Developmental Patterns of Tobacco Product and Cannabis Use Initiation in High School

Abstract

Aims:

To identify prototypical developmental patterns of tobacco product and cannabis use and co-use initiation during adolescence, and determine risk factors for and consequences of these initiation patterns.

Design:

Prospective repeated-measures cohort with eight semiannual assessments during high school. Multiple Event Process Survival Mixture modeling identified latent initiation classes with distinct patterns of variation in timing of use initiation of tobacco products and cannabis. We then estimated: (1) associations of baseline risk factors with membership in initiation classes and (2) differences between initiation classes in frequency of cannabis and tobacco product use at the final assessment.

Setting:

Ten high schools in the Los Angeles, CA, USA metropolitan area, 2013–2017.

Participants:

Students (1031 [45.4%] males; mean [SD] age at baseline = 14.6 [0.39] years) who had never-used any tobacco products or cannabis at baseline 9^th grade assessment (N=2272).

Measurements:

Self-report measures of electronic cigarette (e-cigarette), combustible cigarette, hookah, cigar/cigarillos, and cannabis use were collected at each assessment.

Findings:

Four distinct tobacco and cannabis use initiation classes were identified: (1) Early and High-Risk Cannabis and Poly-Tobacco Initiators (N=116; 5.1%); (2) Early Cannabis and Poly-Tobacco Initiators (N=172; 7.6%); (3) Late Cannabis and e-Cigarette Initiators (N=431; 19.0%); and (4) Abstainers (N=1553; 68.4%). At baseline, older age for the Early and High-Risk Cannabis and Poly-Tobacco Initiators (OR=1.22[95%CI 1.10, 1.35], p<.001) , peer cannabis use (1.60[1.23, 2.08], p<.001), and delinquent behavior (1.30[1.08, 1.55], p=.004) were associated with membership in the three initiation classes (vs. Abstainers). Membership in the Early and High-Risk Cannabis and Poly-Tobacco Initiators class (vs. three other classes) was significantly associated with increased past 30-day frequency and daily intensity of use at the final assessment (p-values<0.001).

Conclusions:

Older age, peer cannabis use, and delinquent behavior appear to be risk factors for the initiation of tobacco/cannabis product use among high school students in the Los Angeles metropolitan region. Early and higher-risk poly-product use initiation appears to be associated with greater escalation of past 30-day and daily tobacco and cannabis use at the end of the high school.

INTRODUCTION

Initiation of tobacco and cannabis use typically occurs during adolescence, with substantial inter-individual variability in the timing and order of tobacco and cannabis use initiation [1–3]. Younger ages of tobacco use onset and cannabis use onset are both strongly associated with increased risk of persistent use of each substance extending into adulthood [4,5]. Co-use of tobacco and cannabis products in adolescence is associated with additive and potentially synergistic adverse health and social consequences [1].

Cigarette smoking has traditionally been conceptualized as a risk factor for the use of cannabis and other illicit substances, and historically, the predominant developmental pattern of adolescent tobacco-cannabis use initiation involved initiation of cigarette smoking followed by initiation of cannabis use [6]. However, the tobacco product marketplace has rapidly evolved this decade [7,8], engendering a greater diversity of tobacco products than previous decades, including new products, such as electronic cigarettes (e-cigarettes), which have markedly increased in popularity in youth [8]. Meanwhile combustible cigarettes have reduced in popularity in youth and the mean onset age of combustible cigarette use has increased this decade [9], legalization of recreational and medical cannabis across a growing number of U.S. states has been coupled with reductions in perceptions of harm of cannabis use among youth [10]. These changes have added new complexities to understanding developmental patterns of use and co-use initiation of tobacco and cannabis products.

While a systematic identification of the patterns of various tobacco products and cannabis use and co-use initiation timing is lacking, some recent data suggests that use of non-cigarette tobacco products (e.g., e-cigarettes, hookah) are associated with cannabis use initiation [11]. Other epidemiological research indicates that cannabis use may precede tobacco product use among some youth [1,2]. The growing popularity of e-cigarettes further complicates use initiation sequences, as e-cigarette use is increasingly becoming the most common first substance used by teens [12], and e-cigarette use not only predicts onset of cannabis use [13,14], but also is prospectively associated with the initiation of combustible cigarette smoking [12].

The numerous configurations of the initiation sequences of multiple tobacco products and cannabis use pose challenges for setting priorities for adolescent tobacco and cannabis control policy. A parsimonious model that identifies typical sequences of use initiation and their relative prevalence is needed to inform whether poly tobacco/cannabis product prevention campaigns are warranted, and the risk factors associated with developmental timing of poly-product use initiation that should be targeted in such prevention measures. While little is known regarding developmental patterns of use and co-use initiation of poly-products as well as risk factors for and consequences of use initiation, prior research suggests that risk of poly-substance use initiation may differ by gender and race/ethnicity [15,16]. For example, males were significantly more likely than females to be in an earlier initiating class of tobacco, alcohol, and cannabis products, relative to a later initiating class [16]. Also, prior research of risk factors for adolescent tobacco or cannabis use initiation have studied each of these outcomes separately and found that tobacco and cannabis use initiation among youth was associated with conduct problems [17], internalizing mental health symptomology (e.g., depression, anxiety) [18,19], and environmental risk factors such as family history of cigarette smoking and living in a single-parent home [20,21]. However, it is unknown if these risk factors are associated with profiles with distinct conjoint trajectories of tobacco products and cannabis use initiation during adolescence.

The current study aimed to identify distinct latent classes (subgroups) of adolescents that differed in their patterns of combustible tobacco products, e-cigarette, and cannabis use initiation across four years of high school (ages 14–18 years old) using a statistical approach, Multiple Event Process Survival Mixture (MEPSUM), that allowed us to characterize adolescents according to their timing of several interrelated events [15,16,22]. Specific aims of this study were to: (1) identify subgroups with distinct initiation trajectories of the multiple events including onset timing of five events (i.e., first use of combustible cigarettes, e-cigarettes, hookah, cigars/cigarillos, and cannabis); (2) examine if baseline risk factors (sociodemographic, environmental, and intrapersonal factors) for tobacco and cannabis use initiation were associated with membership in the initiation subgroups; and (3) examine associations of membership in the initiation classes with past 30-day frequency and daily intensity of cannabis and tobacco product use at the end of high school.

METHODS

Participants

Participants were students from 10 high schools in the Los Angeles, CA metropolitan area who were followed as part of a prospective cohort study of substance use and mental health [23]. Data were collected via paper questionnaires at participants’ high schools during eight semiannual assessments beginning when students were entering 9^th grade (mean [standard deviation (SD)] age = 14.55 [0.39] years; fall of 2013) through the end of their fourth year of high school (mean [SD] age = 17.92 [0.38]; spring of 2017). A total of 3396 participants enrolled in the study; overall retention throughout the entire study period was 92.5% (N=3140). The analytic sample included students who reported never-using any tobacco or cannabis products at baseline and provided valid data for tobacco or cannabis use during at least one of the seven subsequent follow-up assessments (N=2272; Figure S1 and Table S1 in the Online Supplemental Materials). Written or verbal parental consent was obtained and all students assented to participation. The University of Southern California Institutional Review Board approved this study.

Measures

Past 6-Month Tobacco Product and Cannabis Use.

At each assessment, past 6-month use of combustible cigarettes, cigars/cigarillos, hookah, e-cigarettes, and cannabis were assessed (yes/no). Use of big cigars and little cigars or cigarillos were combined to form a composite cigars/cigarillos use variable, and e-cigarette use was assessed by any use of e-cigarettes with or without nicotine. Cannabis use was defined as smoking combustible cannabis (“pot, weed, grass, hash, reefer, or bud”).

Sociodemographic Characteristics.

Sociodemographic characteristics including gender, age, race/ethnicity, and parental education level were assessed with self-report questionnaires to investigator-defined fixed categories at baseline (see response categories given in Table 1).

Table 1.

Sociodemographic, environmental and intrapersonal characteristics by tobacco product and cannabis use initiation classes

	Tobacco Product and Cannabis Use Initiation Classes (N=2272)
Baseline Covariates	Early and High-Risk Cannabis and Poly-Tobacco Initiators (N=116)a	Early Cannabis and Poly-Tobacco Initiators (N=172)b	Late Cannabis and e-Cigarette Initiators (N=431)c	Abstainers (N=1553)d	Test of Group Differencese
	N (%) or M (SD)	N (%) or M (SD)	N (%) or M (SD)	N (%) or M (SD)	χ2 (df) or F (df)	p
Sociodemographic characteristics
Male Gender (vs. Female), N (%)	64 (55.2)1	75 (43.6)1	194 (45.0)1	698 (44.9)1	4.85 (3)	.18
Age, M (SD), year	14.63 (0.39)1	14.57 (0.37)1,2	14.54 (0.39)2	14.54 (0.39)2	2.07 (3)	.10
Race/Ethnicity N (%)					55.25 (12)	<.001
Hispanic	52 (45.2)1,2	102 (59.6)1	192 (45.1)2	656 (42.7)2
Asian	16 (13.9)1,2	15 (8.8)2	58 (13.6)2	366 (23.8)1
African American	5 (4.3)1	7 (4.1)1	23 (5.4)1	72 (4.7)1
Non-Hispanic White	30 (26.1)1	24 (14.0)2	86 (20.2)1,2	251 (16.3)1,2
Otherf	12 (10.4)1	23 (13.5)1	67 (15.7)1	192 (12.5)1
Parents graduated college (vs. Less education), N (%)g	60 (59.4)1	56 (37.1)2	206 (54.5)1	769 (57.1)1	22.78 (3)	<.001
Environmental characteristics
Lives with both parents (vs. Other situation), N (%)	76 (65.5)1,2	110 (64.0)1,2	252 (58.9)2	1110 (71.7)1	27.78 (3)	<.001
Family history of smoking (vs. no history), N (%)	76 (68.5)1,2	98 (59.4)1,2	285 (69.3)1	846 (56.6)2	25.47 (3)	<.001
Family history of drug use (vs. no history), N (%)	25 (22.7)1	30 (18.3)1	102 (24.8)1	160 (10.7)2	59.74 (3)	<.001
Peer smoking, N (%)h	22 (19.5)1	32 (18.8)1	51 (12.2)1,2	127 (8.3)2	31.64 (3)	<.001
Peer cannabis use, N (%)i	48 (42.5)1,2	78 (45.9)1	135 (32.1)2	230 (15.0)3	152.64 (3)	<.001
Intrapersonal characteristics
Depressive symptoms (vs. non-clinical), N (%)j	26 (22.6)1	30 (18.0)1	73 (17.5)1	150 (10.1)2	32.12 (3)	<.001
Generalized anxiety symptoms (vs. non-clinical), N (%)j	30 (26.1)1	18 (10.8)2,3	77 (18.5)1,2	164 (11.0)2	33.92 (3)	<.001
Delinquent behavior, M (SD)k	1.43 (0.33)1	1.41 (0.41)1	1.37 (0.33)1	1.27 (0.41)2	28.67 (3)	<.001

Note. Available (non-missing) data for respective variable and, for categorical variables, denominator for within-column percentages.

^aAvailable data (Ns = 101 – 116).

^bAvailable data (Ns = 151 – 172).

^cAvailable data (Ns = 378 – 431).

^dAvailable data (Ns = 1347 – 1553).

^eDifferences were calculated by the χ² tests for categorical variables or one-way Analysis of Variance (ANOVA) for continuous variables. Groups not sharing superscript numerals are significantly different (p<.05) in Bonferroni-corrected post-hoc pairwise contrasts for χ² tests and ANOVA Least Significant Difference.

^fOther race/ethnicity includes students who selected ‘American Indian/Alaskan Native’ ‘Native Hawaiian/Pacific Islander,’ ‘Multiethnic/Multiracial,’ or ‘Other’ options for the forced-choice race/ethnicity question.

^gStudents (N=295) who did not respond to the survey question or who marked “don’t know” are not included in the denominator.

^hPeer smoking = Has 1 or more friends who smoke combustible cigarettes.

ⁱPeer cannabis use = Has 1 or more friends who smoke cannabis.

^jPositive screen at clinical level (i.e., symptomatic; scoring at or above the clinical cut-off value).

^kRange = 1 – 6.

Environmental Risk Factors.

At baseline, measures of environmental risk factors included family living situation (“Who do you live with most of the time?” [0=both parents; 1=other situation]). Survey measures assessed family history of smoking and drug abuse, which were coded dichotomously (yes/no). Peer smoking behavior (“In the last 30 days, how many of your five closest friends have smoked cigarettes?”) and peer cannabis use were assessed (0=none; 1=one or more friends).

Intrapersonal Risk Factors.

At baseline, symptoms of depression and general anxiety were assessed using the Revised Child Anxiety and Depression Scale (RCADS), a well-validated measure of symptoms of DSM-IV generalized anxiety disorder and major depressive disorder [24]. Participants completed two separate subscales for depression (10-item; α=0.92) and general anxiety (6-item; α=0.89). For each item respondents indicated how often they experienced each individual symptom on a 4-point response scale, from 0 (Never) to 3 (Always). A binary cutoff for experiencing clinically-significant symptoms was used as in previous research [24]. Delinquent behaviors were assessed by the averaged frequency (Never[=1] – 10 or more times[=6]) of 11 past 6-month delinquent behaviors (e.g., stealing, destroying property, physically fighting; α=0.79) [25].

Past 30-Day Frequency and Daily Intensity of Tobacco and Cannabis Product Use.

At the final follow-up, participants reported their past 30-day frequency and daily intensity of use for the five previously assessed products as well as vaporized cannabis (“Electronic device to vape THC/hash oil”)—which was not assessed at baseline. Participants reported the number of days they used each tobacco and cannabis product in the past 30 days (response options: 0, 1–2, 3–5, 6–9, 10–14, 15–19, 20–24, 25–29, all 30 days). These response categories were transformed into quantitative count variables by taking the mean integer value of each ordinal past 30-day frequency variable [26]: 0, 2, 4, 8, 12, 17, 22, 27, 30.

For the daily intensity outcomes, participants reported the number of cigarettes they smoked on each smoking day (0, 1, 2–5, 6–10, 11–15, 16–20, ≥20 cigarettes) as well as the number of combustible cannabis they smoked (0, 1, 2–3, 4–6, 7–10, ≥11 combustible cannabis). Two separate questions characterized the intensity of daily vaping for both nicotine and cannabis: (1) the number of vaping episodes per day (0, 1, 2, 3–5, 6–9, 10–14, 15–20, ≥20 times); and (2) the number of puffs taken during each vaping episode (0, 1, 2, 3–5, 6–9, 10–14, 15–20, ≥20 puffs). These response categories were also transformed into quantitative count variables by taking the lowest value of each ordinal category for use in the daily intensity analyses [27].

Data Analysis

The MEPSUM analytic technique grouped participants by differences in timing and cumulative risk of tobacco product and cannabis use initiation via parallel survival analysis models [22]. The MEPSUM approach used in this study builds upon prior latent trajectory modeling approaches (e.g., growth mixture modeling), and is uniquely suited for parsing use initiation of multiple tobacco and marijuana products. While growth mixture modeling can identify latent subgroups with distinct trajectories of use progression over time, the MEPSUM analysis can identify the specific timing of the occurrence of multiple events and characterize latent initiation subgroups. To investigate different patterns underlying substance use initiation, we focused on a sample excluding students who reported ever use of any tobacco or cannabis products at baseline (N=1111) and conducted the MEPSUM analyses using five indicators: (1) combustible cigarettes; (2) hookah; (3) cigars/cigarillos; (4) e-cigarettes; and (5) cannabis. Given that there was limited variability in the chronological age of the sample, we utilized seven semiannual follow-up assessments (i.e., 6-month interval) as discrete time-points in the survival analysis, and past 6-month use of each tobacco product and cannabis was operationalized as initiation of each respective product. Classes were composed of individuals with similar temporal patterns of tobacco and cannabis product initiation. Participants were censored for all time-points following their initiation of each individual tobacco product or cannabis.

To determine the optimal number of classes, we compared standard fit indices (i.e., Akaike information criterion [AIC], Bayesian information criterion [BIC], entropy) and determined latent classes representing different patterns of tobacco and cannabis initiation over time in the sample [28]. Preliminary analyses calculated descriptive statistics and tested differences in study variables between classes with chi-squared and one-way analysis of variance (ANOVA) tests.

Multinomial logistic regression models tested associations of baseline risk factors and membership in the tobacco and cannabis use initiation trajectory classes. Negative binomial regression models were also tested by adjusting for all baseline covariates to assess associations of the initiation groups with past 30-day frequency and daily intensity of tobacco and cannabis use at the final assessment. The MEPSUM and follow-up analyses were conducted in Mplus version 7 [29] using full information maximum likelihood estimating to account for missing data, and participants’ high school (clustering by school) was considered using complex modeling, which adjusted parameter standard errors for interdependence in the data due to the nesting of students by their school. Benjamini-Hochberg multiple-testing corrections [30] were applied to control the false-discovery rate at 0.05 (based on two-tailed corrected p-value). The analysis was not pre-registered, and the results should therefore be considered exploratory.

RESULTS

Identification of Tobacco Product and Cannabis Initiation Groups

Based on model fit indices criteria (i.e., Akaike Information Criterion [AIC] values; see Table S2 & Supplemental Analyses in the Online Supplemental), we determined that a four-class solution best fit the underlying heterogeneity in hazard of tobacco and cannabis initiation in the sample (i.e., subgroups with distinct time-points and median ages of tobacco product or cannabis use initiation; Table S3). We named the four classes based on their initiation trajectories: (1) Early and High-Risk Cannabis and Poly-Tobacco Initiators (N=116; 5.1%); (2) Early Cannabis and Poly-Tobacco Initiators (N=172; 7.6%); (3) Late Cannabis and e-Cigarette Initiators (N=431; 19.0%); and (4) Abstainers (N=1553; 68.4%). Figure 1 displays the unstructured hazard functions (i.e., probability of each tobacco/cannabis product use initiation at each follow-up time-point; ĥ_t) and cumulative probability (i.e., aggregate probability of initiation across follow-ups; Dˆ_t) of tobacco and cannabis initiation for the four classes across eight assessments.

Figure 1.

Hazard functions and cumulative distributions of tobacco and cannabis use initiation by latent class

Note. N=2272. ^aHazard functions = Probability of tobacco or cannabis use initiation at each follow-up assessment. ^bCumulative probability = Aggregate probability of initiation at each follow-up assessment. X-axis = Assessment wave (Baseline [1] to 42-month follow-up [8]). Y-axis = Probability of initiation.

The Early and High-Risk Cannabis and Poly-Tobacco Initiators class was characterized by an early and large initial risk of initiating all tobacco/cannabis product use that remained moderate-to-high at each time-point throughout the study period. Beginning at the first follow-up assessment, the estimated probability (risk/hazard) of initiation increased sharply for e-cigarettes (ĥ₂=0.40) and hookah (ĥ₂=0.40), while risk of combustible cigarette (ĥ1=0.00; ĥ2=0.19) and marijuana (ĥ1=0.00; ĥ2=0.18) initiation also increased from the baseline to the first follow-up assessment. The estimated hazard probability of cigars/cigarillo initiation in this class was lower than the other tobacco products at the first follow-up assessment (ĥ2=0.09). The risk of e-cigarette (ĥ₂=0.40; ĥ₃=0.45; ĥ₄=0.50; ĥ₅=0.53), combustible cigarette (ĥ₂=0.19; ĥ₃=0.24; ĥ₄=0.28; ĥ₅=0.34) and cigar (ĥ₂=0.09; ĥ₃=0.13; ĥ₄=0.17; ĥ₅=0.22) initiation increased gradually over the next three assessments, whereas risk of hookah initiation experienced a gradual decline from the third through fifth assessments (ĥ₂=0.40; ĥ₃=0.35; ĥ₄=0.32; ĥ₅=0.30). Risk of cannabis initiation increased rapidly over this time period and continued escalating throughout the end of the study period (ĥ₂=0.18; ĥ₃=0.36; ĥ₄=0.54; ĥ₅=0.68; ĥ₆=0.77; ĥ₇=0.81; ĥ₈=0.83). By the eighth and final assessment (the end of high school), the cumulative probability of use initiation in this class equaled or approached 100% for cannabis (Dˆ₈=1.00), e-cigarettes (Dˆ₈=0.99), combustible cigarettes (Dˆ₈=0.95) and hookah (Dˆ₈=0.94), while 84% of the sample ever-used cigars (Dˆ₈=0.84).

The Early Cannabis and Poly-Tobacco Initiators class displayed the highest risk of initiation at the beginning of high school, however after the second assessment risk of initiation of all products decreased over the study period. Risk of combustible cigarette (ĥ₂=0.33), hookah (ĥ₂=0.38), e-cigarette (ĥ₂=0.42) and cannabis (ĥ₂=0.37) initiation peaked at the second assessment. After the second assessment, the risk of e-cigarette (ĥ₃=0.34; ĥ₄=0.26; ĥ₅=0.19; ĥ₆=0.13; ĥ₇=0.07; ĥ₈=0.04), combustible cigarette (ĥ₃=0.21; ĥ₄=0.13; ĥ₅=0.09; ĥ₆=0.06; ĥ₇=0.05; ĥ₈=0.04), hookah (ĥ₃=0.25; ĥ₄=0.16; ĥ₅=0.09; ĥ₆=0.06; ĥ₇=0.02; ĥ₈=0.01), cigar (ĥ₃=0.07; ĥ₄=0.03; ĥ₅=0.01; ĥ₆=0.01; ĥ₇=0.01; ĥ₈=0.01), and cannabis (ĥ₃=0.28; ĥ₄=0.22; ĥ₅=0.17; ĥ₆=0.15; ĥ₇=0.14; ĥ₈=0.14) initiation decreased over the study period. The cumulative probabilities of use initiation at the end of high school were lower than those in the High-Risk Cannabis and Poly-Tobacco Initiator class for e-cigarettes (Dˆ₈=0.82), combustible cigarettes (Dˆ₈=0.64), hookah (Dˆ₈=0.66), marijuana (Dˆ₈=0.82) and cigars (Dˆ₈=0.29).

The Late Cannabis and e-Cigarette Initiators class was characterized by low initial risk of tobacco product and cannabis use initiation throughout the first year of high school before experiencing an escalation in the risk of e-cigarette and cannabis initiation beginning at the fourth assessment. Risks of initiating combustible cigarettes (ĥ₂ = 0.01), hookah (ĥ₂ = 0.05) and cigars (ĥ₂ = 0.00) were low at the second assessment and increased gradually thereafter. Risk of e-cigarette initiation was minimal at the first assessment (ĥ₁ = 0.00; ĥ₂ =0.10), but began increasing at the third assessment (ĥ₃ = 0.14; ĥ₄ =0.17; ĥ₅ = 0.20; ĥ₆ = 0.23; ĥ₇ = 0.25), while risk of cannabis initiation experienced a more rapid increase beginning at the fifth assessment (ĥ₃ = 0.05; ĥ₄ =0.17; ĥ₅ = 0.26; ĥ₆ = 0.39; ĥ₇ = 0.52). The highest risk for initiating both e-cigarettes (ĥ₈ = 0.26) and cannabis (ĥ₈ = 0.62) occurred at the final assessment. The cumulative probabilities for use of e-cigarettes (Dˆ₈ = 0.78) and marijuana (Dˆ₈ = 0.94) in this class approached those of the previous two classes, while the cumulative probabilities for combustible tobacco products (Dˆ₈ = 0.35; 0.42; 0.11; for cigarettes, hookah and cigars, respectively) were lower.

The patterns of tobacco product and cannabis use initiation in the Abstainers class were distinct from the other three classes. Members of this class displayed low initial risk of e-cigarette (ĥ₂=0.02), cannabis or combustible tobacco use initiation (ĥ₂-values<0.01), which did not substantially increase over the study period. At the end of high school, the cumulative initiation probabilities for all products were considerably lower than those in the other three classes (Range of Dˆ₈=0.01–0.08).

Associations of Sociodemographic, Environmental, and Intrapersonal Risk Factors with Tobacco and Cannabis Initiation Trajectories

While descriptive statistics for the sample characteristics and differences among the four initiation subgroups indicated significant differences in baseline factors (except age and gender, Table 1), associations of all baseline factors with membership in three tobacco/cannabis initiation subgroups (vs. Abstainers) were examined (Table 2). In a model adjusting for all sociodemographic, environmental and intrapersonal risk factors as simultaneous regressors, greater age at baseline was significantly associated with increased odds of membership in the Early and High-Risk Cannabis and Poly-Tobacco Initiators class (OR[95% CI]=1.22[1.10, 1.35]; effect size[d]=0.11 ; p<0.001) and Early Cannabis and Poly-Tobacco Initiators class (OR[95% CI]=1.17[1.05, 1.30]; effect size[d]=0.09; p=0.01), compared to the Abstainer class. There were small differences by race/ethnicity, with only Asian (vs. Hispanic) significantly associated with reduced odds of membership (OR[95% CI]=0.74[0.63, 0.87]; effect size[d]=−0.17; p<0.001) in the Early Cannabis and Poly-Tobacco Initiators class (vs. Abstainers).

Table 2.

Associations between baseline risk factors and initiation classes

	Outcomes: Tobacco Product and Cannabis Use Initiation Classesa
Baseline Covariates	Early and High-Risk Cannabis and Poly-Tobacco Initiators (N=116)		Early Cannabis and Poly-Tobacco Initiators (N=172)		Late Cannabis and e-Cigarette Initiators (N=431)
	OR (95% CI)	p	OR (95% CI)	p	OR (95% CI)	p
Sociodemographic Risk Factors
Male Gender (vs. Female)	1.35 (1.01, 1.81)	.04	1.01 (0.92, 1.11)	.87	1.06 (0.96, 1.16)	.24
Ageb	1.22 (1.10, 1.35)	<.001†	1.17 (1.05, 1.30)	.01†	1.00 (0.94, 1.07)	.98
Race/Ethnicity
Hispanic	Reference	−	Reference	−	Reference	−
Asian	0.83 (0.64, 1.09)	.18	0.74 (0.63, 0.87)	<.001†	0.86 (0.70, 1.04)	.13
African-American	0.95 (0.69, 1.29)	.73	0.93 (0.78, 1.11)	.40	0.98 (0.87, 1.11)	.75
Non-Hispanic White	1.20 (0.93, 1.54)	.16	0.96 (0.89, 1.03)	.27	1.09 (0.90, 1.30)	.37
Otherc	0.88 (0.70, 1.10)	.26	1.03 (0.87, 1.22)	.76	1.05 (0.91, 1.22)	.50
Parents graduated college (vs. Less education)	0.83 (0.67, 1.03)	.09g	0.80 (0.66, 0.97)	.03	0.92 (0.74, 1.14)	.44
Environmental Risk Factors
Lives with both parents (vs. Other living situation)	0.94 (0.82, 1.08)	.38	0.83 (0.74, 0.92)	.01†	0.93 (0.79, 1.09)	.36
Family history of smoking (vs. No history)	1.11 (0.89, 1.37)	.35	0.97 (0.86, 1.09)	.59	1.15 (1.03, 1.28)	.02
Family history of drug use (vs. No history)	1.14 (0.94, 1.39)	.18	1.07 (0.85, 1.34)	.56	1.24 (1.13, 1.35)	<.001†
Peer smokingd	1.06 (0.97, 1.16)	.20	0.99 (0.84, 1.16)	.91	0.94 (0.84, 1.05)	.24
Peer cannabis usee	1.60 (1.23, 2.08)	<.001†	1.68 (1.47, 1.93)	<.001†	1.38 (1.26, 1.52)	<.001†
Intrapersonal Risk Factors
Delinquent behaviorsb	1.30 (1.08, 1.55)	.004†	1.34 (1.04, 1.73)	.02†	1.24 (1.08, 1.43)	.003†
Depressive symptoms (vs. non-clinical)f	1.19 (1.01, 1.39)	.04	1.17 (0.96, 1.41)	.11	1.08 (0.99, 1.17)	.08
Generalized anxiety symptoms (vs. non-clinical)f	1.29 (1.02, 1.62)	.03	0.86 (0.72, 1.03)	.09	1.11 (1.01, 1.22)	.04

Note. N=2272. Multinomial logistic regression model including all baseline covariates as simultaneous regressors (OR=odds ratio, 95%CI=95% confidence interval).

^aReference group is Abstainers (N=1553).

^bContinuous scale regressors were standardized (Mean=0, SD=1).

^cOther race/ethnicity = students who selected ‘American Indian/Alaskan Native,’ ‘Native Hawaiian/Pacific Islander,’ ‘Multiethnic/Multiracial,’ or ‘Other’ options for the forced-choice race/ethnicity question.

^dPeer smoking = Has 1 or more friends who smoke combustible cigarettes.

^ePeer cannabis use = Has 1 or more friends who smoke cannabis.

^fPositive screen at clinical level (i.e., symptomatic; scoring at or above the clinical cut-off value).

^gThe finding was inconclusive as to whether or not a significant association was present based on the calculation of Bayes factor (>.03). For parental education level, the Bayes factor was 0.37 based on an expected effect size (OR=1.00) [34], the published effect size (i.e., Ln OR=−0.19), and standard error (SE=0.11) of the parameter using the Dienes online calculator [37].

^†Statistically significant after Benjamini-Hochberg corrections for multiple testing to control false discovery rate at .05 (based on two-tailed corrected p-value).

Participants who reported a family history of drug abuse (vs. no history) had greater odds of being in the Late Cannabis and e-Cigarette Initiators class (OR[95% CI]=1.24[1.13, 1.35]; effect size[d]=0.12; p<0.001), compared to Abstainers. Living with both parents (vs. other living arrangement) was significantly associated with reduced odds of membership in the Early Cannabis and Poly-Tobacco Initiators class (OR[95% CI]=0.83[0.74, 0.92]; effect size[d]=−0.10; p=0.01) but not the two other two classes (p-values>0.36). Having friends who smoked cigarettes at baseline was not associated with membership in any of the initiation subgroups (p-values>0.20), however having friends smoking cannabis was significantly associated with membership in each of the three initiation classes (ORs=1.38–1.68; effect sizes[d]=0.18–0.29; p-values<0.001).

Finally, each increase in one SD-unit of delinquent behaviors was associated with significantly greater odds of membership in the Early and High-Risk Cannabis and Poly-Tobacco Initiators (OR[95% CI]=1.30[1.08, 1.55]; effect size[d]=0.14; p=0.004), Late Cannabis and e-Cigarette Initiators (OR[95% CI]=1.24[1.08, 1.43]; effect size[d]=0.12; p=0.03) and Early Cannabis and Poly-Tobacco Initiators (OR[95% CI]=1.34[1.04, 1.73]; effect size[d]=0.16; p=0.02).

Associations of Initiation Trajectories with Past 30-Day Frequency and Daily Intensity of Tobacco and Cannabis Product Use

After adjusting for all baseline covariates, members of the Early and High-Risk Cannabis and Poly-Tobacco Initiators class displayed significantly greater past 30-day frequency of using each tobacco and cannabis product than the other three classes at the last follow-up assessment (p-values<0.001; Table 3). The Late Cannabis and e-Cigarette Initiators had a significantly higher frequency of past 30-day e-cigarette, smoked cannabis, vaporized cannabis, and hookah use than both Early Cannabis and Poly-Tobacco Initiators and Abstainers (p-values<0.001). There were no significant differences in past 30-day frequency of tobacco or cannabis product use at the end of high school between the Early Cannabis and Poly-Tobacco Initiators and Abstainers. The Early and High-Risk Cannabis and Poly-Tobacco Initiators class displayed significantly greater daily intensity of cannabis and tobacco product use (e.g., cigarettes per day, puffs per vaping episode) than the three other initiation classes (p-values<0.001). The Late Cannabis and e-Cigarette Initiators reported significantly greater daily intensity of combustible cannabis use and vaping episodes than Early Cannabis and Poly-Tobacco Initiators and Abstainers (p-values<0.001).

Table 3.

Associations between initiation classes and past 30-day frequency and daily intensity of tobacco and cannabis product use

	Tobacco Product and Cannabis Use Initiation Classes
Tobacco and Cannabis Product Use at End of High School	Early and High-Risk Cannabis and Poly-Tobacco Initiators (N=116)	Early Cannabis and Poly-Tobacco Initiators (N=172)	Late Cannabis and e-Cigarette Initiators (N=431)	Abstainers (N=1553)	Test of Group Differencesa
	M (SE)	M (SE)	M (SE)	M (SE)	p
Past 30-Day Frequency of Use
Combustible cigaretteb	3.90 (0.20)1	0.10 (0.17)2	0.19 (0.10)2	0.01 (0.05)2	<.001
e-Cigarette with nicotineb	3.92 (0.26)1	0.15 (0.58)3	0.40 (0.13)2	0.11 (0.07)3	<.001
Combustible cannabisb	9.62 (0.39)1	0.66 (0.32)3	2.73 (0.20)2	0.14 (0.10)3	<.001
Vaporized cannabisb	3.62 (0.25)1	0.19 (0.20)3	0.90 (0.13)2	0.08 (0.07)3	<.001
Hookahb	1.27 (0.13)1	0.01 (0.10)3	0.19 (0.07)2	0.01 (0.03)3	<.001
Cigarb	1.03 (0.12)1	0.01 (0.09)2	0.12 (0.06)2	0.01 (0.03)2	<.001
Daily Intensity of Use
Combustible cigarettes per dayc	0.94 (0.07)1	0.14 (0.06)2	0.06 (0.04)2,3	0.01 (0.02)3	<.001
Combustible cannabis per dayd	1.79 (0.08)1	0.19 (0.07)3	0.62 (0.04)2	0.05 (0.02)3	<.001
Vaping episodes per day (nicotine)c	2.00 (0.15)1	0.12 (0.11)2,3	0.25 (0.07)2	0.04 (0.04)3	<.001
Puffs per vaping episode (nicotine)c	1.61 (0.14)1	0.17 (0.11)2,3	0.35 (0.07)2	0.04 (0.03)3	<.001
Vaping episodes per day (THC)c	1.31 (0.11)1	0.10 (0.09)2,3	0.33 (0.06)2	0.02 (0.03)3	<.001
Puffs per vaping episode (THC)c	1.32 (0.11)1	0.15 (0.08)2,3	0.31 (0.05)2	0.03 (0.02)3	<.001

Note. N=2272. Estimated marginal mean (standard error [SE]) of each tobacco and cannabis product use outcome across four initiation classes at final follow-up assessment (Spring 2017, 12^th grade) was calculated after adjusting for all baseline covariates (i.e., gender, age, race/ethnicity, parental education, family living situation, family history of smoking, family history of drug use, peer smoking, peer cannabis use, delinquent behaviors, depressive symptoms, generalized anxiety symptoms).

^aGroups not sharing superscript numerals are significantly different in post-hoc pairwise contrasts across tobacco and cannabis use initiation classes (p<.05).

^bPast 30-day use frequency ranged from 0 to 30.

^cDaily intensity of use ranged from 0 to 20.

^dDaily intensity of use ranged from 0 to 11.

DISCUSSION

This prospective study identified four distinct developmental patterns of combustible tobacco products, e-cigarette, and cannabis use initiation in a longitudinal cohort of adolescents across four years of high school. The four classes of adolescents primarily differed in their timing of peak risk and cumulative risk at the end of high school for the initiation of tobacco products, e-cigarette, and cannabis use, with more subtle differences in the order of combustible tobacco products and cannabis use initiation between each class. In the Early and High-Risk Poly-Tobacco and Cannabis Initiators and Early Cannabis and Poly-Tobacco Initiators classes, risk for tobacco product and cannabis initiation followed similar patterns, with earlier risk of initiation generally coinciding with greater prevalence of tobacco product and cannabis ever-use (i.e., higher overall cumulative probabilities) at the end of high school. However, the Early and High-Risk Poly-Tobacco and Cannabis Initiators and Late Cannabis and e-Cigarette Initiators classes showed higher risk for e-cigarette use initiation than cannabis initiation during first 12-month follow-ups. The risk for cannabis initiation rapidly increased and became higher than the risk for e-cigarette use initiation beginning at the fourth assessment and across later follow-ups. These results may show that e-cigarettes precede cannabis, consistent with the finding that e-cigarette use predicts subsequent marijuana use among youth in the U.S. [13]. However, across the three initiation classes, we found minimal evidence of combustible tobacco product use initiation systematically preceding cannabis use initiation or vice versa.

At baseline, older age, peer cannabis use, and delinquent behavior were associated with membership in each of the three initiation classes (vs. youth who reported negligible tobacco and cannabis product use [Abstainers]), and membership in the initiation classes was associated with progression of tobacco product and cannabis use at the final assessment. In concordance with previous research [31], youth who were older at baseline displayed greater risk of substance use and had greater odds of membership in both the Early and High-Risk Cannabis and Poly-Tobacco Initiators and Early Cannabis and Poly-Tobacco Initiators classes. Family history of drug use was significantly associated with membership in the Late Cannabis and e-Cigarette Initiators class (vs. Abstainers). This finding suggests that family drug use history may confer risk for adolescent substance use, consistent with our previous study on familial transmission of poly substance use trajectories via the development of youth psychopathologic risk factors during adolescence [32]. The impact of family drug use history on substance use is more likely to be distal or indirect and presumably works through other variables [33]. Unlike previous research [16], we did not find evidence of gender differences in patterns of tobacco and marijuana initiation. We also found minimal evidence of race differences between students in trajectories of tobacco and marijuana use initiation, as there were no significant differences between non-Hispanic White and African American students.

We observed differences in past 30-day frequency and daily intensity of tobacco product and cannabis use at the end of high school among the initiation classes. Students in the Early and High-Risk Cannabis and Poly-Tobacco Initiators class (vs. the three other classes) reported the greatest frequency and intensity of use across all tobacco and cannabis products. Students in this class showed both early initiation and high risk for co-use of all tobacco and cannabis products during adolescence. These findings extend prior research, that has primarily focused on the effects of tobacco or cannabis use initiation in isolation [31,34], and suggest that the timing of conjoint tobacco and cannabis product use initiation may affect escalation of use later in adolescence. Co-use of cannabis and tobacco has been shown to exacerbate the adverse and addictive health effects of each substance in isolation, and is associated increased likelihood of experiencing psychosocial difficulties, chronic respiratory problems, and mental health symptomology [6,35]. Such youth may benefit from interventions designed to prevent early initiation of tobacco and cannabis products.

Also, at the final assessment, students in the Late Cannabis and e-Cigarette Initiators class reported greater past 30-day frequency of combustible cannabis, vaporized cannabis, e-cigarette, and hookah use than the Early Cannabis and Poly-Tobacco Initiators, while members of the Early Cannabis and Poly-Tobacco Initiators class did not significantly differ from Abstainers in their frequency or intensity of tobacco and cannabis product use. This finding was not consistent with prior studies to identify significant association between early substance use initiators and the escalation of later substance use [4,5]. We thus conducted additional tests to address the characteristics of Early Cannabis and Poly-Tobacco Initiators and the limitation of our analytic plan in the present study. While the MEPSUM approach is uniquely suited for parsing complex patterns of multiple tobacco and cannabis products use initiation among nonusers, our results are limited by the use of subsample excluding students initiated any tobacco/cannabis use before high school (at study baseline). To address this issue, we tested differences in baseline risk factors and tobacco/cannabis use at the end of high school among early initiators before high school and other four groups identified by the MEPSUM (see the Online Supplemental Analyses; Table S6 & S7).

Strengths of the study include the person-centered statistical modeling approach, large sociodemographically diverse sample and assessment of several of combustible and noncombustible tobacco products popular among youth as well as various environmental and intrapersonal risk factors. Given the growing popularity alternative tobacco and cannabis products as well as increasing poly-tobacco use and co-use of cannabis and tobacco products among youth, parsimoniously and comprehensively assessing the use of novel tobacco and cannabis products is vital. This study thus included vaporized cigarette or cannabis as study variables in the analyses and might contribute to the literature investigating co-use patterns of novel and traditional products among adolescents.

A few limitations of the current study must also be noted. First, e-cigarette use was described as any use of, “electronic cigarettes for nicotine or without nicotine,” because the question of past 6-month e-cigarette use did not distinguish nicotine-containing products from other devices without nicotine during first three assessments. While the term “tobacco products” is generally reserved for those products that are made from and conation nicotine [36], we were to clarify and carefully interpret our findings for tobacco products including e-cigarettes with or without nicotine. Given that the sample was drawn from California, a state that emphasizes tobacco control and has experienced a recent loosening of cannabis restrictions, it is possible that the results may not generalize to other geographic areas. Additionally, the racial and ethnic distribution of our sample (i.e., large percentage of Hispanic and Asian students) may have resulted in distinct findings from previous research.

CONCLUSIONS

This study used an advanced statistical approach to identify four parsimonious classes indicative of prototypical developmental patterns of tobacco product and cannabis use initiation. The subgroups of adolescents were primarily distinguished by their timing of risk for tobacco product and cannabis initiation, rather than the order of substance use onset. Older age, peer cannabis use, and delinquent behavior were risk factors for membership in the initiation classes, and membership in the early and higher-risk poly-product use initiation classes was associated with greater escalation of past 30-day and daily tobacco and cannabis use at the end of the high school. Thus, older students, relative to other students in the same grade, may be an important target group to consider when designing interventions to prevent tobacco product and cannabis use initiation. Peer cannabis use and delinquent behavior were also risk factors for membership in each of the initiation classes, and could serve as targets for intervention. Future research is needed to inform interventions that may prevent the initiation of co-use of tobacco and cannabis products during high school and potentially limit subsequent escalation of use following initiation. Also, establishing different patterns of tobacco and cannabis product use initiation and relating them to subsequent substance use behaviors may inform prevention efforts by clarifying risk substance use behavioral patterns for problematic substance use involvement.