Study design

The Youth Smoking Survey (YSS) is a biennial, national, provincially generalisable, school-based, paper-and-pencil survey that measures knowledge, attitudes and behaviours related to tobacco use among Canadian students.37 The target population is grades 6 through 12 students (ages 11–19) at public and private schools (n=450) in 9 provinces. In the 2012/2013 cycle, those residing in Manitoba, Yukon, Nunavut and Northwest Territories and those living in institutions or on First Nations reserves (representing about 5% of the Canadian population) were excluded. Surveys were pilot tested to assess student understanding of the questions. Approximately three-quarters (73%) of respondents participated with passive parental permission, and 27% participated with active parental permission. The survey was administered during class time, and student participants were not remunerated. Survey development, design, weights, response rates and data collection protocol for the 2008 YSS have been published.37 On average during the 2012/2013 YSS wave, 64% of schools that were approached participated (range, 38% in Ontario to 96% in Newfoundland) and 72% of eligible students completed questionnaires. The 2012/2013 YSS was administered to 47 203 youths in grades 6 through 12 attending schools from 9 (of 10) Canadian provinces (of note, in Quebec, secondary school ends at grade 11.) Given the low prevalence of tobacco use among grades 6–8 students, this study uses data from grades 9 to 12 students (n=27 404). The 2012/2013 YSS survey wave collected data between November 2012 and June 2013 and were analysed in 2015.

Measures and data sources

Only data from non-smokers (defined as those who answered ‘no’ to the question, ‘Have you ever tried cigarette smoking, even just a few puffs?’) who had non-missing values for the primary outcome of interest (smoking susceptibility) were used in the current study (n=17 396). The dichotomous outcome of interest from the 2012/2013 YSS data set was ‘susceptible to smoking’ versus ‘not susceptible to smoking’. Students who were considered not susceptible to smoking were those who responded ‘definitely not’ to three questions: ‘Do you think in the future you might try smoking cigarettes?’; ‘If one of your best friends was to offer you a cigarette, would you smoke it?’, and; ‘At any time during the next year, do you think you will smoke a cigarette?’ Methods of characterising ‘susceptible non-smokers’ were developed by Pierce et al33 and are consistent with those used in the Global Youth Tobacco Survey.36

Two independent variables of interest included ‘ever use’ of ATPs and of flavoured tobacco products. Ever use of ATPs was derived from students’ responses to the question, ‘Have you ever tried any of the following? (mark all that apply)’ with response options: smoking pipe tobacco; smoking little cigars or cigarillos, smoking cigars (not including little cigars or cigarillos, plain or flavoured); smoking roll-your own cigarettes; smoking bidis (little cigarettes that are hand-rolled in leaves, tied with a string at the ends, and may come in different flavours); using smokeless tobacco (chewing tobacco, pinch, snuff or snus); using a water-pipe (hookah) to smoke sheesha (herbal or tobacco); using blunt wraps (a sheet or tube made of tobacco used to roll cigarette tobacco; I have not tried any of these things. Students who marked the ‘I have not tried any of these things’ were considered to have never used an ATP. Those with non-missing data who marked at least one of the other options were considered to have ever used an ATP. Ever use of flavoured ATPs was derived from students’ response to the question, ‘Have you ever used flavoured tobacco products (menthol, cherry, strawberry, vanilla, etc.)?’ with response options yes/no. Students who responded yes were considered to have ever used a flavoured tobacco product.

Two other independent variables of interest were past 30 day tobacco use and past 30 day flavoured tobacco use. Past 30 days tobacco use was derived from the question, ‘In the past 30 days, did you use any of the following?’, with the same response options as those assessing ever use of ATPs (see previous paragraph). Past 30 day flavoured tobacco use was derived from student responses to the question, ‘In the past 30 days, did you use any of the following flavoured tobacco products?’ with response options: menthol cigarette; flavoured little cigar or cigarillo; flavoured cigar; flavoured tobacco in a water-pipe (hookah), and; I did not use any of these things in the past 30 days. Students who marked the ‘I did not use any of these things in the past 30 days’ were considered to have not used a flavoured tobacco product in the past 30 days. Students with non-missing data who marked at least one of the other options were considered to have used a flavoured tobacco product in the past 30 days.

Covariates included the respondent's sex (male, female), grade (9–12), province (where the four Atlantic provinces were grouped together because of cultural similarity and relatively small n), self-reported race/ethnicity (white, black, Asian, Aboriginal, Latin American/Hispanic, or ‘other’), level of urbanisation (urban, rural), amount of weekly spending money received (no money, less than $10/week, $10–40/week, more than $40/week), whether or not any close relatives smoke (yes, no), and whether any close friends smoke (none, one friend, two or more friends).

Two additional covariates included self-esteem and risk of having an alcohol or drug-related disorder. Self-esteem was included as a potential confounder given the generally negative association between self-esteem and smoking.38 ,39 Self-esteem score was derived from participant responses to the following three statements: In general, I like the way I am; When I do something, I do it well; I like the way I look. Response options were on a five-point Likert scale ranging from True to False. Scores were summed (responses to each statement were scored from ‘True’=4 to ‘False’=0) and categorised into three levels of self-esteem: less than median (score ≤8), median to less than 90th centile (score 9–11) and 90th centile or above (score 12+). Risk of having an alcohol or drug-related disorder was included as a potential confounder because of the significant association between tobacco use and alcohol and/or illicit drug use among youth.40 ,41 Risk of having an alcohol or drug-related disorder was considered dichotomous (less than high risk, high risk), and was derived from participants’ yes/no responses to the following questions: ‘Have you ever ridden in a car driven by someone (including yourself) who was ‘high’ or had been using alcohol or drugs?’; ‘Do you ever use alcohol or drugs to relax, feel better about yourself, or fit in?’; ‘Do you ever use alcohol or drugs while you are by yourself, alone?’; ‘Do you ever forget things you did while using alcohol or drugs?’; ‘Do your family and friends ever tell you that you should cut down on your alcohol or drug use?’; ‘Have you ever gotten into trouble while you were using alcohol or drugs?’. Participants were defined as no/moderate risk if they answered yes to zero or one of the questions, and participants were defined as high risk if they answered yes to at least two of the questions.

Statistical analysis

Survey weights were used to adjust for sample selection (school and class levels), non-response (school, class and student levels), and post-stratification of the sample population relative to grade and sex distribution in the total population. Data were weighted so that the variances take account of the sample design. Descriptive statistics were used to show the weighted distribution of smoking susceptibility by sex, grade, geographic region, self-reported race/ethnicity, self-esteem category, weekly spending money, family member smoking, the number of closest friends smoking and the risk of having an alcohol or drug-related disorder. The weekly spending money variable had a high proportion of missing values (20%), so was not included in any models. For all other variables, less than 10% were missing values. Rao-Scott χ² statistics were applied to generate a p value for the statistical difference of each association.

To examine smoking susceptibility among never-smokers, two multiple logistic regression models were fitted to examine associations between sociodemographic characteristics and lifestyle manner and odds of having the outcome, using PROC SURVEY LOGISTICS in SAS 9.4 (SAS/STAT software, SAS Institute Inc, Cary, North Carolina, USA). Model 1 included all sociodemographic variables, and model 2 included all sociodemographic and lifestyle characteristics.

To examine differences in smoking susceptibility by ATP use, we separately conducted five nested case–control analyses within this survey population. The first case–control analysis compared the outcome among the case group ‘Never tried cigarette smoking, but ever tried other types of tobacco’ to that of the control group ‘Never tried cigarette smoking, never tried any types of tobacco’. The second case–control analysis compared the outcome among the case group ‘Never tried cigarette smoking, but ever tried some flavoured tobacco’ to that of the same control group. The third case–control analysis compared the outcome among the case group ‘Never tried cigarette smoking, but used some flavoured tobacco in past 30 days’ to that of the control group ‘Never tried cigarette smoking, did not use any types of tobacco in past 30 days’. The fourth and fifth case–control analyses aimed to examine the extent to which flavoured ATP use was associated with smoking susceptibility controlling for ATP use. Therefore, the fourth case–control analysis compared the outcome among the case group ‘Never tried cigarette smoking but ever tried flavoured tobacco’ with the control group ‘Never tried cigarettes, never tried flavoured tobacco, ever tried other types of tobacco’, and the fifth analyses compared the outcome among the case group ‘Never tried cigarettes, but tried some flavoured tobacco in the past 30 days’ with the control group ‘Never tried cigarettes, no flavoured tobacco in the past 30 days, but tried ATP in the past 30 days.’

All covariates were individually assessed as confounders using the following criteria: (1) how it was associated with the main exposure, (2) how it was associated with the outcome and (3) whether it was theoretically a pathway of disease process. A covariate was included in the final model if it produced a significant association with the outcome with a p value less than 0.05 for any level of the variable. At first, the full models consisted of main exposure, outcome and potential confounders. Assumptions of logistic regression were checked and Goodness-of-fit test was used to check model fit. A backward elimination strategy was employed in order to evaluate each covariate in the presence of others. Potential confounders were removed one at a time if change-in-estimate of effect measure was less than 10%. The final model (reduced model) for each exposure was tailored to include only covariates that appeared to be a confounder in order to avoid over adjustment and a residual confounding effect.