Winters are much milder than they were, and the cold still kills thousands

UKHSA published its cold mortality monitoring report for England yesterday, on 22 October 2026. The figure in it is 2,200 cold-associated deaths last winter, across five cold weather episodes. One nine day spell in early January accounts for 1,563 of them.

The winter before that, 2024 to 2025, produced an estimated 2,788 across three episodes.

Now put that next to the temperature record. Across the 25 towns this site holds a daily series for, the average December to February temperature in the 1960s was 3.18 degrees. Over the six winters from 2020 to 2025 it was 5.46 degrees.

Two and a quarter degrees warmer, and thousands of people are still dying of the cold.

How much milder, precisely

Computed from this site’s stored series, 1950 to 2026, for the 25 published towns. A winter is December plus the following January and February, and a winter counts only where the town has a near-complete record for it.

An air frost day is one where the minimum fell below zero. The 1960s were the coldest decade in the record and are the right comparison for anybody who remembers winters being worse, because they genuinely were.

Every town has lost roughly half its frosts. Perth has gone from 54 air frost days a winter to 32, Manchester from 45 to 21, London from 40 to 20, Plymouth from 14 to 4. Not one of the 25 is colder than it was.

The 1970s step is worth noting and not over-reading: it is a real jump in the record, and a decade is a short enough period that one or two severe winters move it. The direction across the whole 77 years is what the table is for.

Why milder winters have not fixed it

Here is the thing that takes a moment to sit properly in your head. Cold mortality is mostly not caused by extreme cold.

The deaths are not concentrated in the record-breaking spells. They sit across the ordinary cold range, the two to eight degrees that Britain has in abundance from November to March and always will. UKHSA’s own breakdown shows what the mechanism is: an estimated 757 of the 2,200 deaths were from circulatory disease. Cold raises blood pressure, thickens the blood and constricts the vessels. It shows up as heart attacks and strokes, days after the cold rather than during it, which is exactly why it took statistical modelling to find it at all.

A building either holds a reasonable internal temperature during a cold spell or it does not. Losing eighteen air frost days a winter does nothing for a house that cannot hold 18 degrees at an outside temperature of 4. And there are still 38 days a winter below 5 degrees, down from 61, which is a large improvement in a number that is still large.

So the exposure that matters has fallen much less than the headline temperature suggests. The weather got better. A lot of the housing did not.

What the regional figures do and do not tell you

UKHSA puts the North West highest, at an estimated 52 cold-associated deaths per million people, with the East Midlands and Yorkshire and The Humber next. No cold-associated deaths were identified in the North East at all.

It is tempting to read that as a league table of cold homes. Do not. It is one winter, the cold episodes did not fall evenly across the country, and a region with no identified deaths is a region where the model found no detectable signal rather than one where nobody was cold.

What the regional spread does establish is that this is not a story about the far north. The North West and the East Midlands are not the coldest places in Britain. Scotland and the north east of England are colder, as the frost table above shows, and the deaths were not concentrated there. Which points away from outdoor temperature and towards everything else: the age of the housing, the age of the people in it, and what they can afford to run.

The household decision this actually implies

Nothing in a mortality report tells one household what to do. What it does is reframe the question that most people answer by flinching at the smart meter.

Three things worth separating.

Heating a cold home more is expensive and sometimes still the right answer. The cost is calculable, not mysterious. The running cost calculator on this site works out what any given appliance costs per hour at a unit rate you type in, so a decision about a cold week can be made with a number rather than a feeling.

Heating the room you are in beats heating the house, when the house cannot hold it. This is where the arithmetic of a cold old building bites, and we went through what it costs to change that in what it costs to make an old house cope with heat, which deals with the same fabric in the opposite direction.

A warmer winter does not mean a cheaper one, necessarily. Less heating demand is real, and we measured it: every one of the 25 towns now needs fewer degree days than it did in the 1960s, set out in the heating season is getting shorter. What has moved faster than demand is the unit rate, which is why bills have not followed the thermometer down. What the October 2026 price cap changes has the current numbers.

The honest summary

This is the cold half of a pattern we have already written about from the other end. In the heat, the health signal is sharp, concentrated in a handful of days, and visible in the health data within hours. In the cold, it is diffuse, spread over months, lagged by days, and an order of magnitude larger.

Britain is getting better at the first problem and has been slowly stopping measuring the second. The ONS paused its excess winter mortality series in 2023 for a methodological review; UKHSA’s cold monitoring report is the successor and is only in its second year. That is not a conspiracy, but it does mean the cold figures and the heat figures are not on comparable footings, and anybody comparing them should say so.

What is not in doubt is the direction of the two datasets on this page. Winters have warmed by more than two degrees since the 1960s and halved their frosts. The cold is still killing thousands of people a year. Those facts are compatible, and the thing that reconciles them is the building, not the weather.

Temperature figures are computed from this site’s own stored daily series for the 25 published towns, 1950 to 2026, derived from ERA5-Land reanalysis via Open-Meteo and licensed CC BY 4.0. Mortality figures are from UKHSA’s cold mortality monitoring report for winter 2025 to 2026, published 22 October 2026.