Article Text
Abstract
Background: There was a decrease in smoking during early pregnancy in Swedish women between 1982 and 2001. We sought to determine whether there was a parallel decrease in socioeconomic inequality in smoking.
Methods: Registry data indicating educational level and smoking status at first antenatal visit in all 2 022 469 pregnancies in Sweden 1982–2001 were analysed. Prevalence differences, odds ratios based on prevalences and total attributable fractions were compared for five-year intervals.
Results: The prevalence differences of smoking showed a greater decrease at the lowest and middle educational level compared with the highest educational level (14.5%, 15.7% and 10.2%, respectively) indicating reduced inequality in absolute terms. However, odds ratios regarding low educational attainment versus high, increased from 5.6 to 14.2, signifying increased inequality in relative terms. Moreover, the total attributable fraction of low and intermediate educational level regarding smoking at first antenatal visit increased from 61% to 76% during the period studied.
Conclusions: Smoking at first antenatal visit in Sweden between 1982 to 2001 decreased in a way that conclusions regarding trends in inequalities in smoking at first antenatal visit depend on the type of measure applied. However, using the measure of total attributable fraction, which takes into consideration the impact of the exposure on the individual as well as the effect of the varying size of the group of exposed, the growing importance of educational level for the behaviour in the population was demonstrated.
Statistics from Altmetric.com
Inequalities in health have received increasing international attention. One expression of this is the Commission on Social Determinants of Health launched by the World Health Organization in 2005.1 Optimum health for each individual has been accorded growing importance as a human right ever since it was first proposed in the 1978 Alma Ata Declaration by WHO and UNICEF.2 Differences in health among individuals and groups also constitute the starting point of aetiological epidemiological research, the results of which are the foundation of needs-based and evidence-based health promotion, and of disease prevention interventions aimed at decreasing health inequalities.3
In this context, trends over time in inequalities on important risk factors like tobacco smoking, may provide particularly valuable information.4 One segment of the Swedish population has come in for special scrutiny in this respect: pregnant women who have been a favourite target for anti-smoking intervention. They are most likely to respond to such efforts, since health gain will also include their unborn child.5 Smoking during pregnancy has been identified as exerting an independent, adverse effect on a variety of reproductive health outcomes, examples of which are spontaneous abortions, stillbirth, preterm birth, reduced fetal weight and growth and sudden infant death syndrome (SIDS).6–10
The Swedish intervention aiming at reducing smoking during pregnancy has been composed of messages to all women regarding the risk of smoking during pregnancy, as well as actions targeted at pregnant women who attend antenatal care clinics.
The smoking prevalence among pregnant women at first antenatal visit in Sweden has decreased from 31% in 1982 to 10% in 2003. There has been a parallel decrease among women of child bearing age (16–44 years) in the general population, from 35.1% in 1982 to 18.4% in 2003. This decline, however, was not evenly distributed throughout age groups. For instance, the highest prevalence of daily smoking among women in the general population in 1982 was recorded among the age group 25–34 years (39.0%), while in 2003 the prevalence was highest among women of 35–44 years (22.1%).11
Many factors may have contributed to the decline in smoking prevalence during pregnancy. In 1992 a joint project called “Smoke-Free Pregnancy” was launched by the Swedish National Institute of Public Health, the Swedish Cancer Society and the Swedish Heart-Lung Foundation. That project, in turn, began “Smoke-free Children”, an initiative aimed at providing children with a smoke-free start in life by reducing the number of pregnant women who smoke.12 Although these interventions have been successful in reducing the overall prevalence of smoking during pregnancy in Sweden,13 the socioeconomic inequality among those who smoke during pregnancy is still evident.
Previous studies on smoking trends during pregnancy from Sweden,14 the United Kingdom,15 Finland16 and the United States17 18 have confirmed the widening socioeconomic inequalities as well as the association between low maternal education and continued smoking while pregnant. However, these studies covered shorter periods (<10 years) and have not examined the magnitude and trends of educational inequality in relation to pregnancy. Therefore, long-term studies are important in order to evaluate the impact of policies that address tobacco-related disparities.
Since several of the initiatives mentioned specifically addressed the equity issue, the development over time in this respect seems to be an important aspect to evaluate, and a number of methods have been suggested to perform such analyses. We utilised prevalence differences between educational levels from 1982–2001 to reflect inequality on an absolute scale, and odds ratios on a relative scale, and finally the measure of total attributable fraction, which takes into consideration the impact of the exposure on the individual as well as the effect of the varying size of the group of exposed.19
The aim of this study was to examine whether the decrease in smoking at first antenatal visit among Swedish women from 1981 to 2001 was accompanied by a parallel decrease in socioeconomic inequality with regard to educational level by applying three different measures (prevalence differences, odds ratios and total attributable fraction).
METHODS
Study population
Data on all pregnancies in Sweden resulting in a delivery from 1982 to 2001 were collected from the Swedish Medical Birth Registry (MBR), which preserves health profiles, marital status, age, nationality and other demographic factors for all pregnant women in Sweden (who participate in the national health system).20 However, the MBR does not contain information on socioeconomic factors. Therefore, in collaboration with Statistics Sweden, we linked data from the MBR to information from the Longitudinal Database on Education, Income, and Employment (LOUISE), which has recorded and preserved educational level, employment status and income variables for the entire population of Sweden. Educational information obtained from the LOUISE database can be linked with all other sources for the entire period of residence in Sweden by means of every citizen’s personal identification code.21 Such information can be obtained for most individuals from 1982 onwards. Smoking at first antenatal visit has also been reported by the MBR since 1982.22 The linking procedure of the databases was performed by Statistics Sweden with a failure rate of 5.1% (n = 103 336) for education variable, yielding 1 919 133 legible subjects. The linkage failure most probably emerged because of erroneous personal identification codes in the MBR—that is, errors made at the hospital level and/or missing information on education. The number of deliveries in Sweden between 1982 and 2001 ranged between 86 000 and 120 000 per year. Hence, our total study cohort consisted of 2 224 469 individuals. Because of the large number, we found it more manageable in the interest of keeping the tables smaller to present our findings in five-year intervals.
Definitions
Smoking status at first antenatal care visit was categorised and assessed as follows: (1) non-smoker, (2) smoke 1–9 cigarettes per day and (3) smoke 10 cigarettes or more per day.
Outcome variable
A woman who smoked at least one cigarette per day at the time of her first antenatal visit was classified as a smoker.
Exposure variables
The education variable was coded into three levels of education, according to number of years a person was enrolled in school—that is, low educational level (up to 9 years), middle educational level (10–12 years) and high educational level (more than 12 years).
Marital status
Married women or those women co-habiting with a partner were classified as married; other pregnant women were classified as single mothers.
Age was categorised into three ranges: 16–24 years, 25 to 34 years and 35–44 years.
Country of origin options were born in (a) Sweden, (b) other Nordic countries, (c) other European countries or (d) non-European countries.
Statistical analyses
We began by using prevalence as the measure for comparisons over time between the groups studied in order to capture tendencies on an absolute level. Next, multiple logistic regression analyses were performed, yielding odds ratios based on prevalences. In order to eliminate potential confounding from age, marital status and country of origin, a stepwise multiple logistic regression analysis was also performed, yielding adjusted odds ratios. Prevalences and odds ratios (OR) were calculated using SPSS software Version 11.5.
The attributable fraction (AF) was calculated using the formula: AF = (OR–1)/OR, where OR is the adjusted odds ratio generated by multiple logistic regression analysis.23 Total attributable fraction (TAF) was calculated as follows:
where AFi is the attributable fraction for smoking at first antenatal visit for a specific stratum (here: educational level), and Pi represents the proportion of all cases that fall in this stratum. The expression within the parenthesis thus represents the stratum-specific total attributable fraction (sTAF), and Σ indicates the summation of all the strata-specific calculations, which in turn results in the overall TAF. For those with the highest level of education, the AF and sTAF are by definition both zero.23 24 Finally, overall TAF is the summation of the sTAFs and represents the proportion of smoking that would not exist if all pregnant women had had the same prevalence as those with the highest level of education, under the assumption that there is a causal pathway between educational level and the outcome variable.25
RESULTS
Table 1 shows the absolute numbers and their proportion (%) of the main background variables, as well as significant shifts in the sociodemographic distribution in relation to pregnancy during the period studied (1982–2001). The proportion of pregnant women with the lowest level of education (nine years or less) declined from 25% to 14%, while the proportion increased from 26% to 34% in the group with the highest educational level (more than 12 years). Moreover, the proportion of pregnant women in the younger age group decreased, while the proportion in the oldest group increased. The proportion of women classified as single was stable (approximately 2.7% to 5.0%) over the period studied. Moreover, the proportion of pregnant women born in other countries increased from 11% to 16%. The largest increase was seen among women from non-European countries, while the proportion of women from Nordic countries other than Sweden decreased. Owing to the described shifts over time in the variables along with their association with smoking prevalence, the variables were all judged to be potential confounders that must be controlled for in a multiple logistic regression analysis.
Table 2 demonstrates that the prevalence of smoking at the time of the first antenatal visit decreased in all three of the educational groups studied, but more so in the middle and lowest categories, compared with the highest (that is, prevalence differences were 14.5%, 15.7% and 10.2%, respectively). Thus on an absolute scale smoking decreased less in the highest educational group. Regarding associations between other demographic variables and smoking at first antenatal visit, the largest decrease in prevalence was seen among younger age groups (prevalence difference 18.4%), single mothers (prevalence difference 21.8%) and women born in Sweden (prevalence difference 17.7%) and other Nordic countries (prevalence difference 17.9%).
Table 3 illustrates the crude and adjusted ORs and 95% confidence intervals (CI) of the multiple logistic regression analyses for smoking at first antenatal visit in five-year intervals by educational level. We controlled for age, marital status and country of origin.
However, the results demonstrated the same pattern as in the unadjusted analyses—that is, steeply increasing relative differences between the highest and the lowest educational groups. Adjusted OR for women in lowest educational group increased from 5.0 (4.9 to 5.2) between 1982 and 1986 to 14.2 (13.7 to 14.8) between 1997 and 2001, compared to the highest educational groups. Thus, this measure indicates that inequality in smoking increased considerably.
Table 4 shows the relation between smoking at first antenatal visit and educational level, expressed as adjusted OR, AF, sTAF and overall TAF between 1982 and 2001. AF is the proportion of smoking at first antenatal visit in a given educational category that could be attributed to lower education than the reference category, assuming a causal pathway between educational level and smoking at first antenatal visit. Consequently, sTAF could be regarded a measure of the proportion of smoking in given stratum which would not exist if the women in this group had had the same level of education as the women in the highest educational category (more than 12 years).
As table 4 illustrates, TAF increased from 61% to 76% between the periods 1982 to 1986 and 1997 to 2001, indicating an increase of the importance of education lower than the reference category for the total burden of smoking in Swedish women at the time of their first antenatal visit. It should also be noted that in the mid-level educational category, sTAF increased from 29% to 44%, while remaining stable in the lowest educational group over the period studied.
DISCUSSION
Depending on the method of assessment, different conclusions may be reached as to whether educational inequality regarding the habit of smoking among Swedish women at their first antenatal visit decreased or increased. However, the importance of “lack of highest educational level” on the total burden of smoking among Swedish women at their first antenatal visit increased from 61% to 76% (assuming a causal pathway between educational level and the outcome variable). Moreover, only women in the middle educational category contributed to this increase.
Some limitations of this study may be considered. If under-reporting of smoking were associated with educational level, this would lead to differential misclassification. The estimated association of low education with smoking would be too high if a preponderance of highly educated women had under-reported their smoking habits and too low if those with the least education under-reported smoking behaviour. However, many studies have considered self-reported smoking to have high validity as an accurate measure of smoking behaviour.26–29 Since misclassification is considered to be low in general we do not believe that this type of potential error has influenced our analyses to any important degree.
Moreover, the change in distribution of demographic factors such as low age group (which declined) and country of origin other than Sweden (which increased and varied) leaves room for confounding—that is, the association between low educational level and age and/or country of origin could differ considerably from the beginning of the period studied to the end. However, when controlling for the variables mentioned in a logistic regression model, the estimated effect of low educational level on smoking was marginal, implying that the effect of confounding from these variables was low.
A strength of this study is that it is population-based and tries to discern a trend over the course of 20 years involving more than two million participants.
Previous studies have shown low maternal educational level to be the strongest predictor for smoking at first antenatal visit which, in turn, has been considered to contribute to an increasing socioeconomic gap.30–34 The time trends noted in this study follow the general trend of decreasing smoking prevalence in Swedish women, whereby the decline has been much more rapid in the higher socioeconomic groups.35
The decline in the smoking trend in the general population of Swedish women 16–84 years of age was from 27.9% 1981 to 20.4% in 2001.36 However, smoking prevalence at first antenatal care visit declined more steeply, from 32% to 11.2% during the same period.36
The prevalence of smoking among young pregnant women (16–19 years of age) was twofold compared to their counterparts in general population, while there were no differences in smoking prevalence between women 20–24 years old. Additionally, the smoking prevalence among pregnant women aged 25 years and older was halved compared to women of the same age in general population for the period studied.37
Our data show that ethnicity also was linked to smoking at first antenatal visit in a pattern that changed over time, which demonstrates that the relations between age of pregnancy, socioeconomic status and ethnicity are associated in a complex way. This association also varies over time as shown in our data, which all must be handled in the analytical approach when analysing the inequality trend of smoking among women at their first antenatal visit.
It remains unclear whether the observed development is a result of different anti-tobacco intervention strategies applied to the general population, as opposed to those applied to pregnant women. The pregnant women in our study represent a mix of women who have on the one hand been exposed to only general population intervention (all first-time pregnant women) as well as on the other hand women who have been exposed to the more targeted intervention while attending antenatal care (all women who previously have completed a pregnancy in Sweden). The strategies directed towards the population-at-large have been based on general information (for example, warnings on cigarette packages), restrictions on advertising, control on availability (for example, age limits for purchase), price policy (for example, taxation on tobacco products) and smoking restrictions in public places. All of these measures largely target the population en masse, rather than a specific class of individuals.36 However, interventions directed towards pregnant women involve personalised information (for example, motivational counselling, advice on substituting tobacco analogues, etc).13 Since our study was not based exclusively on first time pregnancies, it is not possible to separate the effect of the mentioned two types of intervention.
What this paper adds
The prevalence of smoking at first antenatal visit in Sweden during 1981–2001 decreased for women in all educational levels, with a greater decrease at the lowest and middle compared with the highest.
Odds ratios regarding low level of education and smoking prevalence at first antenatal visit increased, indicating a widening educational inequality in relative terms.
The total attributable fractions increased during the studied period, indicating an increased importance of education for the total burden of smoking during pregnancy among Swedish women.
Using the measure of total attributable fraction, which takes into consideration the impact of the exposure on the individual as well as the effect of the varying size over time of the exposed group, the growing importance of educational level for the behaviour in the population was demonstrated.
Our study utilised total attributable fraction as a measure of inequality, which signifies more than only the gap between the disadvantaged groups. It also takes into account the varying size of the disadvantaged groups, and discloses how much the disadvantage variable (here educational level less than the highest) impacts the outcome variable for the total population. Accordingly, it could be argued that PAF is a more relevant measure than prevalence differences and odds ratios for assessing a policy-relevant quantification of inequality towards a certain exposure.
In addition, using this method also allows us to examine how different strata of the study population, as defined by level of exposure, contribute to inequality. Thus, we noted that it was the group with mid-level education which contributed exclusively to the increasing importance “lack of highest level education” for the overall smoking burden among women at the time of their first antenatal visit. We find this information very valuable from a policy perspective, since the implication would have been different if the group of women with the lowest level of education (a group especially targeted for tobacco control measures in Sweden) had played this part. It is particularly important when balancing the equity issue with the maximised population impact in the formulation of policy.38 In this case, it makes it clear that an important policy issue regarding intervention against smoking at first antenatal visit is whether emphasis should be directed towards the shrinking group of women with the lowest level of education, or towards the women with a middle level of education.
REFERENCES
Footnotes
Competing interests: None.
Funding: This study was made possible by grants from the Swedish Council for Working Life and Social Research, grant number FAS 2002–0920, an ALF grant from the Medical Faculty of Lund University and “Tackling Socioeconomic Inequalities in Smoking” project funded by European Commission, Public Health Directorate, through the ENSP.