Data Source
This study used data from the Population Assessment of Tobacco and Health (PATH) Study [32]. The PATH Study is collaboratively sponsored by the National Institute of Drug Abuse, National Institute of Health, Center for Tobacco Products, and Food and Drug Administration. It consists of longitudinal interview and self-reported survey questions using audio computer-assisted self-interviews administered in English or Spanish to parents, adults, and youth pertaining to tobacco use, behavior, attitudes, beliefs, and health outcomes. It collected data bi-annually in five waves (1,2,3,4 and 4.5) from 2011 to 2019, using weighting procedures to adjust for oversampling and nonresponse which were then further adjusted based on US Census Bureau data to develop a nationally representative study group. About 46,000 people aged 12 years and older, including tobacco users and non-users, were included in the first wave of the PATH Study and followed over time. This study utilized longitudinal data from waves 2, 3, 4, and 4.5 of the PATH Study databases among cigarette users. The wording of wave 1’s questions were less specific and differed from subsequent waves and wave 1’s data was thus excluded from this study. More details regarding PATH can be found at https://www.drugabuse.gov/research/nida-research-programs-activities/population-assessment-tobacco-health-path-study.
Measures
Demographics
The sociodemographic variables included participant age, gender, race, ethnicity, grade level, age of cigarette smoking initiation, and household income from wave 4.5. Wave 4.5 was selected because it contained the most current data for PATH youth participants.
Outcome Measures
The research team reviewed the PATH database and selected six questions related to e-cigarette addiction and three questions related to e-cigarette harm perceptions as outcome measures. Outcome scores were solely derived from the most PATH Wave 4.5. E-cigarette initiation flavor was derived from Wave 2,3,4, or 4.5, depending on when the respondent first reported e-cigarette usage. Only participants who remained in the study from their first reported use of e-cigarettes to the most recent wave were included in the analysis.
E-Cigarette Addiction
Measures of e-cigarette addiction came from wave 4.5 in which participants reported their level of agreement on six variables (i.e., items): (1) I find myself reaching for electronic nicotine products without thinking about it, (2) Frequently crave electronic nicotine products, (3) My electronic nicotine product use is out of control, (4) Using electronic nicotine products helps me feel better if I've been feeling down, (5) Using electronic nicotine products helps me think better, and (6) I would feel alone without my electronic nicotine products. The response options for all six items used a 5-point Likert scale which ranged from 1 (not at all true) to 5 (extremely true).
E-Cigarette Harm Perception
Measures of harm perception came from wave 4.5 data in which participants responded to the following three items: (1) Harmfulness of electronic nicotine products to health (Response options: 1=Not at all, 2=Slightly, 3=Somewhat, 4=Very, 5-Extremely), (2) Thoughts on how much people harm themselves when they use e-cigarettes or other electronic nicotine products (Response options: 1=No harm, 2=Little harm, 3=Some harm, 4=A lot of harm), and (3) Harmfulness of using e-cigarettes or other electronic nicotine products compared to smoking cigarettes (Response options: 1=Less harmful, 2=About the same, 3=More harmful).
Predictor
The predictor was e-cigarette flavor type initiation. Measures about the e-cigarette flavor type used at initiation came from waves 2, 3, 4 and 4.5 of the PATH Study depending on when participants reported previous use of e-cigarettes. Only participants reporting previous ENDS use answered questions about the flavor type initiation. The study examined two general types (traditional and non-traditional) of e-cigarette flavor initiation. Traditional types included standard tobacco, menthol, or mint flavors. Non-traditional types included fruit, clove/spice, alcoholic drink, non-alcoholic drink, and candy/dessert/other sweets. The study excluded respondents that selected more than one initiation flavor type.
Covariates
Sociodemographic factors such as age, sex, race and annual household income, and the age at which they started smoking cigarettes regularly can impact e-cigarette addiction and harm perception [28, 33, 34]. Therefore, this study adjusted for the effects of these covariates in statistical analyses. The survey asked participants to quantify an estimate for the total number of instances they had used an e-cigarette and similarly estimate the age at which they initiated e-cigarette use. The statistical analysis controlled for these estimates.
Statistical Approach
Data analyses were conducted using SPSS for Windows version 28. Descriptive statistics characterized the study sample. Frequency distributions of e-cigarette flavor initiation type of both unweighted and weighted frequencies and proportions were computed and reported. The weighted values were derived from the all-wave youth cohort file and represent national population estimates while unweighted numbers represent sample estimates. Even though the sample is large, it may not accurately represent the entire US without adjusting the sample to represent the population.
Exploratory factor analysis was performed on the six addiction and three harm perception items to assess the factor structure of the items using principal axis factoring and varimax rotation with Kaiser normalization. Investigating the factor structure of items determines whether items associate with each other to form a latent construct (e.g., factor). If the six addiction items have similar patterns of item responses, they will measure the underlying latent construct of addiction and can be used to generate a composite score (e.g., factor score) for analyzing addiction. This facilitates interpretation, since the outcome measures of addiction as a whole is of greater interest than the outcome of each individual addiction item [35].
Factor loading evaluates factor structure and determines how strongly items fit or associate with each other to form one underlying construct. It weighs the correlation of an item with the construct. Factor loading values range from -1 to 1, with values larger than |0.4| regarded as being relevant and having adequate fit for a construct [36]. The Kaiser-Meyer-Olkin Measure of Sampling Adequacy tests for the sampling adequacy of the selected items and the complete dataset. Using this method, a value of >0.6 indicates that factor analysis could be applied to the dataset and a Barlett’s test of sphericity with p<0.05 shows that the selected items were correlated. More detailed descriptions of factor analysis can be found elsewhere [37].
The reliabilities of the addiction factor and harm perception factor were then examined using Cronbach’s alpha with a satisfactory Cronbach’s alpha value set > 0.6 [38]. Cronbach alpha values range from 0 to 1 with larger values representing greater reliability [39]. After each factor demonstrated adequate factor loadings for its items and adequate reliability, composite (factor) scores for both the addiction and harm perception outcomes were created using a linear scale metric. Higher factor scores signified that they had higher levels of addiction and perceived the products as more harmful. Factor scores are essentially a standardized, weighted average of the items’ scores, with the items’ weights coming from the factor loadings. Since most items have unequal correlations with an underlying construct, average item scores should not be used to represent a construct. Using factor scores more appropriately reflects the strength of association with different items.
Linear regression analyses of the composite scores (e.g., factor scores) for addiction and harm perception were used to examine the two research questions: (1) Does e-cigarette flavor initiation type predict e-cigarette addiction, with and without adjustment for a person’s age, age when they first started smoking cigarettes regularly, sex, race and annual household income? (2) Does e-cigarette flavor initiation type predict e-cigarette harm perception, with and without adjustment for a person’s age, age when they first started smoking cigarettes regularly, sex, race and annual household income? The standardized regression coefficient with an associated 95% confidence interval and R [2] were calculated. A two-tailed p-value < 0.05 was considered statistically significant.