Article Text

Download PDFPDF

The potential for reducing differences in life expectancy between educational groups in five European countries: the effects of obesity, physical inactivity and smoking
  1. Netta E Mäki1,
  2. Pekka T Martikainen1,
  3. Terje Eikemo2,3,
  4. Gwenn Menvielle4,5,
  5. Olle Lundberg6,7,
  6. Olof Östergren6,
  7. Johan P Mackenbach2,
  8. the EURO-GBD-SE consortium members
  1. 1Department of Social Research, University of Helsinki, Finland
  2. 2Department of Public Health, Erasmus MC, University Medical Center Rotterdam, Rotterdam, Netherlands
  3. 3Department of Sociology and Political Science, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
  4. 4Department of Social Epidemiology, INSERM, UMR_S 1136, Pierre Louis Institute of Epidemiology and Public Health, Paris, France
  5. 5Department of Social Epidemiology, Sorbonne Universités, UPMC Univ Paris 06, UMR_S 1136, Pierre Louis Institute of Epidemiology and Public Health, Paris, France
  6. 6CHESS, Centre for Health Equity Studies, Sweden
  7. 7Department of Health Sciences, Mid Sweden University, Sweden
  1. Correspondence to Dr Netta E Mäki, Department of Social Research, Faculty of Social Sciences, P.O.Box 59, University of Helsinki, Helsinki, FIN-00014, Finland; netta.maki{at}


Introduction This study assesses the effects of obesity, physical inactivity and smoking on life expectancy (LE) differences between educational groups in five European countries in the early 2000s.

Methods We estimate the contribution of risk factors on LE differences between educational groups using the observed risk factor distributions and under a hypothetically more optimal risk factor distribution. Data on risk factor prevalence were obtained from the Survey of Health, Ageing and Retirement in Europe study, and data on mortality from census-linked data sets for the age between 50 and 79 according to sex and education.

Results Substantial differences in LE of up to 2.8 years emerged between men with a low and a high level of education in Denmark, Austria and France, and smaller differences among men in Italy and Spain. The educational differences in LE were not as large among women. The largest potential for reducing educational differences was in Denmark (25% among men and 41% among women) and Italy (14% among men).

Conclusions The magnitude of the effect of unhealthy behaviours on educational differences in LE varied between countries. LE among those with a low or medium level of education could increase in some European countries if the behavioural risk factor distributions were similar to those observed among the highly educated.

  • Inequalities
  • Health Behaviour
  • Public Health

Statistics from


Higher mortality among those with a lower socioeconomic status is well established. The mechanisms behind this association are not fully understood, but they are commonly thought to be partly due to unhealthy behaviour.1 ,2 Health behaviour is strongly socially patterned, but there is large variation in the extent to which socioeconomic differences in mortality are mediated by corresponding behavioural disparities in health behaviour.3–5 However, interpretation and comparison of the results are difficult because of the wide variation in study design. International comparisons concerning the effects of health behaviours on socioeconomic differences in mortality based on more recent data are few, and most studies include one specific behavioural risk factor at a time, only cover specific nationally non-representative cohorts or vary in the indicator used to measure socioeconomic status.

Several studies have shown that much of the substantial social inequalities in mortality in many Central and Northern European countries and in North America are attributable to the effects of smoking,6 ,7 but that in Southern Europe inequality in smoking-related mortality tends to smaller, or even reversed.8 Stringhini et al9 compared employed cohorts in Great Britain and France, and found that even though four types of health behaviour were strong predictors of mortality in both countries, their association with socioeconomic status was more extensive in Britain mainly because of the stronger social patterning of unhealthy behaviour.

Comparisons of representative samples from different countries are few, due mainly to the lack of data, and this constitutes a limitation in terms of enhancing understanding of the mechanisms linking socioeconomic status, health behaviour and mortality. Even though the association between health-related behaviour and mortality is likely to be similar, the distribution as well as the contribution of different types of behaviour to socioeconomic differences in mortality may vary between countries. Uniform data sets are needed to distinguish to what extent this is due to differences in study design and to what extent to differences in social setting.

Further comparative research would clarify whether health behaviours are equally important mediators of the association between the markers of socioeconomic status—measured most commonly in terms of education, occupation-based social class or income—and mortality in different cultural and social contexts. We assessed the effect of three essential risk factors—high body mass index (BMI), physical inactivity and smoking—on life expectancy (LE) differences between educational groups among men and women in five European countries. The main aim was to estimate the magnitude of socioeconomic inequalities in LE, the effect of the three behavioural risk factors on these inequalities and the generalisability of these findings across countries. In addition, we calculated an alternative, hypothetical scenario LE to those with a low or medium level of education based on a more optimal distribution of risk factors as observed in the highly educated group. These alternative LE estimates can be interpreted as an estimate of the potential increase in LE among those with a low or medium level of education.


We based the sex-specific and education-specific LE calculations on census-linked mortality data sets collected for the recent EURO-GBD-SE project10 comparing health inequalities between European countries. These data covered several years from the early 2000s (from 2001–2002 in Austria to 1999–2005 in France). For Italy and Spain, census-linked mortality data were not available for the whole country but only for a few areas, namely, Turin and Tuscany in Italy and Madrid and Barcelona in Spain. Prevalence data on behavioural risk factors were obtained from the Survey of Health, Ageing and Retirement in Europe (SHARE) in 2004.11 We were thus able to incorporate countries that took part in both of these projects. However, in order to ensure the reliability of the results, we only included countries in which the response rate in the survey exceeded 50%. In sum, the following five countries representing different parts of Europe were included: Denmark, Austria, France, Italy and Spain. Table 1 shows the numbers of person-years and deaths. (See also Appendix.)

Table 1

Distribution (%) of education by sex and country, number of person years and deaths by sex and country, EURO-GBD-SE project database

We calculated partial LE between the ages of 50 and 79 indicating how many years men and women with different educational attainments could expect to live under the current mortality regime,12 the theoretical maximum being 30. The upper age limit we chose corresponded to age spans of special interest, and allowed the exclusion of age groups on which there was unreliable educational data. We calculated partial LE for two reasons. First, the coverage of information on education varies between countries among those aged 80 and over. The proportion of observations with missing values is higher, the older the age group. Second, the proportion of persons in institutional care increases notably at older ages, and as the SHARE data only include the household population the results would have been biased among the oldest.

We categorised education into three groups: those with a low (primary and lower-secondary) level, those with a medium (upper-secondary) level and those educated to the tertiary level (table 1). Both the EURO-GBD-SE project and the SHARE survey pay particular attention to the harmonisation of their data collection and variables.

We used the population attributable fraction (PAF) to quantify the contribution of risk factors to LE differences between educational groups. The PAF is the proportional reduction in mortality that would occur if exposure to risk factors were reduced to an alternative scenario. Its calculation required—in addition to the behavioural risk factor distributions—the mortality rate ratios for the risk factor levels, which we obtained from published meta-analyses and an extensive prospective study. We acquired the BMI mortality rate ratios from McGee13: 1.00 for normal weight (BMI 18.5–24.9), 0.965 and 0.968 for overweight (BMI 25.0–29.9) in men and women, respectively, and 1.201 and 1.275 among obese (BMI≥30) men and women, respectively. In order to avoid reverse causality bias (very low BMI attributable to severe disease), we excluded those with a BMI of less than 18.5 (1.1% on average). The American Cancer Society14 provided the following risk ratios for smoking: 1.00 for never smokers, 1.35 and 1.23 for male and female previous smokers, respectively, and 2.07 and 1.74, for currently smoking men and women, respectively. Finally, we obtained the risk ratios for self-reported physical activity among men and women together from Nocon et al15: 1.00 for the inactive and 0.71 for the active. In line with Nocon et al, we categorised physical activity into two groups: we defined those who reported exercising vigorously several times a week as active, and others as inactive.

We calculated the PAF as shown in the Global Burden of Disease Project16:Embedded Image

where n is the number of exposure categories, Pi is the proportion of the population currently in the ith exposure category, P′i is the proportion of the population in the ith exposure category in alternative exposure scenarios and RRi is the rate ratio of risk-specific mortality for the ith exposure category.

For reasons of multicausality, it is not possible to combine the PAFs obtained for the different risk factors by simple addition, therefore we used the formula developed by Gakidou et al.17Embedded Image

As a starting point for calculating the scenario LE, we set the three risk factors for all educational groups at the level observed for the most highly educated separately in each country and sex group. However, the most highly educated did not have the optimal risk factor level in all country-specific and sex-specific groups, in which case we did not change the observed risk factor level and thus the risk factor did not contribute to the scenario LE. We considered it more interesting to calculate the maximum extent of educational difference in LE that the behavioural risk factors contributed, but a significant finding in itself was that those with the lowest level of education had, in some cases, the most optimal risk factor distribution.


The BMI distribution closest to the optimal was observed among the most highly educated in all cases except for Spanish men and Italian women, where it was among the middle educational group (table 2). Among the men, the highly educated were the most physically active only in Denmark, whereas in Austria and France, those in the middle group were the most active. Physical activity among women was most common among those with the highest education in all countries except Spain, where those in the middle group were the most active. Among men, the highly educated smoked the least in three countries, Denmark, Italy and Spain. Among women those with a high education smoked the least only in Denmark. Otherwise even reverse associations were seen: smoking was notably uncommon among women in the low education group in the Southern European countries: for example, only 8% of Spanish women in this educational category smoked.

Table 2

Distributions (%) of BMI, physical activity and smoking status by sex, country and education, SHARE database

Panel (a) in table 3 shows the current LE among those with the highest (column 1) and lowest (column 2) levels of education and the scenario LE for the latter group (column 3). There were large educational differences in the current LE, up to 2.8 years among men in Denmark, Austria and France (column 4), and smaller differences in Italy and Spain, for example, 1.4 years in Spain. The absolute educational differences in LE were not as large among the women except in Denmark (1.7 years). Column 5 shows the absolute change in LE among those with the lowest level of education in terms of the scenario exposure to the risk factors, and column 6 shows the corresponding relative change in LE. The changes in LE were similar among both sexes (0.6–0.7 years at the most), but partly because the current difference in LE between the educational groups was smaller among women, they showed the largest relative changes. The reduction between LE and scenario LE among the women was as much as 41% in Denmark, and almost 30% in Spain, and even at its smallest the change would be 18% in Austria. The corresponding change between LE and scenario LE among the men was the largest in Denmark (25%) and Italy (14%), and the smallest in France and Austria (3 and 4%, respectively).

Table 3

Partial LE and scenario partial LE among those with a high and low (panel a) or medium (panel b) level of education; 50–79-year-old men and women in five European countries

Panel (b) in table 3 gives the equivalent results for those in the middle educational group. The level of absolute differences was much lower, for example, 1.7 years at most among Austrian men, and the change in LE was smaller (0.3 years at most among Danish men and women), but the relative results were very similar to those in the lowest group. Again, under the hypothetical scenario, the highest potential for reducing the LE differences between educational groups was among women, and the same countries would gain the most (Denmark and Spain for women and Denmark and Italy for men). The proportion of the relative reduction was also very similar (eg, 41% among Danish women and 4% among Austrian men).


We found large differences in LE between educational groups. However, the health behaviours studied were not equally important mediators of the association between education and mortality in different European countries. Furthermore, higher education was not consistently related to the optimal health behaviour pattern.

In some countries, especially Denmark, LE differences between educational groups could potentially be reduced significantly under the scenario of the risk factor distribution as among the highly educated in cases in which those with the highest education indeed had a closer-to-optimal risk factor prevalence. On the other hand, in Austria, France and, to some extent, Italy, the effect seemed rather small. This result is congruent with findings from a previous study comparing British and French employed cohorts and showing that health behaviours are likely to be major contributors to mortality differences between socioeconomic groups only in contexts in which they are strongly socially patterned.9 The current results covering several European countries support the idea that health behaviour may be more closely related to cultural and lifestyle factors not strongly associated with socioeconomic status in Central and Southern Europe, whereas health behaviour and socioeconomic status are more closely linked in Northern European countries.

All three risk factors examined in this study had at least some effect among Danish men and women, but only physical activity among women in Italy, for example, and smoking among men and obesity among women in Spain were of importance.

The BMI distribution was the closest to optimal among the highly educated in most countries. Educational differences in BMI seemed to be larger among women than men, which is consistent with earlier studies reporting larger inequalities in women than in men in Europe.8 ,18 However, given that the differences in mortality rate ratios between the BMI groups were not large,13 the effect of obesity on the potential for reducing LE differences between educational groups was limited, except among Spanish women. Nonetheless, if the obesity epidemic continues to spread and educational differences remain or widen, LE inequalities are also likely to increase.

Likewise, the effect of physical activity in reducing LE differences was rather small in most countries. Our country-specific results on the distribution of physical activity were inconsistent. For example, among those with the highest level of education, the women were physically the most active in four countries, and the men in only one country. However, the effects of physical activity were roughly the same as that of smoking in Denmark, which was mainly because of the particularly large physical activity differences by education in the country. A previous review study19 covering several European countries, North America and Australia reported a higher prevalence of physical activity among those in the top socioeconomic strata but did not find a systematic socioeconomic gradient otherwise. The seemingly contradictory findings in our study and in Gidlow's19 review article may reflect differences in the definition of physical activity in the various studies. According to Gidlow, leisure time physical activity in particular tends to be higher in the more advantaged socioeconomic groups, with narrower differentials for overall physical activity. The SHARE survey, in turn, also includes heavy housework in physical activity and categorises jobs involving physical labour as vigorous physical activity.

Among men, those with the highest level of education were the closest to optimal smoking distribution in three countries, even though the differences were not large in Spain and Italy. As far as the women were concerned, there was a reverse association between education and smoking in all countries except Denmark. Smoking was not very common among females in this age group in Southern Europe, but was much more prevalent among those with a higher level of education. This result of small socioeconomic differences in smoking, or even reverse associations, in the Southern European countries is in line with findings from a study covering an earlier period around 1990.20 These observations are consistent with evidence of the differential progression of the smoking epidemic21 in Europe: the model suggests that without strong antismoking measures during the earlier stages of the epidemic the inequalities would probably be reversed in Southern European countries, increasing the educational differences in LE especially among women. Pricing policies have been given as an example of antismoking measures that could have an effect on the association between smoking and mortality in low socioeconomic groups.20

Overall, the educational differences in LE observed in this study were not as large among women as among men. This finding may be partly attributable to the fact that the variation in educational attainment among those aged 50–79 years was lower for women than for men, and the proportion with tertiary education was particularly low, especially among women in Spain and Italy. However, larger socioeconomic differences have been consistently observed for other indicators with smaller or no distributional differences between the sexes, such as income percentiles. A large part of the gender difference may be attributable to differences in cause of death distributions, men being more likely to die from causes exhibiting large socioeconomic differences (such as accidents and violence, and cardiovascular diseases).22

Methodologically this study has both strengths and limitations. In order to avoid data heterogeneity, we used data from two projects—the EURO-GBD-SE and SHARE—both of which cover representative population samples and pay attention to data and variable harmonisation between countries. However, the study design was cross sectional. The measurement of both health behaviour and mortality covered the same time period, although in reality there is a time delay between health behaviour and both morbidity and mortality. Cross-sectional measurements could have biased our results when there were notable changes in pre-baseline health behaviour. For example, as the smoking epidemic has progressed and there is growing evidence in many countries of a greater decline in smoking especially among the most highly educated men, the estimated educational differences in this study are likely to be underestimates.

We also acknowledge the potential unreliability of self-reported health behaviour. Self-reporting is a potential source of bias, which is discussed more thoroughly elsewhere.23 However, according to a review study, both gender and educational level seem to have little effect on accuracy.24 Given that the focus of our paper is on educational differences in mortality, any biases are likely to lead to an underestimation of the contribution of the risk factors.


The magnitude of the effect of unhealthy behaviours on LE differences between educational groups seems to vary according to the social patterning of these behaviours, and the results obtained in one country are not generalisable to other contexts. There is potential in most of Europe to reduce educational differences in LE through behaviour modification, but different risk factors might need to be targeted in the different countries.

What is already known on this subject

  • Higher mortality among those with a lower socioeconomic status is commonly partly attributed to unhealthy behaviour.

  • However, most studies include one specific behavioural risk factor at a time, or only cover one country.

What this study adds

  • We assessed the effect of three specific risk factors on life expectancy (LE) differences between educational groups among men and women in five European countries.

  • The magnitude of the effects of obesity, physical inactivity and smoking on educational differences in LE varied between countries from a few per cent in Austria and France to 40 per cent in Denmark.

  • In most European countries, there is potential for reducing educational differences in LE through behaviour modification.


This paper uses data from SHARELIFE release 1, as of 24 November  2010, or SHARE release 2.5.0, as of 24 May 2011. The SHARE data collection has been primarily funded by the European Commission through the 5th framework programme (project QLK6-CT-2001-00360 in the thematic programme Quality of Life), through the 6th framework programme (projects SHARE-I3, RII-CT-2006-062193, COMPARE, CIT5-CT-2005-028857 and SHARELIFE, CIT4-CT-2006-028812) and through the 7th framework programme (SHARE-PREP, 211909 and SHARE-LEAP, 227822). Additional funding from the US National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, Y1-AG-4553-01 and OGHA 04-064, IAG BSR06-11, R21 AG025169) as well as from various national sources is gratefully acknowledged (see for a full list of funding institutions).


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Collaborators The EURO-GBD-SE consortium members.

  • Contributors All authors have contributed to planning, conducting and reporting of the results in this article. NEM is the guarantor.

  • Funding This project has been partly funded by the European Commission through the Public Health Programme, grant agreement 20081309. NEM is funded by the Emil Aaltonen foundation and PTM by the Academy of Finland.

  • Competing interests None.

  • Ethics approval SHARE/EURO-GBD-SE: respective national ethics/data safety authorities.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.