Tobacco, cannabis, alcohol, and polysubstance use disparities among sexual identity groups of US young adult women and men Authors: Erin A. Vogel, Katelyn F. Romm, Carla J. Berg Published: 2024 ABSTRACT Highlights•Young adults used tobacco, cannabis, and alcohol in several discernible patterns.•Bisexual women were at risk for tobacco, cannabis, and alcohol polysubstance use.•Bisexual and gay men were at lower risk for polysubstance use.•Substance use interventions for bisexual women should consider co-use patterns. Introduction Young adulthood is a developmental period for forming behavioral patterns with long-term health implications (Harris, 2010), such as initiating or escalating substance use (Richmond-Rakerd et al., 2017), and coincides with increased independence and social pressures that may be conducive to substance use (Stewart et al., 2023). Young adults (YAs) aged 18–25 in the United States (US) have notably high prevalence of past-month (current) use of alcohol (50.1 %), nicotine or tobacco (24.7 %), and cannabis (24.1 %) (Substance Abuse and Mental Health Services Administration, 2023), conferring risks such as reduced engagement in school or work (Amato et al., 2021, Duckworth et al., 2023) and the development of substance use disorders (Substance Abuse and Mental Health Services Administration, 2023). Compared to single-substance use, polysubstance use of tobacco, cannabis, and alcohol in young adulthood may have additive risks, such as more alcohol-related and cannabis-related consequences (e.g., blackouts, risky sexual activity, verbal altercations, financial difficulties, legal problems) (Florimbio et al., 2023, Hayaki et al., 2016) and lower odds of tobacco cessation (Vogel, Rubinstein, Prochaska, & Ramo, 2018). Polysubstance use of tobacco, cannabis, and alcohol may elevate young adults’ risk for substance-related problems. Young adulthood may be a particularly vulnerable time for sexual minority (SM) individuals, who may be developing, defining, and disclosing their sexual identity (Kaestle, 2019). Minority Stress Theory posits that SM individuals face unique minority stressors (e.g., discrimination, internalized stigma) related to their sexual identity, leading to greater mental health problems and associated use of substances to cope (Hatzenbuehler, 2009, Meyer, 2003). SMYAs indeed display disproportionately higher rates of tobacco, cannabis, and alcohol use (The Trevor Project, 2020, Wheldon et al., 2018), use disorders (Goldberg et al., 2013), and polysubstance use (Lee et al., 2023), compared to heterosexual YAs. Prior research has examined common patterns of polysubstance use in YAs, such as low-level use (i.e., low probabilities of tobacco, cannabis, and alcohol use), tobacco and alcohol co-use, and polysubstance use (i.e., co-use of alcohol, cannabis, and illicit substances) (de Jonge et al., 2022). However, differences in patterns of tobacco, cannabis, and alcohol use by SM identity and gender remain largely unexplored. Prior work has largely aggregated across groups of YAs representing distinct sexual identities (e.g., bisexual and gay/lesbian) when examining polysubstance use disparities (Goodwin et al., 2022); however, disparities may vary. For example, bisexual women may be at greater risk relative to other populations (e.g., heterosexual women, lesbian women, SM men) due to a combination of gender-based discrimination (McCabe et al., 2010) and rejection from both heterosexual and SM communities (Callis, 2013). Research suggests greater disparities among SMYA women than among SMYA men (Goodwin et al., 2022), greater prevalence of tobacco, cannabis, and alcohol use among bisexual (versus lesbian or heterosexual) women (Romm et al., 2022, Schuler and Collins, 2020), and greater e-cigarette—but not combustible tobacco—use prevalence among bisexual versus lesbian women (Romm, Huebner, et al., 2022). Tobacco, cannabis, and alcohol are the most commonly used substances in the US (Substance Abuse and Mental Health Services Administration, 2023) and may increase risk for leading causes of preventable death such as heart disease, cancer, respiratory disease, stroke, and unintentional injury (Yoon et al., 2014). Sexual identity disparities in tobacco, cannabis, and alcohol use first emerge in adolescence and young adulthood (Rosario et al., 2014, Schuler et al., 2018), presenting opportunity to intervene before long-term health consequences are incurred. Further research is needed to identify differential patterns of co-occurring combustible tobacco, e-cigarette, cannabis, and alcohol use among specific groups of SMYA women and men, to inform tailored interventions and messaging campaigns aimed at preventing problematic patterns of polysubstance use during this vulnerable period. The current study examined: 1) classes of polysubstance use (i.e., combustible tobacco, e-cigarette, cannabis, alcohol) via latent class analysis (LCA) and 2) associations among sexual identity (distinguishing bisexual, gay/lesbian, and heterosexual) and resulting classes among YA women and men, separately. Methods Participants and Procedures This study analyzed survey data from a 2-year longitudinal study of YAs (aged 18–34) designed to examine correlates of cigarette and e-cigarette use among individuals recruited from 6 metropolitan statistical areas (MSAs: Atlanta, Boston, Minneapolis, Oklahoma City, San Diego, Seattle) with varied tobacco legislative contexts (Public Health Law Center, 2020). This study, detailed elsewhere (Berg et al., 2021), was approved by the George Washington University Institutional Review Board. Participants were recruited in Fall 2018 and surveyed biannually until Fall 2020 (total of 5 waves). To recruit participants, ads posted on Facebook and Reddit targeted individuals by using indicators reflecting those eligible (i.e., ages 18–34, residing in one of the 6 MSAs, English-speaking). After clicking on an ad, individuals were directed to a webpage with a study description, consent form, and eligibility screener, then eligible individuals completed the online baseline survey via Alchemer. Upon completion, participants were asked to confirm their participation in the study by clicking a “confirm” link in an email sent a week later. Purposive, quota-based sampling ensured the sample represented sufficient numbers of cigarette and e-cigarette users (roughly one-third each), roughly equal numbers of women and men, and 40 % racial or ethnic minority individuals (subgroup enrollment was capped by MSA). Of the 10,433 individuals who clicked on ads, 9,847 consented, of whom 2,751 (27.9 %) were excluded due to: (a) ineligibility (n = 1,472) and/or (b) their subgroup target being met (n = 1,279). Among the remaining 7,096 individuals, 48.8 % (n = 3,460) provided complete data, and 86.9 % (n = 3,006) confirmed participation (Berg et al., 2021). The current analyses involve the use of baseline sociodemographic data and Fall 2020 data (n = 2,476, 82.4 %), which assessed past-month use of tobacco products, cannabis, and alcohol. Among participants with baseline data, a greater proportion of those who did not complete vs. completed the Fall 2020 survey identified as male (47.9 % vs. 42.6 %, p = 0.030) and reported past-month combustible tobacco use (50.8 % vs. 32.1 %, p < 0.001), e-cigarette use (55.7 % vs. 33.8 %, p < 0.001), and cannabis use (51.2 % vs. 36.7 %, p < 0.001). Those who did not complete vs. completed the Fall 2020 survey were also younger (M = 24.06 vs. M = 24.66, p = 0.008). Fall 2020 completion was not associated with participant race, ethnicity, sexual identity, or past-month alcohol use. Measures Stratification variable: gender Participants were asked “What is your gender: male; female; or other?” Participants who selected “other” (n = 69) were excluded from primary analyses due to small sample size but explored in descriptive analyses. Primary correlate of interest: sexual identity Participants were asked, “Do you consider yourself: heterosexual or straight; gay or lesbian; or bisexual?” These responses were recoded into dummy variables. Participants could also respond “other” (n = 43) or “prefer not to answer” (n = 37); participants who selected these options were excluded from primary analyses due to small sample sizes but were explored in descriptive analyses. Primary outcome: substance use Participants reported the number of days they used each of the following products, separately, in the past 30 days (0–30 days): combustible tobacco products (i.e., cigarettes, cigars, hookah), e-cigarettes, cannabis, and alcohol. Those who reported ≥ 1 day of use were classified as a current user of each product; those who reported 0 days of use were classified as non-users of each product. Covariates Sociodemographic covariates included age (continuous variable), race (categorized as White, Black, Asian, or another race due to group sizes), ethnicity (Hispanic vs. non-Hispanic), and education (< Bachelor’s degree vs. ≥ Bachelor’s degree). We also controlled for baseline depressive symptoms, which were assessed with the Patient Health Questionnaire – 2 item (PHQ-2), assessing symptoms in the past two weeks (0 = not at all to 3 = nearly every day). Participants with scores of > 3 were classified as having ≥ moderate depressive symptoms (α = 0.88) (Kroenke et al., 2003). Data analysis Descriptive statistics characterized the sample and screened for outliers. Latent class analyses (LCAs) of polysubstance use were fit in Mplus 8.8 and included 4 binary indicators (i.e., past-month combustible tobacco product, e-cigarette, cannabis, alcohol use). The best-fitting latent class solution was determined based on: 1) model interpretability; 2) lower Akaike information criterion (AIC), Bayesian information criterion (BIC), and sample-size adjusted Bayesian information criterion (aBIC) values, which indicate a better balance between model fit and parsimony; 3) higher entropy values, which indicate higher classification utility; and 4) a significant Lo-Mendel-Rubin (LMR) Adjusted Test (LRT), which suggests that the model is a significant improvement from a model with one fewer class. Model identification was confirmed with multiple sets of random starting values (500 for initial stage and 100 for final stage), with replication of the best log likelihood value. Next, bivariate analyses (i.e., ANOVAs, Chi-square tests) assessed associations among sexual identity and polysubstance use for women and men, separately. Finally, multinomial logistic regressions examined associations among sexual identity and polysubstance use class for women and men, separately, including the following covariates: age, race, ethnicity, education, and depressive symptoms. Prior to conducting regressions, we tested for multicollinearity of covariates and primary predictors. All tolerance and Variance Inflation Factor (VIF) values were > 0.25 (range = 0.92–0.97) and < 4 (range = 1.03–1.09), respectively, indicating that multicollinearity was not present (Kim, 2019). Missing data were minimal (<1% for cannabis use only) and were handled via full information maximum likelihood in Mplus when conducting LCA. Results Participant characteristics Participants were 2,343 YAs aged 24.69 on average (SD = 4.70). Participants most commonly reported a heterosexual identity (72.6 %), followed by a bisexual identity (18.0 %), and a gay or lesbian identity (9.4 %). Regarding race and ethnicity, 70.8 % identified as White, 5.5 % Black, 13.3 % Asian, 10.4 % another race, and 11.1 % Hispanic. Most participants had ≥ a Bachelor’s degree (75.9 %) and 22.4 % reported ≥ moderate depressive symptoms. Among women (57.4 %; n = 1,345), 24.8 % (n = 333) identified as bisexual, and 6.1 % (n = 82) as lesbian. On average, women were 24.65 years old (SD = 4.68), 7.1 % Black, 11.9 % Asian, 10.3 % another race, 9.9 % Hispanic; 77.2 % had ≥ a Bachelor’s degree, and 20.1 % reported ≥ moderate depressive symptoms. Past 30-day substance use rates were 20.1 % for combustible tobacco, 22.5 % for e-cigarettes, 42.8 % for cannabis, and 72.9 % for alcohol (Table 1, left panel).Table 1Bivariate analyses examining differences between participants identifying as heterosexual vs. gay/lesbian vs. bisexual, N = 2,343.WomenMenVariablesTotalN = 1,345 (100.0 %)Heterosexual N = 930 (69.1 %)LesbianN = 82(6.1 %)Bisexual N = 333 (24.8 %)TotalN = 998 (100.0 %)Heterosexual N = 771 (77.3 %)GayN = 137 (13.8 %)BisexualN = 89(8.9 %)SociodemographicsAge, M (SD)24.65 (4.68)24.92 (4.68)a24.11 (4.49)a,b24.02 (4.66)b24.74 (4.73)24.74 (4.81)a25.57 (4.45)a23.44 (4.25)bRace, N (%)White951 (70.7)636 (68.4)a62 (75.6)a,b253 (76.0)b709 (71.0)530 (68.7)a109 (79.0)b70 (78.7)a,bBlack95 (7.1)78 (8.4)a4 (4.9)a,b13 (3.9)b34 (3.4)25 (3.2)6 (4.3)3 (3.4)Asian160 (11.9)127 (13.7)4 (4.9)29 (8.7)151 (15.1)135 (17.5)a10 (7.2)b6 (6.7)bOther race139 (10.3)89 (9.6)12 (14.6)38 (11.4)104 (10.4)81 (10.5)13 (9.4)10 (11.2)Hispanic, N (%)133 (9.9)85 (9.1)8 (9.8)40 (12.0)128 (12.8)97 (12.6)17 (12.3)14 (15.7)Education, N (%) 0.80 and 5 conceptually meaningful and distinct classes of sufficient sample size. Finally, the adjusted LMR-LRT test for the 5-class model was significant, indicating significant improvement in model fit for a model with 5 versus 4 classes.Table 2Model fit information and conditional probabilities of past 30-day substance use by latent class.Number of polysubstance use classesAICBICaBICEntropyAdjusted LMR-LRTp-value148407.0748441.6248422.56––246100.3446169.4546131.321.000<.001343822.6043926.2743869.080.9460<.001441998.1642136.3842060.120.9930<.001541061.2941234.0741138.750.9950<.001640796.4141003.7440889.360.9590<.001740808.4141050.3040916.850.9100.499Past 30-day substance use indicatorPrimarily-Alcohol Use, N = 691 (29.5%)Polysubstance Use,N = 569 (24.3 %)Non-use,N = 425 (18.1 %)Cannabis and AlcoholCo-Use, N = 382 (16.3 %)E-cigarette, Cannabis, and Alcohol Co-Use,N = 276 (11.8 %)Combustible tobacco0.001.000.000.000.00E-cigarettes0.000.550.000.000.98Cannabis0.000.560.180.990.57Alcohol1.000.710.001.000.72*Note. The 6-class solution consisted of 2 classes with the same conditional probabilities (i.e., 2 classes reflecting primarily-alcohol use) and thus, the 5-class solution was chosen based on low AIC, BIC, and aBIC, high entropy, a significant adjusted LMR-LRT value, and theoretical interpretability. Table 2Model fit information and conditional probabilities of past 30-day substance use by latent class.Number of polysubstance use classesAICBICaBICEntropyAdjusted LMR-LRTp-value148407.0748441.6248422.56––246100.3446169.4546131.321.000<.001343822.6043926.2743869.080.9460<.001441998.1642136.3842060.120.9930<.001541061.2941234.0741138.750.9950<.001640796.4141003.7440889.360.9590<.001740808.4141050.3040916.850.9100.499Past 30-day substance use indicatorPrimarily-Alcohol Use, N = 691 (29.5%)Polysubstance Use,N = 569 (24.3 %)Non-use,N = 425 (18.1 %)Cannabis and AlcoholCo-Use, N = 382 (16.3 %)E-cigarette, Cannabis, and Alcohol Co-Use,N = 276 (11.8 %)Combustible tobacco0.001.000.000.000.00E-cigarettes0.000.550.000.000.98Cannabis0.000.560.180.990.57Alcohol1.000.710.001.000.72*Note. The 6-class solution consisted of 2 classes with the same conditional probabilities (i.e., 2 classes reflecting primarily-alcohol use) and thus, the 5-class solution was chosen based on low AIC, BIC, and aBIC, high entropy, a significant adjusted LMR-LRT value, and theoretical interpretability. Model fit information and conditional probabilities of past 30-day substance use by latent class. *Note. The 6-class solution consisted of 2 classes with the same conditional probabilities (i.e., 2 classes reflecting primarily-alcohol use) and thus, the 5-class solution was chosen based on low AIC, BIC, and aBIC, high entropy, a significant adjusted LMR-LRT value, and theoretical interpretability. As shown in Table 2, the 5 classes include: 1) Primarily-alcohol use (n = 691, 29.5 %), consisting of individuals who displayed high probabilities for alcohol use, but low probabilities of combustible tobacco, e-cigarette, and cannabis use; 2) Polysubstance Use (n = 569, 24.3 %), with individuals who exhibited moderate to high probabilities of combustible tobacco, e-cigarette, cannabis, and alcohol use; 3) Non-use (n = 425, 18.1 %), characterized by individuals who displayed low probabilities of using all substances; 4) Cannabis and alcohol co-use (n = 382, 16.3 %), consisting of individuals who had high probabilities of cannabis and alcohol use, but low probabilities of combustible tobacco and e-cigarette use; and 5) E-cigarette, cannabis, and alcohol co-use (n = 276, 11.8 %), with individuals who displayed moderate to high probabilities of e-cigarette, cannabis, and alcohol use, but low probabilities of combustible cigarette use. Associations between sexual identity and polysubstance use class among women Most women were classified into the Primarily-alcohol use class (32.6 %), followed by Polysubstance use (20.1 %), Non-use (18.4 %), Cannabis and alcohol co-use (17.7 %), and E-cigarette, cannabis, and alcohol co-use (11.3 %). Bivariate analyses indicated that a smaller proportion of bisexual and lesbian (vs. heterosexual) women belonged to the Primarily-alcohol use class, whereas a larger proportion of bisexual (vs. heterosexual) women belonged to the Polysubstance use and E-cigarette, cannabis, and alcohol co-use classes. Additionally, a larger proportion of lesbian (vs. heterosexual) women belonged to the Cannabis and alcohol co-use class (Table 1, left panel). Multinomial logistic regression results indicated that, relative to heterosexual women, bisexual and lesbian women displayed lower odds for Primarily-alcohol use. Additionally, bisexual women displayed higher odds for Polysubstance use and E-cigarette, cannabis, and alcohol co-use (Table 3, upper panel).Table 3Multinomial logistic regression analyses predicting polysubstance use class among women and men (ref = Non-use).Primarily-Alcohol UsePolysubstance UseCannabis and Alcohol Co-UseE-cigarette, Cannabis, and Alcohol Co-UseVariableaOR95 % CIaOR95 % CIaOR95 % CIaOR95 % CIWomenSexual identity (ref: Straight)Lesbian0.470.22, 0.980.950.44, 2.031.930.97, 3.831.230.54, 2.85Bisexual0.710.47, 0.971.521.10, 2.311.520.98, 2.361.781.11, 2.85Age0.970.94, 1.011.030.99, 1.070.980.94, 1.030.970.92, 1.01Race (ref: White)Black0.530.29, 0.950.830.45, 1.550.420.20, 0.880.310.11, 0.84Asian0.290.18, 0.450.390.23, 0.670.200.11, 0.360.310.16, 0.59Other0.590.35, 0.990.590.33, 1.060.530.29, 0.970.690.36, 1.32Hispanic0.780.46, 1.320.850.48, 1.500.740.41, 1.360.490.23, 1.03Education (ref: