IEA International Journal of Epidemiology, 2021, 931–941 doi: 10.1093/ije/dyaa273 Advance Access Publication Date: 11 January 2021 International Epidemiological Association Original article Smoking The combined impact of smoking, obesity and alcohol on life-expectancy trends in Europe Fanny Janssen ,1,2* Sergi Trias-Llimós3,4 and Anton E Kunst5 1 Netherlands Interdisciplinary Demographic Institute—KNAW/University of Groningen, The Hague, The Netherlands, 2Population Research Centre, Faculty of Spatial Sciences, University of Groningen, The Netherlands, 3Department of Non-communicable Disease Epidemiology, Faculty of Epidemiology and Population Health, London School of Hygiene & Tropical Medicine, London, UK, 4Center for Demographic Studies, Centres de Recerca de Catalunya (CERCA), Bellaterra, Spain and 5Department of Public and Occupational Health, Amsterdam UMC, University of Amsterdam, Amsterdam, The Netherlands *Corresponding author. Lange Houtstraat 19, 2511 CV The Hague, The Netherlands. E-mail: janssen@nidi.nl, f.janssen@rug.nl Editorial decision 12 November 2020; accepted 10 December 2020 Abstract Background: Smoking, obesity and alcohol abuse greatly affect mortality and exhibit a dis- tinct time dynamic, with their prevalence and associated mortality rates increasing and (eventually) declining over time. Their combined impact on secular trends in life expectancy is unknown but is relevant for understanding these trends. We therefore estimate the com- bined impact of smoking, obesity and alcohol on life-expectancy trends in Europe. Methods: We used estimated national age-specific smoking-, obesity- and alcohol- attributable mortality fractions for 30 European countries by sex, 1990–2014, which we aggregated multiplicatively to obtain lifestyle-attributable mortality. We estimated poten- tial gains in life expectancy by eliminating lifestyle-attributable mortality and compared past trends in life expectancy at birth (e0) with and without lifestyle-attributable mortality. We examined all countries combined, by region and individually. Results: Among men, the combined impact of smoking, obesity and alcohol on e0 declined from 6.6 years in 1990 to 5.8 years in 2014, mainly due to declining smoking- attributable mortality. Among women, the combined impact increased from 1.9 to 2.3 years due to mortality increases in all three lifestyle-related factors. The observed increase in e0 over the 1990–2014 period was 5.0 years for men and 4.0 years for women. After excluding lifestyle-attributable mortality, this increase would have been 4.2– 4.3 years for both men and women. Conclusion: Without the combined impact of smoking, obesity and alcohol, the increase over time in life expectancy at birth would have been smaller among men but larger among women, resulting in a stable increase in e0, parallel for men and women. Key words: Health behaviour, lifestyle, Europe, life expectancy, mortality, time trends C The Author(s) 2021. Published by Oxford University Press on behalf of the International Epidemiological Association. V 931 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 932 International Journal of Epidemiology, 2021, Vol. 50, No. 3 Key Messages • We observed a large combined impact of smoking, obesity, and alcohol on trends in life expectancy at birth (e0) in Europe. Without its impact, the increase over time in e0 would have been smaller among men but larger among women. • We identified a stable increase in life expectancy for non-lifestyle-attributable mortality, which occurred in parallel for men and women in most European countries, and was, for men, more similar between countries than for the observed life-expectancy trends. • Distinguishing between non-lifestyle-attributable mortality and lifestyle-attributable mortality is important to better understand current trends in life expectancy and to predict future trends. Introduction Also, various Global Burden of Disease studies esti- Smoking, alcohol abuse and behavioural factors resulting mated trends in risk-attributable deaths and Disability in overweight and obesity (unhealthy diets and insufficient Adjusted Life Years for different (groups of) risks (behav- physical activity) are important contributors to non- ioural, environmental, occupational, metabolic) from 1990 communicable-disease mortality.1 Smoking, alcohol and onwards.31–34 We will complement this work by estimat- obesity are considered to be three key preventable risk factors ing the effects of lifestyle-attributable mortality on secular and public-health problems in Europe,2,3 with a high impact trends in life expectancy. on mortality and life expectancy.4–14 However, the contribu- Previous studies on the impact of smoking-, obesity- tions of smoking, obesity and alcohol are not constant over and alcohol-attributable mortality on trends in life expec- time, because these lifestyle factors tend to evolve as wave- tancy mostly examined the impact for a single lifestyle fac- shaped lifestyle ‘epidemics’ characterized by an initial unprec- tor and usually for only a selection of edented increase followed (eventually) by a decline.15–20 countries.22,23,27,30,35–37 They found that smoking played Given important (recent) stagnations in increases in life ex- an important role in the stagnation of life-expectancy pectancy in Europe over time,21–25 it is important to assess increases among men in many North-western European how these lifestyle factors influenced secular trends in life ex- countries in the 1950s and 1960s, and in other European pectancy in different European countries. countries and among women in more recent deca- Previous research clearly revealed the time-varying nature des.22,23,35 For obesity, an increasingly large effect on of smoking-, obesity- and alcohol-attributable mortality. Past life-expectancy levels over recent decades was ob- trends in smoking-attributable mortality show a clear wave served.27 A similar trend was found for alcohol in pattern of increases followed by declines that occurred ap- Finland.36 In addition, the dissimilar trends in alcohol proximately 30–40 years later than a similar wave pattern for prevalence between Eastern and non-Eastern Europe smoking prevalence.15 This wave pattern has been observed have greatly contributed to diverging (1990–2005) and among men in all European countries, whereas smoking- converging (2005 onwards) life-expectancy levels across attributable mortality appears to still be increasing among Europe.30 Although these previous results suggest that women in the majority of European countries.16 The tripling the combined effect of smoking, obesity and alcohol on of obesity prevalence in Europe since 198026 has resulted in life-expectancy trends could be large, this combined ef- clear increases in obesity-attributable mortality fractions over fect has not yet been studied. the 1975–2016 period,27 albeit with a recent slowing-down Our objective is, therefore, to estimate, for the first of the rates of increase.28 Alcohol prevalence and, time, the combined impact of smoking, obesity subsequently, alcohol-attributable mortality levels, which are (BMI  30 kg/m2) and alcohol abuse on trends in life ex- especially high among Eastern-European men,29,30 had been pectancy for men and women in Europe and to determine mostly increasing in Eastern and North-western Europe, al- how trends in life expectancy would look like without this though recently they have been stagnating or declining in combined impact. We study the majority of European most of these countries. In most South-western-European countries (N ¼ 30) combined, with additional attention on countries, these levels have been steadily declining, albeit differences by region (North, West, South, Central, East) more slowly in recent years.19,20 and country. International Journal of Epidemiology, 2021, Vol. 50, No. 3 933 Data and methods and sex-specific smoking-, obesity-, alcohol- and lifestyle- For the 30 countries under study, we used previously calcu- attributable mortality rates. By multiplying the all-cause lated estimates of smoking-, obesity- and alcohol- mortality rates by one minus the fractions, we obtained the attributable mortality fractions by sex and single adult age- and sex-specific non-smoking-, non-obesity-, non-al- ages (smoking: 35–100 years; obesity and alcohol: 20– cohol- and non-lifestyle-attributable mortality rates. 100 years) (see next paragraph).16,20,38 In addition, we To aid the interpretation of the results from our main used age-, sex-, country- and year-specific all-cause mortal- analysis (see below), we performed two initial analyses. ity and population data from the Human Mortality First, we examined the trends over time in the share of Database.39 We studied the period 1990–2014 because mortality that is attributable to the different lifestyle fac- alcohol-attributable mortality could be estimated only since tors. We obtained shares across adult ages (20–100 years) 1990 and smoking-attributable mortality could not be esti- by applying direct standardization to the age-specific frac- mated after 2014 for several countries due to the unavailabil- tions using the age composition of deaths for the specific ity of lung-cancer-mortality rates for more recent years (see populations in 2010. Second, we examined the impact on Supplementary Data & Methods—Appendix 1, available as life expectancy at birth (e0) of the separate and combined Supplementary data at IJE online, for the data availability lifestyle factors, and the trends therein over time, by calcu- by country for the different elements). lating the potential gain in life expectancy (PGLE)47 if Smoking-attributable-mortality fractions were indi- smoking-, obesity-, alcohol- and lifestyle-attributable mor- rectly estimated using a simplified version of the com- tality were eliminated. The PGLE values are calculated by monly applied Peto-Lopez methodology.16,40,41 The comparing the e0 value for all-cause mortality to the e0 method estimates exposure to smoking based on lung- value based on life-table calculations applied to non- cancer mortality rates that are adjusted for the part not smoking-, non-obesity-, non-alcohol- and non-lifestyle- due to smoking and applies to this exposure the relative attributable mortality rates, respectively. risks (RRs) of dying from smoking. For obesity- For our main aim of estimating the (combined) impact attributable mortality, the estimates stem from the appli- of smoking, obesity and alcohol on life-expectancy trends cation of the population-attributable fraction formula to in Europe, we graphically compared the trends over time (estimated) prevalence data42 using all-cause RRs of dy- in observed e0 (¼ for all-cause mortality) with the trends in ing from obesity,43 in line with previous studies, e.g.27. estimated e0 values (¼ for non-smoking-, non-obesity-, Alcohol-attributable mortality fractions were estimated non-alcohol- and non-lifestyle-attributable mortality) (see based on alcohol-attributable mortality rates obtained before). To compare the overall change in observed e0 from the Global Burden of Disease 2017 study,31,44 ad- over the 1990–2014 period with this change in estimated justed for ages 65þ using the age pattern for alcohol- e0 values, we subtracted the respective e0 values in 1990 related causes of death.20 from the respective e0 values in 2014. To estimate the share of mortality due to smoking, obe- See the Supplementary Data & Methods document, sity and alcohol combined, we used the multiplicative ag- available as Supplementary data at IJE online, for more gregation of the fractions for the individual risk factors, details. using the formula45: Y n Results LAMF ¼ PAF1::n ¼ 1  ð1  PAFi Þ Before describing the results of our main analysis in i¼1 Table 2 and Figure 3, we first describe the results of our where i stands for the individual risk factor and background analyses in Figure 1, Table 1 and Figure 2. PAF1:n stands for the lifestyle-attributable mortality frac- Whereas the tables and figures in the manuscript show the tion for the three lifestyle factors combined, which we refer results by European region, the respective supplementary to as the lifestyle-attributable mortality fraction (LAMF). tables and figures, available as Supplementary data at IJE This aggregation rests on the strong assumption that online, show the country-specific results. risk factors are independent and uncorrelated.45 However, Across all 30 European countries in 2014, the age- validation studies have demonstrated that the resulting standardized share of mortality due to smoking, obesity estimates may closely resemble the true combined and alcohol combined (¼ lifestyle-attributable mortality effect.31,46 fraction) was 38% among men and 19% among women. By multiplying the age- and sex-specific smoking-, obe- Among men, the LAMF mostly declined over the 1990– sity-, alcohol- and lifestyle-attributable mortality fractions 2014 period, albeit only recently in many Northern to the respective all-cause mortality rates, we obtained age- European and Central and Eastern European (CEE) 934 International Journal of Epidemiology, 2021, Vol. 50, No. 3 Figure 1 Trends over time in age-standardized smoking-, obesity- and alcohol-attributable mortality fractions (separately and combined) (%), ages 20– 100 years, 1990–2014*, for the different European regions, by sex. Lifestyle refers to smoking, obesity and alcohol combined. *For 2011 up to 2014, the weighted averages were calculated using the data for the latest available year: Bulgaria (2010), Greece (2013), Ukraine (2012) and Russia (2013). countries (Figure 1 and Supplementary Figure 1, available (decelerating) increases in alcohol-attributable mortality. as Supplementary data at IJE online). Among women, the The increases in the LAMF for women were driven by LAMF generally increased over time. The declines in the increases in smoking-attributable mortality, in obesity- LAMF for men were mostly driven by declines in smoking- attributable mortality and—particularly in CEE—in attributable mortality, which started later in CEE. These alcohol-attributable mortality. declines in smoking-attributable mortality were, however, The potential gain in life expectancy (PGLE) by elimi- counterbalanced by increases in obesity-attributable mor- nating lifestyle-attributable mortality declined from 6.6 to tality and, in Northern and CEE countries, as well by 5.8 years among men and increased from 1.9 to 2.3 years International Journal of Epidemiology, 2021, Vol. 50, No. 3 935 Table 1 Impact of smoking, obesity and alcohol (separately and combined) on life expectancy at birth (e0) in 30 European coun- tries, by sex and region, 1990 and 2014a by means of estimates of the potential gain in life expectancy (PGLE) (in years) from the elimination of mortality linked to the respective lifestyle factor. Lifestyle refers to smoking, obesity and alcohol combined PGLE 1990 (in years) PGLE 2014a (in years) e0 1990 Lifestyle Smoking Obesity Alcohol e0 2014a Lifestyle Smoking Obesity Alcohol Men Europe 69.28 6.58 4.85 0.78 1.52 74.26 5.84 3.36 1.34 1.83 Northern Europe 73.11 4.06 2.83 0.66 0.84 79.42 3.04 1.40 1.03 0.83 Western Europe 72.51 5.51 3.98 0.72 1.27 79.02 3.94 2.19 1.18 0.96 Southern Europe 73.40 5.28 3.61 0.69 1.48 79.99 3.78 2.23 1.06 0.83 Central Europe 66.34 7.72 5.89 0.94 1.59 73.86 6.27 3.88 1.47 1.61 Eastern Europe 64.52 7.74 5.87 0.77 1.72 65.72 8.10 4.56 1.43 2.95 Women Europe 77.28 1.94 0.62 0.99 0.38 81.23 2.30 0.77 1.18 0.45 Northern Europe 79.38 1.86 0.97 0.64 0.32 83.72 2.41 1.40 0.81 0.31 Western Europe 79.16 1.95 0.84 0.74 0.43 83.82 2.38 1.24 0.95 0.32 Southern Europe 80.08 1.47 0.27 0.79 0.45 85.10 1.59 0.52 0.92 0.20 Central Europe 75.07 2.10 0.86 1.03 0.28 81.08 2.81 1.53 1.14 0.31 Eastern Europe 74.57 2.04 0.47 1.27 0.35 76.44 2.31 0.23 1.44 0.72 a For 2014, the weighted averages were calculated using the data for the latest available year for those countries for which 2014 data were missing: Bulgaria (2010), Greece (2013), Ukraine (2012) and Russia (2013). Northern Europe: Denmark, Finland, Iceland, Norway, Sweden. Western Europe: Austria, Belgium, Germany, France, Ireland, Luxembourg, Netherlands, Switzerland, the UK. Southern Europe: Greece, Italy, Portugal, Spain. Central Europe: Bulgaria, Czech Republic, Hungary, Poland, Slovakia, Slovenia. Eastern Europe: Belarus, Estonia, Latvia, Lithuania, Ukraine, Russia. among women across the 30 European countries over the 2014 was 4.0 years, whereas it would have been 4.3 years 1990–2014 period (Table 1 and Figure 2). Among men in (i.e. 0.3 years higher) after excluding lifestyle-attributable Eastern Europe, the combined impact of smoking, obesity mortality (Table 2). An increase in e0 after excluding and alcohol on life expectancy was the greatest and was— lifestyle-attributable mortality can be observed for women in contrast to other regions—increasing up to 2005 primar- in all European regions and countries (Table 2 and ily due to the increases in the PGLE for alcohol (Table 1 and Supplementary Table 2, available as Supplementary data Supplementary Table 1, available as Supplementary data at at IJE online). This impact of lifestyle factors on life- IJE online, Figure 2 and Supplementary Figure 2, available as expectancy trends was biggest in Central Europe (differ- Supplementary data at IJE online). Among women, the im- ence of 0.7 years) and smallest in Southern Europe (differ- pact was the smallest and hardly changing in Southern ence of 0.1 years). Europe because moderate increases in PGLE for smoking and The trend in e0 after excluding lifestyle-attributable mor- obesity were almost offset by small declines in PGLE for tality was not only very similar for men and women across alcohol. the 30 European countries (4.2 and 4.3 years, respectively); Between 1990 and 2014, e0 increased by 5.0 years it was also more similar between countries and regions than among men across the 30 European countries (Table 2). all-cause mortality, at least for men (Figure 3). Among men, Without lifestyle-attributable mortality, this increase the variance between countries in e0 trends after excluding would have been 4.2 years (i.e. 0.8 fewer years). The im- lifestyle-attributable mortality was 2.3, compared with 4.4 pact of lifestyle factors on life-expectancy trends was larg- for all-cause mortality (Supplementary Table 2, available as est among men in Western, Southern and Central Europe, Supplementary data at IJE online). Among women, the re- where the increase in e0 would have been about 1.5 years spective variances (2.0 vs 1.8) did not differ much. The in- lower without lifestyle-attributable mortality (Table 2 and crease in e0 was still far less favourable in Eastern Europe Supplementary Table 2, available as Supplementary data than in the other regions, even after excluding lifestyle- at IJE online). By contrast, among Eastern-European men, attributable mortality (Table 2). This seems mainly driven excluding lifestyle-attributable mortality would have by the declines in e0 in the early 1990s (Figure 3). From resulted in a 0.4-year higher increase in e0. Among around 2000, the increases in e0 were more similar between European women, the average increase in e0 from 1990 to regions/countries. 936 International Journal of Epidemiology, 2021, Vol. 50, No. 3 Figure 2 Trends over time in the potential gain in life expectancy (PGLE) after eliminating smoking-, obesity- and alcohol-attributable mortality (sepa- rately and combined), 1990–2014*, by European region and sex. Lifestyle-attributable mortality refers to mortality that is attributable to smoking, obe- sity and alcohol combined. *For 2011 up to 2014, the weighted averages were calculated using the data for the latest available year: Bulgaria (2010), Greece (2013), Ukraine (2012) and Russia (2013). Discussion risk factors are independent and uncorrelated.45 In case the The combined impact of smoking, obesity and alcohol on effects of these three lifestyle risk factors were to overlap, e0 declined among men from 6.6 years in 1990 to 5.8 years rather than having synergistic relations, their joint contri- in 2014, mainly due to declining smoking-attributable bution would be overestimated. It is unknown how the mortality. Among women, the combined impact increased above-mentioned bias is changing over time and conse- from 1.9 to 2.3 years due to mortality increases in all three quently how this would affect our results on the combined lifestyle-related factors. The observed increase in e0 over impact of the three lifestyle factors on trends in life the 1990–2014 period was 5.0 years for men and 4.0 years expectancy. for women. Without the combined effect of these three fac- The techniques employed to estimate smoking-, obesity- tors, the increase in e0 would have been 4.2–4.3 years for and alcohol-attributable mortality were selected based on both men and women, and would have been more similar a careful assessment of different estimation techniques48–50 between countries for men. and data availability. However, these techniques provide To estimate mortality attributable to smoking, obesity estimates only and should be interpreted as such. and alcohol combined, we multiplicatively aggregated the Comparing our estimates of the PGLE from the elimina- risk-factor-specific mortality fractions, assuming that the tion of smoking-, obesity- and alcohol-attributable International Journal of Epidemiology, 2021, Vol. 50, No. 3 937 Table 2 Impact of smoking, obesity and alcohol (separately and combined) on the change in life expectancy at birth (e0) in 30 European countries from 1990 until 2014a by sex and region Change in e0 1990–2014a (in years) Observed Without smoking Without obesity Without alcohol Without smoking, obesity and alcohol Men Europe 4.98 3.48 5.54 5.29 4.24 Northern Europe 6.31 4.89 6.69 6.30 5.29 Western Europe 6.51 4.72 6.96 6.20 4.93 Southern Europe 6.59 5.22 6.96 5.94 5.09 Central Europe 7.52 5.51 8.04 7.54 6.06 Eastern Europe 1.20 –0.10 1.85 2.42 1.56 Women Europe 3.96 4.10 4.15 4.03 4.32 Northern Europe 4.34 4.77 4.50 4.33 4.89 Western Europe 4.66 5.07 4.87 4.55 5.10 Southern Europe 5.03 5.28 5.16 4.78 5.14 Central Europe 6.01 6.68 6.11 6.03 6.71 Eastern Europe 1.87 1.63 2.04 2.24 2.14 a For countries for which 2014 data were missing, we used data up until the latest available year: Bulgaria (2010), Greece (2013), Ukraine (2012) and Russia (2013). mortality with those using the Global Burden of Diseases smoking epidemic. For example, the large impact of life- 2017 data44 revealed that our estimates in 2014 for men style factors on trends among Central European women were, on average, 0.3 years lower for obesity in Eastern can be attributed to the later start and the greater impact Europe and for alcohol across all regions, but 0.7 years of the smoking epidemic.16 Large international differences higher for smoking in CEE; for women, they were, on aver- in alcohol-prevalence trends, however, have also played a age, largely similar for alcohol, but 0.3–0.4 years lower for role. Among men, the high levels and increasing trends in obesity in CEE and for smoking in Southern and Eastern alcohol prevalence until around 2005 in Eastern Europe (see Supplementary Table 3, available as Europe37 led to a deceleration, instead of an acceleration, Supplementary data at IJE online). Based on this compari- of life-expectancy increases. In contrast, among women son, our estimates seem generally conservative. in Southern Europe, declines in alcohol prevalence54 The differences in the impact of lifestyle on trends in e0 contributed to the relatively small negative impact of life- between men (accelerating the increase) and women (decel- style on life-expectancy trends. These differences can be erating the increase) are mainly attributable to large sex mainly attributed to differences in drinking cultures (spi- differences in the onset of the smoking epidemic. Because rits in Eastern Europe, wine in Southern Europe) and men generally started smoking in large numbers 20– in economic development (economic hardship in Eastern 30 years earlier than women did, trends in smoking- Europe).20 attributable mortality had been declining for several deca- We observed that the e0 trends for non-lifestyle- des for men, but were still increasing for women in most attributable mortality ran parallel for men and women, countries.15,16 This earlier onset of smoking among men and were more similar between countries than the ob- can be attributed to the earlier uptake of risky health served e0 trends for men. Moreover, from a longer-term behaviours by men than by women.51 However, women’s perspective, the e0 trends after excluding lifestyle- behaviours followed those of men when their roles in soci- attributable mortality seem relatively stable over time ety changed due to women’s emancipation and rising fe- (Supplementary Figure 3, available as Supplementary data male labour-force participation.52 The more recent trends at IJE online). This generally uniform trend may capture in obesity- and alcohol-attributable mortality have been the underlying gradual long-term increase in life expec- largely similar for men and women, in line with their rela- tancy in Europe, which, as postulated in the epidemiologi- tively similar prevalence trends.42,53 cal transition theory,55 reflects the long-term effects of The observed differences between regions and countries socio-economic growth and medical progress. In European in the impact of lifestyle factors on trends in e0 can also be countries, large increases in gross domestic product per linked to international differences in the progression of the capita since the late nineteenth century56 drove the secular- 938 International Journal of Epidemiology, 2021, Vol. 50, No. 3 Figure 3 Trends in life expectancy at birth (e0), observed (¼ all-cause mortality) versus after excluding lifestyle-attributable mortality (¼ non-lifestyle- attributable mortality), 1990–2014*, by European region and sex. Lifestyle-attributable mortality refers to mortality that is attributable to smoking, obesity and alcohol combined. *For 2011 up to 2014, the weighted averages were calculated using the data for the latest available year: Bulgaria (2010), Greece (2013), Ukraine (2012) and Russia (2013). mortality decline during the twentieth century. From the mortality remain. Particularly notable are the less favour- 1950s onwards, increases in life expectancy were achieved able trends in Eastern Europe (i.e. former Soviet republics). primarily through declines in adult cardiovascular mortal- The declines in e0 in the early 1990s may be attributable to ity linked to medical improvements (hypertension treat- the economic and political disruptions that followed the ments, statins, thrombolysis, stents).57 dissolution of the Soviet Union.21,58,59 However, from Although country differences in e0 trends can be largely 2000 onwards, the e0 trends without lifestyle-attributable explained by the three lifestyle factors studied, important mortality were again more similar between countries. This country differences in e0 for non-lifestyle-attributable convergence might reflect increasing similarities between International Journal of Epidemiology, 2021, Vol. 50, No. 3 939 countries in factors such as healthcare seeking, health- Supplementary data related policies and socio-cultural trends. Supplementary data are available at IJE online. Smoking, obesity and alcohol likely contribute not only to country differences in e0 trends, but also to socio-economic differences in e0 trends. The three life- Author contributions style factors contribute substantially to socio-economic dif- F.J. conceived of the study. F.J. designed the study with input ferences in mortality60 due to higher current prevalence and from S.T.L. and A.E.K. F.J. ran the first analyses. S.T.L. ran associated mortality among people with low than with high the final analyses. All authors interpreted the results. F.J. SES.61 This contribution has increased over time, due to drafted the manuscript. A.E.K. provided critical input on the socio-economic differences in the uptake of new behaviours, draft manuscript. F.J. revised the manuscript. All authors and those with a higher socio-economic position generally reviewed and approved the final version of the manuscript. being the first to change to healthier behaviours again.15,62,63 Because lifestyle factors have a large effect on life- Funding expectancy trends, it is important to consider them when This work was supported by the Netherlands Organisation for studying the stagnation in the increase in life expectancy. Scientific Research (NWO) in relation to the research programme In Europe—particularly in the UK, but also in France, ‘Smoking, alcohol, and obesity, ingredients for improved and robust Germany, Sweden and the Netherlands—slowdowns in mortality projections’, under grant no. 452–13-001. See www.future life-expectancy increases have been observed since mortality.com. 2011.24,25 Increases in both obesity prevalence25 and alco- hol abuse among UK adults24 have been mentioned as po- Acknowledgements tential causes. For women in France, Germany, Sweden The authors thank Mark van der Broek (Research Master and the Netherlands, rapid increases in smoking- Economics, University of Groningen) for his help in running the analyses in R and making the initial tables and figures. Because the attributable mortality (Supplementary Figure 1b, available data we used consisted of data at the population level, available as Supplementary data at IJE online)16 are likely contribu- from data sets that exist in the public domain, ethical approval was tors as well. Decomposing recent mortality trends into not required. The data that were used as input for the tables and fig- lifestyle-attributable mortality and remaining mortality ures will be made available at the Open Science Framework: https:// may help to explain recent e0 trends. osf.io/j84c6/. The R code can be requested from the author. Similarly, when predicting future trends in life expec- tancy, the time-varying impact of lifestyle factors should be considered. Because mortality forecasts mostly rely on Conflict of interest the extrapolation of past age-specific mortality trends,64 None declared. The funding source had no role in the study design, a major challenge is to identify an underlying mortality collection, analysis or interpretation of the data; in the writing of trend that is sufficiently stable to serve as the basis for ex- the manuscript; or in the decision to submit the paper for publication. trapolation.65,66 At the same time, such extrapolation should not ignore the factors that may cause deviations References from this underlying trend.41,65,66 Our results indicate, first, that the past trends in non-lifestyle-attributable 1. WHO. Global Status Report on Noncommunicable Diseases mortality serve as a better baseline for extrapolations 2014. Geneva: World Health Organization, 2014. 2. WHO. Health in the European Union—Trends and Analysis. than all-cause mortality; and, second, that smoking, obe- Copenhagen: World Health Organization—European sity and alcohol are factors that seem to cause important Observatory on Health Systems and Policies, 2009. deviations that should be predicted separately using more 3. WHO. European Health Report 2018: More than Numbers— advanced techniques. Indeed, over the short term, less Evidence for All. Geneva: World Health Organization— favourable e0 trends are expected because continued European Region, 2018. increases in smoking-attributable mortality for women in 4. Preston SH, Glei DA, Wilmoth Jr. Contribution of smoking to most countries and general increases in obesity- international differences in life expectancy. In: Crimmins EM, attributable mortality are anticipated. Increases in e0 are Preston SH, Cohen B (eds). International Differences in Mortality at Older Ages—Dimensions and Sources. expected to become more favourable again once the Washington: The National Academies Press, 2010, pp. 105–31. declines in smoking-attributable mortality among women 5. Luy M, Wegner-Siegmundt C. The impact of smoking on gender become more widespread and the hypothesized decline in differences in life expectancy: more heterogeneous than often obesity-attributable mortality17,18,28 eventually sets in. stated. Eur J Public Health 2015;25:706–10. 940 International Journal of Epidemiology, 2021, Vol. 50, No. 3 6. Renteria E, Jha P, Forman D, Soerjomataram I. The impact of 23. Lindahl-Jacobsen R, Oeppen J, Rizzi S et al. Why did Danish wom- cigarette smoking on life expectancy between 1980 and 2010: a en’s life expectancy stagnate? The influence of interwar genera- global perspective. Tob Control 2016;25:551–57. tions’ smoking behaviour. Eur J Epidemiol 2016;31:1207–11. 7. Janssen F. The role of smoking in country differences in life ex- 24. Leon DA, Jdanov DA, Shkolnikov VM. Trends in life expectancy pectancy across Europe. Nicotine Tob Res 2020. doi:10.1093/ and age-specific mortality in England and Wales, 1970-2016, in ntr/ntaa011. comparison with a set of 22 high-income countries: an analysis 8. Zatonski W. Closing the Health Gap in the European Union. of vital statistics data. Lancet Public Health 2019;4: Warsaw: Maria Sklodowska-Curie Memorial Cancer Center and e575-82–e582. Institute of Oncology, 2008. 25. Raleigh V. Trends in Life Expectancy in EU and Other OECD 9. McCartney G, Mahmood L, Leyland AH, Batty GD, Hunt K. Countries: Why Are Improvements Slowing? OECD Health Contribution of smoking-related and alcohol-related deaths to Working Papers 2019/108. Paris: Organisation for Economic the gender gap in mortality: evidence from 30 European coun- Co-operation and Development, 2019. tries. Tob Control 2011;20:166–68. 26. WHO. The Challenge of Obesity in the WHO European Region 10. Trias-Llimós S, Janssen F. Alcohol and gender gaps in life expec- and the Strategies for Response: Summary. Copenhagen: World tancy in eight Central and Eastern European countries. Eur J Health Organization, 2007. Public Health 2018;28:687–92. 27. Vidra N, Trias-Llimós S, Janssen F. Impact of obesity on life ex- 11. Olshansky SJ, Passaro DJ, Hershow RC et al. A potential decline pectancy among different European countries: secondary analy- in life expectancy in the United States in the 21st century. N Engl sis of population-level data over the 1975-2012 period. BMJ J Med 2005;352:1138–45. Open 2019;9:e028086. 12. Preston SH, Stokes A. Contribution of obesity to international 28. Janssen F, Bardoutsos A, Vidra N. Obesity prevalence in the differences in life expectancy. Am J Public Health 2011;101: long-term future in 18 European countries and in the USA. Obes 2137–43. Facts 2020;13:514–27. 13. Lhachimi SK, Nusselder WJ, Smit HA et al. Potential health 29. Rehm J, Sulkowska U, Manczuk M et al. Alcohol accounts for a gains and health losses in eleven EU countries attainable through high proportion of premature mortality in central and eastern feasible prevalences of the life-style related risk factors alcohol, Europe. Int J Epidemiol 2007;36:458–67. BMI, and smoking: a quantitative health impact assessment. 30. Trias-Llimós S, Kunst AE, Jasilionis D, Janssen F. The contribu- BMC Public Health 2016;16:734. tion of alcohol to the East-West life expectancy gap in Europe 14. Östergren O, Martikainen P, Tarkiainen L, Elstad JI, Brønnum- from 1990 onward. Int J Epidemiol 2018;47:731–39. Hansen H. Contribution of smoking and alcohol consumption 31. Stanaway JD, Afshin A, Gakidou E et al. 2017 Risk Factor to income differences in life expectancy: evidence using Danish, Collaborators. Global, regional, and national comparative risk Finnish, Norwegian and Swedish register data. J Epidemiol assessment of 84 behavioural, environmental and occupational, Commun H 2019;73:334–39. and metabolic risks or clusters of risks for 195 countries and ter- 15. Lopez AD, Collishaw NE, Piha T. A descriptive model of the ritories, 1990–2017: a systematic analysis for the Global Burden cigarette epidemic in developed countries. Tob Control 1994; of Disease Study 2017. Lancet 2018;392:1923–94. 3:242–47. 32. Reitsma MB, Fullman N, Ng M et al. A systematic analysis from 16. Janssen F. Similarities and differences between sexes and coun- the global burden of disease study 2015. Lancet 2017;389: tries in the mortality imprint of the smoking epidemic in 34 low- 1885–906. mortality countries, 1950-2014. Nicotine Tob Res 2020;22: 33. GBD 2015 Obesity Collaborators; Afshin A, Forouzanfar MH. 1210–20. Health effects of overweight and obesity in 195 countries over 17. Xu L, Lam TH. Stage of obesity epidemic model: Learning from 25 years. N Engl J Med 2017;377:13–27. tobacco control and advocacy for a framework convention on 34. GBD 2016 Alcohol Collaborators. Alcohol use and burden for obesity control. J Diabetes 2018;10:564–71. 195 countries and territories, 1990-2016: a systematic analysis 18. Jaacks LM, Vandevijvere S, Pan A et al. The obesity transition: for the Global Burden of Disease Study. Lancet 2018;392: stages of the global epidemic. Lancet Diabetes Endocrinol 2019; 1015–35. 7:231–40. 35. Janssen F, Rousson V, Paccaud F. The role of smoking in changes 19. Shield K, Manthey J, Rylett M et al. National, regional, and in the survival curve: an empirical study in 10 European coun- global burdens of disease from 2000 to 2016 attributable to alco- tries. Ann Epidemiol 2015;25:243–49. hol use: a comparative risk assessment study. Lancet Public 36. Martikainen P, Mäkelä P, Peltonen R, Myrskylä M. Income dif- Health 2020;5:e51–61. ferences in life expectancy: the changing contribution of harmful 20. Janssen F, El Gewily S, Bardoutsos A, Trias-Llimós S. Past and consumption of alcohol and smoking. Epidemiology 2014;25: future alcohol-attributable mortality in Europe. Int J Environ 182–90. Res Public Health 2020;17:9024. 37. Jasilionis D, Meslé F, Shkolnikov VM, Vallin J. Recent life ex- 21. Vallin J, Meslé F. Convergences and divergences in mortality: pectancy divergence in Baltic countries. Eur J Population 2011; a new approach to health transition. Demogr Res 2004;2: 27:403–31. 11–44. 38. Janssen F, van der Broek M, Bardoutsos A, Vidra N. Obesity-at- 22. Rostron BL, Wilmoth JR. Estimating the effect of smoking on tributable Mortality in the Long-term Future in Europe. NIDI slowdowns in mortality declines in developed countries. Working Paper 2020/06. The Hague: Netherlands Demography 2011;48:461–79. Interdisciplinary Demographic Institute, 2020. International Journal of Epidemiology, 2021, Vol. 50, No. 3 941 39. Human Mortality Database. University of California, Berkeley 51. Rogers R, Everett BG, Saint Onge JM, Krueger PM. Social, be- (USA), and Max Planck Institute for Demographic Research havioral, and biological factors, and sex differences in mortality. (Germany). www.mortality.org (27 August 2018, date last Demography 2010;47:555–78. accessed). 52. Waldron I. Trends in gender differences in mortality: relationships 40. Peto R, Boreham J, Lopez AD, Thun M, Heath C. Mortality to changing gender differences in behaviour and other causal factors. from tobacco in developed countries: indirect estimation from In: Annandale E, Hunt K (eds). Gender Inequalities in Health. national vital statistics. Lancet 1992;339:1268–78. Buckingham: Open University Press, 2000, pp. 150–81. 41. Janssen F, van Wissen L, Kunst AE. Including the smoking epi- 53. WHO. Global Status Report on Alcohol and Health 2018. demic in internationally coherent mortality projections. Geneva: World Health Organization, 2019. Demography 2013;50:1341–62. 54. Gual A, Colom JOAN. Why has alcohol consumption declined 42. NCD, Risk Factor C; (NCD-RisC). Worldwide trends in body- in countries of southern Europe? Addiction 1997;92:S21–31. mass index, underweight, overweight, and obesity from 1975 to 55. Omran AR. The epidemiologic transition theory revisited 2016: a pooled analysis of 2416 population-based measurement thirty years later. World Health Statistics Quarterly 1998;52: studies in 128.9 million children, adolescents, and adults. Lancet 99–119. 2017;390:2627–42. 56. Janssen F, Kunst AE, Mackenbach JP. Association between gross 43. DYNAMO-HIA Consortium. Workpackage 7: Overweight and domestic product throughout the life course and old-age mortal- Obesity: Report on Data Collection for Overweight and Obesity ity across birth cohorts: parallel analyses of seven European Prevalence and Related Relative Risks. London: International countries, 1950-1999. Soc Sci Med 2006;63:239–54. Association for the Study of Obesity, 2010. 57. Mensah GA, Wei GS, Sorlie PD et al. Decline in cardiovascular 44. Global Burden of Disease Study 2017. GBD Results Tool. mortality: possible causes and implications. Circ Res 2017;120: Seattle, WA: Institute for Health Metrics and Evaluation 366–80. (IHME), 2019. https://gbd2017.healthdata.org/gbd-search/ (6 58. Leon DA. Trends in European life expectancy: a salutary view. April 2018, date last accessed). Int J Epidemiol 2011;40:271–77. 45. Ezzati M, Vander Hoorn S, Rodgers A, Lopez AD, Mathers CD, 59. McKee M, Shkolnikov V. Understanding the toll of Murray CJ. Estimates of global and regional potential health premature death among men in eastern Europe. BMJ 2001; gains from reducing multiple major risk factors. Lancet 2003; 323:1051–55. 362:271–80. 60. Petrovic D, de Mestral C, Bochud M et al. The contribution of 46. Lim SS, Carnahan E, Nelson EC et al. Validation of a new pre- health behaviors to socioeconomic inequalities in health: a sys- dictive risk model: measuring the impact of the major modifiable tematic review. Prev Med 2018;113:15–31. risks of death for patients and populations. Popul Health Metr 61. Mackenbach JP. Health Inequalities in Europe: New Insights 2015;13:27. from Comparative Studies. Rotterdam: Erasmus University 47. Tsai SP, Lee ES, Hardy RJ. The effects of a reduction in leading Publishing, 2016. causes of death: potential gains in life expectancy. Am J Public 62. Sobal J, Stunkard AJ. Socioeconomic status and obesity: a review Health 1978;68:966–71. of the literature. Psychol Bull 1989;105:260–75. 48. Vidra N, Bijlsma M, Janssen F. Impact of different estimation 63. Mackenbach JP. The persistence of health inequalities in modern methods on obesity-attributable mortality levels and trends: the welfare states: the explanation of a paradox. Soc Sci Med 2012; case of the Netherlands. Int J Environ Res Public Health 2018; 75:761–69. 15:2146. 64. Stoeldraijer L, van Duin C, van Wissen L, Janssen F. Impact of 49. Trias-Llimós S, Martikainen P, Mäkelä P, Janssen F. F. different mortality forecasting methods and explicit assumptions Comparison of different approaches for estimating age-specific on projected future life expectancy: the case of the Netherlands. alcohol-attributable mortality: the cases of France and Finland. Demogr Res 2013;29:323–54. PLoS One 2018;13:e0194478. 65. Janssen F. Advances in mortality forecasting: introduction. 50. Stoeldraijer L, Bonneux L, van Duin C, van Wissen L, Janssen F. Genus 2018;74:21. The future of smoking-attributable mortality: the case of 66. Janssen F, Kunst A. The choice among past trends as a basis for England & Wales, Denmark and the Netherlands. Addiction the prediction of future trends in old-age mortality. Pop Stud 2015;110:336–45. 2007;61:315–26.