Smoking and physical activity trajectories from childhood to midlife

International Journal of Environmental Research and Public Health Article Smoking and Physical Activity Trajectories from Childhood to Midlife Kasper Salin 1,* , Anna Kankaanpää 2, Mirja Hirvensalo 1, Irinja Lounassalo 1, Xiaolin Yang 2, Costan G. Magnussen 3,4 , Nina Hutri-Kähönen 5, Suvi Rovio 3, Jorma Viikari 6, Olli T. Raitakari 3,7 and Tuija H. Tammelin 2 1Faculty of Sport and Health Sciences, University of Jyväskylä, 40014 Jyväskylä, Finland; [email protected] (M.H.); [email protected] (I.L.) 2LIKES Research Centre for Physical Activity and Health, 40014 Jyväskylä, Finland; [email protected] (A.K.); [email protected] (X.Y.); [email protected] (T.H.T.) 3Research Centre of Applied and Preventive Cardiovascular Medicine, University of Turku, 20520 Turku, Finland; [email protected] (C.G.M.); [email protected] (S.R.); [email protected] (O.T.R.) 4Menzies Institute for Medical Research, University of Tasmania, 7005 Hobart, Australia 5 Department of Pediatrics, University of Tampere and Tampere University Hospital, 33100 Tampere, Finland; [email protected] 6Department of Medicine and Division of Medicine, University of Turku and Turku University Hospital, 20500 Turku, Finland; [email protected] 7 Department of Clinical Physiology and Nuclear Medicine, Turku University Hospital, 20500 Turku, Finland *Correspondence: Kasper[email protected] Received: 5 February 2019; Accepted: 12 March 2019; Published: 18 March 2019   Abstract: Introduction: Despite substantial interest in the development of health behaviors, there is limited research that has examined the longitudinal relationship between physical activity (PA) and smoking trajectories from youth to adulthood in a Finnish population. This study aimed to identify trajectories of smoking and PA for males and females, and study the relationship between these trajectories from youth to adulthood. Methods: Latent profile analysis (LPA) was used to identify trajectories of smoking and PA separately for males and females among 3355 Finnish adults (52.1% females). Participants’ smoking and PA were assessed five to eight times over a 31-year period (3–18 years old at the baseline, 34–49 years at last follow-up). Multinomial logistic regression analysis was used to study the relationship between the trajectories of smoking and PA. Results: Five smoking trajectories and four to five PA trajectories were identified for males and females. Of the PA trajectory groups, the persistently active group were least likely to follow the trajectories of regular smoking and the inactive and low active groups were least likely to follow non-smoking trajectory group. Likewise, inactive (women only) and low active groups were less likely to belong to the non-smokers group. Conclusions: The study suggests that those who are persistently active or increasingly active have substantially reduced probabilities of being in the highest-risk smoking categories. Keywords: physical activity; smoking; cohort study; longitudinal study; trajectory; adults 1. Introduction Smoking has been, and remains, a major risk factor for human health [ 1 ]. In addition, there are many countries, including populous ones like Indonesia and Pakistan, where tobacco smoking is projected to continue increasing at least until 2025, as well as countries where there has been an explosion of smoking, such as Cameroon or Congo [ 2 ]. To promote the reduction and cessation of smoking, it is essential to understand smoking behavior and other health behaviors that may be a gateway for a non-smoking lifestyle. Lack of physical activity (PA) is a risk factor for several major Int. J. Environ. Res. Public Health 2019,16, 974; doi:10.3390/ijerph16060974 www.mdpi.com/journal/ijerph Int. J. Environ. Res. Public Health 2019,16, 974 2 of 14 health problems. These include more than 25 chronic conditions such as cardiovascular disease, diabetes mellitus, depression, and several cancers [ 3 ]. Because smoking is also a major risk factor for several of these diseases [ 4 ], physically inactive smokers are at an even higher risk of developing these diseases [ 5 ]. Smoking has also been seen as a predictor of other unhealthy behaviors [ 6 ], and PA has been suggested to be a “gateway” to healthier habits, meaning that when changes in PA occur, other changes will follow [ 7 ]. On the other hand, smoking and PA have been viewed largely as independent behaviors, with changes in one behavior not necessarily tied to a change in the other [8]. PA in adolescence tracks into adulthood [ 9 ], with low PA and inactivity tending to track better than high PA [ 10 ]. Similarly, smoking in adolescence tends to continue in young adulthood and it has been shown to have the strongest level of continuity into adulthood when compared to other health habits such as PA and alcohol use [ 11 ]. Additionally, previous studies have also found associations between unhealthy habits. For example, smoking in adolescence is associated with alcohol use, physical inactivity, and poor dietary behavior in adulthood [ 6 , 12 ], as well as illicit drug use, unprotected sexual intercourse, and aggressive behavior [ 13 ]. Risky health behaviors have also been found to be associated with gender and sociodemographics, with males showing riskier health behaviors than women [ 14 , 15 ]. To analyze how individual characteristics or behaviors are organized into meaningful patterns that distinguish subgroups of people, mixture modeling is a useful technique [ 16 – 18 ]. This technique allows identifying unobserved subgroups in the population following distinct developmental trajectories or patterns [18]. Previous studies on smoking show evidence for two [ 19 ], three [ 20 ], four [ 21 , 22 ], five [ 23 , 24 ], and six [ 25 ] distinct smoking trajectory groups. In these studies, three common smoking trajectories were found: (1) never or rare smokers, (2) persistent light smokers, and (3) stable or steady heavy smokers. Other trajectories that have been identified are early/late increasers, decreasers [ 22 , 24 , 25 ], and quitters [23]. Likewise, in previous longitudinal studies concerning PA, participants have been grouped into three [ 26 ], four [ 27 ], or five [ 28 , 29 ] trajectory groups. In these studies, three main PA trajectory groups were identified: inactive, moderately active, and highly active [ 26 ]. Other trajectories that have been identified include increasing, decreasing, or curvilinear PA groups [27,28]. Although cross-sectional studies have examined the clustering of health habits, there is a gap in the research literature identifying the linear change of the longitudinal research related to the association between smoking and PA. Even though there are some previous studies, with comprehensive data, about the association of PA and smoking, these rely mainly on longitudinal data collected from youth [ 28 , 30 – 33 ], but not from youth to adulthood. This study fills the gap in the research literature from youth to adulthood. Using data from the 31-year prospective Cardiovascular Risk in Young Finns Study, we aimed to (1) identify different subgroups of participants who shared similar smoking and PA trajectories from youth to mid-adulthood, and (2) determine the associations between the identified smoking and PA trajectories from youth to adulthood. 2. Methods 2.1. Study Design and Participants The data were obtained from the Cardiovascular Risk in Young Finns Study. The population at baseline (1980) consisted of boys and girls aged 3, 6, 9, 12, 15 and 18 years and randomly sampled from all five Finnish university cities with medical schools (Helsinki, Turku, Tampere, Oulu, and Kuopio) and their rural surroundings. In practice, girls and boys of each age cohort in each study community were separately placed in random order by the unique personal identification number. Every kth girl and every kth boy in each community were selected so that the sample consisted of the required number of boys and girls. The varying kfactors were determined by sample size and the total number of boys and girls in the different age cohorts in each community. Int. J. Environ. Res. Public Health 2019,16, 974 3 of 14 Of the 4326 subjects, 3596 (83.2%) participated in the initial survey in 1980. Since baseline, these participants were followed up on seven occasions (1983, 1986, 1989, 1992, 2001, 2007, and 2010–2011), with 2005 of the original participants (55.8%) remaining in the sample in 2011(Supplement 4). The examinations have included comprehensive data collection using questionnaires, physical measurements, and blood tests. In 2010, the participants were adults aged 33, 36, 49, 42, 45, and 48 years. Participants, or their parents, gave written informed consent and the study protocol was approved by the ethics committee of the participating universities. 2.2. Physical Activity and Smoking Smoking and PA were assessed through a self-administered questionnaire five to eight times (depending on age) during the 31 years of follow-up. Questions related to PA in 1980–1989 concerned the frequency and intensity of leisure-time PA, participation in sports club training, participation in sports competitions, and habitual ways of spending leisure time starting from the age of 9 years. From the 1992 follow-up, the questions concerned the intensity of PA, frequency of vigorous PA, hours spent in vigorous PA, the average duration of a PA session, and participation in organized PA during leisure-time. The five items were recorded as inactive or very low activity (= 1) to regular or vigorous activity (= 2 or 3) and then summed to form a physical activity index (PAI; range 5–14/15) [ 10 ]. The validity of the PAI has been studied by correlating child and adult PAIs with indicators of exercise capacity [ 10 , 33 ] and objectively measured data [ 33 ]. Smoking information has been measured four to eight times beginning when participants were 12 years old. Participants were asked to indicate how many cigarettes of different types (factory-made, hand-rolled, pipe, cigar) they smoked per day. The total number of cigarettes per day was calculated as the sum of different types of cigarettes (range 0–30). Questions related to smoking and PA and the coding of these variables are presented in Supplementary Materials supplement 1. 2.3. Ethics and Consent Study participants provided written informed consent in accordance with the Helsinki Declaration and the study protocol was reviewed and approved (88/180/2010) by the ethics committee of the participating universities (5) [34]. 2.4. Statistical Analysis Descriptive statistics were calculated using IBM SPSS Statistics for Windows (version 24.0) and further modeling performed using Mplus, version 7.0 [ 35 ]. Trajectories for smoking and PA over 31 years were defined using latent profile analysis (LPA). Moreover, the relationship between the trajectories was examined via multinomial logistic regression and latent transition probabilities. Participants having at least one assessment of smoking and PA over seven periods were included in the analyses, ending up with a sample of 3355 participants (47.9% men). On average, participants had 5.3 measures for smoking and PA. Multiple cohort data were reorganized in a way that each measurement point represented the participant’s age instead of measurement year. The content of the PA questionnaire was marginally different between the time intervals of 1980–1989 and 1992–2011. In order to avoid overlaps of the two scales, the PA data of the oldest cohort in 1986, the two oldest cohorts in 1989 and three oldest cohorts in 1992 were omitted (recorded as missing). PA data covered ages from 9 to 48. In addition, reorganized smoking data were averaged over three-year periods ending up with six measurement points at ages 15–18, 21–24, 27–30, 33–36, 39–42 and 45–48. Averaged smoking was treated as a censored zero-inflated normal variable in the modeling. First, LPA was performed to identify the optimal number of latent classes for smoking and PA for both genders. The mean profiles were freely estimated and residual variances were fixed to be equal across classes. Based on previous research, one to six class models were fitted to the data. Akaike’s information criterion (AIC), Bayesian information criterion (BIC), and sample-size adjusted Bayesian information criterion (ABIC) were used to evaluate the goodness-of-fit of the LPA Int. J. Environ. Res. Public Health 2019,16, 974 4 of 14 with a different number of classes. A model with lower values of information criteria fitted the data better than an alternative model with higher values. Furthermore, statistical tests were used to determine a sufficient number of classes: Vuong–Lo–Mendell–Rubin likelihood ratio test (VLMR), Lo–Mendell–Rubin adjusted likelihood ratio test (LMR) and parametric bootstrapped likelihood ratio test (BLRT). The estimated model was compared to the model with one class less, with a low pvalue indicating that the model with one classless was rejected in favor of the estimated model [ 35 ]. It has been shown that by identifying the correct number of classes in mixture modeling, BIC outperforms the other information criteria and BLRT the other likelihood-based tests [ 36 ]. In our study, BLRT did not differ between the different solutions for smoking or PA (supplement 2). Hence, for PA we based our judgments on BIC [ 36 ]. Among men, BIC suggested that a four-group PA solution had the best fit to the data. Among women, five groups appeared to represent the data best. The quality of the classification was evaluated using posterior probabilities for the most likely latent class membership, with an average posterior probability of 0.70 for all groups as a minimum indicator of well-separated classes [ 17 ]. Among men and women, the average posterior probabilities dropped below 0.70, when the number of groups was increased to above four groups for men and above five groups for women. For smoking, BIC and ABIC decreased for all models from one to five or six classes. However, average posterior probabilities dropped below 0.70 after the five-class model and therefore the five-class model was considered optimal. Second, the relationship between the trajectories of smoking and PA was studied. A multinomial logistic regression model was specified between the latent class variables; that is, the latent class variable of smoking was regressed on the latent class variable of PA. The number of classes, the mean profiles, and the residual variances were fixed based on the LPAs. Latent transition probabilities, or the probability of belonging to each smoking trajectory conditional on a given PA trajectory, were obtained. The approach is similar to dual trajectory analysis. Missing data were assumed to be missing at random (MAR). Parameters of the models were estimated by using the full information maximum likelihood (FIML) method with robust standard errors (MLR). The method produces unbiased parameter estimates under MAR assumption. 3. Results The characteristics of the study participants in 2011 are presented in Table 1. Participants having at least one measurement of smoking and PA were included in the analyses (1607 males and 1748 females). Table 1. Characteristics of the study sample, year 2011. Males (N= 1607, 47.9%) nFemales (N= 1748, 52.1%) n Age (years), mean (SD) 41.7 (4.9) 41.7 (4.9) Height (cm), mean (SD) 179.8 (6.6) 923 166.1 (6.0) 1115 Weight (kg), mean (SD) 87.5 (16.0) 923 72.0 (15.4) 1117 BMI, mean (SD) 27.0 (4.4) 922 26.1 (5.5) 1114 Education (%) 878 1096 ≤12 years 34.3 18.4 >12 years 65.7 81.6 SES (%) 801 981 Manual 33.3 8.8 Non-manual, low 20.8 53.1 Non-manual, high 45.8 38.1 Smoking status 2011 (%) 882 1105 Non-smoker 78.6 84.3 Smoker 21.4 15.7 Physical activity 2011 (%) 877 1095 Once a month or seldom 24.2 17.4 Once a week 22.2 23.7 2–3 times a week 38.3 38.4 4–6 times a week 13.2 16.5 Every day 2.1 4.0 Int. J. Environ. Res. Public Health 2019,16, 974 5 of 14 3.1. Physical Activity In LPA, four discrete PA trajectories were identified among males. They were named as follows: persistently low active (41.1%), decreasingly active (15.8%), increasingly active (30.7%), and persistently active (12.5%). Among females, five discrete PA trajectories were identified and were named as persistently inactive (17.0%), persistently low active (52.5%), decreasingly active (12.3%), increasingly active (14.9%), and persistently active (3.4%). The proportion of the study population assigned to each group and the posterior probabilities are shown in Table 2, and the mean profiles of PA subgroups are presented in Figure 1. Although some of the identified groups had low proportions of participants, they are highly discriminated with high mean posteriori probabilities, as shown in Table 2. Table 2. Trajectory group distributions and posterior probabilities for males and females. Distribution % Posterior Probability (M) PA trajectory group Males Group 1: Persistently active 12.5 0.87 Group 2: Increasingly active 30.7 0.78 Group 3: Decreasingly active 15.8 0.74 Group 4: Persistently low active 41.1 0.83 Females Group 1: Persistently active 3.4 0.85 Group 2: Increasingly active 14.9 0.75 Group 3: Decreasingly active 12.3 0.78 Group 4: Persistently low active 52.5 0.78 Group 5: Persistently inactive 17.0 0.79 Smoking trajectory group Males Group 1: Persistently non-smokers 44.6 0.91 Group 2: Persistently mild smokers 17.3 0.83 Group 3: Ex-smokers 4.0 0.72 Group 4: Persistently moderate smokers 25.3 0.76 Group 5: Persistently heavy smokers 8.8 0.82 Females Group 1: Persistently non-smokers 56.4 0.94 Group 2: Persistently light smokers 16.0 0.78 Group 3: Persistently mild smokers 16.7 0.76 Group 4: Decreasing smokers 1.8 0.91 Group 5: Persistently moderate smokers 9.2 0.81 Int. J. Environ. Res. Public Health 2019,16, 974 6 of 14 Int. J. Environ. Res. Public Health 2019, 16, x 6 of 14 Figure 1. PA trajectories among (A) males and (B) females. 3.2. Smoking In LPA, five discrete smoking trajectories were identified for both males and females. In line with earlier research [34], the group names were based on the number of cigarettes smoked in a day: light (1–4 cigarettes daily), mild (≤10 cigarettes daily), moderate (11–20 cigarettes daily) and heavy (>20 cigarettes daily). Among both genders, three similar smoking groups were identified and named as persistently non-smokers (men, 44.6%, women, 56.4%), persistently mild smokers (17.3%, 16.7, respectively), and persistently moderate smokers (25.3%, 9.2%, respectively). Two additional groups were found for men: persistently heavy smokers (8.8%) and ex-smokers (4.0%). For females, the additional groups were decreasing smokers (1.8%) and persistently light smokers (16.0%). The proportion of the study population assigned to each trajectory group is shown in Table 2, with mean profiles of smoking subgroups shown in Figure 2. Overall, the smoking trajectories seemed to have good convergent validity (Supplement 3, Figure S1 for male and Figure S2 for female). Figure 1. PA trajectories among (A) males and (B) females. 3.2. Smoking In LPA, five discrete smoking trajectories were identified for both males and females. In line with earlier research [ 34 ], the group names were based on the number of cigarettes smoked in a day: light (1–4 cigarettes daily), mild ( ≤ 10 cigarettes daily), moderate (11–20 cigarettes daily) and heavy (>20 cigarettes daily). Among both genders, three similar smoking groups were identified and named as persistently non-smokers (men, 44.6%, women, 56.4%), persistently mild smokers (17.3%, 16.7, respectively), and persistently moderate smokers (25.3%, 9.2%, respectively). Two additional groups were found for men: persistently heavy smokers (8.8%) and ex-smokers (4.0%). For females, the additional groups were decreasing smokers (1.8%) and persistently light smokers (16.0%). The proportion of the study population assigned to each trajectory group is shown in Table 2, with mean profiles of smoking subgroups shown in Figure 2. Overall, the smoking trajectories seemed to have good convergent validity (Supplement 3, Figure S1 for male and Figure S2 for female). Int. J. Environ. Res. Public Health 2019,16, 974 7 of 14 Int. J. Environ. Res. Public Health 2019, 16, x 7 of 14 Figure 2. Smoking trajectories among (A) males and (B) females. A multinomial logistic regression model between the latent class variables of PA and smoking produced similar mean profiles and class-sizes compared to independently conducted LPA models (see Table 2) for PA classes (males: 11.7%, 31.1%, 15.4%, and 41.8%; females: 3.5%, 15.0%, 12.2%, 52.5%, and 16.9%), and for smoking classes (males: 44.7%, 19.1%, 4.4%, 22.8%, and 9.1%; females: 56.3%, 16.2%, 16.5%, 1.8%, and 9.2%). For the model, the lowest PA group (a persistently low active group for males, and a persistently inactive group for females) was considered as the reference group, and the persistently non-smokers group was considered as the reference group for smoking. Structural parameters of the multinomial regression model are presented in Table 3 and latent transition probabilities for males and females are presented in Figure 3. Figure 2. Smoking trajectories among (A) males and (B) females. A multinomial logistic regression model between the latent class variables of PA and smoking produced similar mean profiles and class-sizes compared to independently conducted LPA models (see Table 2) for PA classes (males: 11.7%, 31.1%, 15.4%, and 41.8%; females: 3.5%, 15.0%, 12.2%, 52.5%, and 16.9%), and for smoking classes (males: 44.7%, 19.1%, 4.4%, 22.8%, and 9.1%; females: 56.3%, 16.2%, 16.5%, 1.8%, and 9.2%). For the model, the lowest PA group (a persistently low active group for males, and a persistently inactive group for females) was considered as the reference group, and the persistently non-smokers group was considered as the reference group for smoking. Structural parameters of the multinomial regression model are presented in Table 3and latent transition probabilities for males and females are presented in Figure 3. Int. J. Environ. Res. Public Health 2019,16, 974 8 of 14 Table 3. Structural parameters of the multinomial logistic regression model of PA and smoking among males (n= 1607) and females (n= 1748). Smoking among Males (Non-Smokers as Reference Group) Persistently Heavy Smokers Persistently Moderate Smokers Ex-Smokers Persistently Mild Smokers b s.e. p b s.e. p b s.e. p b s.e. p Persistently low active (as reference group) Persistently active −3.32 1.39 0.017 * −1.43 0.34 <0.001 *** −0.39 0.69 0.571 −0.30 0.31 0.680 Increasingly active −2.50 0.63 <0.001 *** −0.72 0.24 0.003 ** −0.12 0.60 0.845 −0.14 0.27 0.593 Decreasingly active −1.41 0.58 0.015 * −0.26 0.28 0.364 −0.08 0.86 0.923 0.16 0.33 0.626 Smoking among Females (Rare or Non-Smokers as Reference Group) Persistently Moderate Smokers Persistently Mild Smokers Decreasing Smokers Persistently Light Smokers b s.e. p b s.e. p b s.e. p b s.e. p Persistently inactive (as reference group) Persistently active −2.43 1.21 0.044 * - a- - −1.20 1.28 0.734 0.09 0.52 0.865 Increasingly active - a- - −1.11 0.42 0.007** −1.77 1.04 0.090 0.23 0.39 0.555 Decreasingly active −1.80 0.53 0.00 1** −1.20 0.41 0.004** - - - −0.35 0.44 0.438 Persistently low active −1.08 0.33 0.001 ** −0.60 0.30 0.043* −1.54 0.73 0.036* −0.21 0.38 0.578 Note. a There was an empty cell in joint distribution of latent class variables. The parameter estimates could not be determined. b, regression coefficient; s.e., standard error. Significant difference between groups; * p < 0.005, ** p< 0.01, *** p< 0.001. Int. J. Environ. Res. Public Health 2019,16, 974 9 of 14 Int. J. Environ. Res. Public Health 2019, 16, x 9 of 14 Figure 3. Latent transition probabilities based on the estimated model among (A) males and (B) females. Figure 3. Latent transition probabilities based on the estimated model among (A) males and (B) females.