Jproject

Contents lists available at ScienceDirect

Drug and Alcohol Dependence

journal homepage: www.elsevier.com/locate/drugalcdep

Full length article

Progression to established patterns of cigarette smoking among young adults

Elizabeth Haira,b,⁎, Morgane Bennetta,c, Valerie Williamsa, Amanda Johnsond, Jessica Ratha,b, Jennifer Cantrella,b, Andrea Villantib,d, Craig Enderse, Donna Vallonea,b,f

a Evaluation Science and Research at Truth Initiative, 900 G Street NW, Fourth Floor, Washington, DC 20001, USA b Department of Health, Behavior and Society, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe Street, Baltimore, MD 21205, USA c Department of Prevention and Community Health, Milken Institute School of Public Health, The George Washington University, 950 New Hampshire Ave., NW, Washington, DC 20052, USA d Schroeder Institute for Tobacco Research and Policy Studies at Truth Initiative, 900 G Street NW, Fourth Floor, Washington, DC 20001, USA e Department of Psychology, University of California, Los Angeles, 1285 Franz Hall, Los Angeles, CA 90095, USA f College of Global Public Health, New York University, 41 E. 11th St, New York, NY 10003, USA

A R T I C L E I N F O

Keywords: Tobacco Young adults Use patterns Cigarettes

A B S T R A C T

Background: As tobacco control policies have been implemented across the U.S. over the past decade, patterns of smoking cigarettes have significantly changed, particularly among young adults. For many users, the typical daily use pattern of smoking several packs of cigarettes per day has been supplanted by a variety of use patterns, often referred to as light, intermittent, and occasional smoking. Methods: The aim of this study was to examine progression to established smoking patterns among a nationally representative, longitudinal sample of young adults (n = 9791). Using repeated measures latent class techniques (RMLCA), we modeled the distribution of cigarette smoking intensity over time and latent class categories. Results: Findings demonstrate that young adults fall into three discrete classes that reflect probabilities for never to low use, daily use, and variable cigarette use for progression to established use of cigarettes: 79.3% fall into the class of “never or ever users” of cigarettes (no current use of cigarettes), 11.3% fall into the class of “rapid escalators” or daily users of cigarettes, and 9.4% fall into the “dabbler” class. Smoking patterns were found to be stable by the age of 21. Conclusions: Intervening prior to age 21 has the potential to disrupt progression to established smoking and reduce the long-term health consequences of smoking in this age group.

1. Introduction

As tobacco control policies have been implemented across the U.S. over the past decade, patterns of cigarette smoking have significantly changed, particularly among young adults. For many users, the typical daily pattern of smoking several packs of cigarettes per day has been supplanted by a variety of use patterns, often referred to as light and intermittent smoking (Kozlowski and Giovino, 2014; U.S. Department of Health and Human Services, 2014). These patterns reflect reductions in the number of cigarettes smoked per day as well as non-daily, occasional use of cigarettes, such as smoking only on weekends or when socializing with friends (U.S. Department of Health and Human Services, 2014). Although young adults (ages 18–24) have reported the highest rates for smoking cigarettes of any age group in the U.S., light or intermittent smoking is the most common use pattern among this age group (Lenk et al., 2009; U.S. Department of Health and Human Services, 2014). Among young adults, intermittent smokers smoke

fewer cigarettes per day than daily smokers; are less likely to feel addicted; and are less likely to consider themselves “smokers” (Lenk et al., 2009; Waters et al., 2006).

While many studies have assessed the age at which cigarette use initiation occurs, there is little recent evidence regarding the age at which use patterns become established, or stabilize. In 1984, Kandel et al. found cigarette use patterns stabilize after age 18 (Kandel and Logan, 1984). Later, Chen and Kandel (1995) found the prevalence of current cigarette smoking and the proportion reporting a period of highest use (measured by frequency and quantity) at each age to sharply increase during adolescence with stabilization in late teen years (Chen and Kandel, 1995). A study using data from the 2000 National Health Interview Survey (NHIS) found the proportion of smokers who become regular users by age 18 has decreased over time, while the proportion who become regular smokers between ages 19 and 21 years has increased (Lantz, 2003).

Studies of the trajectories of tobacco use patterns from youth to

http://dx.doi.org/10.1016/j.drugalcdep.2017.03.040 Received 14 October 2016; Received in revised form 16 February 2017; Accepted 20 March 2017

⁎ Corresponding author at: 900G Street NW, Fourth Floor, Washington, DC 20001, USA. E-mail addresses: [email protected], [email protected] (E. Hair).

Drug and Alcohol Dependence 177 (2017) 77–83

Available online 29 May 2017 0376-8716/ © 2017 Elsevier B.V. All rights reserved.

MARK

young adulthood have often focused on changes in frequency or intensity of smoking to help understand how to interrupt or prevent progression to established smoking. Over a decade ago, Chassin et al. (2000) examined subgroups of early stable smokers, late stable smokers, experimenters, and quitters, whereby the majority transition from infrequent smoking to regular weekly smoking by age 18, achieving maximum levels of use by age 24, after which levels of use stabilize (Chassin et al., 2000). White et al. (2002) examined smoking trajectories from adolescence to young adulthood by categorizing subgroups: non/experimental, occasional/maturing out, and heavy/ regular smokers. Findings suggested cigarette use quantity and fre- quency stabilized among heavy/regular smokers in their early 20s. Additionally, among the occasional/maturing out smokers, there were increasing trends in cigarette use until age 18, after which cigarette use decreased (White et al., 2002). Karp et al. (2005) assessed smoking patterns in youth by categorizing subgroups as low-intensity, non- progressing smoking, and slow, moderate, and rapid escalators. Find- ings indicated that the majority of study participants remained in the low-intensity, non-progressing smoking group, however, the study period did not extend further to assess use patterns in later teen years and early adulthood (Karp et al., 2005). Others have reported heavy use patterns to be stable over time, while low intensity or occasional smoking patterns tend to be less stable and vary over time (Kvaavik et al., 2014; Mathur et al., 2014). More recently, Mathur et al. (2014) identified subgroups, including non-smokers, experimental smokers, light smokers, and daily smokers, and assessed their relative stability over time. The never- and daily smoking subgroups were the most stable over time, while light smoking was less stable over time (Mathur et al., 2014).

While use frequency and intensity evidence related to smoking trajectory is key, there is little current evidence aimed at identifying the age at which use patterns become established − another critical variable when designing prevention efforts. Given the recent shifts in the tobacco product landscape and tobacco use prevalence patterns, identifying age and risk factors associated with established use is essential to designing tailored prevention efforts. Until recently, national tobacco prevention education campaigns have been targeted at adolescents who are at risk for smoking initiation, while other efforts have been focused on prompting smoking cessation. However, national surveillance data suggest a shift in the age at which young people initiate tobacco use, with increased initiation during young adulthood (Terry-McElrath & O'Malley, 2015; U.S. Department of Health and Human Services, 2014). The developmental period of young, or emerging, adulthood is a period marked by increased autonomy, identity exploration, and instability as youth transition from adoles- cence into adulthood (Arnett, 2005). The characteristics of freedom and exploration marked by this period place young adults at particular risk for substance use and addiction. Aware of this susceptibility, the tobacco industry has targeted their marketing to young adults with messages focused on rebellion and independence, which resonate with this age group (Davis et al., 2008). This period is a critical transition in the lifecourse when positive behaviors should be promoted to reduce long-term health consequences (Arnett, 2005).

While studies from decades ago have shed light on patterns of cigarette use progression, recent evidence of the emerging cigarette use patterns is scant. The aim of this study is to examine progression to established smoking patterns among a nationally representative sample of young adults. Findings can help inform the design and delivery of tobacco use prevention interventions and future policy recommenda- tions, particularly as the non-cigarette combustible tobacco product category continues to grow and poly use patterns evolve.

2. Methods

2.1. Participants

The Truth Initiative Young Adult Cohort Study was designed to understand trajectories of tobacco use among young adults. Details of the cohort have been described elsewhere (Rath et al., 2012). Briefly, the cohort is comprised of a national sample of young adults ages 18–34 drawn from GfK’s KnowledgePanel®, an online panel of adults ages 18 and older that covers both the online and offline populations in the U.S. (GfK, 2013). The panel was recruited via address-based sampling, a probability-based random sampling method that provides statistically valid representation of the U.S. population, including cell phone-only households. The validity of this methodology has been reported previously, and KnowledgePanel samples have been used broadly in studies in the peer-reviewed literature (Pynnonen et al., 2016; Yeager et al., 2011).

The present study uses data from baseline and 6 waves of follow- ups. The baseline survey was conducted in July 2011, with subsequent assessments occurring every 6 months, with Wave 7 conducted in October 2014. The cohort was refreshed at each wave to retain the initial sample size. The panel recruitment rate ranged from 14.8% at baseline to 13.9% in Wave 7. In more than 64.4% of the identified households, one member completed a core profile survey in which key demographic information was collected. For this study, one panel member per household was selected at random to be part of the sample and no members outside the panel were recruited. The average completion rate was 60.3%, yielding a cumulative average response rate across waves of 5.7%. These sample completion and cumulative response rates are comparable to other samples drawn from the KnowledgePanel (Grande et al., 2013; Kelly et al., 2015). Evidence suggests surveys with a low response rate can still be representative of the sample population, even though the risk of nonresponse bias is higher (Brick, 2011). Moreover, studies assessing nonresponse in the KnowledgePanel have found little evidence of nonresponse bias on core demographic and socioeconomic variables (Garrett et al., 2010; Heeren et al., 2008). In fact, prevalence estimates for ever and current cigarette use in this sample are consistent with estimates from national survey data (Rath et al., 2012).

The first 3 waves of this study were approved by the Independent IRB, Inc. (Protocol #20036-007) and Waves 4 through 7 were approved by the Chesapeake Institutional Review Board, Inc. (Protocol #20036020). For all waves, online consent was collected from partici- pants before survey self-administration.

2.2. Measures

2.2.1. Cigarette use Responses to two survey questions were used to assess cigarette

smoking intensity: “Have you ever used or tried cigarettes (even 1 puff)?” and “During the last 30 days, on how many days have you used cigarettes?” A “never smoker” was defined as someone who never smoked cigarettes. An “ever but noncurrent smoker” was defined as someone who reported having tried smoking but had not smoked in the past 30 days. A “non-daily smoker” was defined as someone who reported having smoked 1–24 days in the past 30 days. A “daily smoker” was defined as someone who smoked 25 days or more in the past 30 days. Respondents in the “never smoker” and “ever but noncurrent smoker” smoking intensity levels reported similar patterns in that there was no to very little change in their smoking intensity during the study period. For this reason, the “never smokers” and “ever but noncurrent smokers” were combined into a single category. Responses were coded as 1 for never or ever tried; 2 for non-daily or occasional use; and 3 for daily use. Given the significant skew in a continuous cigarette use measure, with the majority smoking few to no cigarettes, the measure was reformulated to reflect a categorical

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

78

variable. Various multi-level categorical outcome variables were ana- lyzed, but models using a three-level categorical variable had the best overall fit.

2.2.2. Demographic characteristics Demographic information included age, gender, race/ethnicity,

educational attainment (less than high school, high school, and some college or greater), and household income. Race/ethnicity information was categorized as non-Hispanic White, non-Hispanic Black, non- Hispanic other, and Hispanic. The non-Hispanic other category in- cluded Asians, Pacific Islanders, Native Americans, Native Alaskans, and respondents who self-identified as multiracial. Annual household income was determined using items that yielded a 19-category variable ranging from “less than $5000” to “$175,000 or over”. For the purpose of this analysis, annual household income was collapsed into four categories based on the distribution of the data (quartiles): “ < $20,000”, “$20 − $40,000”, “$41 − $75,000” and “more than $75,000”. These demographic characteristics are known to influence tobacco use patterns, and were therefore included as covariates in the analyses (U.S. Department of Health and Human Services, 2014).

2.2.3. Social influences Parental smoking during childhood was obtained by asking, “did

your parents or guardians smoke during your childhood?” Responses were categorized as “one or both of them” and “neither of them.” The latter option was the reference group for analysis. Parental smoking was included in the models as it is a significant predictor of both smoking initiation and progression to established smoking (Biglan et al., 1995; Flay et al., 1994; Tyas and Pederson, 1998).

2.3. Analytic sample

The analysis sample included data from 9791 respondents who entered the study between age 18 and 34.5 years. Sixty-four percent (64.4%) of the respondents had data at two or more time points. Of the total sample, 21.7% had data at 2 ages, 11.6% had data at 3 ages, 10.0% had data at 4 ages, 8.5% had data at 5 ages, 5.1% had data at 6 ages, and 7.5% had data at 7 ages. The analysis file structure was age- based with each case having a total of 34 age variables in six-month increments.

2.4. Data analysis

Analyses were conducted to assess the development of cigarette use patterns at the individual level using a repeated measures latent class analysis method (RMLCA). RMLCA does not impose a particular functional form on growth (e.g., quadratic, cubic, quartic), but rather lets patterns naturally emerge from the data which allows for greater flexibility in modeling cigarette use as participants age over time. This method has been employed to study longitudinal patterns in the development of substance abuse patterns and a range of other behaviors where the unobserved heterogeneity in the developmental response profiles is captured exclusively by a categorical latent variable (Lanza and Collins, 2006; Reboussin et al., 2007).

For these analyses, we used a full-information maximum likelihood (FIML) estimator with robust standard errors (MLR) to estimate the parameters. All of the available data were used under a missing-at- random (MAR) assumption, which allows that the missingness may be related to variables included in the analysis (Dong and Peng, 2013). The multiple cohort design is responsible for a substantial amount of missing data (e.g., a participant entering the study at age 18 would not have data at age 30), and these values are considered missing completely at random (MCAR) because they are missing by design. The FIML estimator is known to yield consistent (unbiased in large samples) estimates under the less stringent MAR assumption which: 1) allows missingness to be explained by scores at other time points or by

covariates; 2) and not by the missing scores themselves. Although it is not possible to empirically evaluate the MAR assumption, we did explore whether covariates in the model were related to missingness. For example, those with less education and income were less likely to respond, and to a lesser degree, non-Hispanic Black and Hispanic respondents were also less likely to respond. These systematic associa- tions suggest MAR, rather than the MCAR, be used as the relevant assumption for this analysis. However, the FIML estimator adjusts for covariate-dependent missingness, so we can confidently conclude the background variables in the model are not a source of nonresponse bias. Omitted predictors of missingness could, of course, introduce bias, but such variables would only be detrimental to the estimates if they had strong relationships with the outcome, not the covariates. Thus, the MAR assumption has been judged to be reasonable for these analyses.

A series of models were estimated by sequentially increasing the number of latent classes by one, at each step comparing the fit of a K- class model to a simpler model with one fewer class. Following typical class enumeration procedures for mixture-type analyses, we chose the final model by considering a combination of criteria, including fit indices such as the Bayesian Information Criterion (BIC), the sample size-adjusted BIC, the Lo-Mendell-Rubin likelihood ratio test (LRT), classification accuracy (e.g., entropy), and the stability of the solution across many sets of random starting values. This combination of criteria has a strong base of support from the methodological literature (Collins and Lanza, 2013; Nylund et al., 2007; Tofighi and Enders, 2008). In addition to statistical criteria, we considered the interpretability of the final model in comparison to established smoking patterns from previous research, such as those identified by White et al. (2002), Mathur et al. (2014), and Dutra et al. (2017) (Dutra et al., 2017; Mathur et al., 2014; White et al., 2002).

Based on the model fit statistics for all latent class models, the three- class model emerged as superior given its improvement in fit (i.e., lowest BIC, significant LRT) relative to the two- and four-class models; clear class separation (highest entropy); and a stable solution across many random starting values (Table A11). Once the optimal model was identified, a set of covariates was added to the model to identify background variables related to latent class membership. Although adding covariates to a latent class model can affect the composition of classification (Vermunt, 2010), methodologists have proposed proce- dures that effectively separate class enumeration from subsequent steps involving the covariates (e.g., the so-called 3-step procedure) (Asparouhov and Muthén, 2014; Vermunt, 2010). While logistic regression models estimating latent classes and covariate effects produced estimates virtually identical to those from the three-step procedure, we report logistic regression estimates from the three-step procedure (Asparouhov and Muthén, 2014). All analyses were con- ducted using Mplus, version 7.3.

3. Results

Table 1 presents the demographic characteristics of the sample by class membership. For the purposes of this table, we use the highest posterior probability to assign respondents to their most likely class membership. However, it is important to emphasize that we present this tabular breakdown strictly as a heuristic aid for understanding the logistic regression results.

Three classes were identified: Class 1 reflected “rapid escalators” (n = 1109, 11.3%), Class 2 reflected “dabblers” (n = 918, 9.4%), and Class 3 reflected “never or ever triers” (n = 7764, 79.3%). The mean age of those in the “dabbler” and “never or ever trier” class was 26 years, while those in the “rapid escalator” class were slightly older, with a mean age of 27 years. There was a slightly higher proportion of females in the “never or ever trier” class (59%) compared with the “rapid escalators” (55%) and “dabblers” (54%). The proportions of non- Hispanic Blacks were similar across all three classes, while there was a larger proportion of non-Hispanic Whites in the “rapid escalator” class

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

79

(72%) than the “dabbler” (55%) or “never or ever trier” class (57%). More “never or ever tries” reported attaining a college degree or higher education (32%) compared with “rapid escalators” (8%) and “dabblers” (22%). In terms of annual household income, the majority of “rapid escalators” (34%) and “dabblers” (28%) reported incomes of less than $20,000, while the majority of “never or ever triers” (28%) reported annual household incomes of more than $75,000. A greater proportion of the “rapid escalators” reported parental smoking (77%) comparted with “dabblers” (53%) and “never or ever triers” (38%).

Table 2 provides the logistic regression coefficients and odds ratios for each covariate. As noted previously, these regression coefficients derive from the three-step covariate procedure and thus account for uncertainty in latent class membership. Covariates associated with a higher probability of membership in the “rapid escalator” class (Class 1) or the “dabbler” class (Class 2) relative to the “never or ever triers” class (Class 3) included being male (reference group = males; rapid escala- tors: OR = 0.80; dabblers: OR = 0.71), having a parent who smokes (rapid escalators: OR = 4.49; dabblers: OR = 1.61), and having less than a college education (see Table 2). Compared with White respon- dents, non-White respondents had a lower probability of membership in the “rapid escalator” class (Class 1) relative to the “never or ever triers” class (Class 3) (non-Hispanic Black: OR = 0.49; other non-Hispanic: OR = 0.60; Hispanic: OR = 0.18). However, race/ethnicity did not predict membership in the “dabbler” class (Class 2) as compared to the never or ever trier class (Class 3). Compared with respondents who reported higher household incomes (≥$75,000), those with lower income had a higher probability of membership in the “rapid escalator” class (Class 1) relative to the “never or ever triers” class (Class 3) (less than $20,000: OR = 2.63; $20-$40,000: OR = 2.12; $41-$75,000: OR = 1.43). Although the inclusion of covariates could, in principle, alter the class composition, the class proportions were quite stable;

Class 1 11.3%, Class 2 9.4% and Class 3 79.3%. Simultaneously estimating the latent classes and covariate effects had virtually no impact on class membership probabilities.

Figs. 1–3 illustrate the patterns of cigarette use over time for the three latent classes. For example, Fig. 1 plots the estimated probability of a daily smoking response by age and latent class. Stability of smoking patterns was estimated by examining the x-axis (which denotes participant age) in relation to the changes in the probability of cigarette use (as denoted by the y-axis). Class 1 [square line] reaches a stable probability (∼80%) of daily smoking at approximately age 21, three years post baseline. In contrast, the other two classes (Class 2 and Class 3) have relatively low probabilities of daily smoking over the entire study period. Class 2 [triangle line] shows some daily use, but never reaches the probability of daily use as compared to that of the “rapid escalators” (Class 1). Those in Class 2 were estimated to stabilize their daily use in their mid-twenties, (∼23 years old) while Class 3 had a near-zero probability of reporting daily use.

The probabilities of non-daily smoking are illustrated in Fig. 2. Class 2 (“dabblers”) had the highest probabilities (average ∼50%), while Class 1 (“rapid escalators”) had lower overall probabilities of occasional use. Looking at the probabilities at each age, Classes 1 and 2 reflected similar probabilities of occasional smoking at ages 18 and 19, but diverged quickly by age 21 into clear patterns of daily and occasional smoking. As expected, Class 3 (“never or ever triers”) exhibited a near- zero probability of non-daily smoking.

Fig. 3 shows the probability of never or ever smoking (not past 30- day) for each class. Class 3 had a probability close to 100% for being a never smoker or an ever trier, but not a current smoker, and is denoted as the “never or ever triers” class. Use patterns for this class were quite stable, whereas Class 2, the “dabblers”, reflected decreasing probabil- ities of being a never or ever trier (Fig. 3).

The estimated probabilities of both non-daily and daily smoking for “dabblers” (Class 2) increased between the ages of 18–21, but stabilized in the mid-twenties. “Rapid escalators” (Class 1) reached a stable pattern (∼90%) of any smoking by age 21 (data not shown).

Table 1 Demographic characteristics of study sample by most likely class membership.

Participant characteristics

Total Rapid Escalators Class 1

Dabblers Class 2

Never or Ever Triers Class 3

(n = 9791) (n = 1109) (n = 918) (n = 7764)

Age (mean) 26 27 26 26

Gender (% female)

58% 55% 54% 59%

Race/ethnicity (%) Non-Hispanic

White 59% 72% 55% 57%

Non-Hispanic Black

11% 11% 12% 11%

Other Non- Hispanic

7% 6% 8% 7%

Hispanic 24% 11% 26% 25%

Educational attainment (%) Less than high

school 10% 14% 11% 9%

High school graduate

22% 33% 26% 20%

Some college 40% 44% 40% 39% College graduate

or more 29% 8% 22% 32%

Annual household income Less than $20,000 23% 34% 28% 21% $20–40,000 25% 31% 24% 24% $40–75,000 26% 22% 26% 27% More than

$75,000 26% 13% 22% 28%

Parental smoking No 56% 23% 47% 62% Yes 44% 77% 53% 38%

∼2.1% of total respondents were 18.0 at study entry, ∼3.9% were 18.5 at study entry.

Table 2 Logistic regression estimates relating smoking intensity classes to baseline demographic variables.

Est. SE z p OR

Rapid Escalator (Class 1) vs. Never/Ever Triers (Class 3) Parents Smoke 1.502 0.091 16.591 < 0.001 4.491 Female −0.221 0.079 −2.793 0.01 0.802 Non-Hispanic Black −0.717 0.126 −5.697 < 0.001 0.488 Other Non-Hispanic −0.505 0.166 −3.039 < 0.01 0.604 Hispanic −1.714 0.140 −12.264 < 0.001 0.180 Less than high school 1.932 0.181 10.693 < 0.001 6.903 High school graduate 1.825 0.160 11.391 < 0.001 6.203 Some college 1.546 0.154 10.041 < 0.001 4.693 Less than $20,000 0.965 0.131 7.384 < 0.001 2.625 $20–40,000 0.750 0.128 5.874 < 0.001 2.117 $40–75,000 0.354 0.135 2.625 0.01 1.425

Dabblers (Class 2) vs. Never/Ever Triers (Class 3) Parents Smoke 0.475 0.091 5.218 < 0.001 1.608 Female −0.342 0.092 −3.727 < 0.001 0.710 Non-Hispanic Black 0.079 0.151 0.527 0.60 1.082 Other Non-Hispanic 0.204 0.171 1.189 0.23 1.226 Hispanic 0.063 0.115 0.548 0.58 1.065 Less than high school 0.429 0.178 2.415 0.02 1.536 High school graduate 0.526 0.138 3.819 < 0.001 1.692 Some college 0.325 0.120 2.714 0.01 1.384 Less than $20,000 0.321 0.138 2.332 0.02 1.379 $20–40,000 −0.005 0.140 −0.037 0.97 0.995 $41–75,000 0.160 0.128 1.243 0.21 1.174

Note: Reference categories for categorical predictors are as follows: No parental smoking, Male gender, Non-Hispanic White, College graduate, and More than $75,000 income.

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

80

4. Discussion

This study provides evidence related to the development of estab- lished cigarette smoking using a recent sample of young adults. This is of particular importance since today’s young adults came of age after the implementation of the Master Settlement Agreement that mandated strong restrictions to reduce exposure to cigarette marketing among youth (National Association of Attorneys General, 1998). Moreover, these trajectories to established patterns of cigarette smoking reflect the significant changes in general patterns of tobacco use in response to tobacco policy implementation (clean indoor air legislation, tobacco product price via tax legislation), as well as the broad expansion of the tobacco product landscape.

Three discrete classes of cigarette use patterns were identified in this study: 79.3% fall into the class of never or ever users of cigarettes (no current use of cigarettes), 11.3% fall into the class of “rapid escalators” or daily users of cigarettes, and 9.4% fall into the “dabbler” class which reflects non-daily or occasional use. Key findings indicate an established pattern of cigarette use typically occurs by early adulthood, age 20–22. In contrast to other studies on patterns of established smoking, we find those in the non-daily, lighter use class (Class 2, “dabblers”) stabilize their pattern of cigarette use around the same age as those in the daily, heavy use class (Class 1, “rapid escalators”) (Kvaavik et al., 2014; Levy et al., 2009; Mathur et al., 2014). While past prevention efforts have focused primarily on younger teens, these findings indicate the risk for progression to established

Fig. 1. Probability of Daily Smoking by Class.

Fig. 2. Probability of Non-Daily Smoking by Class.

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

81

patterns of cigarette use continues into young adulthood. These findings are consistent with lifecourse studies which describe the increased risk profile among emerging adults as they transition through educational and occupational processes (Arnett, 2005). Findings provide further evidence to support the National Cancer Institute’s recommendation to focus on interventions which interrupt the pathways from experimenta- tion to established patterns of use among those in their later teens and early 20’s (Working Group of the NCI Board of Scientific Advisors, 2016).

Our findings are consistent with several other studies which have identified low socioeconomic status (SES) and parental smoking as risk factors associated with cigarette use progression among young adults (Filippidis et al., 2015; Gilman et al., 2009; Mays et al., 2014). Lower income, lower education, and parental smoking were found to be predictive of membership in the “rapid escalator” or daily use category, and parental smoking lower education were significant predictors of membership in the “dabbler” or occasional use class. The influence of these SES measures is greater for the “rapid escalator” class as compared to the “dabblers” or “never/ever triers”. This is consistent with existing research that non-daily smokers typically have higher levels of education and income compared with daily smokers, making them more similar to non-smokers with regard to these characteristics (Ackerson and Viswanath, 2009; Levy et al., 2009; Reitzel et al., 2014). Similarities between the “dabblers” and the “never/ever triers” could explain the lack of a significant association of income when comparing those groups. Findings indicate tailoring interventions for low-income older teens and young adults will be key to disrupting the processes associated with progression to daily tobacco use. The findings on low SES groups as measured by parental smoking, education, and income reflect a strong and consistent dose-response relationship. This suggests that tax policies which increase the product price could also be particularly effective for preventing initiation and slowing progression among these price-sensitive low SES groups.

These findings also have significant implications in light of recent research indicating the influence of increasing the minimum purchase age for tobacco products to 21 years. This policy initiative is associated with declines in youth smoking (Fidler and West, 2010; Schneider et al., 2015), and has wide support from the public (King et al., 2015; Winickoff et al., 2015). To date, two U.S. states, California and Hawaii, have raised the minimum purchase age for tobacco products from 18 to

21 years. These findings provide support for such policy initiatives in that those who are not smoking by age 21 will most likely remain nonsmokers.

Strengths of the study include its large, nationally representative cohort of young adults, a group typically identified as hard-to-reach. This sample of young adults updates the existing literature related to progression of established smoking patterns, particularly relevant since tobacco use patterns have shifted in recent years. The use of this large, young adult cohort allows for a better understanding of current use patterns. Limitations include the lack of information regarding smoking behaviors during early teens and adolescence. As measurements begin at respondents’ age 18, we cannot assess the age of smoking initiation and its impact on future smoking patterns. Future research should explore the full lifecourse trajectory of combustible tobacco use from early adolescence through adulthood. Restricting the sample to only those within a specified age range limits the ability to generalize results to larger populations. Additionally, this study relies on self-reported data, which is subject to bias. Finally, socioeconomic status was measured by asking for the respondent’s household income. As respondents were in young adulthood and possibly becoming more financially independent, it cannot be determined whether the house- hold income reflects the respondent’s income, or the income of his/her parents/guardians.

5. Conclusions

Findings demonstrate clear patterns of progression to established smoking by age 22. Our results highlight the importance of developing and implementing interventions, including increasing tobacco-related taxes and raising the age limit for purchase of combustible cigarettes to age 21, to help reduce the probability of progression to established patterns of tobacco use.

Role of funding source

This study was funded by Truth Initiative.

Conflicts of interest

No conflicts declared

Fig. 3. Probability of Never or Ever Trying, but Not Current Smoking by Class.

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

82

Contributors

E.H. designed the study. E.H. and M.B. wrote the paper. V.W., A.J., and C.E. performed the analyses and contributed to the methods and results sections. J.R., J.C., A.V., and D.V. contributed to the writing and revisions. All authors have reviewed and approved the final paper.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.drugalcdep.2017.03. 040.

References

Ackerson, L.K., Viswanath, K., 2009. Communication inequalities, social determinants, and intermittent smoking in the 2003 Health Information National Trends Survey.

Arnett, J.J., 2005. The developmental context of substance use in emerging adulthood. J. Drug Issues 35, 235–254.

Asparouhov, T., Muthén, B., 2014. Auxiliary variables in mixture modeling: three-step approaches using M plus. Struct. Equ. Model. 21, 329–341.

Biglan, A., Duncan, T.E., Ary, D.V., Smolkowski, K., 1995. Peer and parental influences on adolescent tobacco use. J. Behav. Med. 18, 315–330.

Brick, J.M., 2011. The future of survey sampling. Public Opin. Quart. 75, 872–888. Chassin, L., Presson, C.C., Pitts, S.C., Sherman, S.J., 2000. The natural history of cigarette

smoking from adolescence to adulthood in a midwestern community sample: multiple trajectories and their psychosocial correlates. Health Psychol. 19, 223.

Chen, K., Kandel, D.B., 1995. The natural history of drug use from adolescence to the mid- thirties in a general population sample. Am. J. Public Health 85, 41–47.

Collins, L.M., Lanza, S.T., 2013. Latent Class And Latent Transition Analysis: With Applications In The Social, Behavioral, And Health Sciences. John Wiley and Sons, Hoboken, NJ.

Davis, R.M., Gilpin, E.A., Loken, B., Viswanath, K., Wakefield, M.A., 2008. The Role Of The Media In Promoting And Reducing Tobacco Use. National Cancer Institute Bethesda, MD.

Dong, Y., Peng, C.-Y.J., 2013. Principled missing data methods for researchers. SpringerPlus 2, 1–17.

Dutra, L.M., Glantz, S.A., Lisha, N.E., Song, A.V., 2017. Beyond experimentation: five trajectories of cigarette smoking in a longitudinal sample of youth. PLoS One 12, e0171808.

Fidler, J.A., West, R., 2010. Changes in smoking prevalence in 16–17-year-old versus older adults following a rise in legal age of sale: findings from an English population study. Addiction 105, 1984–1988.

Filippidis, F.T., Agaku, I.T., Vardavas, C.I., 2015. The association between peer, parental influence and tobacco product features and earlier age of onset of regular smoking among adults in 27 European countries. Eur. J. Public Health 25, 814–818.

Flay, B.R., Hu, F.B., Siddiqui, O., Day, L.E., Hedeker, D., Petraitis, J., Richardson, J., Sussman, S., 1994. Differential influence of parental smoking and friends' smoking on adolescent initiation and escalation and smoking. J. Health Soc. Behav 248–265.

Garrett, J., Dennis, J.M., DiSogra, C., 2010. Non-response bias: recent findings from address-based panel recruitment. Paper Present At: Annual Conference of the American Association for Public Opinion Research.

GfK, 2013. KnowledgePanel Design Summary. Gilman, S.E., Rende, R., Boergers, J., Abrams, D.B., Buka, S.L., Clark, M.A., Colby, S.M.,

Hitsman, B., Kazura, A.N., Lipsitt, L.P., 2009. Parental smoking and adolescent smoking initiation: an intergenerational perspective on tobacco control. Pediatrics 123, e274–e281.

Grande, D., Mitra, N., Shah, A., Wan, F., Asch, D.A., 2013. Public preferences about secondary uses of electronic health information. JAMA Intern. Med. 173, 1798–1806.

Heeren, T., Edwards, E.M., Dennis, J.M., Rodkin, S., Hingson, R.W., Rosenbloom, D.L., 2008. A comparison of results from an alcohol survey of a prerecruited Internet panel and the National Epidemiologic Survey on Alcohol and Related Conditions. Alcohol Clin. Exp. Res. 32, 222–229.

Kandel, D.B., Logan, J.A., 1984. Patterns of drug use from adolescence to young adulthood: I. Periods of risk for initiation continued use, and discontinuation. Am. J. Public Health 74, 660–666.

Karp, I., O'loughlin, J., Paradis, G., Hanley, J., Difranza, J., 2005. Smoking trajectories of adolescent novice smokers in a longitudinal study of tobacco use. Ann. Epidemiol. 15,

445–452. Kelly, B., Squiers, L., Bann, C., Stine, A., Hansen, H., Lynch, M., 2015. Perceptions and

plans for prevention of Ebola: results from a national survey. BMC Public Health 15, 1.

King, B.A., Jama, A.O., Marynak, K.L., Promoff, G.R., 2015. Attitudes toward raising the minimum age of sale for tobacco among US adults. Am. J. Prev. Med. 49, 583–588.

Kozlowski, L.T., Giovino, G.A., 2014. Softening of monthly cigarette use in youth and the need to harden measures in surveillance. Prev. Med. Rep. 1, 53–55.

Kvaavik, E., von Soest, T., Pedersen, W., 2014. Nondaily smoking: a population-based, longitudinal study of stability and predictors. BMC Public Health 14, 1.

Lantz, P.M., 2003. Smoking on the rise among young adults: implications for research and policy. Tob. Control 12, i60–i70.

Lanza, S.T., Collins, L.M., 2006. A mixture model of discontinuous development in heavy drinking from ages 18 to 30: The role of college enrollment. J. Stud. Alcohol 67, 552–561.

Lenk, K.M., Chen, V., Bernat, D.H., Forster, J.L., Rode, P.A., 2009. Characterizing and comparing young adult intermittent and daily smokers. Subst. Use Misuse 44, 2128–2140.

Levy, D.E., Biener, L., Rigotti, N.A., 2009. The natural history of light smokers: a population-based cohort study. Nicotine Tob. Res ntp011.

Mathur, C., Stigler, M.H., Erickson, D.J., Perry, C.L., Forster, J.L., 2014. Transitions in smoking behavior during emerging adulthood: a longitudinal analysis of the effect of home smoking bans. Am. J. Public Health 104, 715–720.

Mays, D., Gilman, S.E., Rende, R., Luta, G., Tercyak, K.P., Niaura, R.S., 2014. Parental smoking exposure and adolescent smoking trajectories. Pediatrics 133, 983–991.

National Association of Attorneys General, 1998. Master Settlement Agreement (MSA). http://www.naag.org/naag/about_naag/naag-center-for-tobacco-and-public-health/ master-settlement-agreement/master-settlement-agreement-msa.php

Nylund, K.L., Asparouhov, T., Muthén, B.O., 2007. Deciding on the number of classes in latent class analysis and growth mixture modeling: a Monte Carlo simulation study. Struct. Equ. Model. 14, 535–569.

Pynnonen, M.A., Handelsman, J.A., King, E.F., Singer, D.C., Davis, M.M., Lesperance, M.M., 2016. Parent perception of newborn hearing screening: results of a US national survey. JAMA Otolaryngol. 142, 538–543.

Rath, J.M., Villanti, A.C., Abrams, D.B., Vallone, D.M., 2012. Patterns of tobacco use and dual use in US young adults: the missing link between youth prevention and adult cessation. J. Environ. Public Health. 2012, 679134.

Reboussin, B.A., Hubbard, S., Ialongo, N.S., 2007. Marijuana use patterns among African- American middle-school students: a longitudinal latent class regression analysis. Drug Alcohol Depen. 90, 12–24.

Reitzel, L.R., Buchanan, T.S., Nguyen, N., Ahluwalia, J.S., 2014. Associations of subjective social status with nondaily and daily smoking. Am. J. Health Behav. 38, 245–253.

Schneider, S.K., Buka, S.L., Dash, K., Winickoff, J.P., O'Donnell, L., 2015. Community reductions in youth smoking after raising the minimum tobacco sales age to 21. Tob. Control.

Terry-McElrath, Y.M., O'Malley, P.M., 2015. Trends and timing of cigarette smoking uptake among US young adults: survival analysis using annual national cohorts from 1976 to 2005. Addiction 110, 1171–1181.

Tofighi, D., Enders, C.K., 2008. Identifying the correct number of classes in growth mixture models. Adv. Latent Var. Mix. Mod. 317–341.

Tyas, S.L., Pederson, L.L., 1998. Psychosocial factors related to adolescent smoking: a critical review of the literature. Tob. Control 7, 409–420.

U.S. Department of Health and Human Services, 2014. The Health Consequences of Smoking: 50 Years of Progress. A Report of the Surgeon General. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health, Atlanta, GA

Vermunt, J.K., 2010. Latent class modeling with covariates: two improved three-step approaches. Polit. Anal. 450–469.

Waters, K., Harris, K., Hall, S., Nazir, N., Waigandt, A., 2006. Characteristics of social smoking among college students. J. Am. Coll. Health 55, 133–139.

White, H.R., Pandina, R.J., Chen, P.H., 2002. Developmental trajectories of cigarette use from early adolescence into young adulthood. Drug Alcohol Depend. 65, 167–178.

Winickoff, J.G., McMillen, R., Tanski, S., Wilson, K., Gottlieb, M., Crane, R., 2015. Public support for raising the age of sale for tobacco to 21 in the United States. Tob. Control.

Working Group of the NCI Board of Scientific Advisors, 2016. Tobacco Control Research Priorities for the Next Decade: Working Group Recommendations for 2016–2025.

Yeager, D.S., Krosnick, J.A., Chang, L., Javitz, H.S., Levendusky, M.S., Simpser, A., Wang, R., 2011. Comparing the accuracy of RDD telephone surveys and internet surveys conducted with probability and non-probability samples. Public Opin. Quart nfr020.

E. Hair et al. Drug and Alcohol Dependence 177 (2017) 77–83

83