Back in January 2022, in a paper titled, ‘Climate-related mortality and hospital admissions, England and Wales: 2001 to 2020’, the UK’s Office for National Statistics (ONS) stated the following:

Over a 20-year period the estimated change in deaths associated with warm or cold temperature was a net decrease of 555,094, an average of 27,755 deaths per year (Table 1). A decrease in deaths from outcomes associated with cold temperature greatly outnumbers deaths associated with warm temperature.

This was, of course, a rather inconvenient truth for those wishing to push the narrative that climate change had heightened the risk of temperature-related mortality within the UK. What made it all the more awkward was the fact that an independent poll from the National Centre for Social Research (NatCen) showed that over 80% of the British public trusts the ONS, widely believing the agency to be free from political manipulation and government interference. This wasn’t a dodgy climate-denying blog blowing a hole in the climate narrative, it was a highly prestigious and trusted organisation.

So something had to be done, and it wasn’t too long before the solution was found. In 2023, the same organisation was able to publish an updated paper that told a very different story. Instead of a summary statement pointing to a steadily falling overall death rate, we had this:

There was some indication that heat-related deaths have increased over recent years; in 2022 an estimated 4,507 deaths (95% confidence interval: 3,363 to 5,587) were associated with the hottest days in England.

Yes, the takeaway message had now become the possibility of a rise in the incidence of heat-related deaths.

But this change of emphasis in an executive summary is the least of the things that should be sounding your alarm bells. The real solution to the problem lay not in a judiciously worded synopsis; the real solution lay in a new way of counting the dead.

The ONS gets creative

In 2022 the ONS embarked on a grand new scheme for analysing climate change health statistics, referred to as the Standards for Official Statistics on Climate-Health Interactions (SOSCHI) project.

SOSCHI is financed by the global charitable foundation Wellcome, and has the purpose described in a 2026 ONS bulletin. This bulletin points out that, working alongside ‘global partners’, the ONS had developed a scheme that promised ‘a transparent and globally usable framework and platform for official statistics on climate and health’. Furthermore, the ‘final outputs of the project’ include the ‘resources and tools to help stakeholders produce high-quality data and statistics, and communicate with a range of audiences’. Which all sounds great, but in practice the real significance lies in the fact that SOSCHI gave the ONS the opportunity to re-analyse the situation. As the ONS said in an August 2026 update to the mortality statistics:

This bulletin incorporates methodological, analysis code and data source improvements, which means these estimates supersede those published in our previous Climate-related mortality, England and Wales: 1988 to 2022 article.

Superseded would be understating things somewhat. What their ‘improvements’ did was reduce the ratio between cold- and heat-related deaths from what was understood to be a massive 20:1 (ref.) to a relatively modest 3.5:1 – admittedly, still an excess of cold over heat, but one that now looks much more vulnerable to rising temperatures. And this absolutely colossal adjustment to the estimates was achieved relatively easily, simply by no longer considering all deaths within a season but counting instead only those deaths associated with the top 2.5% of hottest and 2.5% of coldest days. All of which is quite convenient for those pushing the heat death narrative, because virtually all heat-related deaths will occur within the 2.5% cut-off and therefore be counted, whereas the vast majority of cold-related deaths fall outside of the 2.5% cut-off and therefore will not. To understand why, one has to appreciate differences in what epidemiologists refer to as the exposure-risk curve.

It’s an exposure thing

In epidemiology, temperature mortality is generally mapped on a U-shaped or J-shaped curve that shows how mortality changes with temperature. However, the two sides of the curve act in completely different ways for the two threats. The risk curve for heat features a steep, dramatic spike (a J-shape). Humans tolerate a range of warm temperatures well, but once a specific extreme threshold is crossed (around 29°C in the UK), mortality risk jumps suddenly due to heatstroke, dehydration, and rapid cardiovascular strain. In contrast, the exposure-risk curve for cold is a gentle, prolonged slope (a less dramatic inverse J-shape). Crucially, the risk of dying doesn’t just spike when it is well below freezing (-5°C); it begins to climb steadily at what most people would consider mild or moderately cold temperatures, such as 8°C to 12°C.

In summary, the splendidly new ONS methodology treats heat and cold as if their mathematical distributions of risk are an identical mirror reflection when they are most emphatically not. The heat threat is episodic and acute, but the cold threat is chronic, acting over a wider range of temperatures and over a much longer timescale. The fact is that there are many more moderately cold days than freezing ones in a year, and the cumulative toll of sustained, mild cold on blood pressure, respiratory infections, and strokes across an entire population is massive. To a certain extent, the ONS accounts for this by applying a distributed lag non-linear model (DLNM) that attributes to a coldest day some of the deaths that occur within the following 21 days. Even so, by treating cold-related mortality as something that happens relative to a coldest 2.5% of days, most of the seasonal mortality is overlooked. This looks like an obvious error, so why is the ONS doing it? Well, it has its reasons, but are they good ones?

Taking things to the extreme

The first thing to appreciate is that these estimates are generated by models and there will always be limitations on what can and cannot be easily modelled. If one tries to encompass the full range of factors associated with cold weather, a very messy picture emerges, involving many mediators and confounders. The ONS didn’t like this. In particular, they didn’t like the idea that, in the UK, winter deaths are heavily driven by viral waves (like influenza) and seasonal changes in behaviour. The ONS argued that extending the definition of cold days to include relatively mild days makes it statistically impossible to separate a death caused by a chilly room (presumably a death due to acute hyperthermia) from a death caused by riding on public transport during a flu outbreak. To the ONS, the cold-related statistics have always been massively over-estimated by including deaths that are not truly cold-related (at least not in the same sense that deaths related to extreme heat are). As far as the ONS is concerned, the revised methods are putting right a long-standing error.

Except, there wasn’t an error to correct. You don’t have to die from acute hyperthermia in order for your death to be associated with cold (remember that ‘associated with’ had been the term used by the ONS back in their 2022 paper). The ONS may think that they are doing the right thing by removing confounders but, by treating winter viruses as an unrelated confounding factor to be cancelled from the equation, the ONS methodology fundamentally misrepresents the biology of human exposure to cold. The ONS model tries to separate death-by-thermometer from death-by-virus, but in medicine these two things cannot be separated. Just because it is the virus that gets a mention on the death certificate doesn’t mean that the cold wasn’t an essential, underlying factor in the death.

Firstly, breathing in cold air chills the nasal passages. This directly impairs the body’s local immune response, slowing down the nasal tissue’s ability to fight off respiratory viruses. Secondly, many respiratory viruses are structurally more stable and can survive significantly longer in cold, dry air, increasing the ambient viral load in the community. Thirdly, when temperatures drop even moderately to 8°C or 10°C, people move indoors and close windows. This lack of ventilation drastically increases indoor transmission rates. Finally, living in a moderately cold home does not usually cause immediate hypothermia. Instead, it causes chronic stress on the cardiovascular system and lungs, leaving the resident highly vulnerable to the next respiratory bug going around.

None of this seems to interest the ONS. The political backdrop is such that climate change and its impact on extreme weather events has become the primary focus. Heat waves kill people due to acute heat stress, and so the only meaningful comparison in the eyes of the ONS would be a cold snap resulting in deaths caused by acute cold-stress. This is a nice, neat comparison, but one that entirely misses the point. By chasing a confounder-free indicator that can then be neatly compared to the heat-related scenario, the ONS has committed a classic statistical error: they have prioritized a notion of model integrity over real-world accuracy. They seem far more interested in seeking an extreme weather basis for mortality than they do an accurate indicator of public health. I don’t know exactly what the ONS thinks it is counting, but it is no longer the actual number of cold-related deaths suffered by the British public.

So, is the ONS still to be trusted?

At the very least, one has to say that the new ONS approach to modelling cold- and heat-related deaths is deeply flawed and highly misleading. If this level of misdirection were something that a climate sceptical organisation had engaged in, the leviathan that is the fact-checking industry would have come down on it like a ton of bricks. But this particular misdirection just happens to serve the orthodox narrative. Therefore, by definition, it can’t or won’t be branded as misinformation.

It was almost inevitable that, rather than pounce upon the startlingly dubious methodology undermining the ONS analysis, the Guardian instead would report upon the latest ONS figures by taking them entirely upon trust. For the Guardian, there was a ‘National Scandal’ to be reported but it wasn’t the entirely unacceptable approach taken by the ONS that was scandalous in their view, it was what the new (albeit entirely misleading) figures purported to say about ‘the rising toll of climate crisis’. And more to the point, the Guardian was more than happy to report the ONS’s own spin:

James Tucker at the ONS said: “Our results show that the gap between heat- and cold-related deaths has narrowed in recent years as the proportion of heat-related deaths has increased.”

Perhaps he should have said instead that, over the years, a degree of narrowing of the gap between heat- and cold-related deaths is to be expected, but the only way to explain this particularly dramatic narrowing is to look toward the ONS’s creative accounting and an almost scientifically fraudulent redefinition by the ONS of what it means to cause a particular category of death.

So, no I don’t think they can be trusted any more, and I think it may also be wise to revisit that NatCen poll, with its suggestion that the ONS is free from political manipulation.