NHS England estimates that in 2016 16% of all deaths in England were attributable to smoking. This is the Population Attributable Fraction (PAF) of mortality for smoking. A similar PAF estimate for elevated BMI, again in England, - and just published - is 5.5%. Smoking and elevated BMI are of course just two of several behavioural risk factors that are known to increase the risk of mortality. A total of 14 different behavioural and environmental risk factors were included in a 2011 estimate of the contribution of modifiable risk factors to cancer in the UK (which was reckoned to be 42% of all cancer cases - not mortality).
One of the major outputs from the Global Burden of Disease (GBD) project has been a series of estimates of cause-specific mortality (and DALYs) attributable to measured risk factors in 195 countries. Their 2016 Comparative Risk Assessment (CRA) included 84 risk factors - many of them linked to potentially fatal diseases - which were categorised as (i) behavioural, (ii) environmental or occupational, and (iii) metabolic. As the name suggests, the CRA allows us to compare the magnitude of the health burden that can be attributed to potentially modifiable risk factors. It works by producing PAFs that link cause-specific outcomes with measured risk factors such as prevalence of smoking, high BP and elevated BMI. So, to take one example, it provides estimates of the number of deaths by cause (e.g. deaths from ischaemic heart disease or breast cancer or diabetes) that can be attributed to obesity. Although the scale and ambition of the CRA (250 diagnoses for 84 risk factors in 195 countries) puts it in a different class from the PAF estimate for cancer in the UK, the underlying methods are basically the same. Just as the UK study aggregates PAFs for different risk factors and different cancers (hence the 42%)[*], the CRA aggregates its estimates of cause-specific deaths for various clusters of risk factors and causes of death. There is in fact an estimate of the number and proportion of all deaths that could be attributed to all risks combined. The estimate for 2016 is 59.9%, which is to say that approx. 60% of all deaths (globally) can be attributed to the 84 risk factors included in the model.
The CRA, however, has an important spin-off use that distinguishes it quite fundamentally from the UK cancer estimate. It is intended as an extended demonstration of the importance of variations in risk exposure as drivers of variations in mortality, which means that trends in risk exposure will play a major role in projecting future mortality trends. In other words, the model has the potential to form the basis for life expectancy projections, and to realise this what it requires is a way of incorporating (i.e. a set of equations) variations in mortality that are not explained by variations in risk exposure.
And this is precisely what the GBD published in the Lancet in November 2018, a new set of global life expectancy projections, and it is worth emphasising here that what is new (and interesting) about these projections is the methodology. The full projection model combines
- estimates of the population attributable mortality burden for all risk factors by age and sex as well as cause;
- projections of trends in exposure for measured risk factors for future years ;
- a social development factor which incorporates changes in education, GDP per capita, fertility;
- an autoregressive moving average process for the unexplained latent trends for each location-age-sex-cause.
The Lancet paper itself does not have much to say about country life expectancy projections and it has nothing at all to say about their differences from the widely-used UN projections. The interests of the GBD team lie elsewhere. Firstly, they want to show that the methods underpinning the CRA can be developed and supplemented in such as a way as to produce a full set of life-expectancy projections; it’s a ‘proof of concept’ paper. Secondly, they want to draw attention to the way that the model can be used to generate better and worse health scenarios. They construct a reference scenario which is based on the extrapolation of current trends (not trends in mortality but trends in factors that account for population-level variations in mortality) and two other scenarios that represent (i) strong and positive (ii) weak or negative trends in the various factors in the model that affect population health. They are intended as plausible best case and worst case scenarios.
20 top-performing countries in GBD projections + UK, Ireland and USA
Despite the focus of the paper, however, it is quite hard not to be curious about individual country projections (how the countries ‘perform’), and even harder not to compare these results (which are not reproduced in the paper or indeed in the online supplement) with UN projections. Which is why I compiled the table above. Perhaps the most striking differences between the two sets of projections are
- the rates of mortality improvement are lower (for the top-performing countries) in the GBD model, and
- the rates of mortality improvement vary much less in the UN projection model.
The average rate of improvement for the top 20 countries in the GBD projections is 1.15 mths/year; for the same countries in the UN projections it is 1.7 mths/yr. As for variation, the annual rate of improvement in Portugal (GBD projections) is more than twice that in Norway. In the UN projections, the difference is still there, but it is much smaller. All this results of course from the way the projection models are constructed. For these countries – all of them high-income and with high rates of education and low fertility rates - the big driver of future mortality trends (within the GBD model) is going to be changes in risk factor prevalence.
It should come as no surprise then that in several of the top-performing countries (GBD model), the 2040 projection for life expectancy at birth in the ‘worst case’ scenario is actually lower than their estimate of current (2016) life expectancy. What appears to be at work here is the powerful weighting given to trends in BMI. Failure to make any public health impact on the prevalence of obesity in countries with high levels of obesity/increasing prevalence of obesity points to a decline in life expectancy at birth. So, for example, the countries with the largest difference between current life expectancy at birth and the worst case scenario for 2040 include Canada and Iceland. And the out-and-out winner among high-income countries for potential negative growth in life expectancy is the USA. Life expectancy under the worst case scenario in 2040 is a full 2 years below current life expectancy.
[*] Aggregation in both cases is a complicated business. One single risk factor may be associated with several cancers and each cancer may be associated with many risk factors.
[1] UN projections are for 5 years periods rather than single years. To match with GBD projections for 20040, the UN figures given are the midpoint between the estimates for 2035-40 and 2040-2045.
About the Author:
Kenneth Howse is a Senior Research Fellow at the Oxford Institute of Population Ageing. He is also a key member of The Collen Programme on Fertility, Education and the Environment.
Opinions of the blogger is their own and not endorsed by the Institute
Comments Welcome: We welcome your comments on this or any of the Institute's blog posts. Please feel free to email comments to be posted on your behalf to administrator@ageing.ox.ac.uk or use the Disqus facility linked below.
