Principal findings
This study showed a steady increase in the percentage of deaths attributable to adiposity in England and Scotland over 15 years, and a steady decline in those attributable to smoking over the same period. As a result, adiposity has exceeded smoking as a contributor to deaths since 2014, and the difference has widened.
Whilst the overall trends were fairly consistent across different sub-groups of the population, there were some interesting differences in the timing and magnitude of change. Among older age-groups, the crossover between smoking and adiposity occurred earlier and, therefore, the difference in magnitude is now greater. The earlier crossover in this sub-group reflects their lower prevalence of smoking and higher prevalence of overweight or obese. The trend was also more pronounced in men than women, in spite of men having a higher annual prevalence of smoking than women, and a lower prevalence of obesity.8 9 This is largely due to obesity being associated with a higher mortality risk ratio in men than women [16] as well as more overweight men than women; therefore the same increase in prevalence will produce a greater impact on men.
Strengths and weaknesses of the study
In this study, the internal and external validity of the PAF estimates were enhanced by the use of general population representative data from England and Scotland, and up-to-date risk ratio estimates obtained from meta-analyses of prospective cohort studies. The response rate of both surveys has reduced over time. Any systematic error is likely to take the form of an increasing healthy volunteer effect. Therefore, the estimated PAFs may be underestimates. However, systematic differences in response on the basis of smoking versus obesity are unlikely. Because cohort studies are observational, causation cannot be assumed, and the estimates of risk may be affected by residual confounding. The RRs we extracted for smoking were adjusted for age and sex, and partially for blood pressure, physical activity, and alcohol consumption [15]; while those for adiposity were adjusted for age and sex among never-smokers without pre-existing chronic diseases [16]. Because the effects of lifestyle factors on mortality are mediated through chronic illnesses, the susceptibility to confounding bias between the two sets of RRs should be similar and therefore comparison between the PAFs should be valid. The risk ratio for smoking was based on a 2012 meta-analysis of smokers aged ≥60 years, which may impact its generalisability to our study population as RRs might be different for younger people. Nonetheless, our sensitivity analysis using UK Biobank data provided consistent conclusions. RRs were assumed to remain constant over time and applicable to the UK population even though they were based on studies from various countries. The data from the HSE and SHeS have shown that the median number of cigarettes per day among current smokers has declined from 14.3 and 15.0 in 2003 to 10.0 and 11.3 in 2017 in England and Scotland respectively. This indicates that if the RR of smoking did change during the study, it would likely to be slightly reduced, resulting in our study overestimating the numbers of deaths attributable to smoking in the recent years. The effect sizes of smoking between our selected meta-analysis and studies from UK and Europe were very similar [20], and those of adiposity were slightly stronger in European studies [16]. These indicates that our findings are likely to be a slight underestimate of the increasing contribution of obesity. Smoking deception can result in some current smokers misclassifying themselves as former smokers but this is less common in general population cohorts than smoking-related disease cohorts [21]. BMI is an imperfect measure of adiposity; especially in younger men with high lean body mass. These limitations in the measurement of exposures apply to all study years; therefore, temporal bias is unlikely. Our study was confined to all-cause mortality and the findings will not necessarily apply to other disease-specific outcomes where the relative contribution of the two risk factors may be different.
Strengths and weaknesses in relation to other studies
Previous studies often reported PAFs as a single point estimates because of the difficulty in combining variances in RRs and prevalence.7 10 In our study, we calculated not only PAFs but also the associated confidence intervals, which helps in the comparison of PAFs between sub-groups or across years. For example, the difference in PAFs between adiposity and smoking was only 3.7% in 2017 but the confidence intervals did not overlap giving us greater certainly that the contribution of adiposity was greater than that of smoking in that year. The confidence intervals were wider for smoking than adiposity due to the less precise estimates of RRs [15].
Comparing PAF estimates between studies is problematic because prevalence differs between populations and over time. Nonetheless, our estimates for smoking are not dissimilar to a previous study, which reported that 19% of all deaths in 2005 were attributable to smoking [7]. However, our study’s estimates of the PAF due to adiposity were higher than previous estimates of 2 to 12% [22]. The meta-analysis from which we extracted the RR for adiposity reported the mortality PAF for Europe to be 13.5% [16], based on a prevalence of 33% for overweight and 20% for obesity. In our study the 2017 prevalence of obesity in England and Scotland was much higher at 30.3%. Differences in methodology may also have contributed. For example, in that meta-analysis obesity was treated as a single category whereas, in practice, the RR increases exponentially across obesity classes I, II and III.
The Global Burden of Disease study ranked the top risk factors in the United Kingdom based on disability adjusted life years (DALYs) [23]. It reported that tobacco use was still the top contributor to DALYs in 2017 even though its contribution had fallen by 9.2% since 2007. Over the same period, adiposity rose from fourth to third rank, and its contribution increased by 7.6%. The temporal trends are consistent with our findings. The failure of adiposity to overtake smoking is likely to reflect a greater impact on disability from smoking than adiposity; possibly due to its association with chronic diseases, such as chronic obstructive pulmonary disease (COPD), that impair functioning and wellbeing for a sustained period of time.
Meaning of the study
Historical efforts to protect the public from the harms of smoking have been successful. Adiposity now contributes to more deaths than smoking, which highlights the need to prioritise strategies to address it, including upstream policies and legislation [24], as well as downstream individual interventions. Middle and older age groups and men, in particular, require support in helping them to reduce their weight to a healthy level.
Unanswered questions and future research
The current study included the risk of only active and former smoking. Given that vaping and e-cigarette use are growing among long-term former smokers [25], and passive smoking also contributes to mortality [26], future studies should consider whether these specific legacies of smoking require ongoing focus. Adiposity did not outrank smoking among those under 45 years of age and those finished school early when the surveys were undertaken. However, birth cohort studies should be undertaken in the future to fully understand changing risk over time. The association of adiposity with mortality may differ by ethnicity and this might need to be accounted for in future analysis. The combined associations of smoking and adiposity should be considered in future individual-level studies, as there may be interactions between them.