OUP user menu

Health behaviours as explanations for educational level differences in cardiovascular and all-cause mortality: a follow-up of 60 000 men and women over 23 years

Mikko Laaksonen, Kirsi Talala, Tuija Martelin, Ossi Rahkonen, Eva Roos, Satu Helakorpi, Tiina Laatikainen, Ritva Prättälä
DOI: http://dx.doi.org/10.1093/eurpub/ckm051 38-43 First published online: 14 June 2007

Abstract

Background: Health behaviours are potential explanatory factors for socioeconomic differences in mortality. We examined the extent to which seven health behaviours covering dietary habits, smoking and physical avtivity, can account for relative differences in cardiovascular and all-cause mortality by educational level. Methods: Health behaviour data derived from nationwide Finnish health behaviour surveys from the years 1979 to 2001. These annually repeated cross-sectional surveys were linked to register-based information on educational level and subsequent mortality from the year of the survey until the end of 2001 (average follow-up time 11.9 years). The analyses included 29 065 men and 31 543 women of whom 4263 died. Cardiovascular disease (CVD), coronary heart disease (CHD), stroke and all-cause mortality was studied. Results: Educational level showed a graded association with all mortality outcomes. Health behaviours explained 54% of the relative difference between primary and higher educational level in CVD mortality among in men and 22% among in women. For all-cause mortality the corresponding figures were 45 and 38%. Smoking, vegetable use and physical activity were the most important health behaviours explaining educational level differences in all mortality outcomes, while the effects of type of fat used on bread, coffee drinking, relative weight and alcohol use were small. Conclusions: Smoking, low vegetable use and physical inactivity explained a substantial part of educational level differences in cardiovascular and all-cause mortality among men and women. Socioeconomic trends in these behaviours are of crucial importance in determining whether socioeconomic mortality differences will widen or narrow in the future.

  • cardiovascular diseases
  • health behaviours
  • mortality
  • socioeconomic factors

Introduction

Health behaviours have been ranked as one of the main explanations for socioeconomic differences in health since the 1980s.1,,2 Many unhealthy behaviours like smoking, poor dietary habits and physical inactivity tend to be more prevalent in lower socioeconomic groups, which may explain why people in these groups experience higher rates of morbidity and mortality than those in the higher groups. Reasons for socioeconomic differences in health behaviours are complex and vary between behaviours. The contributing factors include differences in knowledge and skills, cultural norms and material factors.3,,4 Health behaviours have been examined especially in the context of cardiovascular diseases (CVD), which have several well-known behavioural risk factors. In the western countries, cardiovascular deaths cover a large part of all mortality cases and the share of some diseases like stroke is expected to even increase in the future.5

Several previous studies have examined the role of health behaviours as explanations for socioeconomic differences in cardiovascular and all-cause mortality.6–13 There is much variation in how the studies have been conducted and what their results show. The range of health behaviours has varied and many studies have included only a limited number of behaviours. Typically, health behaviours have explained 30–50% of socioeconomic differences in mortality, but the variation between studies has been much larger than that. Some studies have found smoking to be the most important health behaviour explaining socioeconomic differences in mortality,10,,13 while other studies have found physical activity to be the most important.8,,9 It has been a common practice to examine health behaviours in combination with biological risk factors like blood pressure or serum cholesterol. Such biological risk factors are influenced by one's behaviours and may be intervened by behavioural modifications. However, health behaviours are causally preceding to such risk markers. In terms of prevention, it is important to separate biological risk markers from health behaviours that are further from the disease but closer to interventions.

The aim of this study was to examine the effect of health behaviours on relative differences in cardiovascular and all-cause mortality by educational level. We used a series of nationwide Finnish health surveys with prospective linkage to national mortality register. The study is based on a large representative sample, long follow-up period, and a large number of deaths which enables firm analysis of the main causes of death and allows stroke to be analysed separately. Seven health behaviours were chosen for the analyses on the basis of earlier evidence of their contribution to socioeconomic mortality differences or because of their potential role as a risk factor for mortality, especially from cardiovascular diseases.14,,15 Because the prevalence of unhealthy behaviours and their socioeconomic distribution may differ between men and women, and also other, competitive explanations for socioeconomic health inequalities show gender differences, men and women were analysed separately.

Methods

Data on health behaviours came from annual nationwide health behaviour surveys by the National Public Health Institute.16 Postal questionnaires were mailed to a random sample of 5000 permanent residents of Finland aged 15–64 years. In this study, data from the years 1979–2001 was used and the sample was restricted to 25–64-year-old respondents. The response rate decreased over the study period from 83% to 62% among men and from 86% to 77% among women.16,,17 After excluding persons with missing data in any of the health behaviours (8.4%) the data comprised 29 065 men and 31 543 women.

Information on the respondents’ educational level and mortality was linked to the health behaviour survey data at the Statistics Finland. The linkage was done by personal identification codes, assigned to all persons living permanently in Finland. The Register of Completed Education and Degrees was used to divide educational level into three broad categories: primary school or education unknown (primary level), lower or upper secondary education (secondary level) and lower tertiary education or university degree (higher level).18

Mortality data came from the National Causes of Death Register. For each participant, the mortality follow-up stretched from the health behaviour survey year until the end of 2001. The year 1985 was excluded, however, as the personal identification codes were missing from the survey data that year. The mean follow-up time was for 11.9 years and the total number of deaths was 4263. Mortality was examined in the following partly overlapping categories: CVD (cardiovascular diseases) (ICD-9 390-459), CHD (coronary heart disease) (ICD-9 410-414), stroke (ICD-9 430-438) and all-cause mortality.

Seven health behaviours derived from the questionnaire (for details see www.iph.fgov.be/hishes) were classified as follows. Smoking status was divided into four categories on the basis of past and present smoking habits: never smokers, former smokers, current occasional smokers and current daily smokers. Alcohol use was estimated by calculating the sum of reported weekly drinks of beer, cider, wine and spirits. Alcohol use was then classified into three groups separating abstainers and dividing the others into two groups, using the cut-off point of 10 weekly drinks for women and 20 weekly drinks for men. The frequency of leisure time physical activity was asked with a single question. Response categories were 4–7, 2–3, one or less occasions per week and no exercise other than exceptionally. Vegetable use was measured with a question: ‘on how many days did you eat vegetables or root vegetables (potatoes excluded) during the last week either as such or as in a salad?’ The response alternatives were: ‘none’, ‘1–2 days’, ‘3–5 days’, ‘6–7 days’. The type of fat usually used on bread was included as a proxy for saturated fat use in daily diet with the following categories: no spread at all, low fat spread, vegetable margarine, butter or mixture of butter and oil. Usual coffee drinking was divided into the following categories: no coffee, 1–2 cups, 3–4 cups, 5–7 cups or 8 cups or more daily. Body mass index (BMI) was calculated from self-reported weight and height, and classified into four categories: <20 kg/m2, 20–24.9 kg/m2, 25–29.9 kg/m2, and ≥30 kg/m2.

Statistical methods

The analyses were conducted using Cox proportional hazards models. Hazard ratios (HR) with 95% confidence intervals (CI) of the primary and secondary educational levels were calculated relative to the higher educational group. To take into account the strongly increasing mortality by age we adjusted for age and age squared. Secular trend was taken into account by adjusting for study year. Since people may change their behaviours as they get ill we also adjusted for four pre-existing chronic diseases (diabetes, myocardial infarction, coronary disease and heart failure) by calculating summary index and including it in the model as a covariate. We first estimated base models showing relative differences in mortality outcomes by educational level adjusting for the confounders. We then, further adjusted for each of the health behaviours separately and finally for all of them simultaneously. Percentage reduction of the HR due to adjustments was calculated as: [(unadjusted HR−adjusted HR)/(unadjusted HR−1)] × 100. The reduction in the hazard ratio was interpreted to tell how much of the association between education and mortality was accounted for by the health behaviours. The analyses were carried out separately for men and women using SPSS version 14.0.

Results

Table 1 presents characteristics of the follow-up data in the categories of educational level. There were altogether 2727 male and 1526 female deaths during the follow-up. Forty percent of these deaths were due to CVD, a great majority of which was covered by CHD and stroke. Educational level was associated with mortality also when adjusting for age, as shown in the lower part of the table.

View this table:
Table 1

Characteristics of the follow-up data by educational level in men and women

MenWomen
HigherSecondaryPrimaryAllHigherSecondaryPrimaryAll
N755610 00611 50329 065871410 65412 17531 543
PY at follow-up84 782112 444151 177348 40289 070118 419164 724372 213
N deaths (CVD)15220482811842970454553
N deaths (CHD)1001365708061238263313
N deaths (stroke)27341422031124115150
N deaths (all causes)3555221850272715728810811526
Mean FU time (years)11.211.213.112.010.211.113.511.8
CVD/1000 PY*2.22.84.13.40.60.91.81.5
CHD/1000 PY*1.51.92.82.30.20.51.10.8
Stroke/1000 PY*0.40.50.70.60.30.30.40.4
All causes/1000 PY*5.96.49.67.82.83.44.74.1
  • *Mortality by cause of death per 1000 person years, adjusted for age

We next examined the associations of the health behaviours with all mortality outcomes (Supplementary web-only Table W1). Briefly, after adjusting for confounders, smoking was strongly associated with all mortality outcomes in both men and women. High alcohol use was associated with increased mortality from CVD, all causes, as well as CHD among women. Physical activity showed a curvilinear association with many causes of death, but among women mortality from CVD and CHD was especially high for the never physically active. Vegetable use less than twice a week was consistently associated with all mortality outcomes among men and women. Instead, type of fat used on bread was not associated with any of the causes of death. Men who did not drink coffee had lower mortality in CVD and CHD. Underweight increased all-cause mortality in both men and women. Obesity (BMI 30+) but not moderate overweight was associated with mortality from all causes of death except stroke. Furthermore, another set of analyses showed that those with higher education in general tended to follow healthier choices but the differences were not always large (Supplementary web-only Table W2).

Table 2 shows how adjusting for health behaviours affected the relative differences by educational level in CVD mortality among men and women. Among men with primary education the risk for CVD mortality was nearly 50% higher and among men with secondary education a third higher as compared to those with higher education. Adjusting for smoking accounted for 30% of the excess mortality in the primary education group and somewhat less among those with secondary education. Adjusting for vegetable use had almost equally strong effect. Adjusting for physical activity also attenuated the differences between educational groups, whereas the effects of the other health behaviours were nearly non-existent. Altogether, adjusting for all seven health behaviours accounted for about one half of the association between educational level and CVD mortality among men.

View this table:
Table 2

The effect of adjusting for behavioural factors on educational level differences in CVD mortality, men and women. *Hazard ratios (95% CIs) and percent of reduction (%) in CVD mortality among those with primary or secondary education as compare to higher education after adjustments of health behaviours

MenWomen
HigherSecondaryPrimaryHigherSecondaryPrimary
(ref.)HR (95% CI)(%)HR (95% CI)(%)(ref.)HR (95% CI)(%)HR (95% CI)(%)
1:confounders11.30 (1.05–1.60)1.46 (1.22–1.74)11.51 (0.98–2.33)2.16 (1.48–3.17)
1 + smoking11.23 (0.99–1.52)−231.32 (1.11–1.58)−3011.50 (0.97–2.31)−22.09 (1.42–3.06)−6
1 + alcohol use11.31 (1.06–1.61)31.47 (1.23–1.76)211.53 (0.99–2.36)42.20 (1.50–3.23)3
1 + physical activity11.26 (1.02–1.55)−131.39 (1.16–1.66)−1511.47 (0.95–2.26)−82.08 (1.42–3.05)−7
1 + vegetable use11.23 (0.99–1.52)−231.34 (1.12–1.60)−2611.48 (0.96–2.28)−62.06 (1.40–3.04)−9
1 + fat on bread11.29 (1.04–1.59)−31.44 (1.21–1.72)−411.51 (0.98–2.32)02.15 (1.47–3.15)−1
1 + coffee drinking11.30 (1.05–1.60)01.45 (1.21–1.73)−211.50 (0.97–2.31)−22.13 (1.45–3.12)−3
1 + relative weight11.28 (1.04–1.58)−71.43 (1.19–1.70)−711.48 (0.96–2.28)−62.12 (1.45–3.12)−3
1 + all behaviours11.16 (0.94–1.44)−471.21 (1.01–1.45)−5411.40 (0.91–2.17)−221.90 (1.29–2.81)−22
  • *Adjusted for age, age squared, study year and four pre-existing chronic diseases

Women with primary education had double the risk for CVD mortality and those with secondary education 50% excess mortality compared to women with higher education. Also among smoking women, vegetable use and physical activity accounted for some of the association between educational level and CVD mortality, while the effects of other health behaviours were negligible. Altogether, adjusting for all health behaviours accounted for a fifth of the association between educational level and CVD mortality among women.

Among men, the relative differences in CHD and stroke mortality by educational level were identical to those for CVD. In addition, health behaviours explained a similar share of the associations than for all CVD. Among women, the results for CHD and stroke were slightly different (Table 3). The relative differences between educational level groups in CHD mortality were slightly larger than those in all CVD. Smoking, vegetable use, physical activity and relative weight explained a small part of the differences, while alcohol use slightly increased the differences. The net explanation by all health behaviours together was 14% in the primary education group. For stroke the baseline association tended to be weaker than for all CVD, but more of the association was explained by the health behaviours. Physical activity and vegetable use were the most important explanatory factors while relative weight slightly increased the differences. Altogether the seven health behaviours explained a third of the excess mortality in the primary education group and a fourth among those with secondary education.

View this table:
Table 3

The effect of adjusting for behavioural factors on educational level differences in CHD and stroke mortality in women. *Hazard ratios (95% CIs) and percent of reduction (%) in CHD and Stroke mortality among those with primary or secondary education as compare to higher education after adjustments of health behaviours

CHDStroke
HigherSecondaryPrimaryHigherSecondaryPrimary
(ref.)HR (95% CI)(%)HR (95% CI)(%)(ref.)HR (95% CI)(%)HR (95% CI)(%)
1:confounders11.93 (1.01–3.70)2.75 (1.53–4.94)11.40 (0.69–2.87)1.59 (0.84–2.99)
1 + smoking11.93 (1.01–3.70)02.67 (1.48–4.79)−511.39 (0.68–2.84)−31.54 (0.82–2.90)−8
1 + alcohol use11.99 (1.04–3.82)62.84 (1.58–5.12)511.39 (0.68–2.85)−31.57 (0.83–2.96)−3
1 + physical activity11.88 (0.98–3.60)−52.67 (1.49–4.80)−511.37 (0.67–2.79)−71.50 (0.79–2.82)−15
1 + vegetable use11.89 (0.99–3.63)−42.62 (1.45–4.73)−711.38 (0.67–2.82)−51.50 (0.79–2.85)−15
1 + fat on bread11.95 (1.02–3.73)22.76 (1.54–4.97)111.38 (0.67–2.81)−51.55 (0.82–2.93)−7
1 + coffee drinking11.93 (1.01–3.70)02.70 (1.50–4.87)−311.38 (0.67–2.82)−51.55 (0.82–2.93)−7
1 + relative weight11.87 (0.98–3.59)−62.66 (1.48–4.78)−511.43 (0.70–2.92)81.63 (0.86–3.09)7
1 + all behaviours11.84 (0.96–3.54)−102.50 (1.38–4.53)−1411.30 (0.63–2.69)−251.37 (0.71–2.63)−37
  • *Adjusted for age, age squared, study year and four pre-existing chronic diseases

Among men, relative differences by educational level in all-cause mortality were somewhat larger than those in CVD (Table 4). Among women, the differences were considerably weaker. Smoking, vegetable use and to a lesser degree physical activity were important explanatory factors for the differences in both men and women. Adjusting for alcohol use increased the differences. Altogether the health behaviours explained nearly half of the excess mortality in primary educated men and more than a third in primary educated women.

View this table:
Table 4

The effect of adjusting for behavioural factors on educational level differences in all cause mortality, men and women. *Hazard ratios (95% CIs) and percent of reduction (%) in all cause mortality among those with primary or secondary education as compare to higher education after adjustments of health behaviours

MenWomen
HigherSecondaryPrimaryHigherSecondaryPrimary
(ref.)HR (95% CI)(%)HR (95% CI)(%)(ref.)HR (95% CI)(%)HR (95% CI)(%)
1:confounders11.36 (1.19–1.56)1.64 (1.46–1.84)11.22 (1.01–1.49)1.32 (1.11–1.57)
1 + smoking11.27 (1.11–1.46)−251.46 (1.30–1.64)−2811.20 (0.98–1.45)−91.25 (1.05–1.48)−22
1 + alcohol use11.39 (1.21–1.59)81.69 (1.50–1.90)811.25 (1.03–1.52)141.36 (1.14–1.62)13
1 + physical activity11.31 (1.14–1.50)−141.55 (1.38–1.74)−1411.20 (0.99–1.46)−91.29 (1.08–1.53)−9
1 + vegetable use11.26 (1.10–1.45)−281.45 (1.29–1.64)−3011.19 (0.98–1.45)−141.26 (1.06–1.50)−19
1 + fat on bread11.36 (1.18–1.55)01.62 (1.44–1.82)−311.22 (1.00–1.48)01.32 (1.11–1.57)0
1 + coffee drinking11.35 (1.18–1.55)−31.62 (1.44–1.82)−311.22 (1.00–1.48)01.31 (1.10–1.56)−3
1 + relative weight11.36 (1.19–1.56)01.62 (1.44–1.82)−311.23 (1.01–1.49)51.32 (1.11–1.58)0
1 + all behaviours11.22 (1.06–1.40)−391.35 (1.20–1.53)−4511.18 (0.97–1.44)−181.20 (1.00–1.44)−38
  • *Adjusted for age, age squared, study year and four pre-existing chronic diseases

Discussion

Educational level showed a graded association with all studied mortality outcomes. The associations were stronger among women, which is in agreement with Finnish mortality statistics.19 Among men, the relative differences by educational level were even slightly larger in all-cause mortality than in CVD mortality. In contrast, among women the relative differences CVD were much larger, suggesting that other causes of death show only a weak socioeconomic gradient. In breast cancer mortality a positive association with educational level is commonly found. However, in Finland such an association has been disappearing,20,,21 and there were only 120 breast cancer cases in our data, suggesting that the weak gradient in other causes of death than CVD is not due to breast cancer alone. Among women, educational level differences in all-cause mortality are primarily due to CVD, but this is not true for men.

Among men, health behaviours explained more of the associations between educational level and mortality outcomes than in women. Even if the health behaviours examined in our study are commonly thought of as cardiovascular risk factors, among women they explained even more of the educational level differences in all-cause mortality than in CVD mortality. This suggests that these health behaviours may not be only aetiological risk factors but also markers of generally risky lifestyles.

Stroke is a common cause of death but previous studies examining explanations for socioeconomic differences in mortality have not examined stroke separately, probably because they have been limited by sample size. In our study, the relative mortality differences by educational level in CHD, stroke and all CVD were identical among men. However, among women the relative mortality differences were weaker for stroke than for CHD. Although the total number of strokes among women was reasonable, most deaths occurred in the primary education group which increases random error in our analysis. However, also nationwide mortality statistics show much weaker educational level differences for stroke than for CHD.19 Among women, it was also clear that vegetable use and physical activity explained more of the educational level differences in stroke than in CHD. About one third of all stroke cases are haemorrhagic and two thirds ischaemic strokes. Risk factors for ischaemic stroke are the same as for CHD but for haemorrhagic strokes blood pressure is more important.22 It has been shown that physical activity and also dietary habits are related to blood pressure,23,,24 which offers a plausible aetiological explanation to the differences in our results for stroke and CHD.

In our study, smoking, vegetable use and physical activity were the most important health behaviours explaining relative educational level differences in all mortality outcomes. In most previous studies, smoking has had a clear effect10,,11,13 but other health behaviours have been studied less often and the results have been mixed. However, two previous studies have found physical activity to be the most important health behaviour explaining socioeconomic differences in all-cause mortality.8,,9 To our knowledge, vegetable use has not been previously included in studies examining health behaviours as explanations for socioeconomic differences in mortality. In our study, a very simple indicator of vegetable use explained a part of the mortality differences. Vegetable consumption reflects the intake of antioxidants, fibre, and minerals but frequent vegetable users may also otherwise follow a healthy diet. Persons who regularly use vegetables and fruits have been shown to have a lower risk of developing CVDs.14

The type of fat used on bread, coffee drinking and relative weight did not explain educational level differences in mortality in our data. Adjusting for alcohol use even slightly increased the relative differences between educational groups, except in stroke among women. The total volume of alcohol use is often higher in the higher socioeconomic groups although hazardous drinking styles (binge drinking) show the opposite pattern.25 Our measure of alcohol use captures relatively moderate drinking which, however, was associated with mortality in these data. In addition, the heaviest drinkers are likely to be missing from our data. Previously it has been shown that alcohol-related deaths cover a large part of the excess mortality of the lower socioeconomic groups.26

In addition to health behaviours, many other factors contribute to socioeconomic differences in mortality. Such factors include material circumstances, psychological and social risk factors and biological risk factors, many of which are related to health behaviours.7,,9,12 These factors are interrelated in various ways: for example, the effect of material and psychosocial circumstances may partly go through health behaviours and the effect of health behaviours is eventually mediated through biological risk factors.27,,28 If the individual effects of all explanatory factors are simply summed together the common explanation may reach 100% or even exceed it.7,,29 Our study shows that health behaviours are involved in the explanation of socioeconomic differences in mortality but a complete understanding of these differences requires consideration of several kinds of factors.

In previous studies that have included several types of explanatory factors the group consisting of health behaviours has not been the most powerful explanation for socioeconomic differences in mortality.7,,9 However, in many of these studies only few behaviours have been included. Our study included seven health behaviours, which explained about half of the differences among men and one fourth to one third among women. However, their true effect may be even larger since survey responses are always imprecise which is likely to dilute their effects. In a large multicountry study based on indirect estimation from disease-specific death rates, half of socioeconomic differences in male mortality could be explained by smoking alone.30 In our study, all health behaviours were measured only once: were they measured several times over time their explanatory power would probably have been larger. Furthermore, in our study some of the measures, for example those of diet and physical activity, were quite simple. With their more precise and extensive measurement more of the mortality differences had probably been explained.

Methodological considerations

This study was based on a large nationwide population data of men and women with a long follow-up period. The large size of the data allowed us to divide the health behaviours into several categories to maximize their explanatory potential and provided us a reasonable number of deaths to separately examine various causes of death, including stroke. However, a drawback from a long follow-up is that people may change their behaviours which is likely to dilute their effects as explanatory factors. Furthermore, the distribution of educational level has changed as the overall level of education in the population has increased. However, because the relative mortality differences between educational groups have been fairly stable over time31 this is unlikely to have largely affected the results.

The overall response rate to our surveys was similar to what is usually achieved by postal questionnaires,32 but decreased over time during the study period. A non-response study from these data showed that the decrease was larger among those with lower education.17 It has also been shown that the level of mortality is higher among non-respondents than respondents.33 Because the non-respondents are also more likely to represent the lower socioeconomic groups and have more unhealthy behaviours, it is possible that the true mortality differences between educational level groups are even larger and the effects of health behaviours may have been underestimated.

Conclusion

Educational level differences were found in cardiovascular and all-cause mortality in both men and women. Smoking, low vegetable use and physical inactivity explained a substantial part of socioeconomic differences in mortality. For socioeconomic differences in mortality, it is therefore of crucial importance in which direction socioeconomic differences in these behaviours develop in the future. However, the reasons behind such behaviours are complex, and population trends and socioeconomic differences in health behaviours are not easily affected by political decisions and health interventions.

Supplementary Data

Supplementary data are available at EURPUB online.

Acknowledgements

M.L. (#204894), O.R. (#210435) and R.P. (#214126) have been supported by the Academy of Finland.

Conflict of interest: None declared.

Key points

  • Health behaviours are considered as one of the main explanations for socioeconomic differences in mortality but empirical studies on the subject are relatively few.

  • Most previous studies have included a limited number of health behaviours, not examined men and women separately, and not included a separate analysis for stroke.

  • Smoking, vegetable use and physical activity were the most important behaviours explaining the differences in both men and women.

  • Several health behaviours were involved in the explanation for socioeconomic differences in cardiovascular and all-cause mortality. For future mortality differentials, it is of crucial importance in which direction socioeconomic differences in health behaviours develop.

References

View Abstract