Long et al. Population Health Metrics (2021) 19:3 https://doi.org/10.1186/s12963-021-00247-2 RESEARCH Open Access Smoking and inequalities in mortality in 11 European countries: a birth cohort analysis Di Long1* , Johan Mackenbach1, Pekka Martikainen2, Olle Lundberg3, Henrik Brønnum-Hansen4, Matthias Bopp5, Giuseppe Costa6, Katalin Kovács7, Mall Leinsalu8,9, Maica Rodríguez-Sanz10,11, Gwenn Menvielle12 and Wilma Nusselder1 Abstract Purpose: To study the trends of smoking-attributable mortality among the low and high educated in consecutive birth cohorts in 11 European countries. Methods: Register-based mortality data were collected among adults aged 30 to 79 years in 11 European countries between 1971 and 2012. Smoking-attributable deaths were estimated indirectly from lung cancer mortality rates using the Preston-Glei-Wilmoth method. Rate ratios and rate differences among the low and high-educated were estimated and used to estimate the contribution of inequality in smoking-attributable mortality to inequality in total mortality. Results: In most countries, smoking-attributable mortality decreased in consecutive birth cohorts born between 1906 and 1961 among low- and high-educated men and high-educated women, but not among low-educated women among whom it increased. Relative educational inequalities in smoking-attributable mortality increased among both men and women with no signs of turning points. Absolute inequalities were stable among men but slightly increased among women. The contribution of inequality in smoking-attributable mortality to inequality in total mortality decreased in consecutive generations among men but increased among women. Conclusions: Smoking might become less important as a driver of inequalities in total mortality among men in the future. However, among women, smoking threatens to further widen inequalities in total mortality. Keywords: Birth cohort, Smoking, Mortality, Educational inequalities Introduction many countries [7], partly as a result of tobacco control While people’s health in Europe has improved in the last policies [8, 9]. few decades, the improvements are not experienced However, it has been less studied how changes in equally by all. Reducing educational inequalities in smoking behavior have influenced trends in educational health has been one of the most important targets in inequalities in mortality. The studies that measured the public health [1]. Among all health risks, smoking is be- effect of smoking on inequalities in mortality have sug- lieved to be the greatest avoidable risk factor [2] and one gested increases in relative inequalities in smoking- of the leading contributors to inequalities in mortality related mortality in many European countries in recent [3–6]. In Europe, smoking prevalence has declined in decades [10–13]. Although with different methods, these studies all used a period approach. However, other stud- ies have established that the evolution of the smoking * Correspondence: d.long@erasmusmc.nl 1 Department of Public Health, Erasmus MC, P.O. Box 2040, 3000 CA epidemic and lung cancer epidemic are phenomena re- Rotterdam, The Netherlands lated to birth cohorts [14, 15]. The habit of smoking is Full list of author information is available at the end of the article © The Author(s). 2021 Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Long et al. Population Health Metrics (2021) 19:3 Page 2 of 9 adopted in the early stages of life and is likely to persist were available for the ages 30-49), and educational level throughout the life course [16]. The likelihood of picking over the period 1971-2012 in 11 European countries up smoking varies over time, due to historical differ- (Table 1). For those of age 30-39 and 40-49 in Norway, ences in the physical or social environment, such as ac- we estimated the mortality number in 5-year age groups cess to cigarettes, socioeconomic conditions, or social using the Human Mortality Database (www.mortality. acceptance of smoking in time of uptake smoking [17, org). Most data comprise national populations, except 18]. These effects are likely to be only partially captured for Italy, where we obtained data from Turin only; for by period effects. Therefore, a birth cohort perspective Spain, from Barcelona only; and for the UK, from Eng- can offer us a more comprehensive understanding not land and Wales only. Details on data sources can be only of the evolution of mortality due to smoking in the found in Supplementary file 1. population as a whole but also of educational inequal- Educational level was measured as the highest level ities in mortality from smoking. It may also offer better achieved and coded into three groups based on the insights into how inequalities in mortality are likely to International Standard Classification of Education (ISCE evolve in the future. D-97): Up to lower secondary education (ISCED 0, 1 In this article, we studied trends in smoking- and 2; “low”), completed upper secondary education attributable mortality among the low and high-educated (ISCED 3 and 4; “mid”) and tertiary education (ISCED 5 in consecutive birth cohorts in 11 European countries. and 6; “high”). In the UK (England and Wales), only two We assessed whether educational inequalities in groups could be distinguished, namely the “low” (ISCED smoking-attributable mortality and their contribution to 0, 1, 2, and 3) and “high” (ISCED 5 and 6). educational inequalities in total mortality have changed Birth cohorts were reconstructed using age and period between older and younger generations in these coun- (birth cohort = period − age) in the mortality dataset by tries. We used the Preston-Glei-Wilmoth (PGW) country, sex, and level of education. Because data were method to estimate smoking-attributable mortality. collected in 5-year age groups and 1- to 5-year periods, the birth cohorts reconstructed were in 6- to 10-year Data and methods ranges with overlaps. For example, those of age 30 to 34 Data who died during 1980-1984 belong to birth cohort 1946- Register-based mortality data were collected and harmo- 1954, and those of age 35 to 39 who died during the nized for all-cause mortality and lung cancer mortality same period belong to birth cohort 1941-1949. In each by sex, age (5-year age group from 30 to 34 to 75-79 country-sex-education level-age-period group, we ex- years, except Norway for which only 10-year age groups cluded cells in which the number of deaths due to lung Table 1 Characteristics of the mortality data and population data used in the analyses in each country: Types of data set (longitudinal data or cross-sectional data: unlinked or repeated); geographic coverage of the dataset (National or Urben); the period covered in the dataset; the number of birth cohorts constructed based on the dataset; the oldest birth cohort constructed; the youngest birth cohort constructed; person-years at risk followed up; number of deaths due to lung cancer; and the percentage of lung cancer deaths with low education in each gender Population Type of Geographic Period Number Oldest Youngest P-Y2 Number % of lung cancer deaths with dataset1 coverage of birth birth birth follow- of low education cohorts cohort cohort up deaths3 Male Female Belgium Longitudinal National 1991-2006 6 1927-1936 1952-1961 78819 77795 77.8 74.3 Denmark Longitudinal National 1991-2010 6 1926-1934 1951-1959 47377 53491 47.2 63.4 Estonia CS, unlinked National 1987-2012 5 1928-1936 1948-1956 9551 11715 59.0 47.4 Finland Longitudinal National 1971-2011 11 1902-1910 1952-1960 84266 71774 81.0 74.1 Hungary CS, unlinked National 1971-2010 9 1909-1916 1949-1956 101767 106780 83.9 76.9 Italy (Turin) Longitudinal Urban 1971-2011 8 1907-1916 1942-1951 13983 17722 85.7 91.5 Lithuania CS, unlinked National 1988-2011 5 1932-1940 1952-1960 16449 14428 56.6 45.3 Norway Longitudinal National 1971-2006 9 1902-1910 1942-1950 53347 44959 55.6 61.1 Spain (Barcelona) CS, repeated Urban 1992-2011 7 1923-1931 1953-1961 13723 14313 72.4 75.2 Sweden Longitudinal National 1990-2005 6 1921-1929 1946-1954 66980 39996 56.5 54.5 Switzerland Longitudinal National 1991-2010 7 1922-1930 1952-1960 59446 43679 33.8 49.1 1 CS represents cross-sectional 2 P-Y represents person-years. Person-years at risk is given in thousands 3 Number of deaths refers to deaths due to lung cancer Long et al. Population Health Metrics (2021) 19:3 Page 3 of 9 cancer was less than 10 in order to avoid erratic results deaths due to smoking and person-years in each due to small numbers. We further excluded birth co- country-sex-education level-cohort group. Details of cal- horts with data for less than three age groups. As a re- culation can be found in Supplementary file 3. sult, our analysis covers 18 birth cohort groups, ranging from 1902-1910 to 1952-1961. We present birth cohorts The CCMF method by their median birth year. Detailed descriptions of the We used the CCMF method, introduced by Gardner birth cohort construction are provided in Supplementary et al. [21], to compare mortality between birth cohorts file 2_a and 2_b. that have different age compositions. The CCMF method is a form of direct standardization and is de- Methods scribed in Supplementary file 4. We summed the The method consisted of three steps. The first step used person-years in all countries by sex and age to obtain a the PGW method to estimate the period- and age- standard population. Briefly, the resulting CCMF is the specific number of deaths attributable to smoking, by ratio of the expected deaths in a birth cohort if its age- country, sex, and level of education, from the number of specific observed mortality rates would apply to the lung cancer deaths. Total numbers of deaths in the same standard population in the same age groups, to the ob- groups were also obtained for later use. The second step served deaths in the standard population in the same performed a form of direct standardization method cohort. yielding comparative cohort mortality figures (CCMF) to X Expected deaths in standard population at cohort rates X Y si r cseki summarize overall mortality from smoking in each birth CCMF ¼ Observed deaths in standard population at standard rates ¼ Dsi i i cohort. The final step calculated rate ratios (RR), rate differences (RD), and contributions of inequalities in X Y si r cseki X r cseki smoking-attributable mortality to educational inequal- ¼ ¼ Y si Rsi Rsi ities in total mortality. In the following, we describe each i i of these steps in more detail. where i = age group, s = sex, c = country, e = level of The PGW method education, k = birth cohort, The PGW method, introduced by Preston et al. [19], is Ysi = standard population in gender s and age group i based on a regression analysis which uses the lung can- rcseki = observed death rate in country c, gender s, edu- cer death rate to estimate smoking-attributable mortality cation level e, cohort k, and age group i from all causes, and was originally performed on data of Dsi = number of deaths in standard population in gen- people aged 50 and higher in 21 high-income countries der s and age group i from 1950 to 2007. For the age group 30-49, we used an Rsi = standard death rate in standard population in extended PGW model developed by Martikainen et al. gender s and age group i [20]. We applied this method to each country-sex- education level-cohort group in our data. Rate ratios, rated differences, and the contributions First, we compared lung cancer death rates (M) in our To estimate relative inequalities in mortality, we calcu- study to lung cancer death rates among non-smokers (λ) lated RR, by taking the ratio between the CCMFs of the in the Second Prospective Cancer Prevention Study low-educated and the CCMFs of the high-educated in (CPS-II) to get the proportion of lung cancer deaths at- each birth cohort by sex and country. To estimate abso- tributable to smoking (AL ¼ MM− λ). Second, we calculated lute inequalities, we calculated RD, for which we first calculated smoking-attributable mortality rates by sex in the proportion of other deaths attributable to smoking ( 0 the standard population. Next, we multiplied these AO ¼ 1 − e − β ðM − λÞ ). The parameters (β′) were pub- standard smoking-attributable mortality rates with the lished in the original articles and can be directly used CCMF of every birth cohort to obtain the absolute without re-performing the analysis in further studies to smoking-attributable mortality rates for every birth co- obtain the proportion of other deaths. Finally, we com- hort in each country-sex-education group. RDs were bined the two fractions calculated above, weighted by then calculated by subtraction of the smoking- the numbers of deaths from lung cancer (DL) and other attributable mortality rate of the high educated from that causes (DO), to generate the overall smoking-attributable of the low educated. In the same manner, we calculated fraction for deaths from all causes ( A ¼ AL DDLLþAO DO þDO ). We RDs of the total mortality rates, which we used to obtain obtained the overall number of deaths due to smoking the fraction of smoking-attributable mortality from in- by multiplying the overall smoking-attributable fraction equalities in total mortality. to the number of total deaths. We calculated smoking- We repeated our analyses using less strict exclusion attributable death rates using the overall number of criteria, which also included birth cohorts with data for Long et al. Population Health Metrics (2021) 19:3 Page 4 of 9 only two age groups. We present the results in Supple- for low-educated men in Estonia and Hungary, where mentary file 5. smoking-attributable mortality increased, and in We calculated 95% confidence intervals (95% CI) using Lithuania, where a decrease followed after an increase. parametric bootstrapping, assuming Poisson-distributed Among women, smoking-attributable mortality was not death counts and setting the number of repetitions to always higher in the low educated than the high edu- 1000. All analyses were performed using Stata V.15.1 SE. cated. Among high-educated women, smoking- attributable mortality decreased in consecutive genera- Results tions in most countries. Among low-educated women, Smoking-attributable mortality however, smoking-attributable mortality gradually in- Table 1 shows the characteristics of the mortality and creased in consecutive cohorts in most countries, except population data used in the analyses. Figure 1 presents for Denmark where it declined from older generations at birth-cohort trends in age-standardized smoking- a very high level. attributable mortality in 11 countries, by education and gender. When a CCMF value is higher than 1, the age- Relative and absolute inequalities in smoking-attributable standardized smoking-attributable mortality for that mortality birth cohort is higher than the average age-standardized Relative inequalities were present among men in all gen- smoking-attributable mortality of 11 countries in the erations in all countries (Fig. 2a, Supplementary file 7). same sex group. Relative inequalities increased greatly in consecutive For men, smoking-attributable mortality was higher birth cohorts except for Italy where relative inequalities among the low educated than among the high educated reached a peak among the generation born around 1931, in every country. Smoking-attributable mortality grad- and declined in the younger generations. Among ually decreased in consecutive birth cohorts among both women, relative inequalities were present among all gen- low and high-educated men in most countries, except erations in most countries, but reverse inequalities, with Fig. 1 Birth cohort trends in age-standardized smoking-attributable mortality in 11 European countries, by education and gender Long et al. Population Health Metrics (2021) 19:3 Page 5 of 9 Fig. 2 Relative inequalities (a), absolute inequalities (b) in smoking-attributable mortality, the contribution of smoking to absolute inequalities in total mortality (c), in 11 European countries by sex and by educational leve. (We excluded generations born before 1936 in Hungary because they had higher smoking-attributable mortality among the high educated and higher total mortality rates among the low educated in older generations). higher smoking-attributable mortality among the high- where reverse inequalities were present for those born educated than among the low-educated, were present before 1936. among the generations born before 1936 in Hungary. Compared to men, relative inequalities increased more Contribution of smoking to inequalities in total mortality dramatically among women in consecutive birth cohorts: Absolute inequalities in total mortality (Supplementary they started with smaller relative inequalities in older file 6) increased in consecutive generations in all coun- generations but reached larger inequalities in the youn- tries among both men and women. Figure 2c (Supple- ger generations. mentary file 7) shows the extent to which inequalities in Absolute inequalities (Fig. 2b, Supplementary file 7) smoking-attributable mortality contributed to these in- were stable in consecutive birth cohorts in many coun- equalities in total mortality in consecutive generations. tries, except in Hungary, Estonia, and Lithuania where Among men, the contributions declined in consecutive absolute inequalities increased largely, and in Finland birth cohorts in most countries. In Norway, the contri- and Italy where absolute inequality decreased slightly in bution reached a peak among the generation born consecutive generations. Among women, absolute in- around 1931 and declined in younger generations. In equalities increased subtly in consecutive birth cohorts Hungary, the contributions increased over consecutive in most countries and increased largely in Hungary generations. Among women, the contributions increased Long et al. Population Health Metrics (2021) 19:3 Page 6 of 9 in consecutive birth cohorts in many countries, except underestimated. However, it is noteworthy that we used for Denmark where it reached a peak with the gener- the coefficients presented in the Preston et al. paper, and ation born around 1936, followed by a decline. that regression analyses were based on several European Overall, we are confident in most of our results except countries including Eastern European countries. The for RR for the youngest generation of women in Finland PGW method further assumes the overall damage and the contribution for men in Italy, where the range caused by the mix of prevalence, duration, and intensity of 95% CIs are excessively large. of smoking based on lung cancer mortality is propor- tionally the same for other causes. This assumption may Discussion be unreasonable. However, it is very difficult to measure Main findings these factors, let alone estimate their damage to deaths We found that in most countries, smoking-attributable [24]. In our application of the PGW method, we also as- mortality decreased in consecutive birth cohorts among sume that the relationship between lung cancer mortal- low- and high-educated men and high-educated women, ity and smoking-related mortality from other causes is but not among low-educated women in whom it in- the same in all education groups. This may not be true. creased. In terms of educational inequalities in smoking- For example, low-educated non-smokers might suffer attributable mortality, we found steep increases in rela- from a higher risk of lung cancer than high-educated tive inequalities among both men and women, stable ab- non-smokers due to other exposures [25], which could solute inequalities among men and subtle increases in have contributed to an overestimation of smoking- absolute inequalities among women in consecutive gen- attributable mortality among the low educated. The erations in most countries. The contribution of inequal- PGW method also does not distinguish between expos- ities in smoking-attributable mortality to inequalities in ure and susceptibility to smoking, both of which may total mortality decreased among men and increased vary by level of education. Although several studies have among women in consecutive generations in most coun- assessed some of these issues and found that the PGW tries. The decreasing trends in the contribution of in- provides robust results [13, 26, 27], we recommend rep- equalities among men are the result of stable absolute lication of our findings with other methods, e.g., survey- inequalities in smoking-attributable mortality and larger based smoking prevalence data by level of education an- increases of absolute inequalities in total mortality; the alyzed in a cohort perspective. increasing trends among women are the results of larger Our study has several other limitations. First, it is a increases of absolute inequalities in smoking-attributable simple descriptive analysis, without a formal age-period- mortality and smaller increases of absolute inequalities cohort analysis, which does not allow us to quantify the in total mortality. extent of the cohort-related changes, as compared to period- and age-related changes in mortality. Although Limitations and strengths our results are highly suggestive of cohort effects, it will Our study applies the innovative PGW method, which be interesting to see this confirmed in a formal age- uses lung cancer deaths to estimate the overall number period-cohort analysis [28, 29]. of deaths due to smoking. Essentially, this method as- Second, despite the large size of our dataset, the scope sumes that there is a constant relationship between lung of our analyses was sometimes restricted by small num- cancer mortality and smoking-related mortality from ber problems. Due to the slower evolution of higher other causes of death. It is based on the CPS-II studies educational attainment among women, especially in which provide an estimate of lung cancer mortality Mediterranean countries, we could not always closely among non-smokers in the USA [22], but whether the follow the evolution of inequalities in smoking- same rates apply elsewhere is unknown. The PGW attributable mortality among women. Moreover, to cal- method may overestimate the damage from smoking in culate age-standardized smoking-attributable mortality countries with unusually low mortality from other and inequalities for different birth cohorts, we stratified causes. This might be the case in some Mediterranean mortality by sex, level of education, age, and period, countries where cardiovascular disease mortality has causing small numbers of deaths in some cells. We had much less impact on total mortality [23]. The PGW to apply strict exclusion criteria to avoid erratic results method may misjudge the impact of smoking with un- due to these small numbers, which reduced the number usually high mortality from other causes. High mortality of birth cohorts and countries included in the analysis. rates in middle age in central European countries were In a sensitivity analysis, we applied less strict exclusion observed in our data, where middle-aged people may criteria, which allowed us to include more birth cohorts, have died from other causes before potentially develop- both early and recent generations and more countries. ing lung cancer. Therefore, the proportion of lung can- However, the results remain largely unchanged in this cer deaths attributable to smoking may have been extended analysis (Supplementary file 5). Long et al. Population Health Metrics (2021) 19:3 Page 7 of 9 Third, the constructed birth cohorts do not contain and increased or plateaued in women in the last decades the same age groups. We used the CCMF to maximize of the twentieth century [37]. Low-educated women comparability between birth cohorts that consist of dif- were a few generations behind low-educated men in the ferent age groups. In this method, we compared mortal- cigarette epidemic: when generations of low-educated ity in each cohort with the standard population of the women still showed increases of smoking-attributable same age range. This may have introduced bias in the mortality, the same generations of low-educated men comparison of relative inequality between birth cohorts have already reached the decline stage. Exceptions were of different age ranges because relative inequalities in in Denmark, where low-educated women already mortality tend to be larger at younger ages [30]. The reached the stage of decline in the generation born after CCMF method does not fully adjust for these differences the first quarter of the twentieth century. in age structure. Several studies have examined the trends of smoking- Finally, the coverages of data were different between related mortality in European countries by birth cohort countries. Italy and Spain did not offer national-level and sex using age-period-cohort models [14, 38–40]. datasets, but mortality patterns have been shown to dif- These studies showed a prominent role of cohort effects fer little between these (large) urban regions compared in shaping smoking-related mortality trends, together with the national average [31, 32]. Moreover, there were with little evidence for period effects. These studies only two levels of education in the data from the UK, showed decreasing trends in smoking-attributable mor- resulting in some mid-educated (ISCED 3) populations tality in consecutive cohorts among men and increasing categorized as low-educated. This might have caused trends among women in most countries but did not dif- underestimation of the inequalities of the UK. ferentiate by level of education. Our results show similar The strengths of our study include using a unique set trends for low-educated men in most countries except of data on mortality by education, covering a wide range for Denmark, Estonia, Lithuania, and Spain, and among of time periods and countries. Most mortality data were low-educated women in most countries except for based on individual mortality follow up, avoiding Denmark and Sweden. Our results of high-educated numerator-denominator bias [33]. The approach of the men in some countries and high-educated women in PGW method avoids several limitations with the popula- most counties are not in line with previous studies, tion attributable fraction (PAF) method, which is based showing that population averages may hide important on the prevalence of self-reported smoking and relative educational differences. risks. Firstly, there are possible differences in self-reports Gregoraci et al. [13] compared inequalities in between different surveys [34], between different educa- smoking-attributable mortality between the low and high tion groups [35], and between different countries. Sec- educated in 14 European countries in two time periods ondly, currently available relative risks of smoking- 2000-2004 and 1990-1994, using the PGW-method like related mortality are not specific by countries and level we do, but without applying a cohort perspective. They of education. Using general relative risks may over(un- found increases in relative inequalities and decreases in der)estimate smoking attributed mortality if the propor- absolute inequalities among men, and increases in both tion of non-smoking-related causes in total mortality is relative and absolute inequalities among women. Our re- high (low) in that specific country [36]. Third, the preva- sults are based on a longer time period and we found lence and relative risk used in the PAF approach gener- larger relative inequalities among men and women in ally do not account for the impact of duration and younger generations and increasing absolute inequalities. intensity of smoking on mortality. A final strength is that The difference between their study and ours may be due we focused on birth-cohort trends in smoking- to their period perspective that mixes older generations attributable mortality by education and birth-cohort with lower absolute inequalities together with younger trends in inequalities. The birth cohort perspective en- generations with higher absolute inequalities. ables us to estimate the damage by smoking cumulated Several studies examined cohort patterns in smoking over the life-course and can provide useful insights into initiation and cessation and found rising disparities be- the impact of smoking on future inequalities in tween groups of low and high educated [18, 41, 42], mortality. which lends support to our findings. Interpretation Conclusion The trends in smoking-attributable mortality in consecu- Our study found, in most counties, decreasing smoking- tive generations by educational level for men and attributable mortality trends in consecutive birth cohorts women were, to some extent, consistent with the up- among both low- and high-educated men and among dated cigarette epidemic model, which shows that high-educated women, but increasing smoking- smoking-attributable mortality has decreased in men attributable mortality among low-educated women. For Long et al. Population Health Metrics (2021) 19:3 Page 8 of 9 men, based on these birth cohort-specific patterns, we Funding expect that in most non-Eastern European countries This study was conducted as part of the LIFEPATH project, which has received financial support from the European Commission (Horizon 2020 smoking will become less important as a driver of in- grant number 633666), and as part of the project “Longer life, longer in equalities in total mortality in the future, despite a con- good health, working longer? Implications of educational differences for the tinuing widening of relative inequalities in smoking- pension system,” which has received financial support from Netspar (Network for Studies on Pensions, Aging and Retirement). Data were partly attributable mortality. Among men in Hungary, Estonia, collected as part of the DEMETRIQ project, which also received support from and Lithuania, smoking may well remain an important the European Commission (grant number FP7-CP-FP grant no. 278511). The driver of inequalities in total mortality in the near future. funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Among women, however, smoking threatens to further Pekka Martikainen was funded by the Academy of Finland and MINDMAP, a widen inequalities in total mortality. In order to avoid European Commission HORIZON 2020 research and innovation action grant this, it is necessary to step up efforts to promote smok- 667661. Mall Leinsalu was supported by the Estonian Research Council (grant PRG722). ing prevention and cessation while not increasing health inequalities. Before implementation, interventions Availability of data and materials should be equity-checked and evaluated [43]. Available upon request. Population-level interventions such as price increase to Ethics approval and consent to participate tobacco and sale restriction to minors have the potential Not applicable. to reduce inequalities [44]. Consent for publication Not applicable. Supplementary Information The online version contains supplementary material available at https://doi. Competing interests org/10.1186/s12963-021-00247-2. None. Additional file 1: Supplementary file 1 Overview of mortality data Author details 1 source. Supplementary file 2_a. Detailed birth cohorts constructed by Department of Public Health, Erasmus MC, P.O. Box 2040, 3000 CA country. Supplementary file 2_b. Age composition of each birth Rotterdam, The Netherlands. 2Population Research Unit, Faculty of Social cohort. Supplementary file 3. Calculations of smoking-attributable mor- Sciences, University of Helsinki, Helsinki, Finland. 3Department of Public tality from the Preston-Glei-Wilmoth method. Supplementary file 4 Cal- Health Sciences, Stockholm University, Stockholm, Sweden. 4Department of culations of CCMFs. Supplementary file 5. CCMF results from less strict Public Health, University of Copenhagen, Copenhagen, Denmark. 5 exclusion criteria. Supplementary file 6. Birth cohort-specific trends of Epidemiology, Biostatistics and Prevention Institute, University of Zurich, total mortality. Additional file 8: Supplementary file 7. Table for rate ra- Zurich, Switzerland. 6Department of Clinical Medicine and Biology, University tio, rate difference and the fraction with 95% CI of Turin, Torino, Italy. 7Demographic Research Institute, Budapest, Hungary. 8 Additional file 2. Stockholm Centre for Health and Social Change, Södertörn University, Huddinge, Sweden. 9Department of Epidemiology and Biostatistics, National Institute for Health Development, Tallinn, Estonia. 10Agència de Salut Pública Abbreviations de Barcelona, Barcelona, Spain. 11CIBER Epidemiología y Salud Pública PGW: Preston-glei-wilmoth; ISCED: International standard classification of (CIBERESP), Madrid, Spain. 12Sorbonne Université, INSERM, Institut Pierre Louis education; CCMF: Comparative cohort mortality figures; RR: Rate ratio; d’Epidémiologie et de Santé Publique, Paris, France. RD: Rate difference; CPS-II: Second prospective cancer prevention study; CI: Confidence intervals; PAF: Population attributable fraction; P-Y: Person- Received: 21 September 2020 Accepted: 21 January 2021 years; CS: Cross-sectional Acknowledgements References The permission of the Office for National Statistics to use the Longitudinal 1. M. Marmot et al., “WHO European review of social determinants of health Study is gratefully acknowledged, as is the help provided by the staff of the and the health divide,” Lancet, vol. 380, no. 9846, pp. 1011-1029, Sep 15 Centre for Longitudinal Study Information & User Support (CeLSIUS). CeLSIUS 2012, doi: https://doi.org/10.1016/S0140-6736(12)61228-8. is supported by the ESRC Census of Population Programme (Award Ref: ES/ 2. Eurobarometer S. Attitudes of Europeans towards tobacco and electronic K000365/1). The authors alone are responsible for the interpretation of the cigarettes. TNS Opin Soc. 2015;429:214. data. This work contains statistical data from ONS which is Crown Copyright. 3. U. D. o. H. a. H. Services, “How tobacco smoke causes disease: what it The use of the ONS statistical data in this work does not imply the means to you. Atlanta: Centers for Disease Control and Prevention,” endorsement of the ONS in relation to the interpretation or analysis of the National Center for Chronic Disease Prevention and Health Promotion, statistical data. This work uses research datasets that may not exactly Office on Smoking and Health 2010. reproduce National Statistics aggregates. 4. Eikemo TA, et al. How can inequalities in mortality be reduced? A The mortality data for Switzerland were obtained from the Swiss National quantitative analysis of 6 risk factors in 21 European populations. PLoS One. Cohort, which is based on mortality and census data provided by the 2014;9(11):e110952. https://doi.org/10.1371/journal.pone.0110952. Federal Statistical Office and was supported by the Swiss National Science 5. J. P. Mackenbach et al., “Socioeconomic inequalities in health in 22 Foundation. European countries,” N Engl J Med, vol. 358, no. 23, pp. 2468-2481, Jun 5 We thank Patrick Deboosere, Bjørn Heine Strand, Eva-Maria Asamer, Ramunė 2008, doi: https://doi.org/10.1056/NEJMsa0707519. Kalėdienė, and Bogdan Wojtyniak for providing data for this study. 6. Ostergren O, Martikainen P, Tarkiainen L, Elstad JI, Bronnum-Hansen H. Contribution of smoking and alcohol consumption to income differences in Authors’ contributions life expectancy: evidence using Danish, Finnish, Norwegian and Swedish Material preparation, data collection were performed by all authors, data register data. J Epidemiol Community Health. 2019;73(4):334–9. https://doi. analyses were performed by Di Long. The first draft of the manuscript was org/10.1136/jech-2018-211640. written by Di Long and revised by Wilma Nusselder and Johan Mackenbach. 7. Islami F, Torre LA, Jemal A. Global trends of lung cancer mortality and All authors commented on previous versions of the manuscript. All authors smoking prevalence. Transl Lung Cancer Res. 2015;4(4):327–38. https://doi. read and approved the final manuscript. org/10.3978/j.issn.2218-6751.2015.08.04. Long et al. Population Health Metrics (2021) 19:3 Page 9 of 9 8. Feliu A, et al. Impact of tobacco control policies on smoking prevalence 27. L. Blue and A. Fenelon, “Explaining low mortality among US immigrants and quit ratios in 27 European Union countries from 2006 to 2014. Tob relative to native-born Americans: the role of smoking,” (in eng), Int J Control. 2019;28(1):101–9. https://doi.org/10.1136/tobaccocontrol-2017- Epidemiol, vol. 40, no. 3, pp. 786-793, Jun 2011, doi: dyr011 [pii]https://doi. 054119. org/10.1093/ije/dyr011. 9. So VH, Best C, Currie D, Haw S. Association between tobacco control 28. Clayton D, Schifflers E. Models for temporal variation in cancer rates. I: Age- policies and current smoking across different occupational groups in the EU period and age-cohort models. Stat Med. 1987;6(4):449–67. https://doi.org/ between 2009 and 2017. J Epidemiol Community Health. 2019;73(8):759–67. 10.1002/sim.4780060405. https://doi.org/10.1136/jech-2018-211935. 29. Clayton D, Schifflers E. Models for temporal variation in cancer rates. II: Age- 10. P. Jha, R. Peto, W. Zatonski, J. Boreham, M. J. Jarvis, and A. D. Lopez, “Social period-cohort models. Stat Med. 1987;6(4):469–81. https://doi.org/10.1002/ inequalities in male mortality, and in male mortality from smoking: indirect sim.4780060406. estimation from national death rates in England and Wales, Poland, and 30. M. Huisman et al., “Educational inequalities in cause-specific mortality in North America,” Lancet, vol. 368, no. 9533, pp. 367-370, Jul 29 2006, doi: middle-aged and older men and women in eight western European https://doi.org/10.1016/S0140-6736(06)68975-7. populations,” Lancet, vol. 365, no. 9458, pp. 493-500, Feb 5-11 2005, doi: 11. Stringhini S, et al. Health behaviours, socioeconomic status, and mortality: https://doi.org/10.1016/S0140-6736(05)17867-2. further analyses of the British Whitehall II and the French GAZEL 31. Regidor E, Kunst AE, Rodriguez-Artalejo F, Mackenbach JP. Small socio- prospective cohorts. PLoS Med. 2011;8(2):e1000419. https://doi.org/10.1371/ economic differences in mortality in Spanish older people. Eur J Public journal.pmed.1000419. Health. 2012;22(1):80–5. https://doi.org/10.1093/eurpub/ckr051. 12. M. C. Kulik et al., “Educational inequalities in three smoking-related causes of 32. Marinacci C, et al. Social inequalities in total and cause-specific mortality of death in 18 European populations,” (in eng), Nicotine Tob Res, vol. 16, no. 5, a sample of the Italian population, from 1999 to 2007. Eur J Public Health. pp. 507-518, May 2014, doi: ntt175 [pii]https://doi.org/10.1093/ntr/ntt175. 2013;23(4):582–7. https://doi.org/10.1093/eurpub/cks184. 13. G. Gregoraci et al., “Contribution of smoking to socioeconomic inequalities 33. Valkonen T. Problems in the measurement and international comparisons of in mortality: a study of 14 European countries, 1990-2004,” (in eng), Tob socio-economic differences in mortality. Soc Sci Med. 1993;36(4):409–18. Control, vol. 26, no. 3, pp. 260-268, May 2017, doi: tobaccocontrol-2015- https://doi.org/10.1016/0277-9536(93)90403-q. 052766 [pii]https://doi.org/10.1136/tobaccocontrol-2015-052766. 34. Kulik MC, Eikemo TA, Regidor E, Menvielle G, Mackenbach JP. Does the 14. F. I. Bray and E. Weiderpass, “Lung cancer mortality trends in 36 European pattern of educational inequalities in smoking in Western Europe depend countries: secular trends and birth cohort patterns by sex and region 1970- on the choice of survey? Int J Public Health. 2014;59(4):587–97. https://doi. 2007,” (in eng), Int J Cancer, vol. 126, no. 6, pp. 1454-1466, Mar 15 2010, doi: org/10.1007/s00038-014-0560-z. https://doi.org/10.1002/ijc.24855. 35. Wagenknecht LE, Burke GL, Perkins LL, Haley NJ, Friedman GD. 15. Pampel FC. Diffusion, cohort change, and social patterns of smoking(). Soc Misclassification of smoking status in the CARDIA study: a comparison of Sci Res. 2005;34(1):117–39. https://doi.org/10.1016/j.ssresearch.2003.12.003. self-report with serum cotinine levels. Am J Public Health. 1992;82(1):33–6. 16. Troost JP, et al. An updated global picture of cigarette smoking persistence https://doi.org/10.2105/ajph.82.1.33. among adults. J Epidemiol Glob Health. Sep 2012;2(3):135–44. https://doi. 36. C. o. P. National Research Council. International differences in mortality at org/10.1016/j.jegh.2012.06.003. older ages: dimensions and sources. Washington, D.C.: National Academies 17. Ahacic K, Parker MG, Thorslund M. Aging in disguise: age, period and cohort Press; 2011. effects in mobility and edentulousness over three decades. Eur J Ageing. 37. Thun M, Peto R, Boreham J, Lopez AD. Stages of the cigarette epidemic on Jun 2007;4(2):83–91. https://doi.org/10.1007/s10433-007-0049-1. entering its second century. Tob Control. 2012;21(2):96–101. https://doi.org/ 18. S. Legleye, M. Khlat, F. Beck, and P. Peretti-Watel, “Widening inequalities in 10.1136/tobaccocontrol-2011-050294. smoking initiation and cessation patterns: a cohort and gender analysis in 38. C. La Vecchia, E. Negri, F. Levi, A. Decarli, and P. Boyle, “Cancer mortality in France,” Drug Alcohol Depend, vol. 117, no. 2-3, pp. 233-241, Sep 1 2011, Europe: effects of age, cohort of birth and period of death,”(in eng), Eur J doi: https://doi.org/10.1016/j.drugalcdep.2011.02.004. Cancer, vol. 34, no. 1, pp. 118-141, 1998, doi: S0959804997003353 [pii]. 19. S. H. Preston, D. A. Glei, and J. R. Wilmoth, “A new method for estimating 39. J. Franco, S. Perez-Hoyos, and P. Plaza, “Changes in lung-cancer mortality smoking-attributable mortality in high-income countries,” (in eng), Int J trends in Spain,” Int J Cancer, vol. 97, no. 1, pp. 102-105, Jan 1 2002, doi: Epidemiol, vol. 39, no. 2, pp. 430-438, Apr 2010, doi: dyp360 [pii]https://doi. https://doi.org/10.1002/ijc.1575. org/10.1093/ije/dyp360. 40. K. Ahern, “Hemoglobin’s moving around (to the tune of “Santa Claus is 20. Martikainen P, Makela P, Peltonen R, Myrskyla M. Income differences in life Coming to Town”),” (in eng), Biochem Mol Biol Educ, vol. 35, no. 6, p. 478, expectancy: the changing contribution of harmful consumption of alcohol 2007, doi: https://doi.org/10.1002/bmb.118. and smoking. Epidemiology. 2014;25(2):182–90. https://doi.org/10.1097/EDE. 41. Federico B, Costa G, Kunst AE. Educational inequalities in initiation, 0000000000000064. cessation, and prevalence of smoking among 3 Italian birth cohorts. Am J Public Health. 2007;97(5):838–45. https://doi.org/10.2105/AJPH.2005.067082. 21. Gardner MJ, Osmond C. Interpretation of time trends in disease rates in the 42. Rohrmann S, Becker N, Kroke A, Boeing H. Trends in cigarette smoking in presence of generation effects. Stat Med. 1984;3(2):113–30. https://doi.org/ the German centers of the European Prospective Investigation into Cancer 10.1002/sim.4780030205. and Nutrition (EPIC): the influence of the educational level. Prev Med. 2003; 22. D. R. Shopland, D. M. Burns, and National Cancer Institute (U.S.). Smoking 36(4):448–54. https://doi.org/10.1016/s0091-7435(02)00047-6. and Tobacco Control Program., Changes in cigarette-related disease risks 43. C. Bambra, “First do no harm: developing interventions that combat and their implication for prevention and control (Smoking and tobacco addiction without increasing inequalities,” Addiction, vol. 113, no. 5, pp. control monograph, no. 8). Bethesda, Md.: National Institutes of Health, 787-788, 2018, doi: https://doi.org/https://doi.org/10.1111/add.14116. National Cancer Institute, 1997, pp. xxxvii, 565 p. 44. Thomas S, et al. Population tobacco control interventions and their effects 23. Kromhout D. Epidemiology of cardiovascular diseases in Europe. Public on social inequalities in smoking: systematic review. Tob Control. 2008;17(4): Health Nutr. 2001;4(2B):441–57. https://doi.org/10.1079/phn2001133. 230–7. https://doi.org/10.1136/tc.2007.023911. 24. M. Wensink, J. A. Alvarez, S. Rizzi, F. Janssen, and R. Lindahl-Jacobsen, “Progression of the smoking epidemic in high-income regions and its effects on male-female survival differences: a cohort-by-age analysis of 17 Publisher’s Note countries,” BMC Public Health, vol. 20, no. 1, p. 39, Jan 10 2020, doi: https:// Springer Nature remains neutral with regard to jurisdictional claims in doi.org/10.1186/s12889-020-8148-4. published maps and institutional affiliations. 25. Hovanec J, et al. Lung cancer and socioeconomic status in a pooled analysis of case-control studies. PLoS One. 2018;13(2):e0192999. https://doi. org/10.1371/journal.pone.0192999. 26. Martikainen P, Ho JY, Preston S, Elo IT. The changing contribution of smoking to educational differences in life expectancy: indirect estimates for Finnish men and women from 1971 to 2010. J Epidemiol Community Health. Mar 2013;67(3):219–24. https://doi.org/10.1136/jech-2012-201266.