The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Department of Public Health, PO Box 41, University of Helsinki, 00014 Helsinki, Finland. E-mail: firstname.lastname@example.org
Objective: To examine the association of smoking with recurrent dieting and BMI among Finnish adults.
Research Methods and Procedures: We used questionnaire data from 1990 on 11,055 subjects from the Finnish Twin Cohort who were 33 to 61 years of age. Multinomial logistic regression analysis was carried out using lifetime dieting as the outcome variable and smoking as the main explanatory variable, adjusted for BMI and age. Twin pairs discordant for dieting and smoking were studied to examine the effect of environmental and genetic factors.
Results: Among women, current smokers [odds ratio (OR), 1.09 to 1.41 at different ages] and former smokers (OR, 1.52 to 2.82) were more likely to have dieted recurrently than never smokers. Among men, current smokers were less likely (OR, 0.69; 95% confidence interval, 0.55, 0.87) and former smokers were more likely (OR, 1.30; 95% confidence interval, 1.05, 1.61) to have dieted recurrently at different ages. The differences between the discordant pairs were consistent with this, although not statistically significant.
Discussion: Recurrent dieting was associated with former smoking in both sexes and with current smoking in women.
Smoking and obesity are central public health problems and represent a huge challenge to health promotion throughout the world. Escalating rates of obesity (1) have paralleled the prevalence of dieting, especially among younger generations (1, 2), and unhealthy dieting methods have become more prevalent (3, 4, 5).
Smoking has been cited as a weight control method by women in many studies (6, 7, 8), and a history of recurrent intentional weight loss has been associated with current smoking in young adults (9). Body dissatisfaction and intention to lose weight are associated with smoking initiation among adolescents (10, 11), and many adolescents justify their smoking initiation with weight control concerns (3, 12). Moreover, heightened levels of weight concerns and body dissatisfaction are more frequently observed among adolescent female smokers than non-smokers (13, 14, 15).
Studies on the effect of weight concerns on smoking behavior among adults have yielded contradictory results. In some studies, weight concerns are associated with a greater smoking relapse rate or reduced likelihood of quitting in women (16, 17, 18). Other studies have found no association between concern about weight gain and smoking cessation among women (19, 20), or in either sex (21, 22), or they have found weight concerns even to predict better quitting outcome (20, 22).
Research on the association of smoking with body weight has also yielded contradictory results. In many studies, current smokers have had lower body mass than non-smokers, and former smokers have been the heaviest group (23, 24, 25, 26, 27). Educational level seems to modify the effect of smoking on body weight: among more highly educated people, current smokers tend to be heavier than never smokers, whereas the opposite is observed among subjects with lower education (25, 28). On the other hand, recent studies have shown smokers to be heavier than non-smokers in many Western populations, Finns included (29, 30). It also has been found that, in older cohorts, smokers tend to be leaner, and in younger cohorts, smoking is associated with higher body weight and an unhealthy diet (29, 30). In a large Australian population-based sample (n = 8726), young female smokers were more prone to weight gain than never smokers (31). Short-term weight gain after smoking cessation is well documented (32), but in longer follow-ups and some cross-sectional studies (33, 34, 35, 36), the effect of smoking cessation on body weight is more controversial. The lack of consistent findings may be because both smoking and weight loss attempts occur in specific cultural and social contexts, which may have changed over time even within a society.
On the other hand, given that there are familial, including genetic, influences on both smoking (37) and dieting (38), it is possible that the association between the two may be mediated by common underlying genetic factors. Studies on weight loss attempts and smoking in families, particularly twins, can help elucidate the role of underlying genetic and familial factors.
This study examines the association among smoking, BMI, and recurrent dieting in a large population-based sample of Finnish adult twins assessed in 1990. Based on previous studies, we hypothesized that both current and former smoking would be associated with a higher frequency of recurrent dieting. We also studied how this association varies between sexes and age groups. The availability of a twin sample allowed us to study the dependence of this association on familial factors shared by twins.
Research Methods and Procedures
The Finnish Twin Cohort was established to examine genetic, environmental, and psychosocial determinants of chronic diseases. It includes virtually all Finnish same-sex twin pairs born before 1958 in which both co-twins were alive in 1975 (39). In 1990, only twin pairs born in the period 1930 to 1957 were contacted. Twin individuals who replied to the questionnaire mailed in the autumn of 1990 were included in this study. The study population thereby consisted of 12,793 twin individuals (response rate, 77%); the youngest subject was 33 and the oldest was 61 years of age at the time of response; reminder letters prompted responses well into 1991. We excluded from the analyses individuals with missing or ambiguous data on height (n = 341), weight (n = 368), dieting (n = 345), or smoking status (n = 445; numbers are not mutually exclusive). We also excluded those reporting a history of malignant disease (n = 168), a diagnosis of diabetes (n = 251), or use of antihypertensive medication for at least 2 months during the past year (n = 887). People with these diseases were excluded because they might have changed their eating patterns, smoking habits, or dieting behavior as a result of knowing they had such a condition. The final data included 5046 male and 6009 female twin individuals.
Subjects were asked how many times in their entire lifetime they had tried to lose weight (dieted; never, 1 to 2, 3 to 5, 6 to 10 times, or more often). Overall, 12% reported having dieted 3 to 5 times, 4.6% reported 6 to 10 times, and 6.3% reported more often. Because the smoking proportion of those who reported dieting more than six times did not differ from those who had dieted three to five times, and the number of those who had dieted more than six times was relatively small, we combined these two dieting categories. We classified those who had dieted at least three times in their entire lifetime as recurrent dieters.
Smoking was divided into three categories based on the questionnaires: current smokers, former smokers, and never smokers. Those reporting never having smoked >5 to 10 packs (100 to 200) of cigarettes during their life were classified as never smokers (33). Subjects who had smoked >5 to 10 packs of cigarettes during their life and had smoked on a daily or almost daily basis were classified as regular smokers. They were further classified as former smokers or current smokers depending on whether they smoked at the time of the questionnaire. Those who reported having smoked >5 to 10 packs of cigarettes but never on a daily basis were initially classified as occasional smokers. This group was substantially smaller (n = 19 men, n = 14 women) than the other smoking categories and was combined with the former smokers group, which it resembled very closely in BMI, age, and dieting behavior.
BMI was calculated from self-reported weight and height. The validity of self-reported BMI was assessed in a sample of twins (100 men and 125 women) from these data. A clinical examination some years after the questionnaire showed that the correlation between self-reported and measured BMI was 0.89 among men and 0.90 among women, suggesting good reliability of self-reported BMI (40).
Exact age at response was calculated using the subjects’ birth date, available from the Finnish central population register, and the date of response. Because the response date was missing for some subjects (n = 652) because of clerical error, the mean date of response of all responders was used for these subjects.
Subjects who indicated that they had school age or younger children in the household were classified as having children.
Educational level was divided into four classes based on years of schooling: ≤6, 7 to 8, 9 to 11, and ≥12 years.
Twin zygosity was determined by a validated questionnaire method. The reliability of the questionnaire was assessed by 11 blood markers in a sample of 104 twin pairs classified as monozygous (MZ)1 or dizygous (DZ) (41). Agreement between the blood test and the questionnaire was 100%. The probability of misclassification of a twin pair based on blood tests in this subsample was 1.7%.
The association between smoking and recurrent dieting was analyzed using multinomial logistic regression analysis, with dieting as the outcome variable with three levels (never, one to two times, three or more times), and never-smokers as the reference category for smoking status. All statistical analyses were carried out with Stata statistical software (42). Because BMI was strongly associated with recurrent dieting, all models were adjusted for BMI and age. Because of evidence of age effects in women, we examined the association between recurrent dieting and smoking in three age groups in both sexes. Age adjustment was also performed within these age categories. All individual level analyses were carried out by treating the subjects as singletons. Because the study population was formed by identifying and including twin pairs, the subjects (twin individuals) do not represent fully independent observations but may correlate for study traits within pairs. The effect of the twin-sampling design on standard error was taken into account in the individual level analysis by computing the robust estimate of variance level using the cluster-option in Stata (43).
The association between smoking and recurrent dieting within twin-pairs was examined using conditional logistic regression analysis. If the association is found only in individual level analyses but not within discordant twin pairs, i.e., smoking twins of discordant pairs do not run a higher risk of recurrent dieting than their non-smoking co-twins, this indicates that the association is caused by shared familial or genetic factors affecting both smoking and recurrent dieting. Correspondingly, if the association is found in both individual level and discordant twin pair analyses, it suggests a direct association between smoking and recurrent dieting, or they are both caused by environmental factors or experiences unshared by the twins. Furthermore, because twin pairs share the same family background, MZ twin pairs also possess identical genes, whereas DZ twins share, on average, 50% of their segregating genes, (as do full siblings). Thus, a comparison of MZ and DZ twin-pairs provides further information about the role of genetic and environmental factors behind this association. If genetic factors are the main explanation for the association between recurrent dieting and smoking, this association should be seen only within DZ and not within MZ twin-pairs (44, 45).
To have a clear discordance between pairs, we classified a twin pair as discordant for dieting if one twin reported no dieting and the other reported recurrent dieting (at least three times). Twin pairs were classified as smoking discordant if one twin had never smoked and the other was currently smoking (never-current discordant pairs) or one had never smoked and the other was a former smoker (never-former discordant pairs).
Almost 30% of women and 50% of men were overweight (BMI > 25 kg/m2). Recurrent dieting was more than twice as common among women as men (31.5% vs. 13.1%), whereas current smoking was less common among women (22.3% vs. 32.4%).
Table 1 presents the overall association of dieting with smoking status, education, and BMI. Recurrent dieting was most prevalent among both sexes in the group of former smokers (women, 38.8%; men, 17.3%). Among women, the lowest prevalence of recurrent dieting was in the group of never smokers (29%), whereas among men, it was in the group of current smokers (10.3%). Compared with never smokers, former smokers had significantly more often dieted both one to two times and recurrently in both sexes (p < 0.01), whereas recurrent dieting was more common in currently smoking women but less common among men (p ≤ 0.05). Recurrent dieting was also more prevalent among the higher educated and was associated with greater BMI.
Table 1. . Dieting prevalence (%) in different smoking and educational categories
One to two times
Three or more
p ≤ 0.05
p ≤ 0.005, tested with multinomial regression, with never dieters, never smokers, and lowest education as comparison group.
The association between smoking and body weight differed between the sexes and age groups (Figure 1). Among women, the current smokers were leaner than never smokers in the age groups of 40 to 49 and 50 to 61 years. Former smokers were heavier only in the 40- to 49-year age group. Compared with never-smoking men, the former-smoking men were clearly heavier in all age groups. Moreover, currently smoking men were heavier in the youngest age group, but there was no statistically significant difference in BMI between the current and never smokers in older age groups. Overall, mean BMI increased with age.
Because the association between history of recurrent dieting and smoking status was considered likely to be sex specific, analyses were carried out separately for men and women. This was supported by the finding of a statistically significant interaction between sex and smoking (p < 0.001) in a model using all of the data. Among women, the interaction between smoking and age was statistically significant (p < 0.01). Age was inversely associated with recurrent dieting among never and current smokers, whereas among former smokers, the association between age and recurrent dieting was positive. In men, there was no statistically significant interaction between age and smoking.
The results of the multinomial logistic regression analysis are summarized in Table 2. All of the logistic regression analyses were adjusted for age (also within age groups) and BMI. The association between current smoking and recurrent dieting (three or more times) was different in men and women [odds ratio (OR) range: women, 1.09 to 1.41; men, 0.57 to 0.78 in different age groups].
Table 2. . Results of the multinomial logistic regression (ORs and 95% CIs) for recurrent dieting according to smoking in different age categories
Age 33 to 39 years
Age 40 to 49 years
Age 50 to 61 years
One to two times
Three or more
One to two times
Three or more
One or two times
Three or more
One to two times
Three or more
OR, odds ratio; CI, confidence interval. All analyses were adjusted for age and BMI. Subjects with diagnosed diabetes, cancer, and hypertensive medication for at least 2 months during the previous year were excluded.
1.45 (1.21, 1.74)
1.83 (1.53, 2.19)
1.39 (1.08, 1.80)
1.52 (1.17, 1.97)
1.35 (0.98, 1.85)
1.94 (1.44, 2.63)
1.95 (1.28, 2.98)
2.82 (1.86, 4.26)
1.07 (0.91, 1.27)
1.29 (1.09, 1.54)
1.16 (0.91, 1.48)
1.41 (1.10, 1.81)
0.85 (0.65, 1.12)
1.09 (0.83, 1.42)
1.50 (0.95, 2.37)
1.28 (0.76, 2.15)
0.95 (0.94, 0.96)
0.94 (0.93, 0.95)
1.26 (1.22, 1.29)
1.39 (1.35, 1.43)
1.14 (0.95, 1.38)
1.30 (1.05, 1.61)
1.32 (0.97, 1.80)
1.38 (0.91, 2.09)
1.13 (0.85, 1.51)
1.12 (0.73, 1.72)
0.91 (0.61, 1.36)
1.32 (0.97, 1.80)
0.80 (0.67, 0.96)
0.69 (0.55, 0.87)
0.87 (0.66, 1.15)
0.78 (0.53, 1.16)
0.75 (0.56, 1.00)
0.57 (0.34, 0.98)
0.76 (0.50, 1.15)
0.66 (0.46, 0.94)
0.96 (0.95, 0.98)
0.97 (0.96, 0.99)
1.29 (1.26, 1.33)
1.39 (1.35, 1.43)
The currently smoking women were more likely to have dieted recurrently than never smokers in all age groups, although the finding was statistically significant only in the youngest age group [33 to 39 years; OR, 1.41; 95% confidence interval (CI), 1.1, 1.8]. Among women, the former smokers were significantly more likely to have dieted recurrently in all age groups than never smokers, with ORs ranging from 1.52 to 2.82. In contrast, current-smoking men were less likely to have dieted recurrently than the never-smoking men, with ORs ranging from 0.57 to 0.78. Former-smoking men were more likely to have dieted recurrently than others, but this trend was not significant. The results for those who had dieted one to two times and those who had dieted recurrently were mostly parallel.
BMI was a strong explanatory factor for recurrent dieting in all age groups in both sexes (OR for one unit increase in BMI = 1.39; 95% CI, 1.35, 1.43 for both sexes). We also found that more educated women and men were leaner (mean BMI for more vs. less educated: women, 22.6 vs. 24.5; men, 24.5 vs. 25.6; age-adjusted p < 0.001) and smoked less than those with less education (prevalence of current smoking in more vs. less educated: women, 11.4% vs. 42.0%; men, 12.2% vs. 40.1%; age-adjusted p < 0.001; data not shown). The risk for recurrent dieting in women increased statistically significantly with the presence of children in the household (OR, 1.3; 95% CI, 1.14, 1.47 for women; OR, 1.1; 95% CI, 0.92, 1.31 for men; data not shown). Therefore, we also studied models adjusted for educational level and presence of children, but these adjustments had only minor effects (i.e., no change in statistical significance levels in Table 2) on the association between smoking and recurrent dieting (data not shown).
Finally, we studied the association between smoking status and recurrent dieting in twin pairs discordant for smoking and recurrent dieting behavior (Table 3). We found 458 twin pairs in which one co-twin was a current smoker and the other had never smoked. Of the never-current smoking discordant pairs, 50 were also discordant for recurrent dieting, i.e., one twin had never dieted and the co-twin reported recurrent dieting. Furthermore, we found 497 twin pairs in which one co-twin was a former smoker and the other had never smoked, and 78 of these pairs were also discordant for their dieting status. The results of the conditional logistic regression analyses of these discordant pairs paralleled the results of individual-based analyses, but statistical significance was reached only in the analysis of all never-current smoking discordant pairs (OR, 1.56; 95% CI, 1.20 to 2.03).
Table 3. . Results of discordant pair analysis
OR (95% CI)
Number of pairs
OR, odds ratio; CI, confidence interval; DZ, dizygous; MZ, monozygous. ORs for recurrent dieting among the current/former smoking twins.
Never-current smoking discordance
All discordant pairs
1.56 (1.20 to 2.03)
1.13 (0.50 to 2.55)
0.99 (0.14 to 7.18)
Never-former smoking discordance
All discordant pairs
1.66 (0.91 to 3.04)
1.48 (0.69 to 3.18)
2.80 (0.74 to 10.52)
Our main finding in this large population-based sample of adult Finnish twins is that current smoking was positively associated with recurrent dieting among the youngest women, whereas the opposite was observed among older men. The result concerning current smoking differs from previous studies of the association between smoking and dieting status. French et al. (22) and Serdula et al. (46, 47) reported a reduced likelihood of dieting among current smokers compared with non-smokers of approximately the same magnitude in both sexes. Moreover, in our previous study (9) from survey data collected during 2000 to 2002 in Finland on young adults (23 to 27 years of age), we found an increased likelihood of recurrent intentional weight loss among current smokers compared with never smokers in both sexes. The disparities compared with the results of this study might be caused by differences in the participants’ age and time of data collection, because, in this study, we used data collected in 1990 on older subjects. This suggests that there may be an ongoing transition in the cultures of smoking and dieting so that smoking is more often associated with intentional weight loss among younger generations and women, who, in general, are more prone to dieting and health-compromising dieting methods than older generations and men (13, 14, 15). The youngest female smokers’ higher risk of recurrent dieting might be explained by greater concerns about body image. The sex and age group differences found in this study support the idea of such a transition. Menopausal status might have an effect on both smoking and dieting behavior. Although we had no information on menopausal status of women, the results concerning women younger and older than 50 years of age were much alike.
We found a somewhat higher dieting prevalence than that reported recently by Lahti-Koski et al. (48) in a Finnish population. The definition of weight cycling and recurrent dieting varies in the literature (49), and this difference in prevalence is probably caused by our broader definition of dieting. Lahti-Koski et al. did not report dieting by smoking status, and changes in smoking rates may also account for some of the prevalence differences.
In 1990, when the data were collected, quitting smoking occurred nearly always without professional or pharmaceutical assistance. Nicotine replacement therapy became available in Finland by prescription in 1983 and over the counter in 1989, while practically no treatment groups existed. Multifactorial behavioral modifications (e.g., weight management, nutrition, and exercise) used in treatment groups currently might nowadays confound the intertwining of smoking and dieting. In this study, we found that the former smokers of both sexes, particularly women, were more likely to have dieted recurrently at all ages. This accords with our finding that former-smoking women were not (age classes 33 to 39 and 50 to 62 years) or only slightly (age class 40 to 49 years) heavier than their never-smoking counterparts. This might indicate that women who quit smoking put a lot of effort into weight management and experience some success, whereas quitting men do not. The cross-sectional nature of our data cannot reveal which comes first: smoking cessation or recurrent dieting. We are, therefore, unable to determine whether recurrent dieting is a possible consequence of weight gain induced by smoking cessation or even multiple attempts to quit or whether both smoking cessation and recurrent dieting reflect a long-term process of lifestyle changes.
Korkeila et al. (38) have reported dieting to be more frequent among former smokers than non-smokers, and among young women, heavy smokers were more often reporting dieting than others. Furthermore, dieting was found to be more often reported by subjects with the highest alcohol consumption (young men and older women) and by men with high social class. Various psychological and behavioral factors may explain age and sex differences and need to be addressed in future studies. Although educational level was a strong explanatory factor for both smoking and being overweight in our data, it is interesting that adjusting for educational level did not change the association between smoking and recurrent dieting. This indicates that the association is not dependent on education despite the fact that the more educated groups were leaner and smoked less than the less educated groups.
The use of twin data provides the possibility of controlling for the effect of both common environmental and genetic factors and, thus, yields more information on this association than data on singletons only. Previous studies have shown the importance of genetic and common (familial) background factors in smoking behavior (50). Korkeila et al. (38) reported that weight loss attempts have a common family background, which may, moreover, be partly mediated through the genetic component of BMI. In younger adult twins, the genetic effects on recurrent weight loss attempts were only partially correlated with genetic effects on BMI (51). The number of pairs discordant for smoking and recurrent dieting was relatively small in our dataset, and no statistical significance was reached in the analysis of these pairs. However, for the smoking discordant pairs, the point estimates for current-smoking women and former smokers of both sexes were greater than unity and, thus, consistent with the results from the individual level analyses. This suggests that the association is not caused by factors shared by the twins. The power of further statistical analyses of MZ and DZ twins was inadequate to permit definite conclusions about the role of genetic factors.
The strengths of this study include a large and representative population with a high participation rate. Self-reported smoking status has been shown to be reliable in the Finnish population-based data (52), as has self-reported BMI (40). The demographic factors of population-based twin samples can be reliably compared with those of the population at large, and smoking rates among adult twins and singletons are reportedly very similar (52). Because of the cross-sectional nature of our study, no firm conclusions about causal relationships between smoking and recurrent dieting behavior can be made, but the pairwise analyses support such a relationship. Unfortunately, we had no information on the motivational reasons for smoking or smoking cessation, nor did we have detailed information on dieting methods used by the subjects.
In conclusion, our results suggest that different approaches may be needed to promote healthy weight control methods and to prevent smoking in middle-age and young adult men and women, because smoking and dieting habits are associated differently between sexes and age classes. The finding that the association of current smoking and recurrent dieting was independent of educational level indicates that healthy weight control interventions are needed at all levels of society. Further longitudinal studies of the associations among recurrent dieting, body weight, and smoking behavior would be needed to effectively plan interventions.
S.E.S. received financial support from the Yrjö Jahnsson Foundation (Grant 5074), K.S. was supported by the Academy of Finland (Grant 108297), and S.S.-L. was supported by the Eurobese project (SAS6 Contract 016646). This paper originates from the GenomEUtwin project supported by the European Union, Contract QLG2-CT-2002-01254. The Finnish Twin Cohort study is part of the Academy of Finland Centre of Excellence in Complex Disease Genetics.