Original Article

Differentiating between girls and boys in transition through smoking stages: A sex-specific growth mixture modeling

Introduction

Many studies reports the burden of smoking-related diseases and their mortality, ¹ the prevalence, ^2-4 theoretical models, ⁵ and risk factors ⁶ for smoking initiation in adolescents.

Nevertheless, smoking is a complex process, and adolescents pass through a number of different stages on the way to become smokers. ⁷ Few studies, especially few longitudinal ones, have been conducted on the transition through smoking stages and the predictors. ^7-11 Determining the process of smoking stages and modeling for predictors is necessary to implement policies.

On the other hand, the smoking behavior is different between boys and girls, ¹² and the rate of smoking experimentation is sex-dependent. This rate is higher in boys than girls, ¹³ and boys tend to initiate smoking earlier than girls. ^13-15 It was found that, girls who do not live with their families are more likely to experiment with smoking. Among them, the process of smoking initiates earlier than those who are living with their families. ¹³ Besides, when friends smoke, boys are five times more likely to smoke, and this rate approximately increases to 9 times in girls. ¹³ Therefore, sex-specific modeling of transition in different stages of smoking appears necessary.

In another point of view, few studies model the transition through smoking stages, which are limited to the single value probability to describe the transition process. Secondly, they are restricted to the certain statistical assumptions in the modeling process. ^8,9,16 On the other hand, the advantages of newly established model, are: (1) GMMs have less bias and error, (2) give more detailed information, and (3) produce findings that are more accurate. ^17-19 GMMs also relaxed some of the restricting statistical assumptions of the traditional models, so GMMs are feasible in practice. ¹⁹

Moreover, due to sex differences, ^13,20 it seems that transition through smoking stages and the predictors should also be modeled sex-specifically. Based on a general search, no studies (if any) have been found that use sex-specific multi-group GMMs to examine the transition through smoking stages. This study aimed to use sex-specific multi-group GMMs to determine the transition through smoking stages and its predictors in adolescents in a longitudinal study.

Materials and Methods

The study procedure details have been published elsewhere. ^10,16 This study pursued an innovative perspective to differentiate the girls and boys in the transition through smoking stages using the sex-specific GMMs.

Study design

We used a cohort study to assess the transition through smoking stages of an illustrative sample of 10th-grade students in Tabriz metropolitan, Iran.

Procedures

A total of 2959 (47.1%) girls and 2241 (42.9%) boys (4903 students) participated in this study, with 82 and 114 clusters (out of 865) respectively of male and female students randomly, and proportionally, selected using a multistage sampling procedure. The pupils filled out an anonymous multiple-choice questionnaire in the first and second year of study.

Sample size

The sample size (of 4903) was much higher than the suggested rule-of-thumb of 10 samples per variable required for GMMs. ²¹ Subsequently, the samples were allocated to schools and then classes proportional to the size of schools and classes, and separately for girls and boys.

Variables and measures

The variables and measures description are presented elsewhere, ¹⁶ though, we describe them here briefly. The participants’ demographic characteristics, including age, gender, living with parents, grade, major and socioeconomic status (SES) using a checklist. Also, the smoking-related variables, including having smoker friends (by a no/any binary response), having the smoker in the family (no/any), self-injury (no/any), and substance abuse (no/any), were measured using a validated checklist. We measured attitude toward smoking score by a 6-item questionnaire of Hill et al (in the range of -12 to 12) ²² and self-esteem by a 10-item Rosenberg questionnaire. ²³ Also, we assessed general risk-taking behavior by one question (“Do you enjoy the exciting things that are a little risky?” with (0: No, 1: Yes) binary response). The main outcomes, including smoking status, intention to start smoking, and smoking during the past month/week, were assessed using an algorithm ¹⁶ (see Table 1).

Statistical analyses and model development:

Data were analyzed by Mplus 7.4 ¹⁸ and SPSS 17 (SPSS Inc., Chicago, IL, USA) and expressed using mean (SD) and n(%) for numeric and categorical variables, respectively. The numeric variables were checked and confirmed for normality by distribution measures, namely skewness within ±1.5, and kurtosis within ± 2. Changes were assessed using McNemar and marginal homogeneity tests within the phases of the study.

We used growth mixture models (GMMs) to model the change of primary outcomes within two points of the study by estimating the intercept, showing the outcome level at zero time, and the slope, showing the change rate in outcome over time. ^17,19 Akaike information criteria (AIC), Bayesian information criteria (BIC), and adjusted BIC (ABIC) as well as the entropy values (>0.80) were utilized to select the optimal model were used to choose among the intercept only and 2- to 6-class linear models. ^24,25 Furthermore, we used the likelihood ratio test to compare the model of interest with a model with one smaller class. ²⁴ Finally, we fit the best-fitted GMMs with covariate to model the underlying demographic and smoking-related variables. ¹⁸

Results

Participants’ profile

The details of the recruitment and participants’ profile are presented elsewhere. ^10,16 Of 4903 participants, 2956 (56.9%) were girls. The girls’ and boys’ mean age was 15.6 (SD 0.7) and 15.8 (SD 0.8) years, respectively. The girls had significantly higher grades than boys did, and the major of most girls were science and humanity. Most of the girls had a higher SES level compared to boys. The prevalence of cigarette smoking among girls was lower than among boys. The girls had a lower percent of the smoking friend and smokers in their family and had a lower prevalence of substance abuse than boys had (all P < 0.001). The girls had a lower attitude toward smoking and higher self-esteem scores than boys had (all P < 0.001).

Besides, for the primary outcomes of the study, the girls reported a higher percentage of “never smoked” in their smoking status. Also, the girls reported a higher percentage of “sure to never start smoking” in their smoking status than boys did. In contrast, the girls reported a lower percentage of “smoking during the past month” and “smoking during the past week” than boys did (all P < 0.001). Furthermore, there were no significant differences between girls and boys regarding other variables (Table 1).

Distribution of changes in primary outcomes

There were 151 (5.9%) changes for girls vs. 300 (16.6%) changes for boys toward higher levels in smoking status (P < 0.001), and 141 (5.6%) changes for girls vs. 192 (10.4%) changes for boys toward higher intention to start smoking. However, there were 84 (3.3%) changes for girls vs. 126 (6.8%) changes for boys toward lower levels in intention to start smoking (P < 0.001). Furthermore, boys significantly tend to smoke more during the past month/week in the second assessment compared with the first assessment (both P < 0.05), but the changes were not significant among girls (both P > 0.05).

Selecting the optimal GMMs class number

For the 2-class model, smaller AIC, BIC, and ABIC were observed compared with 1-class model, whereas the amount of change was negligible compared to the models with a higher number of classes. Besides, the entropy value was acceptable for this model (0.959). Therefore, we chose the optimal 2-class model for further assessments (Table 2).

Name and the prevalence of the classes in the 2-class GMMs

presents the sex-specific loading of the 2-class model for the primary outcomes. Class 1 showed similar patterns between girls and boys for all outcomes, except for “Never smoked” and “Tried only a puff or 1–2 cigarettes” items, wherein light differences could be observed. Both girls and boys responded “Never smoked” and “Sure to never start smoking” and “not smoked” items with a significantly high probability (all P < 0.05). We named this class “Non-smokers,” which had a prevalence of 48.7% among girls vs. 34.4% among boys.

Close

Figure 1. The loading of outcomes in the probability scale for two classes in girls (n=2956) and boys (n=2241). SmkStat: Smoking Status; IntSmk: Intention to start smoking.

Class 2 showed different patterns between girls and boys for nearly all outcomes. For girls, the smoking status had significant loadings on lower and middle levels. However, boys had significant loadings on the middle and higher levels of the items. Among the girls, the “never smoked,” “Tried only a puff or 1–2 cigarettes,” and “Smoked > two but < 100 cigarettes in lifespan” items, while among the boys, “Tried only a puff or 1–2 cigarettes” and “Smoked > two but < 100 cigarettes in lifespan” and “Smoke daily or almost every day” items had significantly high loadings. Besides, the loadings of “not smoked” items for smoking during the past week were significant for girls, while the loading of both “Not smoked” and “Smoked” items was significant for boys (All P < 0.05). On the other hand, in class 2, the same pattern was observed for girls and boys for intention toward smoking and smoking during the past month, with significant loadings on nearly all items for both girls and boys (all P < 0.05). The values of loadings on “Smoked” item were considerably larger among boys than that among girls. We named this class “Occasional/Intending smokers”, which had a prevalence of 8.1% among girls vs. 8.8% among boys, with a similar pattern for both assessment times ().

Change in primary outcomes based on GMMs

For all outcomes, in both classes, the intercepts of the model were negative and significant compared with the intercept of the “Occasional/Intending smokers” among boys (all P < 0.05), indicating a lower level of the outcomes compared with the reference level.

Besides, the slope of the model was positive and significant for all outcomes in both “Non-smokers” and “Occasional/Intending smokers”, indicating an increased rate of change in the outcomes over time. There was an exception for smoking during the past month/week among girls in “Non-smokers”, wherein a negative slope was observed (Table 3).

Background characteristics and changes in outcomes

Higher SES, having non-mathematics majors, having a smoker friend, higher levels of attitude toward smoking, and general risk-taking behavior were directly related to positive changes in smoking status. Also, the results indicated that higher age, living without parents, higher SES, having non-mathematics majors, having a smoker friend, and higher levels of attitude toward smoking were directly related to positive changes in intention to start smoking (all P < 0.05). Besides, higher age, higher SES, having a mathematics major, having a smoker friend, having a smoker in family, and higher levels of attitude toward smoking were directly related to changes to smoking during the past month/week (all P < 0.05) (Table 4).

Discussion

A general overview of the most important results

This study is one of the rare longitudinal studies that use sex-specific GMMs to describe the transition through smoking stages, and model the transition predictors. GMMs presented an optimal 2-class model, utilizing information criteria: the “Occasional/Intending smokers” with 8.1% and 8.8% and the “Non-smokers” with 48.7% and 34.4% prevalence among girls and boys, respectively. Remarkably, negative intercepts and the positive slopes (except for smoking during the past month/week among girls in the “Non-smokers”), indicated negative level and the increasing rate of change in the outcome categories over time, respectively. Besides importantly, age, living with parents, SES, major, having smoker friends, having a smoker in family, self-injury, substance abuse, attitude toward smoking, and general risk-taking behavior were significantly related changes in outcomes and affected the speed of transition. Also, significant differences were observed between girls and boys in terms of changes in smoking stages, intention to start smoking, smoking during the past month/week over time. To sum up, it can be concluded that GMMs provide relevant findings of the sex-specific transition through smoking stages and the predictors.

Utilizing GMMs

GMMs comprise distinguishing features over conventional statistical methods that have been used to study the transition in smoking stages as a latent variable. ²⁶ GMMs, like traditional methods, classify the heterogeneous population into homogenous subgroups and construct latent classes. However, in our data, GMMs, by fitting a tailored transition model, provided more information than conventional methods. In contrast, the traditional models are restricted to a single probability of transition and do not produce the statistical significance for the effect of the covariates. ^7,8,11,16 Furthermore, GMMs are not limited to the restricting statistical assumptions for estimating the parameters, which may not be satisfied in many practical situations. ^17-19 These features propose GMMs in studies aiming at assessing transition through smoking stages, taking an advantage over the conventional procedures. ^7-9,11,16

The process of sex-specific transition in smoking stages

In general, an increasing trend of transition to higher stages of smoking over time was observed. In the “Non-smokers”, girls moved more rapidly towards higher levels of smoking than boys did in terms of smoking status and intention to start smoking, and the difference in smoking status was considerable. Also, there was a positive slope for smoking among boys during the past week/month, but this slope became negative among girls. In the “Occasional/Intending smokers”, the slopes of all outcomes were positive among both girls and boys, with a negligible difference. Other studies have suggested various possibilities for differences between girls and boys, which can be explained by the verity in smoking definitions and instruments, and stages of smoking. Nevertheless, the results of a longitudinal study, which examined the transition process in adolescents, are consistent with the results of the present study showing a positive trend in the transition through smoking stages, and the transition process is usually progressive, and return is unlikely. ⁷

The predictors of transition in smoking stages

Transition through smoking stages accelerated by increasing in age, not living with parents, SES, having a smoker friends, having a smoker in family, self-injury, substance abuse, attitude toward smoking, and general risk-taking behavior. The major plays an interesting role so that non-mathematics students create a positive slope in terms of smoking status and intention to start smoking. However, in terms of smoking during the past week/month, non-mathematics students create a negative slope. Almost similar results have been reported. ^{7,9,11,13,16,27-29} Of course, there were differences in the predictors and the size of the reported effects, which can be explained by various definitions of the outcomes and measurement tools, variety in participants, and dissimilarities in study settings.

Strengths, limitations, and future directions

The present study’s main strengths are as follows: this is one of the few studies that present the transition process through smoking stages using sex-specific optimal GMMs. Moreover, the length of the study and the large sample size can be noted. On the other hand, the results should be interpreted, taking into account the existing limitations: First, the two measurements in the present study make it impossible to draw higher order models. Second, the results depend on the self-reported measurement tool, although similar studies have used the same approach. ^7,27-29 Third, the generalizability of the results is limited to the tenth-grade students of Tabriz. Fourth, time-independent covariates were utilized in the study baseline. Future studies are strongly recommended to explore these issues.

Conclusions and Implications

In general, based on sex-specific optimal GMMs model, a sex-dependent transition through smoking stages was observed. On the other hand, predictors accelerated this rising trend, and there was a significant difference between girls and boys in terms of these predictors. The study highlighted the importance of stopping the initiation and preventing transition through smoking stages with special sex-specific planning for adolescent girls and boys. These programs should be designed with a focus on essential predictors of transition.

Acknowledgements

We would like to appreciate the collaboration of Research Deputy of Tabriz University of Medical Sciences for their appreciated contribution in this study. This work was supported by the Research Deputy of Tabriz University of Medical Sciences [Grant no 5/D/509208].

Funding

This work was supported by the Research Deputy of Tabriz University of Medical Sciences [Grant no 5/D/509208].

Competing interests

Mohammad Asghari-Jafarabadi is an Associate Editor for Health Promotion Perspectives. Other authors declare that there is no conflict of interest.

Ethics approval

The institutional review board of Tabriz University of Medical Sciences approved the protocol (ethics code: IR.TBAMED.REC.1396.448). The participants were free to participate in the study, and privacy was preserved. All participants filled and signed the informed consent and assent. We have complied with the World Medical Association Declaration of Helsinki regarding the ethical conduct of research involving human subjects.

Authors’ contributions

All authors read and approved the final manuscript. AMA conceived of the study and participated in the design and data collection. MAJ, AMA, and NJ participated in data analyses and manuscript preparation.

References

1. Thun MJ, Carter BD, Feskanich D, Freedman ND, Prentice R, Lopez AD. 50-year trends in smoking-related mortality in the United States. N Engl J Med 2013; 368(4):351-64. doi: 10.1056/NEJMsa1211127 [Crossref]
2. Allahverdipour H, MacIntyre R, Hidarnia A, Shafii F, Kzamnegad A, Ghaleiha A. Assessing protective factors against drug abuse among high school students: self-control and the extended parallel process model. J Addict Nurs 2007; 18(2):65-73. doi: 10.1080/10884600701334820 [Crossref]
3. Barati M, Allahverdipour H, Hidarnia A, Niknami S. Predicting tobacco smoking among male adolescents in Hamadan city, west of Iran in 2014: an application of the prototype willingness model. J Res Health Sci 2015; 15(2):113-8.
4. Jafarabadi MA, Allahverdipour H, Bashirian S, Jannati A. Modeling the underlying predicting factors of tobacco smoking among adolescents. Iran J Public Health 2012; 41(5):46-57.
5. Wellman RJ, Sylvestre MP, O’Loughlin EK, Dutczak H, Montreuil A, Datta GD. Socioeconomic status is associated with the prevalence and co-occurrence of risk factors for cigarette smoking initiation during adolescence. Int J Public Health 2018; 63(1):125-36. doi: 10.1007/s00038-017-1051-9 [Crossref]
6. Okagua J, Opara P, Alex-Hart BA. Prevalence and determinants of cigarette smoking among adolescents in secondary schools in Port Harcourt, Southern Nigeria. Int J Adolesc Med Health 2016; 28(1):19-24. doi: 10.1515/ijamh-2014-0066 [Crossref]
7. Fergusson DM, Horwood LJ. Transitions to cigarette smoking during adolescence. Addict Behav 1995; 20(5):627-42. doi: 10.1016/0306-4603(95)00023-6 [Crossref]
8. Khosravi A, Holakouie-Naieni K, Mansournia MA, Mahmoodi M, Pouyan AA. Predictors of transition in smoking stages in Iranian adolescents: a probabilistic discrete event system model. Iran J Psychiatry Behav Sci 2017; 11(2):e7670. doi: 10.5812/ijpbs.7670 [Crossref]
9. Mohammadpoorasl A, Fakhari A, Rostami F, Shamsipour M, Rashidian H, Goreishizadeh MA. Predictors of transition in different stages of smoking: a longitudinal study. Addict Health 2010; 2(1-2):49-56.
10. Mohammadpoorasl A, Nedjat S, Yazdani K, Fakhari A, Foroushani AR, Fotouhi A. An algorithm of smoking stages assessment in adolescents: a validation study using the latent class analysis model. Int J Prev Med 2013; 4(11):1304-11.
11. Yu M, Whitbeck LB. A prospective, longitudinal study of cigarette smoking status among North American Indigenous adolescents. Addict Behav 2016; 58:35-41. doi: 10.1016/j.addbeh.2016.02.007 [Crossref]
12. Kim H, Kim EK, Choi ES, Kim YJ, Lee HJ, Kim JJ. [The determinants of adolescent smoking by gender and type of school in Korea]. J Prev Med Public Health 2006; 39(5):379-88.
13. Kim EM, Park E, Kim H. Sex differences in multilevel factors of smoking experimentation and age of initiation in Korean adolescents. J Sch Nurs 2020; 36(5):348-59. doi: 10.1177/1059840519840805 [Crossref]
14. Breslau N, Peterson EL. Smoking cessation in young adults: age at initiation of cigarette smoking and other suspected influences. Am J Public Health 1996; 86(2):214-20. doi: 10.2105/ajph.86.2.214 [Crossref]
15. Pierce JP, Gilpin E. How long will today’s new adolescent smoker be addicted to cigarettes?. Am J Public Health 1996; 86(2):253-6. doi: 10.2105/ajph.86.2.253 [Crossref]
16. Mohammadpoorasl A, Nedjat S, Fakhari A, Yazdani K, Fotouhi A. Predictors of transition in smoking stages in Iranian adolescents: latent transition analysis. East Mediterr Health J 2014; 20(5):330-9.
17. Berlin KS, Parra GR, Williams NA. An introduction to latent variable mixture modeling (part 2): longitudinal latent class growth analysis and growth mixture models. J Pediatr Psychol 2014; 39(2):188-203. doi: 10.1093/jpepsy/jst085 [Crossref]
18. Muthén LK, Muthén BO. Mplus User’s Guide. Los Angeles, CA: Muthén & Muthén; 2012.
19. Ram N, Grimm KJ. Growth mixture modeling: a method for identifying differences in longitudinal change among unobserved groups. Int J Behav Dev 2009; 33(6):565-76. doi: 10.1177/0165025409343765 [Crossref]
20. Jaber R, Mzayek F, Madhivanan P, Khader Y, Maziak W. Predictors of cigarette smoking progression among a school-based sample of adolescents in Irbid, Jordan: a longitudinal study (2008-2011). Nicotine Tob Res 2016; 18(4):403-9. doi: 10.1093/ntr/ntv100 [Crossref]
21. Muthén LK, Muthén BO. How to use a Monte Carlo study to decide on sample size and determine power. Struct Equ Modeling 2002; 9(4):599-620. doi: 10.1207/s15328007sem0904_8 [Crossref]
22. Hill AJ, Boudreau F, Amyot E, Déry D, Godin G. Predicting the stages of smoking acquisition according to the theory of planned behavior. J Adolesc Health 1997; 21(2):107-15. doi: 10.1016/s1054-139x(97)00039-6 [Crossref]
23. Rosenberg M. Society and the Adolescent Self-Image. Princeton University Press; 2015.
24. Nylund KL, Asparouhov T, Muthén BO. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct Equ Modeling 2007; 14(4):535-69. doi: 10.1080/10705510701575396 [Crossref]
25. Greenbaum PE, Del Boca FK, Darkes J, Wang CP, Goldman MS. Variation in the drinking trajectories of freshmen college students. J Consult Clin Psychol 2005; 73(2):229-38. doi: 10.1037/0022-006x.73.2.229 [Crossref]
26. Pallonen UE, Prochaska JO, Velicer WF, Prokhorov AV, Smith NF. Stages of acquisition and cessation for adolescent smoking: an empirical integration. Addict Behav 1998; 23(3):303-24. doi: 10.1016/s0306-4603(97)00074-9 [Crossref]
27. Ayatollahi SA, Mohammadpoor Asl A, Rajaee Fard A. Psychological predictors of transition in different stages of cigarette smoking. J Ardabil Univ Med Sci 2005;4(14):13-9. [Persian].
28. Hoving C, Reubsaet A, de Vries H. Predictors of smoking stage transitions for adolescent boys and girls. Prev Med 2007; 44(6):485-9. doi: 10.1016/j.ypmed.2007.02.011 [Crossref]
29. Khosravi A, Emamian MH, Hashemi H, Fotouhi A. Transition in tobacco use stages and its related factors in a longitudinal study. Environ Health Prev Med 2018; 23(1):39. doi: 10.1186/s12199-018-0728-x [Crossref]