Data from this study come from the 2017 U.S. Centers for Disease Control and Prevention (CDC) Youth Risk Behavior Surveillance System (YRBS), a biennial survey to monitor health behaviors including substance use, violence, sexual risk behaviors, and mental health among 9th-12th graders. The survey utilizes a three-stage cluster sampling design to yield representative estimates of the prevalence of risk behaviors among high school students (30). Our study used the 2017 survey and included students who self-identified as non-Hispanic Black (i.e., multi-racial youth were not included), and excluded respondents missing data on key variables (n=2,782) from the total sample of 14,765. The analysis of this publically available non-identifiable data was not subject to review by institutional review board pursuant to United States Code of Federal Regulations §46.101 and §46.104.
The 2017 YRBS included 95 items covering a range of domains including substance use, mental health and other health-related behaviors. Tobacco use was measured by the following items:
Cigarette Use. Participants were asked “Have you ever tried smoking, even one or two puffs?” Those who indicated cigarette use were asked to identify the number of days of cigarette use in the last 30 days. Responses were recoded to indicate “Never,” “Ever,” and “Recent (past 30 days)” use.
E-cigarette Use. Participants were asked to identify whether they had ever used any electronic vapor products. Then, among those who indicated e-cigarette use, youths were asked to identify the number of days within the past 30 days they used e-cigarette products. Responses were recoded to indicate “Never,” “Ever,” and “Recent (past 30 days)” use.
Cigar Use. Participants were asked on how many days in the past 30 days they had smoked cigars, cigarillos, or little cigars. Responses were dichotomized to indicate “No,” or “Yes” use.
Dip/chewing tobacco Use. Participants indicated how many days they used chewing tobacco, snuff, dip, snus, or dissolvable tobacco products in the past 30 days. Responses were dichotomized to indicate “No,” or “Yes” to indicate use in the past 30 days.
Covariates for latent class membership
Variables used to examine correlates of membership in latent classes included sex, grade (9th through 12th) and marijuana use. Sex was determined by asking respondents “What is your sex?” Response options were limited to “Male” or “Female.” Students indicated what grade in school they were in. Students were asked two questions related to marijuana use, which were combined and recoded to indicate “Never,” “Ever,” or “Recent” marijuana use.
Statistical Analysis
LCA was utilized to explore and identify tobacco use profiles among Black adolescents using cigarette smoking, e-cigarette use, other combustible tobacco products, and other non-combustible tobacco products as latent class indicators. A series of latent class models specifying one to five classes was tested. Optimal model selection was based upon recommended indices including low Adjusted Bayesian Information Criterion (aBIC) relative to other models, significant Lo-Mendell-Rubin Likelihood Ratio Test (LMR/LRT), and acceptable quality of classification (31). aBIC is based on the loglikelihood of each model with the lowest value providing support to select a particular model. The LMR/LRT tests for improvement of fit for the model under consideration compared with a model with one less class. A p-value greater than 0.05 indicate that the model with one less class fits best. All analyses were conducted using Mplus Version 8.1 (32). The Mplus tools stratification, cluster, and weight were used to calculate the correct standard errors for the complex survey design of the YRBS; data were weighted to represent the U.S. population. Missing data for latent class indicators were accounted for using the full information maximum likelihood (FIML) capabilities of Mplus. After determining the appropriate number of classes, multinomial logistic regression was used to assess the role of sex, grade level, and marijuana use in association with tobacco use class membership (33). Covariates were treated as auxiliary variables using the R3STEP function in Mplus, which initiates the multinomial regression and maintains the class structure while controlling for uncertainty in class assignment (33). Post hoc cross tabulations to explore tobacco product use within class were conducted in SAS 9.4 and represent unweighted data.