OBJECTIVE To analyse the dynamics of smoking prevalence, initiation, and cessation in relation to sex, age, birth cohort, study year, and educational level.
DESIGN Six independent cross-sectional population surveys repeated every five years between 1972 and 1997.
SETTING The provinces of North Karelia and Kuopio in eastern Finland.
SUBJECTS Independent random samples of 18 088 men and 19 200 women aged 25–64 years. Those comprising the oldest birth cohort were born in 1913–17 and those in the youngest were born in 1968–72.
RESULTS Among men the prevalence of smoking decreased over time, but the cohort effect observed in smoking initiation was obscured by the changes in smoking cessation. Differences between the educational categories were small. Among women the prevalence of smoking increased during the study period. This was mainly caused by the less highly educated, in whom smoking initiation clearly increased in successive birth cohorts, but a more moderate cohort effect was also present among the more highly educated women.
CONCLUSIONS In men decreased initiation and increased cessation contributed to the downward trend in smoking prevalence, whereas among women, changes in smoking were mostly caused by augmented initiation in successive birth cohorts. During the study period educational inequalities in smoking widened, as the less highly educated came increasingly to form the smoking population.
- smoking prevalence
- smoking initiation
- smoking cessation
Statistics from Altmetric.com
The health hazards of smoking have been well established. Smoking is one of the most important risk factors for cardiovascular disease (CVD) and also increases the risk for dozens of other diseases.1 2 During the early 1970s eastern Finland was reported to have the highest mortality from CVD in the world. High levels of the main CVD risk factors, smoking among them, have been reported to be contributing factors to this.3
Thirty years ago half the men in eastern Finland smoked, in contrast to women, among whom only one in 10 smoked. Since then the prevalence of smoking has decreased notably among men but among women it has almost doubled. Among men the decrease has been similar in all age groups, whereas among women the increase has been greatest in the youngest age group.4
One of the most important determinants of smoking behaviour is educational level. In eastern Finland less highly educated men have been more likely to smoke than the more highly educated, and male smoking has decreased comparably across all educational categories. Among women the more highly educated used to smoke more than the less highly educated until the 1970s, after which the less highly educated have been smoking more.5
Birth cohort analysis may be used to gain further information on the development of smoking in the population. The aim of the present study is to examine the dynamics of smoking in eastern Finland in relation to birth cohort, calendar year, and age during the period 1972–97. We used all three variables in our analyses but did not try to quantify their separate effects, because of their interaction. As the changes in smoking prevalence result either from initiation in successive birth cohorts or from changes in cessation within the birth cohorts, we analysed these components separately. The analyses were done separately for men and women, and performed in two educational categories.
The data were derived from six cross-sectional risk factor surveys conducted in eastern Finland since 1972. Independent population samples were studied every five years in the provinces of North Karelia and Kuopio. In 1972 and 1977 random samples of 6.6% of the population born between 1913 and 1947 were drawn in both provinces. In 1977 an additional sample of 6.6% of those born between 1948 and 1952 was taken randomly in North Karelia only. For the subsequent surveys (1982, 1987, 1992, and 1997) the sampling method was modified to comply with the protocol of the World Health Organisation MONICA project (monitoring trends and determinants incardiovascular disease).6 The sample was stratified according to sex and 10-year age group so that at least 250 people were included in each subgroup in both provinces. Because in the earliest surveys the population aged 25–64 years was equally distributed to each 10-year age group, the results remained comparable. Our analyses included 25–64 year olds, except in 1972 when 60–64 year olds were not studied. In 1977 the age group 25–29 years old consisted of North Karelians only. The number of respondents in the six surveys was 37 288, and the response rates varied between 75% and 92%.
As the overall development of smoking in the two provinces has been basically similar, despite the somewhat steeper decrease in male smoking in North Karelia during the first surveys,4 the data from the two areas were combined for our analyses. Twelve synthetic five-year birth cohorts were constructed; those included in the oldest birth cohort were born in 1913–17 and those in the youngest cohort were born in 1968–72. The number of respondents decreased progressively from the earliest to later surveys. The smallest cell defined by birth cohort, age, study year, and sex included 135 respondents.
Respondents’ smoking status was assessed with standardised questions in a self administered questionnaire. Those who had smoked regularly for at least one year and continued to do so during the previous month were regarded as current smokers. Smoking initiation was defined on the basis of ever having smoked regularly; those who declared they had smoked for at least one year at some point during their lifetime were considered as smoking initiators. Quitters were those who had smoked regularly but stopped at least one month before the study.
The analyses were performed in two educational categories. Among the indicators of socioeconomic status, educational level is the one showing the strongest and most consistent associations with CVD risk factors.7 8 Here, educational level was measured by the total number of school years. As the average length of education in Finland has increased remarkably during the recent decades, the respondents were divided into two educational categories on the basis of their position below or above the median length of education in their own birth cohort.
The results are presented in the tables, stratified by sex and educational level, using five-year birth cohorts, five-year study periods and five-year age groups as classifying variables. Birth cohorts appear in rows and study years in columns, and trends at each age level can be traced by reading diagonally down and to the right. Even if the results do not allow complete separation of age, period, and birth cohort from each other, simultaneous use of the three variables can produce tentative evidence about their relative contribution to the development of smoking.
Logistic regression analysis was used to test the statistical significance between the explanatory factors and the smoking variables. The order of the modelling follows the order in the text. The models birth cohort, age group, and study year were treated as continuous variables. First-level interactions between the main effects were also tested. Odds ratios with significance levels for five-year increase in age, five-year period, five-year difference in birth cohort, and change from lower to higher educational level are shown in the table footnotes. The non-linear development of smoking cessation by study year and age was corrected by adding study year 1972 and age group 60–64 to the models as dummy variables. For the same reason, birth cohort was fitted in two parts into the model of women’s smoking initiation. The statistical software package SAS was used for all the analyses.9
In men the prevalence of smoking decreased markedly between 1972 and 1997 (table 1). The decrease was observed in all age groups, but was slightly greater in the older groups. With the exception of the 1972 survey, younger subjects had a higher smoking prevalence than older subjects in cross-sectional analyses. Prevalence declined with advancing age within each birth cohort. In both educational groups the development of smoking was virtually equal, but less highly educated men had consistently higher smoking prevalence than the more highly educated men.
In contrast to men, the development of smoking prevalence over time differed between the educational categories in women (table 2). Among less highly educated women the prevalence of smoking doubled during the study period, whereas among the more highly educated it remained unchanged. As a result, smoking prevalence among less highly educated women has exceeded that of the more highly educated since the 1980s. Among less highly educated women the prevalence of smoking increased over time in every age group. In cross-sectional analyses smoking decreased markedly with age, as it did within birth cohorts, although more moderately; thus an increase in smoking prevalence in successive birth cohorts was clear. Among more highly educated women, periodical changes were small in every age group. The differences between age groups were parallel, but smaller in magnitude, to those in the lower educational category. The increase in smoking prevalence was clear from one birth cohort to the next, which was similar to the situation in the lower educational category.
In men smoking initiation diminished quite steadily over time since the second survey in 1977 (table 3). The decreasing trend applied to all age groups. In cross-sectional analyses the proportion of smoking initiators increased when moving from the youngest to the oldest age groups. In contrast, no increase was noted as age advanced within the birth cohorts. In the oldest birth cohorts smoking initiation was more common than in the youngest cohorts. Birth cohort appeared to be an important factor contributing to decreasing initiation rate; the decrease occurred in both educational categories, but the more highly educated showed consistently lower initiation rates than the less highly educated.
In women smoking initiation increased over time in both educational categories, but the increase was greater at the lower educational level (table 4). In cross-sectional analyses the initiation rate of younger women was clearly higher than that of older women, whereas the change with age within the cohorts was minimal. Together, these two observations indicate a clear birth cohort effect.
Figure 1 shows the development of smoking initiation by birth cohort age groups combined. Among men smoking initiation declined steadily from one birth cohort to the next. The development was parallel in both educational categories until the cohorts born during the 1950s, after which the more highly educated experienced a marked decrease. Among women smoking initiation increased in both educational categories. In those cohorts born during the 1950s the increase levelled off, such that initiation among the less highly educated settled at a considerably higher level than the more highly educated. When moving from the oldest to the youngest birth cohorts, the difference between men and women became a difference between educational categories.
In men smoking cessation increased notably from 1972 to 1977, after which the increase became more moderate (table 5). The increase took place throughout the age groups. The proportion of quitters increased with advancing age in cross-sectional analyses and within the birth cohorts. The development followed a similar pattern in both educational groups, even though the general level of quitting was higher among the more highly educated.
In women smoking cessation developed similarly to men; between 1972 and 1977 it increased strikingly and then became a more moderate but constant increase (table 6). The pattern was similar in both educational categories, although the increase over time was even more marked among the more highly educated. In cross-sectional analyses the differences between age groups were small, but within cohorts an increasing rate of cessation was evident. Cessation increased across age groups, and the differences between successive birth cohorts were small and inconsistent.
In men the prevalence of smoking decreased markedly during the study period, but consistent change between successive birth cohorts was not observed. Among women, in contrast, temporal increase in smoking prevalence arose indisputably from changes in successive birth cohorts. The increase was mainly caused by smoking in less highly educated women, but also among the more highly educated smoking was clearly augmented in successive birth cohorts. An especially large increase was observed between female cohorts born during the late 1930s and early 1950s.
In smoking initiation a birth cohort effect was evident in both sexes, inducing a decreasing trend in men and a more prominent increase in women. Smoking cessation, on the other hand, showed an increasing trend, but no cohort effect, in both sexes. Among men the development of smoking cessation appeared to obscure the cohort effect in current smokers. Among women smoking initiation and cessation were augmented, but a greater increase in initiation produced an increase in smoking prevalence.
Birth cohort, period, and age are linearly interrelated such that any one of the variables can be derived given the other two. Separating these variables has been a constant problem in birth cohort studies. Because smoking behaviour is dependent on age, separating birth cohort effects and periodical effects is particularly difficult. At a given point in time, people in birth cohorts are at different ages and at different phases of their smoking career, and therefore the effects can be seen more clearly in some birth cohorts than in others. A change in the pricing policy, for example, constitutes a typical periodic phenomenon, but its influence may be seen more clearly in some birth cohorts than in others; in other words, the change can not be detected independently from the age effect.
Various strategies have been developed to separate these three effects statistically, but thus far none has gained collective acceptance.10 11 Our statistical results (see table footnotes) do not allow separation of these effects. We presented all three variables simultaneously in the tables but did not attempt to isolate and quantify their separate effects; logistic regression analysis was used only to test the significance of the explanatory variables.
In addition to the tables, the figure describes smoking initiation in successive birth cohorts. Smoking initiation was defined as ever smoking regularly during one’s lifetime; date or age of initiation were not taken into account. In the figure the respondents were combined into cohorts, disregarding their ages at the time of the study. Of course, some people have not yet initiated their smoking but will do so later, and this possibility is greater in the youngest birth cohorts than in the oldest ones. However, as smoking initiation in adulthood is uncommon12 and our subjects were 25–64 years of age, we assumed that the possible bias caused by age differences between the cohorts was small. Tables 3 and 4 support this assumption: within the birth cohorts smoking initiation was quite stable, and systematic increase with advancing age could not be observed in any of them.
Smoking as a topic introduces other possible sources of bias; higher mortality rates of current smokers compared with those of non-smokers and former smokers may distort the results. Some of the previous cohort studies on smoking have estimated the effect of selective mortality13-15; this effect remained minimal, especially in younger birth cohorts. A previous Finnish study also found the effect of selective mortality to be small.16 In addition, the effect of selective mortality in our study was reduced by the low overall mortality in the age groups examined.
Previous birth cohort studies on smoking have been predominantly retrospective,13-23 one of them combining retrospective and prospective approaches,24 whereas our data were derived from repeated cross-sectional surveys. In both study designs, possible problems caused by selective mortality are equal. In contrast, recall errors are much more likely in retrospective studies; differences in the ability to recall dates of smoking initiation and cessation between older and younger cohorts may confound inter-cohort comparisons.
In the assessment of smoking status we relied on self reporting, the validity of which has sometimes been questioned. The data from the 1992 survey were validated elsewhere (Vartiainen E, Seppälä T, Korhnonen HJ, et al, submitted).25 Self reporting and biochemical markers have shown a high degree of correspondence; thus, misclassification of smoking status due to self reporting is not likely to be great.
As far as comparison is possible, the results of our study are consistent with those of a previous Finnish study16 which located the peak in male smoking in those cohorts born in 1911–25, whereas the prevalence in female smoking was highest in the most recent cohorts (born during the 1960s). For the most part, the changes followed the variation in smoking initiation, but among men smoking cessation was also augmented.
Compared with Norway, another Nordic country, Finnish men experienced their peak smoking prevalence earlier than Norwegian men, whereas Norwegian women were ahead of their Finnish counterparts. After a period of decline in smoking prevalence in both sexes, Norwegian women experienced another peak around 1970, at which time about half of the women born in 1940–49 smoked. From 1970 to 1990 smoking declined again in all birth cohorts among men and women. The gender gap diminished until no difference between men and women was observed among those born after 1950.24
The previous Finnish study16 also showed a change in smoking prevalence between the educational categories in women. Moreover, the change was dated to the same birth cohorts as in our study—to those born at the beginning of the 1940s. In our study the change was observed in smoking prevalence and in smoking initiation. As previously reported, smoking is today more common among the less highly educated in both sexes.5 25 The educational discrepancy already begins to emerge at the upper stage comprehensive level (12-15 years).26 In Finland, less highly educated women have smoked as much as more highly educated men since the late 1980s.25 Our results indicate that in the youngest birth cohorts, smoking initiation among less highly educated women clearly exceeds that of more highly educated men and approaches that of less highly educated men. Smoking initiation is thus determined as much by educational level as by gender in the youngest birth cohorts. In addition, the sex difference also appeared to be diminishing with regard to smoking cessation: among women the age gradient steepened and began to resemble that observed in men.
Comparisons with German results provide an intriguing contrast. Brenner13 examined the prevalence of smoking in successive birth cohorts at two educational levels. As in Finland, smoking prevalence was higher among less highly educated men than among the more highly educated. From birth cohort 1941–50 onwards the gap widened because of a decrease in smoking among the more highly educated. Interestingly, this seemed to be mainly due to increased cessation among the more highly educated. Among women, the prevalence increased in each successive birth cohort, but, in contrast to Finland, the development was comparable in both educational categories.
We observed a decreasing trend in smoking prevalence among men and an increasing trend among women. Previous birth cohort studies reporting similarly diverging trends have considered differing responses to anti-smoking campaigns (more or less broadly conceived) or tobacco advertising as reasons for the sex difference.13 17 19 20 22 Our data originate from an area that has been a target for a major community-level risk factor intervention,27 initially directed more towards men than women, that may have resulted in some of the difference between the sexes. The harmful effects of smoking came into public awareness after the late 1950s, but women have been found to be less responsive to tobacco related health publicity than men.28 The influence of advertising, in contrast, is likely to be negligible because advertising directed explicitly towards women has been rare, and since the late 1970s direct advertising has been banned.29
Similarly, the influence of pricing policy presumably remains small. Pricing policy is known to have a strong influence on tobacco consumption28 30 and lower socioeconomic groups have been found to be more responsive to price fluctuations than higher socioeconomic groups, but no such difference has been observed between genders.28 Moreover, in Finland pricing policy has not been used to control tobacco consumption; the price of tobacco has paralleled the consumer price index.29 31
The outcome of this dissimilar development between men and women has been the convergence of smoking behaviour between the genders. We presume that the most likely explanation is the change in women’s social position. Social norms restricting women’s behaviour relaxed, and within that development female smoking became more widely accepted.32 33 Urbanisation, accelerating especially since the 1960s, led to eroding of the norms of the agrarian society and gave rise to women’s increasing participation in the labour force.34 In addition, the influence of the family has diminished and youth culture has become increasingly important, factors that plausibly have promoted the adoption of smoking among young women.
Explanations concerning advertising or the lack of health education have been criticised for failing to take into account the positive aspects of smoking35—although people are aware of the negative health effects, smoking’s positive social and cultural connotations may be more important at the point when the smoking behaviour is adopted or abandoned. It appears that in the case of women these factors have been more influential and have truncated anti-smoking efforts.
Despite the reduced difference between the sexes, men still smoke more than women, and the trend in men has not been completely positive. The number of male smokers in the youngest age groups is still large, especially among the less highly educated, and in recent surveys the increasing trend in female smoking has levelled off. Our results indicate that smoking initiation had already begun to decrease in those born during the 1950s. Smoking prevalence in women will not yet decline, however, because people in the the oldest, lightly smoking cohorts are being replaced by those in the cohorts where smoking is heavier. Overall, the development of smoking has been similar throughout the Western countries,36 37 and other birth cohort studies have found trends parallel to ours. Although few of these studies have reported the more recent decrease in smoking prevalence in women,20 22 the decline of female smoking may be a more general, Western phenomenon as well.
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.