Sample

The U.S. Bureau of Labor Statistics NLSY97 [35] began in 1997 with a sample of 8,984 youth aged 12 to 16 and provides 15 years of annual data on smoking (ages 26 to 30 at wave 15). The NLSY includes a nationally representative sample (n = 6,748) and an oversample of Hispanic/Latino and black respondents (n = 2,236) divided into five cohorts based on age (12 to 16) on December 31, 1996. Eighty-three percent (7,423 of 8,984) of those sampled in 1997 remained in 2011.

The analysis, which is based on the unrestricted public use NLSY97 dataset, included the 8,791 (98%) NLSY97 respondents with at least three non-missing years of data for 30-day cigarette use because a minimum of three points is necessary for latent class growth analysis (LCGA) [36]. All analyses used Bureau of Labor Statistics baseline weights [37] to adjust for NLSY sampling techniques, including the oversample of racial/ethnic minorities. This is an analysis of a de-identified dataset and was ruled exempt by the University of California San Francisco Committee on Human Research.

Measures

Analysis

LCGA is a person-centered approach to modeling that can be used to identify individuals with similar behavioral trajectories in adolescence and young adulthood [44]. Never smokers (34.1% of sample, n = 3,147), those who reported 0 days smoked in the past 30 days for all non-missing years of data collection, were specified as an a priori class and excluded from the LCGA [14]. This established approach [45] reduced the computational burden of the large number of zeros (from never smokers) and provided a better fit to the distribution.

We used Mplus Version 7.3 because it accounts for the longitudinal structure of the NLSY and for missing data using full information maximum likelihood [46].

Overall sample characteristics

Approximately half of the sample was female (48.7%) and was primarily non-Hispanic white (66.6%) (Table 1). About half the respondents (52.9%) had mothers whose highest education was GED/high school diploma. At age 16, 88.6% were in school, including 54.7%) who were employed while attending school. Mean household income was about twice (mean = 192%, median = 200%) the poverty level. In 1997, mean peer smoking was 1.63 (on average, participants estimated that approximately between 25% and 50% of their peers smoked), mean likelihood of breaking rules in school was 2.26 (between “disagree a little” and “neither agree nor disagree”), 47.3% did not live with both biological parents in 1997, 45.1% had ever drunk alcohol, 21.0% had ever tried marijuana, and 7.1% had ever used cocaine or hard drugs in 1997. Mean depression score in 2000 (range from 0 to 4) was 0.93 (symptoms of depression “some of the time”).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Individual-level characteristics by trajectory of cigarette smoking.

https://doi.org/10.1371/journal.pone.0171808.t001

By age 26, most of the respondents (59.3%) had never been married and had no children (54.4%). At age 26, 9.4% had no degree, 55.3% had a GED/high school diploma, 30.6% had a college degree, and 4.7% had a graduate or professional degree.

Trajectories of smoking from 1997–2011

We tested one to five trajectories. The BIC consistently declined across solutions (S2 Table). The LMR-LRT p-value increased across trajectories, reaching non-significance at five trajectories. Since this result suggested five trajectories was not superior to four and all posterior probabilities were over 0.7 for all solutions (0.90–0.98), we selected four classes.

Based on trajectory shapes and the literature, we named the four trajectories (in addition to never smokers, trajectory 0) experimenters (trajectory 1), quitters (trajectory 2), early established smokers (trajectory 3), and late escalators (trajectory 4) (Fig 1). Never smokers comprised 34.1% (n = 3,147) of the sample, experimenters 13.6% (n = 1,205), quitters 8.1% (n = 701), early established smokers 39.0% (n = 3,205), and late escalators 5.2% (n = 533).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Four trajectories of smoking behavior among ever smokers in the National Longitudinal Survey of Youth 1997 (NLSY97).

Analyses of 19 years (15 waves) of smoking data (days smoked per month) for participants in the NLSY97 revealed four patterns of smoking behavior (in addition to never smokers): experimenters, quitters, early established smokers, and late escalators.

https://doi.org/10.1371/journal.pone.0171808.g001

Days smoked per month peaked at age 16 for experimenters (mean = 1.22; 95% CI: 0.96–1.48), with mean days smoked per month remaining below one for 16 of the 19 years. For quitters, days smoked peaked at age 17 (10.64; 9.61–11.67) and decreased to less than one day per month at age 25. For early established smokers, days smoked peaked at age 22 (22.60; 22.14–23.05) and plateaued by age 30 (18.49; 17.21–19.78). Late escalators’ highest smoking was 9.00 (6.11–11.89) days per month at age 30 and still trending upward.

Only 2.0% of experimenters, 49.5% of quitters, 93.5% of early established smokers, and 24.0% of late escalators ever smoked daily. Median years of daily smoking was 0 for experimenters (10th–90th percentile: 0–0; only 2.0% ever smoked daily), 0 for quitters (0–3), 6 for early established smokers (1–12), and 0 for late escalators (0–2). Median years from first cigarette to daily smoking was 0 (0–1) for the 2.0% (23 of 1,205) of experimenters who made this transition, 1 year (0–4) for the 48.4% (339 of 701) of quitters who made this transition, 1 year (0–5) for the 92.5% (2,966 of 3,205) of early established smokers who made this transition, and 3 years (0–11) for the 26.3% (140 of 533) of late escalators who made this transition. Median age of transition to daily smoking (for those who transitioned) was 15 (13–16) for experimenters, 17 (15–20) for quitters, 17 (14–21) for early established smokers, and 26 (24–28) for late escalators.

Predictors of trajectory membership

Our adjusted model included all covariates (Table 2). Compared with never smokers, experimenters were more likely to have ever tried alcohol in 1997 and less likely to be non-Hispanic black (than non-Hispanic white) and to be employed and enrolled in school at age 16 (versus not enrolled in school and unemployed). Quitters were more likely to be unemployed at age 16, to have a mother who had at least completed high school/GED, and to have ever drank alcohol or used marijuana in 1997. They were less likely to be black, to be enrolled in school at age 16, and to have ever been married by age 26. Early established smokers were more likely to have a mother whose highest level of education was a GED/high school diploma (versus no degree), to have used alcohol or marijuana in 1997, have peers who smoked in 1997, to be depressed in 2000, have broken rules in school in 2008, and to have children at age 26. They were less likely to be black or Hispanic, to be enrolled in school or employed at age 16, to live with both biological parents in 1997, and to have ever been married at age 26. Late escalators were more likely to be Hispanic, to have drank alcohol in 1997, and to have broken rules in school and less likely to have ever been married at age 26 than never smokers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Adjusted odds of trajectory membership for all significant covariates (with 95% confidence intervals).

https://doi.org/10.1371/journal.pone.0171808.t002

All of the variance inflation factors were 1.44 or under in the final adjusted model, indicating little multicollinearity. The results of the unadjusted analyses (see Table 1) were similar to the adjusted results (see Table 2). The only variables that were significant in bivariate analyses but not significant in adjusted models were household income, which was significantly lower for early established smokers than members of the other trajectories (except late escalators), and ever use of cocaine/hard drug use, which was more common for experimenters and quitters than other trajectories.

Trajectory characteristics

We found several qualitative and quantitative differences between our trajectories and those identified in LCGAs of AddHealth (S3 and S4 Tables) [30,31,32]. Fuemmeler et al. [31] did not identify a quitter class; this group is key to understanding why some individuals spontaneously quit. In addition, neither Fuemmeler et al. [31] nor Pollard et al. [32] identified a true late escalator class (started smoking after age 20 and trending upward at end of follow-up), thereby failing to identify an important minority group that has been ignored in the past. Fuemmeler et al. [31] and Pollard et al.’s [32] experimenter classes peaked much later (ages 24 and 25, respectively) than ours (age 16). Costello et al.’s [30] quitter class peaked at age 20, later than our quitter class, which peaked at age 16.

Some of these differences may be due to our specification of never smokers as an a priori class (as opposed to including never smokers in the LCGA). As a result, our never smoker category was smaller than those identified in analyses of AddHealth (S4 Table). In addition, our use of days smoked per month (versus cigarettes per day or a combined measure of frequency and intensity) distinguishes our analysis from those completed with AddHealth. Most importantly, the great number of waves of data collection in the NLSY97 (annually between 1997 and 2011) resulted in a level of detail unavailable when using AddHealth’s three or four waves of data (1995, 1996, 2001–2002, 2008–2009). This important difference enabled us to identify specific ages at which key transitions occur.

For example, early established smokers escalated between age 12 and 22, late escalators escalated between age 22 and 30, and quitters de-escalated between 18 and 24. As a result, we concluded that late escalators should be targeted in young adulthood. In the Fuemmeler et al. [31] analysis, in particular, with the exception of the late heavy user group, the most distinguishing graphical characteristic between the classes was the intercept. In contrast, our analysis revealed trajectories with differing intercepts, slopes, and shapes.

Like the current analysis and analyses of Add Health [30,31,32], LCGAs of smoking of the sub-national samples [4,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29] provides support for the existence of multiple smoking trajectories. The findings of the sub-national studies, however, are less likely to be generalizable to the entire United States, so have limited applicability to the development of national policies and interventions for smoking prevention and cessation.

Characteristics of individuals in trajectories

Analyses of AddHealth [30,31,32] revealed that alcohol or drug use, deviance, maternal smoking, peer smoking, conduct problems, depressive symptoms, and state prevalence of adolescent smoking distinguished one or more smoking trajectories from never smokers. With the exception of state smoking prevalence, which we did not include in our analysis, our results demonstrated similar findings for these variables.

We also identified several characteristics that were not identified in previous analyses (S4 Table). Members of all trajectories except experimenters were less likely to have been married at age 26 compared with never smokers. Compared with never smokers, experimenters were less likely to be black and less likely to be enrolled in school or working at age 16. However, overall, experimenters were similar to never smokers. Our results suggest that tobacco control programs should expand their efforts beyond preventing experimentation to also preventing escalation to established smoking.

Quitters were less likely to be black, tended to have mothers with higher education, and were more likely to have ever used marijuana. Compared with early established smokers, quitters were more likely to be Hispanic and less likely to break rules in school. Additional research should identify further characteristics about this group and their reasons for quitting. However, the existence of this group likely indicates that traditional youth-based smoking prevention and cessation programs are successful for some adolescents.

Early established smokers were less likely to be black or Hispanic, to have educated mothers, and to grow up in a two-parent household; they were more likely to have ever used marijuana and to drop out of school by age 16. As a result, they are probably less likely to be reached by school-based interventions in adolescence.

Late escalators only exhibited some traditional smoking risk factors, were more likely to be Hispanic, and were still enrolled in school at age 16. Because their smoking escalates after age 18, they are more likely to benefit from interventions that extend into young adulthood.

Public health policy implications

This is the first analysis that can be used to create targeted national public health and clinical interventions for smoking prevention and early cessation that extend through young adulthood. Population-level interventions, including tobacco taxes, smokefree laws (particularly workplace) and rules (such as car), smokefree housing, and media campaigns are key to reaching all types of smokers [56,57,58]. People covered by smokefree laws are more likely to have smokefree cars [56] and homes [59], which increase the likelihood of successful smoking cessation [57,60]. In addition, as recommended by the American Academy of Pediatrics [61], all movies and video games depicting smoking should be given an R rating [62,63] or eliminated entirely. Similarly, restricting all tobacco advertising, including point-of-sale and product placement, would deter smoking initiation and adolescent smoking and progression to established smoking. In addition to these approaches, we identified specific methods of reaching early established smokers and late escalators. Because early established smokers are more likely to have children, one method of reaching them is by incorporating smoking cessation interventions into early intervention programs, such as Women, Infants, and Children, an assistance program for pregnant women and children under age 5, and California’s “First Five,” a county-based early intervention program [40], which is funded by tobacco taxes. To reach late escalators in young adulthood when their smoking frequency is increasing, we recommend smokefree schools, smokefree college campuses, college-based smoking cessation programs, and bar-based smoking cessation and prevention interventions [64].

The Truth Initiative, which targets youth and young adults [65], and the Centers for Disease Control and Prevention’s Tips From Former Smokers media campaign, which targets 18- to 54-year-olds [66], are likely to be effective for early established smokers and late escalators. Media campaigns, such as the Food and Drug Administration’s The Real Cost and Fresh Empire campaigns [41,67], should be extended to include individuals into their late 20s, particularly since youth-centric programs are an industry tactic [68].

Because of the immense gains made in tobacco control in the past 50 years, including denormalizing tobacco and revolutionizing social norms, it is necessary to reevaluate the efficacy of the established approach to tobacco prevention, particularly as new databases emerge. The Population Assessment of Tobacco and Health (PATH) study should be used to continue to identify patterns of smoking to prevent progression to established smoking and encourage smoking cessation and identify the effects of rising dual use of tobacco products (including e-cigarettes) and marijuana on trajectories of cigarette smoking.

Limitations

The primary limitation of this analysis is that it does not include policy variables that affect smoking. It is possible that individuals shift trajectories over time, but LCGAs do not accommodate such shifts.

Conclusion

Days smoked per month is a useful measure of smoking for identifying different types of smokers and should therefore be included in tobacco use surveys. Also, multiple patterns of smoking are necessary to fully describe adolescent and young adult smoking uptake and patterns of use. Just as the tobacco industry identified different types of smokers (market segments) to target, tobacco control programs can use information about types of smokers to develop more effective interventions [69]. Targeting young adults (18 to 24) is important because this group is most likely to quit and do so successfully. Pediatricians should collect information on smoking intensity and frequency and attempt to prevent smoking escalation among light smokers by helping them identify triggers, referring them to counseling resources, and encouraging youth to advocate a smokefree lifestyle for parents and peers. Because smoking is still dynamic beyond age 18, pediatricians and other physicians should ensure that information about tobacco use is transferred to general medicine practitioners as their patients get older. Our analysis also supports increasing the smoking age from 18 to 21 [61]; 16.9% (954) of the ever smokers in our analysis initiated smoking between the ages of 18 and 21.