Culturally Tailored Smoking Cessation Intervention

Key Specific Aims

The purpose of the present analysis is to (a) examine the relationship between participation in a culturally tailored smoking cessation program on smoking outcome, compared to a standard non-tailored intervention; (b) examine if success in the intervention varied by gender or educational level; (c) examine if depressive symptoms was associated with smoking cessation.

For the binary outcome of smoking cessation, we used logistic regression models to examine six models. Consequently, prior distributions for the effect of the predictors are on the logit scale.

The first model was an intercept only model. The second model included only the treatment condition (MPACT or Freedom from Smoking). The third model added education as a covariate. Model four replaced education with gender as the only covariate. Model five included both gender and education as covariates. Model six added depression as a second predictor.

After fitting the models, each model will be examined for Markov Chain Monte Carlo (MCMC) convergence by examining the trace plot, Gelman-Rubin statistics (r-hat), effective sample size and autocorrelation function. The MCMC trace plot traces the sampled parameter as a function of the step in the chain (Kruschke, 2014). The Gelman-Rubin statistics, Rhat or potential scale reduction factor is a good indicator of model convergence. It calculates both the between-chain variance and within-chain variance and assesses whether they are different enough to worry. Values close to 1 and below 1.1 indicate model convergence (Gelman et al., 2013; Kruschke, 2014). Autocorrelation function provides information about the autocorrelation across subsequent steps in the sampling. Autocorrelation function that drops rapidly to zero indicates that the MCMC chains are sampling accurately from the posterior distribution (Kruschke, 2014). Effective sample size accounts for autocorrelation in the posterior samples. It is the number of relatively independent samples obtained from the posterior distribution (Gelman et al., 2013). Examining the metrics above provides information about the accuracy, preventiveness and efficiency of the MCMC sampling.

To evaluate whether our model explains the data by investigating posterior parameter distributions and posterior predictive checking. Posterior parameter distribution shows a histogram of the parameter and provides a quick visual representation of the estimates and the uncertainty around them. Posterior predictive check is used to assess whether predicted values from the model are similar to the actual data (Gelman et al., 2013). If the model fits the data well, the predicted values should look similar to the observed data and recover important patterns in the observed data.

Model comparison and selection will be done by model averaging via stacking of predictive distributions using the r-package loo. Model stacking combines all candidate models by maximizing the leave-one-out predictive density of the combination distribution and returns a model weighting for each model (Yao, Vehtari, Simpson, & Gelman, 2018). The r-package LOO uses Pareto-smoothed importance sampling (PSIS) to conduct a leave one-out cross validation and estimate the expected log predictive density (ELPD) (Vehtari, Gelman, & Gabry, 2017a). Model stacking regularizes based on the LOO-ELPD scores of all candidate models. The model with the highest ELPD is selected as the best model. However, the reliability of the PSIS approximation is determined by, and can be assessed by the estimated shape parameter (k-hat). Values over 0.7 may be problematic (Vehtari et al., 2017a).

The prior represents the current state of belief about the phenomenon of interest. The likelihood reflects the probability of the phenomenon as observed in the current data. Therefore, the posterior distribution can be said to be a process updating a prior belief about a phenomenon with the observed data.

After specifying the prior, a Markov Chain Monte Carlo method was used to approximate posterior distributions using BRMS and stan (Kruschke, 2014). Stan uses a Hamiltonian Monte Carlo sampling algorithm to sample from the posterior distribution of the parameter (Carpenter et al., 2017). For the analysis, we specified four (4) chains of 5000 iterations, with a warmup period of 1000. Adapt delta was set to 0.99. Using the equations above, we sample from the posterior distribution of the parameters using brm() code below. Detailed BRMS code is provided in Appendix A.

Background and Significance Tobacco smoking remains the leading cause of preventable death in the United States (U.S.), with over 480,000 deaths annually(United States Surgeon General, 2014). Although Asian/Pacific Islander (API) are less likely to smoked cigarette, compared with other major ethnic categories in the U.S., when broken into subgroups, some API groups have a high burden of cigarette use (Mukherjea, Wackowski, Lee, & Delnevo, 2014; L. T. Wu, Swartz, Burchett, Workgroup, & Blazer, 2013). API subgroups such as Native Hawaiian/Pacific Islanders and Filipinos are at high risk for cigarette smoking, and experience a high burden of the mortality and morbidity associated with cigarette smoking such as cancer (Mukherjea et al., 2014; L. T. Wu et al., 2013). Thus, smoking prevention and cessation interventions that work are needed for Native Hawaiian/Pacific Islanders and other API subgroups.

Majority of adult smokers have a desire to quit smoking. Among current U.S. adult smokers, about 68.0% reported in 2015 that they wanted to quit completely, 55.4% made a past-year quit attempt (Jamal et al., 2018). Among Asian Americans, about half of all smokers reported making an attempt to quit smoking completely (Romero & Pulvers, 2013). Despite the high levels of desire to quit smoking, less than one third of people used an evidence based cessation method and only about 7.4% successfully quit smoking (Jamal et al., 2018). Among all smokers, successful smoking cessation was associated with increasing level of educational attainment, less than 45 years, being Hispanic or Asian, having private health insurance, or having no serious psychological distress (Jamal et al., 2018). Thus, it is important to understanding the barriers to and facilitators of successfully quitting following participation in smoking cessation program. In addition, evidence-based tobacco cessation programs must be made widely available and accessible to people who want to quit.

Culturally specific interventions have been recommended for ethnically diverse populations. The Tobacco Use and Dependence Clinical Practice Guidelines stressed the need for additional research to determine the effectiveness of culturally targeted smoking cessation interventions for racial and ethnic minorities (Fiori et al., 2008). Culturally specific interventions may be more acceptable to the targeted population, thereby increasing participation in treatment and decreasing discontinuation rates. Rodr?­guez Esquivel, Webb Hooper, Baker, and McNutt (2015) examined the efficacy of a culturally specific smoking cessation intervention for US Hispanic. The result showed that the culturally specific intervention successfully reduced the number of cigarettes smoked per day, however, it had no significant effect on actual smoking cessation. Nollen et al. (2007), examined the efficacy of targeted versus standard care smoking cessation materials among urban African American smokers. Despite greater use of the targeted, there were no significant differences between groups on the smoking outcomes. A systematic review of adapted smoking cessation interventions for African-American, Chinese-American and South-Asian populations found that adapted interventions were more acceptable to the target population, however, they found no clear evidence of adapted interventions being more effective (J. J. Liu et al., 2013). The greater acceptability of adapted interventions is important as it exposes more people to smoking cessation interventions. However, studies should seek to understand other factors that are associated with successful participation in these interventions.

Although disparities in smoking cessation can be partly explained by differences in tobacco use behaviors, health care utilization, and access to cessation treatments, and knowledge about these treatments (Jamal et al., 2018). Mood and depressive symptoms are associated with higher odds of tobacco use and relapse, and lower likelihood of stopping cigarette smoking (Sonne et al., 2010). Stepankova et al. (2017), in examining the association between baseline depression and 1-year smoking abstinence also found that patients with mild depression were less likely to be abstinent and patients with moderate to severe depression were considerably less likely to be abstinent compared to those without depression. However, other studies found no association between depression and smoking cessation (Schnoll, Leone, & Hitsman, 2013). Understanding the relationship between depression and outcome of smoking cessation may help guide the development of future interventions.

Did you like this example?

1034

54