Overweight and perception of overweight as

RESEARCH REPORT

doi:10.1111/j.1360-0443.2009.02506.x

Overweight and perception of overweight as predictors of smokeless tobacco use and of cigarette smoking in a cohort of Swedish adolescents Maria Paola Caria1, Rino Bellocco1,2, Antonella Zambon1, Nicholas J. Horton3 & Maria Rosaria Galanti4,5 Department of Statistics, University of Milano-Bicocca, Milan, Italy,1 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden,2 Department of Mathematics and Statistics, Smith College, Northampton, MA, USA,3 Stockholm Centre for Public Health/Tobacco Prevention, Stockholm, Sweden4 and Clinical Epidemiology Unit, Department of Medicine, Karolinska Institutet, Stockholm, Sweden5

ABSTRACT Aims To study the association between measured or perceived overweight in adolescence and subsequent uptake of cigarette smoking and of the Swedish smokeless tobacco ‘snus’ (oral moist snuff). Design Population-based prospective cohort study with 7 years’ follow-up. Setting Self-administered questionnaires and school nurses’ visits. Participants A total of 2922 children of both sexes and mean age 11.6 years at recruitment, resident in the Stockholm region, Sweden. Measurements Tobacco use was self-reported at baseline and on six subsequent surveys. Subjects’ height and weight were measured by school nurses during the first 4 years, self-reported thereafter. Overweight perception was self-reported at the age of 15 years. Findings Overweight and perception of overweight were not associated with subsequent uptake of either smoking or snus among males. Among females, overweight at baseline was associated with uptake of smoking [adjusted hazard ratio (HR) 1.34, 95% confidence interval (CI) 1.09–1.63], but not of snus. A similar pattern was found with overweight status during follow-up. Among girls with low-educated parents, overweight at baseline predicted the uptake of both snus and smoking. Among 15-year-old females who never used tobacco perceived overweight was associated with subsequent uptake of smoking (adjusted HR 1.71, 95% CI 1.20–2.46), but not of snus. Conclusions In Sweden, adolescent girls with actual or perceived overweight are at increased risk to start smoking, while indications that this increased risk applies to smokeless tobacco (snus) are limited to girls of low socio-economic status. Keywords

Adolescence, body perception, longitudinal study, overweight, smokeless tobacco, smoking, Sweden.

Correspondence to: Maria Rosaria Galanti, Department of Medicine, Clinical Epidemiology Unit, T2, 17176 Stockholm, Sweden. E-mail: [email protected] Submitted 18 June 2008; initial review completed 9 September 2008; final version accepted 5 December 2008

INTRODUCTION The association of tobacco use with body weight is not understood completely. First, while the increase in weight among those who quit smoking is well established [1,2], whether smoking initiation is followed by weight change is far less clear, because carefully designed longitudinal studies are rare [3]. In fact, smoking may be perceived as a tool to control weight, at least among young women [4], and overweight adolescents seem to take up smoking to a higher extent than their leaner peers [5]. Thus, the relationship between initiation of tobacco use and subsequent changes in body weight may be confounded by

pre-existing weight and lead to inconsistent results, especially in cross-sectional studies. Secondly, there are unanswered questions concerning the relation between weight, weight concerns and use of smokeless tobacco (SLT). In cross-sectional studies, adults SLT users have been found to have higher body weight than non-users [6,7], but it is unclear whether this represents selection of subjects wishing to control weight [8,9] or a causal effect of SLT use on metabolic processes [10]. In the last decade, low-nitrosamine SLT, such as the Swedish brands of snus (oral moist snuff), has attracted attention as a potential harm-reduction tool in the process of controlling the smoking epidemic [11]. While the debate in the public

© 2009 The Authors. Journal compilation © 2009 Society for the Study of Addiction

Addiction

2

Maria Paola Caria et al.

health community is still ongoing [12], manufacturers of snus have begun marketing their products in countries outside Europe [13]. In Sweden new portion-bag brands introduced in the 1970s led to widespread use of snus among men of all social groups, and rising among women [14]. This suggests that use might rise elsewhere. It can be hypothesized that health-conscious individuals may use SLT [15] because they perceive it as safer than cigarettes, while conveying the same advantages in terms of weight control. To our knowledge, the relation between weight, perceived weight and SLT use among adolescents has never been explored. Sweden presents an ideal setting for studying behavioural determinants of SLT use in adolescence, because in 2006 the prevalence of snus use was 20% among 9th grade male students and 7% among same-age females; the latter representing a twofold increase during 10 years [16]. In this study, we present an analysis of actual and perceived overweight as predictors of tobacco uptake in a longitudinal study of youths with follow-up between 11 and 18 years of age. We hypothesized that youths being or perceiving themselves to be overweight would have higher risk of initiating tobacco use compared to non-overweight, with similar risks for cigarette smoking and use of snus.

METHODS The BROMS cohort study (Swedish acronym for ‘Children Smoking and their Environment in Stockholm region’) was approved by the Karolinska Institutet’s Ethical Board at Huddinge University Hospital, in the Stockholm region of Sweden. Subjects The characteristics of the design and of the study cohort at baseline have been described in detail elsewhere [17,18]. Briefly, the BROMS cohort encompasses 3020 adolescents of both sexes, who in January 1998 participated in a school-based survey of tobacco use based in the urban area of the Stockholm region of Sweden. The baseline assessment was conducted in grade 5, at a mean age of 11.6 years. The adolescents were invited to participate in a similar annual survey up to the age of 15 years (four follow-up waves during compulsory school education). Thereafter, two further follow-up surveys were conducted, at the ages of 17 and 18 years. The annual retention rate ranged from 96% (at follow-up 1, age 12) to 82% (at follow-up 6, age 18). The proportion participating in all follow-up waves was 69%. Participants who remained in the study did not differ significantly from subjects who failed to participate in one or more surveys for either mean body mass index (BMI) at baseline or classification

of overweight. Completers were more likely than dropouts to have parents with college education, to endorse the perception of overweight at the age of 15 years, and were less likely to have started smoking or using snus by the end of follow-up. The present analysis was based on 2922 subjects who at baseline reported never regular or current smoking/ use of snus.

Information on tobacco use, perception of body mass and other covariates Tobacco use was self-reported by the adolescents in a paper-and-pencil questionnaire, with similar, but separate questions addressing the use of cigarettes and of snus. A study of the validity of self-reports against a biomarker of nicotine intake (cotinine in saliva) indicated high reliability of the adolescents’ reports of current tobacco use [19]. Questions inquired about: (1) ever trying cigarette or snus; (2) life-time number of cigarettes or snus dips; (3) regular use of tobacco ever in life, defined as weekly smoking/snus use for at least 3 consecutive months; and (4) use at present (none or less than monthly, monthly, weekly, daily). For the purpose of this analysis, current use of either cigarettes or snus was defined as use at least monthly. At the fourth follow-up survey (age 15), the adolescents were asked to report on their perceived body mass by choosing between five response alternatives on a Likert scale ranging from ‘I think I am much too thin’ to ‘I think I am much too fat’. We employed a survey question from the multi-national survey Health Behavior among School Children, Swedish version [20]. An out-of-scale further response option was added—‘I never thought of it’—to allow for unawareness or lack of interest. A score of four (‘a bit too fat’) or five (‘much too fat’) on the Likert scale was categorized as perception of overweight. Missing information for this variable concerned 11.4% of the subjects, 80% of which was accounted for by nonparticipation in the survey and 20% by item missingness. Observation of this variable was not associated with other relevant predictors used in these analyses (i.e. BMI/ overweight at baseline, initiation of smoking/snus use, parental education). Because in Sweden, as in many other developed countries, socio-economic status is a strong predictor both of tobacco use and of overweight [21], we considered parental education both as potential confounder of the association between offspring’s overweight and later smoking and as a possible effect modifier. Parental education (years of schooling) was self-reported by each parent at baseline, categorized into ‘compulsory school’, ‘senior high school’ or ‘college’. We used mother’s education, unless information on mother’s education was not


Addiction

Overweight and use of tobacco among adolescents

available, in which case the father’s information was used. Current cigarette smoking and snus use was selfreported separately at baseline by both parents. For the purpose of this analysis it was categorized dichotomously as ‘at least one parent’ versus ‘neither parent’ currently using any type of tobacco. Information on weight and height At baseline and during the three following surveys, the adolescents’ weight and height were measured by the school nurses using a standardized protocol (standing, without clothes and shoes). During the third follow-up survey the students, now of average age 14 years, selfreported in the questionnaire their weight and height under the same specified conditions of measurement. After the third follow-up survey, the information on anthropometric measures was self-reported by the adolescents only. Based on the growth charts for a Swedish sample [22], increases exceeding 10% in height and 30% in weight during 1 year were considered unreliable. We also considered unreliable yearly decreases exceeding 2% in height or 20% in weight. Overall, 269 unreliable values from 243 subjects were set to missing. These subjects did not differ from subjects with completely reliable measures with regard to BMI/overweight status at baseline, perception of overweight and parental education. However, they were significantly more likely to have initiated smoking and snus use by the end of follow-up. BMI was estimated as the ratio weight (kg)/height (m)2, using nurses’ measurements until available (third follow-up), students’ self-reported height and weight thereafter. The cut-off for overweight for each sex and year of age was equalized to the adult cut-off of BMI ⱖ 25, based on measurements derived from an international sample [23]. Information on BMI was missing for 4.8% of the subjects at the first follow-up, up to a maximum of 33.5% of the subjects at the fourth follow-up (age 15). At the third follow-up (average age of the participants 14 years) the total concordance between the classification of overweight based on nurse measurements and that based on students’ self-reports was 94.7%; kappa statistic for inter-rater agreement was 0.76. The concordance was higher for females (96.8%, kappa = 0.80) than for males (92.6%, kappa = 0.73). Statistical analysis As gender differences concerning both tobacco use and overweight were observed at baseline in our cohort, analyses were conducted separately by gender. In order to explore the association between overweight or perception of overweight and subsequent tobacco use, standard statistical methods were implemented to estimate the

3

survival function, specifically the Kaplan–Meier technique, and assess group differences through the log-rank test. In the second step, Cox proportional hazard regression models were fitted to estimate hazard ratios (HR) and the 95% confidence interval (CI) of initiation of any or daily smoking, snus use or use of any tobacco. The assumption of proportional hazards was assessed, both graphically and analytically, using the scaled Schoenfeld residuals [24]. Time of smoking initiation was assigned at the first year when the child reported either being a current smoker or having been a past regular smoker. The same definition was applied to initiation with snus. Time of initiation with any tobacco was set at the first year when use of cigarettes or snus was reported, whichever occurred first. In the case of missing information on current or past regular use in one or several consecutive surveys, followed by a report of current use in the subsequent survey, time of onset of use was set back at the year when the subject reported having smoked at least 10 cigarettes or having used snus at least 10 times, respectively. Time of initiation of daily use was set at the year when the subject reported for the first time current daily use of cigarettes and/or snus. In order to study the simultaneous effect of being overweight at baseline and during time of follow-up, an extended version of the Cox model was implemented to adjust for time-dependent variables. The same models were used to analyse perception of overweight reported at the age of 15 years as predictor of the onset of any or daily tobacco use between age 16 and 18. In this latter analysis, the study sample consisted of adolescents who at the age of 15 years had not initiated tobacco use and reported on their perception of overweight. In all models, the estimates were adjusted for parental tobacco use (neither parent uses tobacco/at least one parent uses tobacco) and for parental education (compulsory; high school; college and above). Finally, all analyses, performed using the statistical package Stata version 10 [25], were conducted on two data sets: (i) excluding unreliable values for weight and height and (ii) preserving the original information.

Missing data Missing data affected primarily weight and height, although other covariates were sometimes missing. Therefore, we assessed whether the main findings were robust to different assumptions regarding missingness, using a series of sensitivity analyses. First, missing data on parental education, parental use of tobacco, weight, height and body mass perception were handled by means of multiple imputation (MI) [26,27]. All variables used in


Addiction

4

Maria Paola Caria et al.

the model for the complete case analysis were used for the chained equation main effects imputation model [26,27], which generated 20 different data sets with imputed values for the missing data. The estimated model coefficients from each imputed data set were then combined using Rubin’s procedure [28]. A secondary sensitivity analysis imputed the dichotomous variable for overweight under the extreme assumptions that missing values were either all positive or all negative. Finally, we also imputed the missing data for overweight by carrying forward the value last observed (LOCF).

RESULTS The baseline and outcome characteristics of the study sample are shown in Table 1. The mean BMI was 18.50 for both males [standard deviation (SD) = 2.83] and females (SD = 2.96). Altogether 17.7% of the subjects were classified as being overweight at baseline; this proportion was significantly higher among males than among females. The associations between being overweight and subsequent uptake of smoking and snus are displayed in Tables 2 and 3. Because the results obtained from Cox proportional hazard regression models and discrete-time hazard

models were relatively consistent, only the former are reported. We report the results obtained from the analysis of the data set after the exclusion of unreliable values, as these did not differ from those based on the original data set. The hazard of snus uptake was not affected significantly by being overweight at baseline or during follow-up in either gender group. On the other hand, being overweight at baseline was associated with uptake of smoking among girls (results were similar when overweight status was treated as a time-dependent variable; Table 2). The results were also similar when the outcome was represented by onset of daily smoking or daily snus use (data not shown). In a separate analysis in subgroups of parental education the hazard for overweight at baseline was twice that of the non-overweight for the uptake of smoking (HR = 2.23, 95% CI = 1.33–3.72) and three times the hazard for the uptake of snus (HR = 3.44, 95% CI = 1.25–9.48) among girls whose mothers achieved only compulsory school education. To test formally if the effect of being overweight at baseline differed across levels of parental education, an appropriate interaction term was included in the model as a new variable, equal to the product of the two factors. However, this test did not reach statistical significance (P = 0.18 for snus uptake; P = 0.11 for smoking uptake). Girls who perceived themselves as somewhat or definitely overweight at the age of

Table 1 Characteristics of the study sample, the BROMS (Swedish acronym for ‘Children Smoking and their Environment in Stockholm region’) Cohort Study, Stockholm, 1998–2005.

Parental education at baseline Compulsory school Senior high school College Parental use of tobacco at baseline No Yes Initiated tobacco use during follow-up Not initiated Only cigarettes Only snus Both cigarettes and snus BMI at baseline based on weight and height measured by nurses I quart (BMI ⱕ 16.4) II quart (16.4 < BMI ⱕ 17.9) III quart (17.9 < BMI ⱕ 19.9) IV quart (BMI > 19.9) Overweight at baseline according to reference [21] No Yes

Females (n = 1446)

Males (n = 1476)

na

na

%

%

170 531 730

11.9 37.1 51.0

171 576 712

11.7 39.5 48.8

676 719

48.5 51.5

722 718

50.1 49.9

777 545 20 104

53.7 37.7 1.4 7.2

786 150 156 384

53.2 10.2 10.6 26.0

362 356 378 350

25.0 24.6 26.2 24.2

359 395 343 378

24.3 26.8 23.3 25.6

1213 233

83.9 16.1

1192 283

80.8 19.2

c2

P-value 1.79

0.41

0.80

0.37

490.04