England recorded an estimated 1,504 heat-associated deaths during five defined heat episodes in summer 2025, according to the UK Health Security Agency. That was far below the 3,039 deaths predicted by a separate model using the temperatures recorded and recent relationships between heat and mortality.
The difference was 1,535 deaths. It is an important result, but not proof that a particular warning system, policy or change in behaviour prevented that exact number of deaths. UKHSA said the causes could not be established from this analysis and called for further evaluation.
The finding came from a summer that set a UK temperature record. The Met Office measured a mean temperature of 16.10C from June through August, 1.51C above the 1991-2020 average and higher than the previous record of 15.76C in 2018. The season included four meteorological heatwaves, each relatively short, and a peak of 35.8C in Faversham, Kent.
Two Estimates Answered Different Questions
UKHSA's observed estimate was not a count of death certificates listing heat as the cause. Analysts identified periods when the mean Central England Temperature reached at least 20C, added one day on either side, and compared daily deaths during those periods with deaths in 14-day baseline windows before and after each episode.
That calculation produced 1,504 heat-associated deaths, with a 95% confidence interval from 936 to 2,072. The interval shows the uncertainty around an estimate derived from changes in daily mortality. UKHSA treats an estimate as statistically significant when its interval does not overlap zero.
The 3,039 figure came from another calculation. A statistical model used England's temperature-mortality relationship over the previous five years to estimate the relative risk at temperatures above the median summer temperature. Researchers then applied that relationship to the actual temperatures and dates of the 2025 heat episodes. Its estimate was 3,039 deaths, with a 95% confidence interval from 2,731 to 3,293.
Calling one figure observed and the other predicted can make the comparison sound simpler than it is. Both are statistical estimates. One looks for deaths above nearby seasonal baselines; the other asks what recent temperature-mortality patterns would have predicted for the heat that occurred.
The Lower Figure Did Not Establish a Cause
The observed estimate stayed below the modelled estimate in all five public-health heat episodes. The difference was statistically significant in the first three episodes. That consistency is why UKHSA considered the result noteworthy rather than dismissing it as a single unusual period.
Possible explanations include national and local implementation of the Adverse Weather and Health Plan, heat-health alerts, action across health and social care, emergency response and protective behaviour by the public. Warmer conditions in late spring may also have prompted people to change their behaviour before the main alert periods.
None of those possibilities was measured as a separate intervention. The analysis did not compare alerted and unalerted communities, assign people to different protective measures or calculate how many deaths each response prevented. UKHSA stated directly that the work could not attribute the reduction to specific causes.
The language matters because the 1,535-death gap is not a verified tally of lives saved by alerts. It is the difference between two estimates produced by different methods. Treating it as an intervention effect would turn a plausible explanation into a measured result.
Older Adults and Care Homes Carried the Highest Burden
Lower-than-modelled mortality did not mean the heat was harmless. Adults aged 85 and over had an estimated 793 heat-associated deaths, or 364 per million people. Those aged 75 to 84 had a rate of 116 per million. UKHSA found no statistically significant heat-associated mortality in younger age groups during the identified episodes.
Care homes recorded an estimated 677 heat-associated deaths and the largest increase above baseline, at 9%. Hospitals recorded 485. The final episode produced the highest overall estimate, 697 deaths, and was the only episode with statistically significant heat-associated mortality in hospitals.
The geographic pattern followed exposure. Mortality was concentrated in southern, central and eastern England. UKHSA did not detect statistically significant heat-associated mortality in the North East, North West, or Yorkshire and the Humber. At smaller local-resilience-forum level, the agency warned against overinterpretation because modest changes in daily deaths can look large when population and event counts are small.
Circulatory diseases were the leading underlying cause associated with the estimated excess, followed by cancer and dementia or Alzheimer's disease. The cancer finding appeared for the first time in this annual analysis and UKHSA said it required further investigation.
A Better Season Is Evidence to Study, Not a Shield
The comparison supports a focused conclusion: England experienced fewer heat-associated deaths than a recent temperature-mortality model predicted during the defined 2025 episodes. It does not show that heat risk was solved, identify one successful intervention or guarantee that the same pattern will recur.
Even the lower estimate exceeded the 1,311 reported for the cooler summer of 2024. The 2025 total was below the 2,295 reported in 2023 and the 2,985 estimated in 2022, when the UK reached a record 40.3C. Those annual figures provide context, but differences in weather, timing, geography and population exposure prevent a simple ranking of policy performance.
The Met Office found that a summer as warm as 2025 was around 70 times more likely in the current human-influenced climate than in a climate without that influence. Four meteorological heatwaves and five UKHSA heat episodes are not contradictory counts: the agencies use different definitions for weather events and health monitoring.
A serious public-health response should investigate the gap rather than claim ownership of it. If alerts, care-sector preparation or earlier protective behaviour reduced harm, evaluation should show where and for whom they worked. If other conditions explain part of the result, those limits matter just as much. The strongest lesson from 2025 is not that England built a shield against heat. It is that a favourable estimate created a testable question that the next season may answer differently.