Contribution of Cold Versus Climate Change to Mortality in London, UK, 1976–2019

Abstract

Objectives. To quantify past reductions in cold-related mortality attributable to anthropogenic climate change.

Methods. We performed a daily time-series regression analysis employing distributed lag nonlinear models of 1 203 981 deaths in Greater London, United Kingdom, in winter months (November–March) during 1976 to 2019. We made attribution assessment by comparing differential cold-related mortality impacts associated with observed temperatures to those using counterfactual temperatures representing no climate change.

Results. Over the past decade, the average number of cold days (below 8 °C) per year was 120 in the observed series and 158 in the counterfactual series. Since 1976, we estimate 447 (95% confidence interval = 330, 559) annual cold-related all-cause deaths have been avoided because of milder temperatures associated with climate change. Annually, 241 cardiovascular and 73 respiratory disease deaths have been avoided.

Conclusions. Anthropogenic climate change made some contribution to reducing previous cold-related deaths in London; however, cold remains an important public health risk factor.

Public Health Implications. Better adaptation to both heat and cold should be promoted in public health measures to protect against climate change. In England, this has been addressed by the development of a new year-round Adverse Weather and Health Plan. (Am J Public Health. 2024;114(4):398–402. https://doi.org/10.2105/AJPH.2023.307552)

Although deaths associated with low ambient temperatures reduced over much of the past century, continued reductions in cold-related vulnerability have generally not been observed in more recent years, meaning that wintertime weather remains an important public health concern for the United Kingdom and elsewhere.¹ As well as improvements in housing, health care and nutrition, historical reductions in cold-related mortality were also likely because of milder winters caused by climate change; however, to our knowledge, this has never been quantified. The fraction of heat-related deaths attributable to climate change has recently been estimated by comparing differential impacts associated with observed temperatures to those using counterfactual temperatures modeled in the absence of anthropogenic climate change.² These attribution studies provide the basis for better climate change risk management³; however, a similar approach can also be applied to attribute reductions in cold-related deaths to past climate change. Such information can also inform the likelihood of future reductions in cold deaths under climate change scenarios, over which there remains much uncertainty and debate.⁴

The United Kingdom experiences greater cold-related health impacts compared with many of its colder European neighbors, partly from poor insulation of its housing stock,⁵ and London is more vulnerable than other UK regions despite milder temperatures.⁶ We analyzed extended time-series mortality data sets in relation to observed and counterfactual climate data to quantify past reductions in cold-related deaths in London attributable to anthropogenic climate change.

METHODS

We obtained daily counts of all-cause, cardiovascular, and respiratory disease deaths in London between 1976 and 2019 from the Office for National Statistics. To characterize exposures, we considered 2 daily mean temperature series. First, we extracted reference data of average temperature from the W5E5 data set representing a bias-corrected reanalysis data set that can be considered quasi-observational for temperature (containing observed trends in global temperatures).⁷ Second, we applied a counterfactual version of this data set estimated through the ATTRICI (ATTRIbuting Climate Impacts) approach.⁸ This data set approximates a temperature series preserving interannual variability of the quasi-observations but removing long-term global warming trends using a quantile mapping approach, thus representing a non–climate change scenario. Both data sets are available from the Inter-Sectoral Impact Model Intercomparison Project (https://www.isimip.org) on a 0.5° grid. We conducted additional bias correction by using deviations from the observed W5E5 temperature in the quasi-preindustrial period (1901–1920) as reference through quantile mapping to remove local biases remaining in the statistically derived counterfactual data set. (Codes used to bias-correct the counterfactual temperature series are available from the authors upon request.)

We used quasi-Poisson time-series regression to assess short-term associations between daily mean temperature and mortality, adjusting for trend and within-season variability using natural cubic spline (ncs) functions with 4 degrees of freedom (df) per season. We used indicator terms to model day-of-week variations. We then employed distributed lag nonlinear models to flexibly model nonlinear and delayed effects of temperature using cross-basis functions.⁹ The model is summarized here:

$$\text{Log}\left[\mathrm{E}\left({\mathrm{Y}}_{\mathrm{i}}\right)\right]=\mathrm{\alpha }+{\mathrm{\beta }}_1{\mathrm{T}}_{\mathrm{i},\mathrm{j}}+{\mathrm{\beta }}_2\text{ncs}\left({\text{time}}_{\mathrm{i}},\text{df}=4/\text{season}\right)+{\mathrm{\beta }}_3\left({\text{dow}}_{\mathrm{i}}\right)$$ (1)

where E[Y_i] is expected mortality on day i; T_i,j is the cross-basis matrix of temperature and lag j up to 21 days, using ncs functions for both domains with 5df and 4df, respectively; ncs = ncs functions of time; and dow = day-of-week indicator. We conducted analysis using the observed temperature series, and we then also applied the estimated exposure–response function from this model to the counterfactual temperature series representing no climate change. We then estimated the relative risk (RR) of cold-related death at selected temperatures compared with the minimum mortality temperature (MMT). The MMT is defined as the temperature value at which risk of death is lowest and is determined using statistical model fit (Akaike information criterion [AIC]). We used the most recent decade of data (2010–2019) to derive the risk function because cold risk was relatively stable during this period, thus enabling a direct comparison of differential impacts between the observed and counterfactual temperature series without other secular changes. We derived the percentage of cold-attributable deaths below the MMT by using (RR-1)/RR. Cold effects were restricted to months November through March when evidence for a causal association with mortality is strongest.¹⁰ We conducted analyses with Stata version 17 (StataCorp LP, College Station, TX) and R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Figure 1a shows the annual number of cold days (below 8 °C) in the observed and counterfactual temperature series during the study period. Unsurprisingly, the observed series is associated with milder winters compared with the counterfactual, although a slight reduction in cold days is also evident in the counterfactual series, potentially attributable to the urban heat-island. The negative trend in cold days was statistically significant in the observed series (Mann–Kendall test; P < .05) but not in the counterfactual. During 2010 to 2019, the average number of cold days per year was 120 in the observed series and 158 in the counterfactual.

FIGURE 1—

Annual (a) Average Wintertime Temperature and (b) Cold-Related Deaths Associated With the Observed (obs.) and Counterfactual (cf.) Temperature Series: Greater London, UK, November‒March 1976‒2019

There were 1 203 981 deaths in London during winter months (November–March). Figure A (available as a supplement to the online version of this article at https://ajph.org) shows the seasonally adjusted temperature– mortality relationship in winter during 2010 to 2019. The MMT was estimated at 8 °C, and there was a 27.9% (95% CI = 18.8, 37.7) increased risk of death at −1 °C (approximately the 1st percentile) compared with the MMT. From this model, we estimated the total number of cold-attributable deaths each winter associated with both observed and counterfactual series (Figure 1b). Cold-related deaths have reduced over time, reflecting improved adaptation to wintertime climate, but there was wide year-to-year fluctuation and some degree of leveling-off in recent years. Differences in mortality between the 2 temperature series have also widened over time, with the average annual number of cold deaths avoided in the counterfactual series being 447 (95% CI = 330, 559). The annual number of deaths avoided from cardiovascular and respiratory diseases was 241 (95% CI = 190, 289) and 73 (95% CI = 52, 107), respectively.

DISCUSSION

The damaging effects of anthropogenic climate change on public health are unquestionable, and studies have demonstrated substantial numbers of heat-related deaths already attributable to such changes.² Although health impact assessments have previously reported reductions in future cold-related mortality under climate change scenarios,¹¹ to our knowledge, this is the first study to quantify historic reductions in cold-related mortality attributable to the degree of anthropogenic climate change that has already occurred.

In the most recent winter in our data set (winter 2018–2019), we estimate 639 fewer cold-related deaths in London associated with counterfactual temperature, amounting to 70.1 wintertime deaths per million people in the population avoided because of climate change. Any continuing gains are, however, likely to be impacted by current trends in other determinants of wintertime health (e.g., the current volatility in global fuel prices). Better housing and other adaptation strategies will almost certainly remain more important in counteracting the negative health impacts of these. For example, Taylor et al. estimated that 168 to 174 annual cold-related deaths per million population in London could be avoided by the 2050s based on current rates of retrofit home energy efficiency measures, and 261 to 269 deaths per million under more ambitious retrofit rates.¹² Improved home energy efficiency also offers other health benefits, including reducing dampness and mold that contribute to asthma, as well decreasing residential space heating demands to help reach Net Zero targets.¹³

Limitations

Some study limitations are acknowledged. Our assessment only considered 1 pathway by which milder winters affect public health; there are also negative impacts such as enhanced wintertime transmission of zoonotic pathogens.¹⁴ The degree of seasonal control in our statistical models was decided a priori rather than based on AIC; however, model fit was good, and results were robust to greater dfs in ncs and alternative model specifications. Comparison of the observed series with the statistically constructed counterfactual one allows attribution of differential health impacts to climate change in general, but does not provide direct attribution to increased anthropogenic greenhouse gas emissions or changes in aerosol concentrations. Furthermore, the relatively coarse resolution of data, the extraction procedure, and bias-correction of the counterfactual series impose additional sources of uncertainty in the temperature data not characterized in this assessment. Nevertheless, our findings are likely robust to such data considerations.

Public Health Implications

Although climate change has played a role in reducing wintertime deaths in London, its contribution has been modest, and cold remains an important public health risk factor in the United Kingdom. Better adaptation to both heat and cold should be prioritized in public health plans to protect against climate change. In England, this has been addressed by the development of a new year-round Adverse Weather and Health Plan by the UK Health Security Agency.¹⁵ Wherever cold remains an important determinant of ill health, better longer-term strategies are needed to protect public health and improve resilience to future wintertime weather.