Climate change and winter deaths – experts respond

UK SMC: A paper published today in the journal Nature Climate Change looking at mortality during the winter in England and Wales, suggest climate change would likely not decrease winter mortality in those areas. It suggests that volatile winters, with swings from cold to mild weather linked to climate change, may even increase death rates.

Screen Shot 2014-02-24 at 11.53.55 AMThe UK SMC provides commentary on the paper below including a “Before the Headlines” analysis of the paper by independent experts.

Explanatory comment from the lead author of the paper: Dr Philip Staddon, Research Fellow in Climate Change, Sustainability and Human Health, University of Exeter Medical School, comments:

“To clarify the thrust of this paper, we are interested in explaining the year to year variation in excess winter deaths – not the daily variation – and we are not saying that temperature does not play a role, if it didn’t there would be no excess winter deaths. What we aim to demonstrate is that how harsh a winter is no longer predicts how many excess winter deaths there will be.

“An extensive literature attests to the fact that changes in daily temperature influence health outcomes at the local levels, and that EWDs are influenced by temperature. However, our data suggest that year to year variation in EWDs is no longer explained by the year to year variation in winter temperature: winter temperatures now contribute little to the yearly variation in excess winter mortality so that milder winters resulting from climate change are unlikely to offer a winter health dividend.

“And the reasons behind the apparent contradiction (strong daily temp link, but not very strong whole winter temp link): First, let us reiterate the association between cold temperatures and increased deaths is not in question; that is a fact and explains the existence of the EWD phenomenon. Clearly more research must be carried to identify why this apparent contradiction is occurring. However, we hypothesise that this can occur through the following mechanisms: (1) help to the most vulnerable people is becoming more targeted (this is certainly true in many parts of the UK, e.g. Cornwall Council’s Winter Wellbeing activities); (2) it’s not low temperature per se but rather the drop in temperature (after the first day of cold, many people are prepared for the following days – this explains why cold spells in early winter often lead to greater mortality than similarly cold but later spells); (3) increased temperature volatility, making cold spells more unpredictable, catches people off guard (this is part of the explanation as to why European with warmer winters such as the UK, Portugal, Italy suffer greater EWDs than Northern Europe). These ideas are briefly discussed in the latter part of the paper.”

BEFORE THE HEADLINES

COMMENTARY

Title, Date of Publication & Journal
‘Climate warming will not decrease winter mortality’ by Philip L. Staddon et al published inNature Climate Change on Sunday 23 February.

Claim supported by evidence?
The paper does not provide robust support for the claim that climate warming will not decrease winter mortality, because the statistical methods used mean that there could still be an effect of cold temperature on number of deaths that is masked by an effect of influenza. There are other limitations (see below).

Summary

– The study investigated the relationship between cold temperatures and excess winter deaths (EWDs).

– The authors conclude that cold temperatures were previously associated with EWDs (until the 1970s), but no longer are.

– The authors conclude that from the 1970s onwards, influenza has been the main factor associated with EWDs.

– Flaws in the statistical analysis mean that their conclusions may not be reliable because:

o   Influenza and cold temperatures may not be independent of one another (if cold temperatures leave people more vulnerable to infections then cold may still affect EWDs via its effect on influenza).

o   The statistical models the authors used contained both number of cold days (<5?C) and number of days with a big temperature drop – these two measures are likely to be highly correlated with each other (we call this over-adjustment of the regression analysis).  The result is that there may actually be effects on EWDs of either cold days or days with a big temperature drop, but by including them both in the model their effects could be masked by one another.  Hence, there may still be an effect of cold temperature on EWDs beyond the 1970s but we are not seeing the effect because of the model used.

o   The group of people looked at in this study is the over 65s.  Comparing those aged over 65 in the 1950s with those aged over 65 now may not be without problems because today there are many more people in this age group than there were in the 1950s, and because health improvements mean that many people aged 65 are healthier now than they were in the 1950s.

Study Conclusions

The authors report that there was previously a clear relationship between the number of cold winter days (defined as days with a temperature < 5°C) and EWDs, but that that relationship no longer holds.  The authors suggest that this is because improvements in healthcare and housing over recent decades have made people less vulnerable to the effects of cold.This explanation seems plausible but the statistical analyses are not robust enough to conclude this for definite (see limitations section).The conclusions that CAN be drawn are:

– There has been a substantial decrease in RATE of EWDs over the last six decades, but because authors do not report the absolute numbers of EWDs (rather they report the number as a proportion of the population of over 65s – the rate) we can’t tell whether there are now fewer EWDs or whether the population of over 65s has increased in size so there are PROPORTIONALLY fewer EWDs.

-Cold temperatures certainly were associated with EWDs until the 1970s.

– Since the 1970s influenza incidence has been strongly associated with EWDs.

Strengths/Limitations

Strengths:

– The study used data from a period of over 60 years – lots of data.

– The authors used a reliable source for the measurements of temperature (data from the UK’s Meteorological Office Hadley Centre).

– The authors used a reliable source of death data (data from the Office for National Statistics).

Limitations:

The conclusion that there is no relationship between cold temperatures and EWDs in recent years may not be sound, because:

– The model contains both cold days and influenza incidence.  These two factors may not be independent – If increased influenza activity is affected by cold weather, then this adjustment is inappropriate, as cold temperature could affect EWDs via its effect on influenza activity. The paper does not present data on the correlation between cold temperatures and flu activity, so it is not possible to know how serious a problem this is from the paper.

– The model contained both number of days <5?C and number of days with a big temperature drop – these two measures are likely to be highly correlated so although neither appear to be associated with EWDs this may be because each one is masking the other’s effect.  A model containing only one or the other measure of cold days may reveal a significant association with EWDs, but such a model has not been presented in this paper.

– The authors do not report the absolute numbers of EWDs – rather they report the number of EWDs as a proportion of the total population of over 65s (the rate of EWDs).  While results do suggest a substantial decrease in the rate of EWDs over the last six decades, it is possible that there is little change in the absolute number of EWDs if the population of over 65s has grown substantially.  Therefore, we don’t really know if the reduction in rate of EWDs is due to a decrease in absolute numbers of EWDs or due to an increase in the size of the population of over 65s.

– The dataset used a single geographic measure of temperature – that for central England. It is possible that this may have masked important regional temperature variations.  The measure of numbers of deaths was also a general population measure – there may be regional variations in this too.

– The authors chose to measure cold days as those <5?C.  Other definitions of cold temperature could have been looked at too, such as mean temperature over the whole winter, minimum winter temperature, etc.

– It is not clear whether the time periods they looked at separately were decided in advance or once the data had been looked at – if it is the latter then results are less robust.

– EWDs are measured in the over 65s.  this was designed to account for demographic changes but may in fact introduce bias – the population of over 65s is not only much larger than in the 1950s but also healthier so may not all be at great risk of EWDs.

Glossary

EWDs = excess winter deaths

Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry(PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.

Experts respond

The UK Science Media Centre gathered the following commentary on the research from experts.

Commenting on the statistics:

Prof Kevin McConway, Professor of Applied Statistics, The Open University, comments:

“This paper makes some very interesting points, but the research that it reports does not really support the title of the paper. If it had been called “Climate warming may not reduce winter mortality”, I’d have been happier.

“Statisticians like quoting a few mantras, and this study breaks two of them.

“First, “Correlation in not causation”. The research is entirely based on correlations – situations where one quantity (say, excess winter deaths) goes up and down roughly in step with another quantity (say, flu activity). But you can’t conclude from such data that the changes in one quantity are causing the changes in the other one. There are always other possible explanations – maybe some other quantity, that wasn’t included in the study, is independently causing changes in flu activity and excess winter deaths. A study of correlation can indicate interesting things to investigate in researching what causes what, but it can’t show what is causing what.

“The second mantra, which isn’t confined to statisticians, is “You can’t prove a negative”. The paper’s title sounds as if the researchers have proved a negative. But all they have really shown is that it’s possible that climate warning won’t decrease winter mortality. In the paper itself, for instance, the researchers point out that climate change may lead to winters where the temperatures vary much more than they did in the past. They speculate that this increased variability might increase winter deaths – but, although that’s very plausible, they do not really have data to support it. Maybe this change will work alongside others so that winter deaths increase – or maybe they will decrease, or remain unchanged. We still don’t really know which.

“It’s good that these researchers have pointed out that winter deaths are related to more than just cold temperatures. But their work has made it clear the prediction of future winter deaths is difficult, because of all the factors potentially involved, rather than establishing for certain what the effect of climate change will be.”

Prof Patrick Wolfe, Professor of Statistics, UCL, comments:

“It’s important to remember that no amount of data analysis can prove a negative. The authors of this study are attempting to isolate the effects of the number of cold winter days on additional deaths during wintertime – and they rightly point out that a number of potential factors are at play, including improvements in housing and healthcare from the 1950s onward.

“But I think it’s very difficult to extract the signal from the noise in this situation, given the data the authors are working with relative to the problem they are tackling: a long-term decline in the incidence of excess winter deaths relative to the UK population aged 65 and above, and a number of potential factors that have shifted and varied over these same long-term time scales.

“And of course, just because the authors found no evidence in the data they considered does not mean for certain that no evidence exists. In light of these facts, I’d interpret this study as evidence that the association between cold winter days and excess winter deaths is worth a closer look. A finer granularity of data for recent years (concomitant with the declining incidence of excess winter deaths) might enable to enable a more comprehensive analysis, and help to tease out any potentially spurious associations.”

Commenting on the influenza aspect:

Dr Michael Skinner, Senior Lecturer in Virology, Imperial College London, comments:

“As a virologist I cannot argue with the conclusion in the paper that “particular attention should also be paid to public health initiatives to reduce the risk of infection with flu-like illnesses”, not because of the findings of this study about winter temperatures but because of what we have long known about influenza virus variability – in fact most of the influenza virus peaks shown here correspond to pandemic shifts or major seasonal influenza drifts in antigenicity. We would expect the impact of such outbreaks on the incidence of EWDs to be far higher than any of the complicated temperature-dependent changes on normal, seasonal influenza virus transmission or susceptibility.

“The authors say “Improving [inactivated influenza vaccine] uptake in the over 65s would be very beneficial”; although I cannot disagree with the sentiment, as a virologist I cannot see that this paper makes a significant contribution to the debate about how best to protect the vulnerable from seasonal influenza, and the paper they cite on this point is not intended to be an authoritative investigation of that particular issue.”

Dr Ben Neuman, Virologist, University of Reading, comments:

“Flu viruses are unwelcome and sometimes deadly winter guests. These findings show that while better housing, cheaper heating and government policy may have reduced winter deaths, flu-related deaths are the next challenge we need to meet. Modern flu vaccines are cheaper and more effective than ever. This study reinforces the importance of vaccination programmes, especially in elderly people who are most at risk of severe disease.”