Cross-substance patterns of alcohol, cigarette, and cannabis use initiation in Black and White adolescent girls

Abstract

Characterizing variations in the timing of alcohol, cigarette, and cannabis use onset both among and between Black and White youth can inform targeted prevention. The current study aimed to capture cross-substance initiation patterns in Black and White girls and characterize these patterns with respect to substance use related socioeconomic, neighborhood, family, community, and individual level factors. Data were drawn from interviews conducted at ages 8 through 17 in an urban sample of girls (n = 2172; 56.86% Black, 43.14% White). Discrete-time multiple event process survival mixture modeling was used to identify patterns (i.e., classes) representing timing of alcohol, cigarette, and cannabis use initiation, separately by race. Class characteristics were compared using multinomial logistic regression. Among both Black and White girls, four classes, including abstainer and cross-substance early onset classes, emerged. Two classes characterized by mid-adolescence onset (Black girls) and variation in onset by substance (White girls) were also observed. Class differences centered around cannabis for Black girls (e.g., preceding or following cigarette use) and alcohol for White girls (e.g., (in) consistency over time in greater likelihood of initiation relative to cigarette and cannabis use). Several factors distinguishing the classes were common across race (e.g., externalizing behaviors, friends’ cannabis use); some were specific to Black girls (e.g., intentions to smoke cigarettes) or White girls (e.g., primary caregiver problem drinking). Findings underscore the need to recognize a more complex picture than a high-risk/low-risk dichotomy for substance use initiation and to attend to nuanced differences in markers of risky onset pathways between Black and White girls.

1. Introduction

By 12th grade, 60.2% of girls have consumed alcohol, 18.8% have smoked cigarettes, and 42.9% have used cannabis (Miech et al., 2020). Whereas Black youth are less likely than White youth to use alcohol or cigarettes, they are more vulnerable to later health and social consequences of alcohol and tobacco use (Heron, 2019; Zapolski et al., 2014) and more likely to use cannabis (Miech et al., 2020). There is growing evidence as well that it is more common for Black youth than White youth to initiate cannabis use at the same age or prior to alcohol or tobacco use (Banks et al., 2017; Fairman et al., 2019; Sartor et al., 2013), counter to the typical, i.e., ‘gateway’ sequence of alcohol and tobacco before cannabis. Establishing a more complete picture of the heterogeneity within and between Black and White youth in the co-occurrence of alcohol, cigarette, and cannabis use onset – which substances, at what age, and in what order – may inform targeted prevention efforts. Capturing simultaneously the timing of initiation of all three substances, which have many common sources of risk (Ramo and Prochaska, 2012; Whitesell et al., 2013), can contribute to this goal.

A limited number of studies have applied a statistical approach integrating survival analysis and latent class analysis, discrete-time multiple event process survival mixture (MEPSUM) modeling (Dean et al., 2014), for this purpose. In addition to identifying patterns in the timing of initiation, it captures cumulative risk of use over time. Classes may be differentiated by a combination of the likelihood of ever initiating use, the timing of onset, and the number of substances/products used. In Dean et al.’s (2015) investigation using National Survey on Drug Use and Health data (n = 55,772; ages 12 and older), the onset of nine substances was modeled, revealing six classes: an overall low risk class, two early and two late onset classes (one each for ‘hard’ and ‘soft’ drugs), and a class characterized by cross-substance consistent risk. Richmond-Rakerd and colleagues’ study (Richmond-Rakerd et al., 2016) of alcohol, tobacco, and cannabis use onset based on Add Health data (n = 18,923; adolescent to adult years) revealed four classes, similarly distinguished by overall risk and stability of risk for initiation: a low risk class, two cross-substance early initiation classes with slightly different peaks in risk, and a late adolescent onset class. Likewise, Cho et al. (2021) examined initiation of cannabis and five tobacco products in a longitudinal study of high school students (n = 2272) and identified four classes: a low risk class, two cross-substance early onset classes (one high-risk), and a late onset class specific to cannabis and e-cigarette use.

All three studies included female and male participants and assessed differences between Black and White participants. Cho et al. (2021), who, notably, did not examine alcohol use initiation, found no distinctions between Black and White youth in class membership, but the other two investigations found that Black participants were less likely than their White counterparts to fall into the early initiation classes. However, Cho and colleagues identified several substance use related risk factors that differentiated early and late onset classes from the low risk class, including peer cannabis use and anxiety and for early onset specifically, socioeconomic status (SES) indicators and depression.

Current study aims

The current study aimed to identify cross-substance patterns of initiation of alcohol, cigarette, and cannabis use in Black and White girls via MEPSUM. It further aimed to characterize these patterns (i.e., classes) with respect to an even broader range of SES indicators as well as neighborhood, family, community, and individual level factors linked to adolescent substance use than was explored in Cho et al.’s study. Many are highly prevalent among Black youth (e.g., socioeconomic disadvantage [Shrider et al., 2021] and religious involvement [Wallace et al., 2007]) and thus may differentiate classes to a lesser degree among Black than White girls. Although each has been established as a risk or protective factor for substance use onset among adolescents, it is not yet known which are most closely associated with overall cross-substance patterns of initiation. We therefore hypothesized that we would observe greater risk factors (e.g., conduct problems) and lower protective factors (e.g., parental monitoring) in early onset class(es) and the reverse in overall low risk class(es), but did not hypothesize specific profiles of risk and protective factors for other initiation patterns. Analyses were stratified by race to identify initiation patterns specific to Black girls and specific to White girls, rather than assume common patterns and test for differences in representation of Black versus White girls in those patterns. We expected to observe at least one additional class other than overall low and overall high risk classes among both Black and White girls, with race differences in class structures broadly reflective of documented distinctions in prevalence and timing of alcohol and cannabis use onset (e.g., lower prevalence of alcohol use in Black compared to White girls).

2. Methods

2.1. Participants

The Pittsburgh Girls Study (PGS; N = 2450) is a longitudinal study of four female age cohorts ascertained at ages 5–8 from an urban community, oversampling low-resourced neighborhoods. Recruitment was conducted from 1999 to 2000, with 85.2% of eligible families completing the first wave of data collection. Girls and their primary caregivers (94% mothers) completed assessments annually. Sample retention was high: over 85% in the years that data used in the current analyses were collected (10 annual assessments per girl), when girls in each cohort were 8–17 years old. Details of sampling and study procedures are in prior publications (Hipwell et al., 2002; Keenan et al., 2010). Girls’ primary caregivers identified the girl’s race. Given our focus on differences between Black and (non-Hispanic) White girls, we did not use data from the small number of participants (n = 144) identified as another race or Hispanic/Latinx ethnicity. An additional 134 participants were excluded due to missing alcohol, cigarette, and cannabis data in all waves, resulting in a final analytic sample of 2172 (56.86% Black, 43.14% White). A larger proportion of White than Black participants were excluded (7.14% vs. 4.78%; χ²(1) = 5.75; p = 0.02), reflecting the overall PGS attrition pattern, but the mean number of missing waves of data for participants in the analytic sample did not differ by race (2.37 (SD = 1.38) for Black girls; 2.58 (SD = 1.55) for White girls).

2.2. Procedure

The University of Pittsburgh’s Human Research Protection Office approved the protocol. Trained research staff conducted separate face-to-face interviews with the girl and her primary caregiver in the family’s home after obtaining written informed consent from the primary caregiver and verbal assent from the girl. Families received compensation for their participation, with small increases in payment over time. Thus, caregivers received $40 in wave 1, increasing to $53 when the girl was age 17. Girls received $20 in wave 1, increasing to $45 at age 17.

2.3. Assessment

2.3.1. Alcohol, cigarette, and cannabis use

The Nicotine, Alcohol, and Drug Use questionnaire (Pandina et al., 1984), a self-report measure of past year frequency and usual quantity of consumption, administered starting at age 11, was the primary measure of substance use. Age at initiation represented the participant’s first report of any past year use, including a sip for alcohol and ‘a few drags’ for cigarettes and cannabis. If a participant endorsed cigarette smoking on the Nicotine Dependence Questionnaire (Fagerström, 1978) at age 11 or younger, age at first cigarette was coded as 11.

2.3.2. SES indicators and neighborhood, family, community, and individual level factors

Measurement and operationalization of SES indicators and neighborhood, family, community, and individual level factors are described in Table 1. SES was indexed by household receipt of public assistance, single parent-headed household, and primary caregiver educational attainment. Neighborhood measures captured safety, physical disorder, and cohesion. Family and community level factors consisted of primary caregiver problem drinking, parental monitoring, trauma exposure (within or outside the family environment), four indicators of religious involvement, and substance-specific indicators of friends’ use of alcohol, cigarettes, and cannabis. Individual level factors consisted of early puberty status, depression, anxiety, conduct problems, self-control, sensation seeking, and substance-specific indicators of prior year intentions to use alcohol, cigarettes, and cannabis.

Table 1.

Measurement of socioeconomic status indicators and neighborhood, community, family, and individual level factors.

Construct	Measure(s)	Scale/operationalization	Reporter	Alpha across waves
Socioeconomic status indicators and neighborhood factors
Low primary caregiver education level	Interview question	<12 years vs. >12 years	P	-
Household receipt of public assistance	Interview question	Yes/No	P	-
Single parent headed household	Interview question	Yes/No	P	-
Low neighborhood safety	Your Neighborhood (17 items; Loeber et al., 1998)	1 = ‘Not a problem’ to 3 = ‘Big problem’	P	0.95–0.96
Community cohesion	10 items from Community Survey (Gorman-Smith et al., 2000)	1 = ‘Strongly disagree’ to 5 = ‘Strongly agree’	P	0.91–0.92
Neighborhood physical disorder	5 items from Interviewer Impressions of the Neighborhood (Wei et al., 2005)	Dichotomously coded (e.g., graffiti) and summed	I	0.69–0.73
Family and community level factors
Primary caregiver problem drinking	10-item AUDIT (Babor et al., 1992)	0=‘Never’/’1 or 2’/’No’ to 4=‘4 or more times a week’/’10 or more’/’Daily or almost daily’/’Yes’	P	0.72–0.77
Low parental monitoring	4 items from Supervision and Involvement Scale (Loeber et al., 1998)	1 = ‘Almost always’ to 3 = ‘Almost never’	C	0.61–0.74
Frequency of religious service attendance	Interview question	0=‘Never’ to 3=‘More than once a week’	C	-
Frequency of participation in other religious activities	Interview question	0=‘Never’ to 5=‘Just about every day’	C	-
Importance of religion	Interview question	0=‘Not important’, to 2 = ‘Very important’	C	-
Frequency pray	Interview question	0 = ‘Never’ to 3 = ‘About once a day’	C	-
Friends’ alcohol use	Interview question	0=‘None’ to 3=‘All’	C	-
Friends’ cigarette use	Interview question	0=‘None’ to 3=‘All’	C	-
Friends’ cannabis use	Interview question	0=‘None’ to 3=‘All’	C	-
Family and community level factors
Trauma exposure	Child PTSD Symptom Scale (Foa et al., 2001); Conflict Tactics Scale - Romantic Partner (Straus et al., 1996); Child Police Contact, Health Questionnaire (Loeber et al., 1998); Abuse Questionnaire, Victimization Under the Influence of Alcohol or Drugs (Pittsburgh Girls Study)	Yes (endorsement of any event)/no (no event endorsed); note: Items assess trauma exposure within and outside the family environment	P, C	-
Individual level factors
Early puberty	1 item from Pubertal Development Scale (Petersen et al., 1988)	Yes (Menarche <age 11)/ No (Menarche >age 12)	C	-
Depression	11 items from Adolescent Symptom Inventory-4th edition ( Gadow and Sprafkin, 1997)	0=‘Never’ to 3=‘Very often’	C	0.74–0.84
Anxiety	29 items from Screen for Child Anxiety and Related Disorders ( Birmaher et al., 1997)	0=‘Not true or hardly ever true’ to 2=‘Very true’	C	0.90–0.92
Conduct problems	15 items from Adolescent Symptom Inventory-4th edition ( Gadow and Sprafkin, 1997)	0=‘Never’ to 3=‘Very often’	C	0.64–0.75
Self-control	8 items from Social Skills Rating Scale (Gresham and Elliott, 1990)	0=‘0ften’ to 2=‘Never’	C	0.74–0.80
Sensation seeking	4 items from Child and Adolescent Dispositions Scale (Lahey et al., 2010)	0=‘Not at all’ to 3=‘Very much or very often’	C	0.63–0.77
Intentions to use alcohol	Interview question	1=‘Definitely not’ to 4=‘Definitely yes’	C	-
Intentions to smoke cigarettes	Interview question	1=‘Definitely not’ to 4=‘Definitely yes’	C	-
Intentions to use cannabis	Interview question	1=‘Definitely not’ to 4=‘Definitely yes’	C	-

Note P = primary caregiver; C=Child; I=Interviewer. All indicators were coded according to responses at the age at which the participant reported first using alcohol, cigarettes, or cannabis, with the exception of intentions to use, which was coded according the prior year response, given its future orientation. For girls who never used any substances, responses at last interview were used.

As our modeling approach does not accommodate time-varying covariates, indicators were coded according to responses at the age at which initiation status was established, thus capturing concurrent risk and protective factors. For girls who never used any substances (n = 734), this was age at last interview. For all others, it was age that the participant first reported using alcohol, cigarettes, or cannabis (e.g., depression symptoms at age 14 for a participant whose first use of any substance occurred at age 14). The only exception was intentions to use, which was coded according to the prior year response, given its future orientation. Means/prevalences are reported by race in Table 2.

Table 2.

Sample characteristics by race.

	Black girls (n = 1235)	White girls (n = 937)
Socioeconomic status indicators and neighborhood factors
Household receipt of public assistance	53.77%	17.40%
Single parent-headed household	63.29%	23.90%
Primary caregiver education <12 years	16.35%	10.67%
Low neighborhood safety: M (SD)	24.18 (8.44)	20.56 (5.70)
Community cohesion: M (SD)	30.18 (8.70)	37.03 (8.18)
Neighborhood physical disorder: M (SD)	1.27 (1.34)	0.49 (0.86)
Family and community level factors
Primary caregiver AUDIT score: M (SD)	2.33(3.24)	2.20 (2.69)
Low parental monitoring: M (SD)	5.13(1.51)	4.61 (1.00)
History of trauma exposure(s)	20.00%	9.18%
Religious involvement: M (SD)
Frequency attend religious services	1.34(0.97)	1.22 (0.89)
Frequency participate in other religious	1.13(1.42)	1.13 (1.35)
activities
Importance of religion	1.38 (0.67)	1.10 (0.76)
Frequency pray	1.89 (1.01)	1.57 (1.10)
Friends’ substance use: M (SD)
Alcohol	1.16 (1.06)	1.07 (1.04)
Cigarettes	0.91 (0.95)	0.91 (0.95)
Cannabis	1.18 (1.07)	0.71 (0.93)
Individual level factors
Early puberty	36.84%	19.96%
Depression: M (SD)	7.49 (5.13)	7.34 (4.93)
Anxiety: M (SD)	14.58 (9.64)	13.91 (9.08)
Conduct problems: M (SD)	1.69 (2.25)	1.00 (1.55)
Self-control: M (SD)	11.77 (3.46)	12.87 (3.51)
Sensation seeking: M (SD)	5.04 (2.62)	5.39 (2.60)
Prior year intentions to use: M (SD)
Alcohol	1.30 (0.57)	1.25 (0.51)
Cigarettes	1.10 (0.35)	1.07(0.30)
Cannabi	1.25 (0.52)	1.08(0.29)

Note M = mean; SD = standard deviation; values reflect status at age at first use of alcohol, cigarettes, or cannabis (age 17 for never users).

2.4. Analytic approach

2.4.1. Derivation of classes

Analyses were stratified by race, with modeling conducted separately by subsample (i.e., Black girls and White girls). All analyses used full information maximum likelihood estimation in Mplus Version 8 (Muthén and Muthén, 2017), which accounts for missing data attributable to non-response and/or dropout in the longitudinal sample. MEP-SUM was applied to each subsample to identify cross-substance patterns in the timing of alcohol, cigarette, and cannabis use initiation from ages 11 to 17. Under this analytic framework, the multivariate hazard distribution is approximated using a finite mixture and components of the mixture (latent classes) reflecting discrete different patterns of event occurrence over time. We followed Richmond-Rakerd and colleagues’ approach (Richmond-Rakerd et al., 2016), approximating unstructured hazards with a logit link function, which can better approximate the hazard if it is nonparametric and the shape changes over time. Model fitting was conducted for up to five classes, in the interest of maximizing generalizability of findings from a sample of this size for identifying patterns of initiation of three substances.

2.4.2. Characterization of classes

Within each subsample, classes were compared on SES indicators and neighborhood, family, community, and individual level factors using multinomial regression analyses. Analyses were conducted in two stages to derive the most parsimonious models. First, each variable was entered individually into the regression model. Next, variables that were statistically significant after applying the False Discovery Rate for dependent tests were retained and entered simultaneously into a single model. The statistical significance of each variable was based on the p-value that was adjusted by False Discovery Rate to control the overall Type 1 error rate across numerous dependent tests of significance (Benjamini and Hochberg, 1995).

3. Results

3.1. Class identification

Results of model fit for two through five classes are shown in Table 3. Four-class solutions were selected for both subsamples based on model fit, entropy, and clear interpretability (e.g., evidence for a qualitatively distinct class with increasing classes, avoiding small classes [<5% of participants; Nylund-Gibson and Choi, 2018]). Figs. 1 and 2 show survival curves for alcohol, cigarette, and cannabis use initiation by class for Black girls and White girls, respectively. Class intercepts and slope parameters are reported in Supplemental Table 1. Classes were numbered by onset age and overall risk from latest/lowest to earliest/highest. Classes for Black girls were characterized as: (1) Relative Abstainers (58.46%), (2) Mid-adolescence Cross-substance - Moderate Risk (26.07%), (3) Mid-adolescence Cross-substance - High Risk (5.02%), and (4) Pre-adolescence Cross-substance (10.45%), with cigarettes and cannabis following the same trajectory in Classes 1 and 3 and cannabis preceding cigarettes in Classes 2 and 4. Classes for White girls were characterized as: (1) Relative Abstainers (59.02%), (2) Early-adolescence Alcohol, Late-adolescence Cigarettes and Cannabis (16.86%), (3) Pre-adolescence Alcohol, Mid-Adolescence Cigarettes and Cannabis (16.44%), and (4) Pre-adolescence Alcohol and Cigarettes, Mid-adolescence Cannabis (7.68%).

Table 3.

Fit indices for 2-class to 5-class solutions for Black and White girls.

	Black girls				White girls
Model fit/adequacy	2-class	3-class	4-class	5-class	2-class	3-class	4-class	5-class
AIC	9641.000	9539.606	9495.370	9467.606	7611.815	7514.204	7449.630	7427.275
BIC	9738.257	9688.052	9695.004	9718.428	7703.826	7654.641	7638.495	7664.570
Sample size-adjusted BIC	9677.905	9595.935	9571.123	9562.783	7643.484	7562.540	7514.634	7508.950
Entropy	0.768	0.754	0.780	0.801	0.780	0.720	0.743	0.781
Proportion of sample in smallest class (%)	36.275	14.170	5.020	3.205	31.590	20.671	7.684	3.417
Lo-Mendell-Rubin adjusted LRT (p-value)		0.001	0.012	0.207		0.002	1.000	0.005
Parametric bootstrapped LRT (p-value)		0.000	0.000	0.000		0.000	0.000	0.000

Note AIC = Akaike’s Information Criterion; BIC=Bayesian Information Criterion. * p-values <0.05 indicate improvement in fit from modeling additional class.

Fig. 1.

Cross-substance use initiation classes: Black girls.

Fig. 2.

Cross-substance use initiation classes: White girls.

3.2. Differences between Black and White girls in classes paired by approximate onset age

Classes derived for Black girls were paired with classes derived for White girls by approximate onset age (Class 1: consistently low likelihood; Class 2: 14–16; Class 3: 13–14; Class 4: 12–13) to assess for (a) statistical support of separate class solutions and (b) differences in likelihood of initiation by age 17 (prevalence) and age at onset of each substance in Black vs. White girls who fell into classes with roughly equivalent onset ages. Tests of equivalence of intercept and slope parameters in paired classes, using a multi-group approach, revealed statistically significant different class structures, supporting stratification by race. (See Supplemental Table 2.) Comparisons of prevalence and mean age at onset of alcohol, cigarettes, and cannabis in paired classes revealed several distinctions. (See Table 4.) For alcohol, prevalence was lower among Black girls in Classes 1 and 2; for cigarettes, prevalence was lower among Black girls in Classes 2 and 4 but higher in Class 3; and for cannabis, prevalence was higher among Black girls in Classes 1, 2, and 3. Mean age at onset differed by race only for cannabis, with earlier onset for Black girls in Classes 2, 3, and 4.

Table 4.

Prevalence of use by age 17 and mean age at first use of alcohol, cigarettes, and cannabis by class from models conducted separately for Black and White girls.

Class	Class description	Race	Proportion	Alcohol		Cigarettes		Cannabis
				Prevalence	Age at 1st use M (SD)	Prevalence	Age at 1st use M (SD)	Prevalence	Age at 1st use M (SD)
1	Relative Abstainers	Black	58.46%	27.40%**	14.47 (2.05)	3.46%	14.40 (2.80)	5.68%**	17.00 (0.00) 15.00 (- —)a
	Relative Abstainers	White	59.02%	47.73%**	14.02 (2.09)	3.25%	15.28 (1.84)	0.18%**	15.58(1.15)**
	Mid-adolescence Cross-substance - Moderate Risk	Black	26.07%	82.61%**	14.74 (1.87)	43.17%**	15.85 (1.36)	92.55%**	15.58 (1.15)**
2	Early-adolescence Alcohol, Late-adolescence Cigarettes and Cannabis	White	16.86%	93.67%**	14.28 (2.06)	72.15%**	16.01 (1.39)	64.56%**	16.73 (0.79)**
	Mid-adolescence Cross-substance - High Risk	Black	5.02%	100.00%	13.48 (1.47)	100.00%	14.26 (1.09)	100.00%	14.47 (0.67)**
3	Pre-adolescence Alcohol, Mid-Adolescence Cigarettes and Cannabis	White	16.44%	100.00%	13.32 (1.50)	78.57%**	14.64 (1.04)	83.77%**	14.99 (0.71)**
	Pre-adolescence Cross-substance	Black	10.45%	97.67%	12.40 (1.24)	75.97%**	12.86 (1.70)	89.15%	13.03 (1.12)**
4	Pre-adolescence Alcohol and Cigarettes, Mid-adolescence Cannabis	White	7.68%	94.44%	12.07 (1.04)	100.00%	12.58 (0.96)	83.33%	13.70 (0.77)**
Across classes		Black	-	52.84%**	14.08 (2.00)**	26.23%**	14.53 (2.02)	41.78%**	14.99 (1.58)
		White	-	67.70%**	13.70 (1.98)**	34.69%**	14.70 (1.75)	31.20%**	15.33 (1.36)

Note M = mean; SD = standard deviation. Comparisons across race addressed potential differences in likelihood of initiation by age 17 (prevalence) and age at onset of each substance in Black vs. White girls who fell into classes with roughly equivalent onset ages.

^**difference across race statistically significant at p < 0.001;

^aSD cannot be calculated (n = 1).

3.3. Class distinctions by SES indicators and neighborhood, family, community, and individual level factors

Results of multinomial regression analyses comparing class characteristics, using Class 1 as the reference category, are reported in Table 5. For both Black and White girls, conduct problems and sensation seeking were associated with elevated odds of membership in Classes 2, 3, and 4, and self-control with reduced odds of membership in two of the three classes. Friends’ cannabis use was also positively associated with class membership across race: Classes 2 and 4 for Black girls and Class 3 for White girls, as was depression: Class 4 for Black girls and all three classes for White girls. For Black girls only, low parental monitoring was associated with reduced odds of Class 3 membership, prior year intentions to smoke cigarettes with elevated odds. For White girls only, primary caregiver problem drinking was associated with elevated odds of Class 4 membership and prior year intentions to use cannabis with Class 2 membership.

Table 5.

Results of multinomial regression analyses comparing class characteristics from models conducted separately for Black and White girls.

	Black girls			White girls
	2. Mid-adolescence Cross-substance Moderate Risk	3. Mid-adolescence Cross-substance High Risk	4. Preadolescence Cross-substance	2. Early-adolescence Alcohol, Late-adolescence Cigarettes and Cannabis	3. Pre-adolescence Alcohol, Mid-Adolescence Cigarettes and Cannabis	4. Pre-adolescence Alcohol and Cigarettes, Midadolescence Cannabis
	n = 322	n = 62	n = 129	n = 158	n = 154	n = 72
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Family and community level factors Primary caregiver problem drinking	Not included in models			1.41 (0.44–4.54)	1.40 (0.34–5.71)	3.89 (1.13–13.41)
Low parental monitoring Friends’ cannabis use	0.97 (0.87–1.08) 1.28 (1.05–1.56)	0.74 (0.61–0.91) 0.90 (0.66–1.22)	0.91 (0.78–1.06) 0.72 (0.54–0.96)	1.19 (0.90–1.59)	Non-significant across models 1.40 (1.02–1.93)	0.75 (0.46–1.21)
Individual level factors Depression	1.03 (1.00–1/07)	1.06 (1.00–1.13)	1.10 (1.05–1.16)	1.06 (1.01–1.10)	1.09 (1.04–1.14)	1.10 (1.04–1.16)
Conduct problems	1.41 (1.27–1.57)	1.72 (1.49–1.99)	1.72 (1.52–1.94)	1.30 (1.09–1.S6)	1.43 1.20–1.70)	1.S2 (1.24–1.86)
Self-control Sensation seeking Prior year	0.93 (0.88–0.98) 1.10 (1.03–1.17)	0.95 (0.86–1.04) 1.22 (1.10–1.34)	0.88 (0.82–0.96) 1.14 (1.04–1.26)	0.97 (0.91–1.04) 1.11 (1.01–1.21)	0.91 (0.85–0.98) 1.18 (1/08–1.30)	0.86 (0.78–0.96) 1.21 (1.07–1.36)
intentions to smoke cigarettes	1.39 (0.88–2.19)	1.98 (1.05–3.71)	1.74 (0.95–3.20)	Non-significant across models
Prior year intentions to use cannabis	Not included in models			2.75 (1.40-5.39)	1.23 (0.54–2.79)	1.40 (0.48–4.11)

Note Reference group = Class 1 (n = 722-Black girls; n = 553-White girls); OR = Odds Ratio; CI=Confidence Intervals; bold indicates statistically significant at p < 0.05. Only variables statistically significant in omnibus bivariate tests after dependent false discovery rate adjustment were included in models. ORs are reported for all classes if the variable was statistically significant for any class.

4. Discussion

In a rare effort to capture within-racial group heterogeneity in substance use onset among Black and White youth, focusing exclusively on girls, the current study applied a novel analytic approach to identify patterns in the timing of alcohol, cigarette, and cannabis use initiation in the pre- to late-adolescent years. Our hypothesis that we would identify at least one additional class other than the overall low risk and cross-substance early initiators that have been identified across prior studies applying MEPSUM (Cho et al., 2021; Dean et al., 2015; Richmond-Rakerd et al., 2016) was supported. Two additional classes emerged among Black girls and White girls. Consistent with the study covering roughly the same age range (Cho et al., 2021), which also revealed two additional classes (early cannabis and polytobacco initiators as well as late cannabis and e-cigarette initiators), classes were distinguished by variation in both timing and overall likelihood of initiating a given substance by late adolescence. Both common and racial group-specific family, community, and individual level factors distinguished classes.

4.1. Classes of initiation patterns

Key distinctions among classes for Black girls were whether cigarette and cannabis use initiation patterns mapped onto each other (Classes 1 and 3) or diverged (Classes 2 and 4) and whether initiation of cannabis use preceded (Classes 2 and 4) or followed (Classes 1 and 3) cigarette use. Notably, initiation patterns involving cannabis use preceding cigarette use were observed only among Black girls, in keeping with prior work demonstrating greater likelihood of initiating use of cannabis before licit substances (tobacco and/or alcohol) among Black compared to White youth (Banks et al., 2017; Fairman et al., 2019; Sartor et al., 2013). By contrast, key distinctions among classes for White girls were related to alcohol, specifically whether likelihood of alcohol use initiation was consistently greater than likelihood of cigarette or cannabis use initiation (Classes 1 and 2 vs. 3 and 4) and whether it preceded (Classes 1, 2, and 3) or occurred concurrently with cigarette and cannabis use initiation (Class 4). Comparisons across race in classes paired on approximate onset age revealed higher prevalence of cannabis use by age 17 among Black girls in all but the earliest onset class (Class 4) and higher prevalence of alcohol use by age 17 among White girls in the relative abstainer and mid-adolescent onset classes (1 and 2), consistent with documented differences by race in prevalence of adolescent cannabis and alcohol use (Miech et al., 2020). These findings suggest substantive differences across race both in substance use onset pathways and in the extent to which use of alcohol or cannabis use better distinguishes these pathways. For example, whereas alcohol use by mid-adolescence is likely to be accompanied by cigarette and cannabis use among Black girls, the co-occurrence of cigarette and cannabis use among White girls who have used alcohol by mid-adolescence is highly varied.

4.2. Distinctions by class in family, community, and individual level factors

Several family, community, and individual level factors examined in the present study distinguished the higher risk classes from the relative abstainers, some common across race, others specific to Black girls or White girls. Common differentiating factors included externalizing traits and behaviors consistently linked to early alcohol, cigarette, and cannabis use: low self-control (King et al., 2011; Koning et al., 2014), sensation seeking (Adachi-Mejia et al., 2012; Scheier and Griffin, 2021; Stautz and Cooper, 2013), and conduct problems (Erskine et al., 2016; Groenman et al., 2017) as well as depression. Evidence for links between depression and early substance use, specifically, independent of externalizing problems, is also strong (Hussong et al., 2017). It includes findings from Cho and colleagues’ study (Cho et al., 2021), in which depression was associated with membership in early and high-risk initiation classes. Notably, in the present study, depression was associated with all initiation classes for White girls but only the earliest initiation class for Black girls, suggesting that depression may be a global marker of risk for adolescent substance use initiation for White girls but more specific to very early onset for Black girls. Friends’ cannabis use, another established risk factor for early onset substance use (D’Amico and McCarthy, 2006) that distinguished classes in Cho et al.’s study (Cho et al., 2021), was also associated with class membership in both Black and White girls. The specific classes differed across race, but taken together, the results suggest that friends’ cannabis use may be most closely linked to initiation during mid-adolescence (Class 2 for Black girls, Class 3 for White girls).

The two significant risk factors specific to Black girls, prior year intentions to smoke cigarettes and low parental monitoring, were associated with membership in Class 3. The elevated odds associated with prior year intentions to smoke is consistent with Class 3 having the highest prevalence of cigarette use by age 17. The reduced odds associated with low parental monitoring suggests a high degree of parental monitoring in response to (rather than preceding) the acute onset of substance use among girls following this initiation pattern. Among White girls only, primary caregiver problem drinking, a robust risk factor for early onset alcohol use (McGue et al., 2001), was associated with the class with the youngest mean age (12.07) at first alcohol use (Class 4) and prior year intentions to use cannabis were associated with the class characterized by the latest mean age (16.73) at cannabis use initiation (Class 2). Its specificity to White girls likely reflects the somewhat less normative use of cannabis relative to Black youth (Miech et al., 2020), and thus, a later age at development of intentions to use.

The lack of distinctions by SES indicators that differentiated classes in Cho et al.’s (2021) study merits comment as well. Cho et al. found associations of single parent household and high parental education with lower likelihood of membership in the early cannabis use risk class. However, they did not include alcohol in their models and whereas indicators of low SES have fairly consistently been associated with increased risk for adolescent substance use (Assari et al., 2020; Leventhal et al., 2015), an inverse association with alcohol has also been found (Blum et al., 2000). The non-significant associations of SES indicators with initiation patterns in the current study may be attributable in part to inconsistencies in the directions of associations with SES indicators for alcohol versus cigarette and cannabis use.

4.3. Limitations

Results should be interpreted with certain limitations in mind. First, participants were recruited from an urban area, oversampling low-resourced neighborhoods, so findings may not generalize to Black and White girls from non-urban or higher income populations. Second, we operationalized initiation as first use at any frequency or quantity; findings may differ under other definitions (e.g., first full cigarette). Third, since substance use assessments started at age 11 (alcohol, cannabis) or were recoded (n = 10) to age 11 (cigarettes), onset age may have been underestimated for a small proportion of participants. Fourth, participants whose first use of a substance occurred in a year that they did not provide study data would have an inaccurate recorded age at first use of that substance. Fifth, as some individuals do not try alcohol, cigarettes, or cannabis until adulthood, the current investigation should be viewed as capturing initiation patterns among girls from pre- to late adolescence.

4.4. Conclusions

The present investigation took a novel approach to identifying pathways of substance use onset among Black and White girls, revealing distinctions across race in initiation patterns as well as nuanced differences between Black and White girls in the prevalence of certain patterns and how they are distinguished. Findings underscore the need to recognize a more complex picture than a dichotomy of ‘high-risk’ and ‘low-risk.’ They further suggest that markers of risky substance use pathways may differ by race, with alcohol use being a stronger predictor for Black girls and cannabis use for White girls.

4.5. Future directions

In addition to applying this approach to identify substance use initiation patterns in boys, youth from other racial/ethnic minority groups (e.g., Asian, Latinx), and youth living in rural areas, future directions include characterizing initiation patterns by factors of particular relevance to Black youth, such as racial discrimination. Exploration of protective factors, such as racial socialization and racial identity, would be especially valuable for translation of this work into prevention efforts. Additionally, assessment of modes of tobacco and cannabis use, including vaping, would provide a finer-grain characterization of variations in timing and sequences of substance use initiation, potentially yielding more specific intervention targets.