures and model specifications [s1].

The headline number

Suboptimal temperature exposure accounted for 3.0% of total health-care expenditure (95% CI 1.2–4.7) [s1].

Cold drove 64.4% of that burden [s1].

Why heat looks bigger than it is

The most useful part of the paper is a methodological correction it applies to its own results.

Very hot days — above 23°C — produced sharp same-day surges in accident and emergency attendances and in prescriptions [s1]. That is the pattern that generates headlines during a heatwave, and it is real.

But the cumulative effect was attenuated once the dips in the following days were included [s1]. In other words, a substantial share of what looks like heat-generated demand is demand displaced forward by a few days rather than created. Someone who would have attended on Thursday attends on Monday instead.

Cold behaved in the opposite way. Colder days — average temperature between 0°C and 9°C — were associated with cumulative increases in GP consultations, inpatient admissions and deaths, with disproportionately larger effects among people over 65 [s1]. And A&E attendance actually declined on cold days [s1].

That combination is the interesting one. The single most visible indicator of health-system stress — emergency department attendance — moves in the wrong direction during exactly the temperature range that is producing the most excess consultations, admissions and deaths. A monitoring system watching A&E queues would read cold weeks as quiet ones.

What this can and cannot support

This is an observational economic analysis using routine records, and temperature is not randomly assigned across days in the way an experiment would require. Fixed-effects models of this kind control for seasonality and stable practice characteristics; they do not control for whatever else moves with cold weather, including respiratory virus circulation, which is both temperature-linked and a major driver of winter consultations. The paper's framing — temperature variability disrupting health-care delivery — is compatible with cold acting largely through infection, and the analysis does not adjudicate that.

The window also closes in June 2019, before the pandemic reorganised English primary care and before the record heat of 2022 [s1]. An estimate of the heat share drawn from a period ending in 2019 is a conservative one for a warming country.

The study was funded by the Oxford Martin School's Future of Cooling Programme at the University of Oxford and the Spanish Ministry of Science, Innovation and Universities [s1].

The same issue, in a different currency

Published the same day in the same journal, a nationwide case-crossover study measured what temperature does to a specific vulnerable group in mortality terms rather than cost terms. It analysed 289,902 diabetes-related deaths across mainland China from 2013 to 2019, using records from the China Cause of Death Reporting System [s2].

Exposure to extreme high temperatures — the 97.5th percentile, 31.0°C — compared with the minimum mortality temperature was associated with an increase in overall diabetes mortality, with an odds ratio of 1.25 (95% CI 1.22–1.29) over a 0–6 day lag [s2]. The risk was higher in cooler regions [s2].

The regional pattern inverted between diabetes subtypes. In the warmest zone studied, the subtropical monsoon zone, people with type 2 diabetes were at higher risk than those with type 1 (OR 1.21, 95% CI 1.16–1.26, versus 1.14, 1.04–1.26); in the coldest, the temperate continental zone, the relationship reversed (1.31, 1.09–1.58, versus 1.65, 1.17–2.33) [s2].

Projected forward to the 2090s under a high-emission scenario, the study estimates the heat-attributable fraction of diabetes deaths at 11.16% (empirical 95% CI 6.11–18.01) [s2]. Population ageing and growth were projected to add roughly one percentage point; a 50% adaptation scenario was projected to subtract about five [s2].

Those are projections dependent on emissions pathway, population model and the assumption that exposure–response relationships hold for seventy years. The wide interval on some of the complication estimates — 28.65% for peripheral vascular disease deaths, with bounds of −22.60 to 46.95 — is the honest signal of how much uncertainty sits inside them [s2].

Why it matters

Two papers, two currencies, one shared implication: the temperature burden is concentrated where the health system is least likely to be looking. In England it is in winter consultations rather than summer emergency queues [s1]. In China it is in the coldest zone, where heat is rarest and adaptation weakest [s2].

Adaptation spending tends to follow visibility. Both studies suggest visibility is a poor guide.

What to watch

Whether the English analysis is repeated on a window that includes 2020–2025, and whether NHS winter planning begins to treat cumulative cold-related primary care demand as a tracked indicator alongside emergency attendances.

This article is informational and is not medical advice.

Sources

• [s1] Fahr P, Cohen F, Schiff J, et al. "Quantifying the health-care burden of temperature in the National Health Service in England: an economic analysis of resource use and costs." The Lancet Planetary Health, 9(12), published online 18 December 2025. https://doi.org/10.1016/j.lanplh.2025.101373
• [s2] Luo H, Yu T, Gao Y, et al. "Heat-related mortality burden of type 1 diabetes, type 2 diabetes, and diabetes complications in mainland China amid global warming: a nationwide, case-crossover study." The Lancet Planetary Health, 9(12), published online 18 December 2025. https://doi.org/10.1016/j.lanplh.2025.101384