Ill health and winter death rates – are we even measuring it right?

I have just written an article on excess winter deaths and their link to housing quality in the UK. It’s published here and here. It was a fairly short article discussing a complicated issue, and there was a thorny problem that didn’t make the final edit… Namely, whether the usual calculation method is any good or not. The following discussion is largely based on a journal paper by Lidell et al (2016), so if you prefer journal articles to blogs, here’s the link.

The number of excess winter deaths (in northern hemisphere countries) is taken as the number of deaths in the December to March period, compared to those in the four preceding months (august to November) and the following four months (April to July). The equation for calculating the excess winter death rate is:

EWD

There are some fairly obvious problems with this rather arbitrary calculation

  • If the winter period is longer or shorter than December to March, the calculation fails. For example, if someone dies on November 30th or April 1st, it is not a winter death; in fact, since the excess winter death rate is a ratio, not only would these deaths not count towards winter deaths, they would actually contribute to making the winter death rate lower. The arbitrary definition of winter is also problematic if someone becomes seriously ill due to winter conditions, but doesn’t actually die until a month or so later.
  • If the country suffered from heat waves in the summer, which are also known to increase deaths, this would also change the ratio of winter:summer deaths, without telling us anything very useful about whether people are dying unnecessarily in winter.

So whilst there may be countries in which this calculation is a good way of considering winter deaths because the above factors are not important, there will be other countries where the EWDi is not a particularly reasonable calculation. What alternative methods exist for calculating excess winter deaths? What Lidell et al (2015) suggest is to use heating degree days (HDD). This is an overall measure of how much heating is required. It takes the number of days when the average temperature outside was below a pre-defined comfortable indoor temperature (i.e. days when a building might need heating), and the number of degrees below that pre-defined indoor temperature. HDD therefore give an idea of both the number of days of cold weather, and also how cold it is. Examples of HDD for some European countries are given in the table below, along with the percentage of these days that occur during ‘winter’ as defined by the conventional calculation method.

HDD
Heating degree days (HDD) by country and the percentage that occur in ‘winter’ or in the 6 colder months of the year. From Lidell (2016).

The authors go on to discuss how the EWDi calculation may drastically under-estimate the true excess of winter deaths in many countries. The use of HDD is somewhat different to EWDi; it’s a measure of cold weather per se, and there might be other factors occurring in winter that we need to consider (e.g. humidity levels are much higher in winter than summer, and air pollution levels are also much higher). Intriguingly, for the UK there is a very poor correlation between the EWDi calculated according to the conventional method and the equivalent index derived from HDD.

Europe also operates a sophisticated mortality monitoring programme that records weekly deaths according to age and gender across regions (EuroMOMO). These deaths are fitted to a model that allows one to determine at a glance if mortality is higher or lower than expected. Its remit is much wider than simply measuring excess winter deaths; it was set up to allow accurate and rapid mapping and monitoring of pandemics, such as flu, AIDS and SARS using a common method across Europe, although it has so far only been adopted by 17 countries. Data is available weekly, for each individual age group and gender in each country. It remains to be seen what public health measures will result from this data collection initiative; it will certainly allow a much more sophisticated and immediate response to any future pandemic, but whether or not it will be applied to analysing winter death rates remains to be seen.

EuroMOMOweek51
Example data from EuroMOMO. The last week of 2017. Very high mortality in Scotland and Spain, high in England and Portugal, above expected in Ireland, Northern Ireland and France. 
EuroMOMOannualsinecurve
Example data from EuroMOMO. Modelled baseline mortality with the actual numbers of deaths (green line) indicating fluctuations above and below the number of deaths predicted by the model.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: