Study Population and Data Collection
The authors obtained data from a survey study focusing on health risk behaviors of Chinese sexual minority men conducted by School of Health Sciences, Wuhan University, China. Details of the survey methodology have been published elsewhere . Briefly, between September 2017 and January 2018, we collected baseline survey data of 755 self-identified gay and bisexual men using venue-based sampling from college campus-based sexual minority-serving organizations in four cities, namely Beijing, Wuhan, Nanchang, and Changsha, as these cities have high concentration of LGBT populations. These data are among the most recent data available regarding tobacco use among sexual minority men in China. This convenient sample was mainly comprised of urban, well-educated, and high-income gay/bisexual men, and might not be representative to all sexual minority men. After screening for eligibility (self-identified as gay/bisexual men and aged 16 years or older), participants were given a brief explanation of the survey’s purpose. Informed consent was obtained from every participant. Each participant received a one-time compensation of 50 Chinese RMB (approximately $8 US dollars) for their time. For this study, we excluded 79 (10.5%) participants who reported “heterosexual/unsure/other” sexual orientation. Participants involved in tobacco cessation programs, including HIV-positive individuals who were linked to HIV care and provided with behavioral intervention (i.e., tobacco cessation services) , were excluded from this study. The analytical sample size for this study was 676. This study was approved by the Institutional Review Board (IRB) at [blinded for review].
Primary outcome: Cigarette smoking. Participants were asked, “During the past 30 days, on how many days did you smoke cigarettes? Little cigars or cigarillos? Traditional pipe? Chewing tobacco? E-cigarettes? Hookah?” Given that fewer than 10 participants reported on alternative tobacco use (i.e., little cigars or cigarillos, traditional pipe, chewing tobacco, e-cigarette, and hookah), this study only focused on the past 30-day cigarette smoking outcome, dichotomized as “no” versus “any smoking.”
Primary grouping variable: Sexual orientation. Sexual orientation was assessed by asking, “Is your sexual orientation: heterosexual; gay or homosexual; bisexual; or unsure?” All responses were coded dichotomously (0 = gay and 1 = bisexual).
Sociodemographic characteristics. Participants were asked to provide sociodemographic information, including age, education (high school/below vs. college/above), place of origin (urban vs. rural), employment (dummy coded into student, employed, and unemployed), marital status (unmarried/divorced vs. married), monthly income (≤ 3000 RMB [73 USD] vs. > 3000 RMB, note that the minimum monthly wage ranges between 1580 and 2000 RMB in 4 sampled cities ), and nature of their health insurance (yes vs. no/unsure).
Depressive symptoms as mediator. Depressive symptoms were hypothesized as a mediator of the relationships between minority stressors and cigarette smoking (Figure 1) and were assessed with the Center for Epidemiological Studies Depression (CES-D) scale.  The CES-D is a well-established and widely used  20-item scale designed to measure depressive symptoms experienced by the individual within the past week, and its Chinese version has been validated [67, 68]. Items were answered on a 4-point scale ranging from 0=less than a day or never to 3=5–7 days. The Cronbach’s alpha was 0.89.
Minority Stressors. Informed by Minority Stress Theory [26, 27], we measured distal minority stressors (everyday discrimination), proximal minority stressors (outness, rejection anticipation, identity concealment, and internalized homophobia), general stressors (adverse childhood experiences), and stress-moderating factors (social support and resilience).
Everyday discrimination was assessed using the Everyday Discrimination Scale . This scale asks about the frequency of 9 types of hassles and prejudice events that sexual minority people may encounter. This 9-item scale is rated on a 6-point Likert scale with response options of 1=happens daily to 6=never happened. All responses were reverse coded and averaged to create a mean score, with higher scores indicating severer everyday discrimination. The Cronbach’s alpha coefficient was 0.94 for this scale.
Outness was assessed by asking respondents “Have you ever ‘come out’ to anyone?” All responses were coded dichotomously (0=no and 1=yes).
Rejection anticipation was assessed with a scale which was originally used to assess stigma of mental illness . Later, this scale was modified and adapted by Forst et al. (2015) for assessing sexual minority’s state of hypervigilance and worry about being rejected. This 6-item scale is rated on a 5-point Likert scale, ranging from 1=applies very strongly to 5=does not apply at all. All responses were reverse coded and averaged to create a mean score, with higher scores indicating higher expectation of rejection. The Cronbach’s alpha of this scale in this study was 0.88.
Identity concealment was assessed using a subscale on nondisclosure developed and validated by Testa et al.(2015) . This 6-item scale asks about the intentions and behaviors of sexual minority individuals to avoid disclose their sexual minority identities. This scale is rated on a 5-point Likert scale, ranging from 1=applies very strongly to 5=does not apply at all. All responses were reverse coded and averaged to create a mean score, with higher scores indicating greater identity concealment. The Cronbach’s alpha was 0.91.
Internalized homophobia was assessed using the Internalized Homophobia Scale which was originally developed by Martin and Dean (1987)  and further modified and validated by Forst et al. (2009; 2015). [31, 73] This scale asks about the negative attitudes sexual minorities hold against their own sexual identities. This 8-item scale is rated on a 4-point Likert scale, ranging from 1=never to 4=always. Responses were averaged to create a mean score, with higher scores indicating greater internal homophobia. The Cronbach’s alpha was 0.89.
Adverse childhood experiences (ACEs) was assessed using a 10-item ACEs index developed by the U.S. Centers for Disease Control and Prevention [74, 75]. This index asks about the physical and mental abuse and traumatic experiences of participants prior to 18 years old. Participants answered 0=no or 1=yes to each item. Total score of all responses was summed; higher scores indicate more ACEs. The Cronbach’s alpha was 0.63.
Social support was assessed with the Multidimensional Scale of Perceived Social Support, which measures perceived support from family, friends, and significant others. This 12-item scale is rated on a 7-point Likert scale, ranging from 1=very strongly disagree to 7=very strongly agree. The total score of this scale was calculated for each participant, with higher scores indicating more social support. The Cronbach’s alpha was 0.94.
Resilience was assessed with the 10-item Connor-Davidson Resilience scale.[77, 78] This scale is a measure of stress coping capabilities. It is rated on a 5-point Likert scale ranging from 1=not true at all true to 5=true nearly all of the time. The responses are summed to derive a total score, with higher scores indicating more resilience. The Cronbach’s alpha was 0.95.
Univariate analyses were conducted to examine the distribution of each variable. ANOVA and chi-square tests were conducted to assess the bivariate relationships between sociodemographic and psychosocial variables and the cigarette smoking outcome. We also examined the multi-collinearity between all variables. In consideration of study power, we excluded selected sociodemographic variables (i.e., education, marriage, monthly income, and health insurance) from the modeling analyses, as these variables either were not associated with the outcome or showed considerable collinearity (data not shown).
For the two-group SEM, we specified the SEM model in Figure 2 based on Minority Stress Theory, preliminary analyses, and correlation matrix results. SEM is a process that allows for testing one or more theories that are hypothesized a prior to explain the characteristics of measured variables . SEM can be used for model confirmatory purpose, testing alternative models, or model generation . Two-group SEM can be used to examine inter-group differences across sexual minority subgroups with increased rigor. The advantage of two-group SEM is that it allows the comparison of the extent of associations based on path coefficients  and uses model fit indices to determine which tested paths best fit the data. This method could help examine whether the underlying pathway are significantly different among sexual minority subgroups. Given the goals of this study, two-group SEM is a particularly useful tool for examining how different pathways might vary across sexual orientation. A two-phase modeling approach was used for the two-group SEM. First, we examined measurement invariance (i.e., item-scale relationships) between gay and bisexual participants using confirmatory factor analysis (CFA). Second, we examined structural invariance (i.e., hypothesized relationships among variables) between gay and bisexual participants. Chi-square differences between these two models were examined and indicated non-significant results and thus no group differences in the measurement models .
Then, we examined the model fit and path-coefficients of the final two-group structural SEM. The model fit indices included: chi-square test, standardized root mean square residual (SRMR), comparative fit index (CFI), Tucker–Lewis fit index (TLI), the root mean-square error of approximation (RMSEA), and weighted root mean square residual (WRMR). The indicators of goodness of fit were: χ2 p< 0.05; SRMR > 0.08; CFI > 0.90; TLI > 0.90; RMSEA < 0.05; and WRMR < 1 . The bootstrapping method was used to test 95% Confidence Interval (95% CI). Standardized regression (β) coefficients, the standard errors, and p-values for β were reported in the final model.
Data were double-entered and cleaned using EpiData 3.1 (The EpiData Association, Odense, Denmark) software. Descriptive analysis and sequential regressions were conducted using SAS 9.4 (SAS Institute Inc.: Cary, NC, USA). Two-group SEM was conducted using Mplus 8.0 (Muthén & Muthén: Los Angeles, CA, USA).