Due to the unequal time points between Week 16 and Week 26, we allowed for a free autoregression between these last time points.Lastly, initial models indicated large standardized residual covariances associated with the drinks per drinking day at Week 4. Due to the smoking quit date occurring at the start of the active medication phase, it is possible that greater changes in smoking, and possibly drinking, occurred during the first 4 weeks of the trial. Therefore, we allowed for a second-order autoregression as a free path between Week 4 and Week 12 to allow for any changes during Week 4 to influence Week 12. To account for the relationship between smoking and drinking variable within a time period, error covariances between drinking variables and cigarettes per smoking day at each time point were fixed to be equal across time points. By specifying error covariances, we specifically aimed to account for the correlated errors due to the strong evidence highlighting the relationship between drinking and smoking behaviors concurrently. For the first model, this included one set of fixed error covariances between drinks per drinking day and cigarettes per smoking day. For the second model, three sets of error covariances were specified: one for percent heavy drinking days and cigarettes per smoking day, a second for percent days abstinent and cigarettes per smoking day,cannabis flood table and a third for percent heavy drinking days and percent days abstinent. We did not anticipate the relationship between these variables to differ over time, therefore fixed the error covariances across time.
Due to multiple paths in the cross-lagged panel models, we constrained our indirect paths of interest to those in which drinking variables were present first, followed by subsequent drinking variables or smoking variables, and ending with cigarettes per smoking day at Week 12 and Week 26. These indirect paths are displayed in bold in Figure 1A and 2A for Week 12 and Figure 1B and 2Bfor Week 26. When testing drinks per drinking day as a mediator, there were a total of 3 indirect paths that were aggregated in our test of indirect effects at Week 12 and a total of 11 paths were aggregated at Week 26 . The same approach of selecting indirect effects to aggregate was applied to the second model testing percent heavy drinking days and percent days abstinent as mediators. A total of 6 paths were aggregated when cigarettes per smoking day at Week 12 was the outcome , and a total of 22 paths for Week 26 . Of note, these mediators were tested simultaneously in the model such that the indirect paths for percent heavy drinking days and percent days abstinent were aggregated. In order to properly account for the fact that our outcome was positively skewed by virtue of participants who abstained from smoking at various points throughout the trial, maximum likelihood estimation was used with the SatorraBentler scaled chi-square test statistics and the associated sandwich-type standard error estimates . For the individual path estimates, we set our a-priori cut-off at p < .0125 to correct for the 4 sets of cross-lagged paths. In addition, percentile bootstrap confidence intervals were used to conduct inference on the aggregated indirect effect. An advantage of bootstrapping in our sample is that this method makes no assumption regarding the shape of the sample distribution of the indirect effect .
An initial model was estimated followed by a total of 2,000 bootstrapped estimates of the aggregated indirect effect with 95% confidence intervals. If the confidence interval did not contain zero, then the test of our indirect effect was considered statistically significant. At baseline our sample consisted of 165 heavy drinking smokers who were randomized to one of the two medication conditions. However, due to the maximum likelihood estimation method with the Satorra-Bentler correction described above, only those with complete data were used in our cross-lagged panel models. While there was some attrition throughout the trial, the sample size based on the availability of complete data across medication groups remained relatively consistent throughout the trial, ending with 115 total participants at Week 26. Thus, our final sample size for the results described herein was 115. To examine any changes when using the full sample, we also conducted the same models with full maximum likelihood estimation , with no Satorra-Bentler correction, to make use of all available data. SAS does not allow for simultaneous use of the Satorra-Bentler correction and FIML, therefore the Satorra-Bentler correction was not used when employing FIML. The fit indices and path estimates for these models are presented in Supplementary Materials.In a sample of treatment-seeking heavy drinking daily smokers enrolled in a smoking cessation and drinking reduction clinical trial, we examined the directional influence drinking and smoking variables have on each other over time. We hypothesized that smoking outcomes are influenced by pharmacotherapy through a sequence in which pharmacotherapy influences drinking behaviors, which then influences smoking behavior at a subsequent time point. We did not find a significant medication effect across our primary or secondary drinking outcomes. We did find consistent significant autoregression over time across all drinking variables, and cigarettes per smoking day.
While each of our drinking and smoking variables were measured concurrently, the study design allowed us to examine the presence of cross-lagged associations. We did not find significant cross-lagged associations between our primary or secondary drinking variables and cigarettes per smoking day that met our a-priori alpha cutoff. There was no significant indirect effect of pharmacotherapy on cigarettes per smoking day through our primary or secondary drinking variables. These results indicate that while there was some degree of stability in our constructs over time, as indicated by the autoregressions, drinking variables did not explain the effect of pharmacotherapy on cigarettes per smoking day. Notably, these findings may be distinct from the primary trial in that only 115 individuals provided complete data and were therefore included in these mechanistic models. This study is one of the first to examine concurrent and lagged smoking and drinking in the context of a combined smoking cessation and drinking reduction treatment. Other studies have shown that drinking episodes are associated with smoking lapses , which provided initial support for our hypothesis that changes in drinking behavior could occur prior to changes in smoking behavior. Contrary to our hypothesis, the effect of drinking on smoking was not a pathway through which medication exerted its effect on smoking outcomes, nor was there a direct effect of medication condition on cigarettes per smoking day during the active medication phase. It may be that this phenomenon is observed in a smaller time scale than our study design allowed for us to test. Methods such as ecological momentary assessment may be able to further understand the role of drinking in this context. Previous studies utilizing EMA have highlighted how alcohol can predict smoking after covarying for contextual factors such as time of day, location, and presence of other smokers , and how co-use is associated with greater urge to use both alcohol and cigarettes . Our results aligned broadly with these findings,cannabis grow supplier as we found a significant fixed error covariances between drinking variables and cigarettes per smoking day at each time point. These results highlight the relationship between drinking and smoking within a single time period. The application of EMA methods to pharmacotherapy treatment studies may shed new lights on how these medications may disrupt alcohol, and subsequent smoking behaviors. Given the context of our study with a focus on smoking cessation, and drinking reduction or cessation, it is possible that participants in our sample did not have a strong desire to substantially reduce their drinking given that the range of drinking throughout the trial was arguably limited. And that the addition of naltrexone on varenicline was not sufficient to promote the clinical level of drinking reduction that was required to observe a direct effect during the active medication phase, or an indirect effect on cigarettes per smoking day through drinking variables. Future studies may benefit from examining varying levels of drinking reduction or cessation, as well as larger sample sizes, to further understand the sensitivity in the combination of varenicline plus naltrexone. A previous study by Anton and colleagues found naltrexone was more efficacious in reducing percent heavy drinking days among those who were smokers versus non-smokers. However, naltrexone did not exert a direct effect on smoking behavior itself . It is possible that naltrexone may still exhibit an added benefit for heavy drinking smokers, albeit in comparison to non-smoking heavy drinkers. While approved for smoking cessation, varenicline has been shown to reduce alcohol consumption in heavy drinking smokers . However, there have been studies to suggest that varenicline does not reduce alcohol consumption . Pre-clinical research has shown the combination of varenicline and naltrexone to be more effective than low doses of either medication alone in reducing alcohol use in mice.
Human laboratory research has also shown a benefit of combined varenicline and naltrexone reducing drinks per day and cigarettes per smoking day . A study comparing varenicline versus varenicline plus naltrexone on ability to resist smoking after completing an alcohol challenge did not find a benefit of the combined medication on delayed smoking . Our results add to the somewhat mixed literature on potential benefits of combined varenicline and naltrexone on smoking outcomes in the presence of alcohol. Mixed literature notwithstanding, our results do underscore the degree to which drinking and smoking behaviors fluctuate together over time as indexed by respective stability in drinking and smoking variables over time and error covariance between drinking and smoking variables. The clinical implications of this study, based on our significant fixed error covariances, are that throughout the entire cessation attempt, the drinking patterns continue to predict smoking patterns across the medication and follow-up periods. In other words, as individuals attempt to reduce their drinking, their smoking behavior at one time point continues to influence smoking at the next time point. In the context of treatment, there is evidence to support why it is feasible to address more than one substance at a time . An examination of barriers to smoking cessation among alcohol dependent patients made salient a concern that quitting smoking may increase urges to drink or use other substances to an overwhelming degree, or make it more difficult to stay sober during substance abuse treatment . These concerns are contradicted by studies showing that smoking cessation does not have a negative effect on substance use outcomes, rather it can often have a positive outcome . Specific to alcohol, smoking cessation does not offset abstinence from alcohol and may go as far as increasing the likelihood of maintaining alcohol abstinence . A recent study supported these results showing that smoking cessation does not increase binge drinking among patients with serious mental illness . In addition, individuals who are motivated to reduce their drinking during the smoking cessation attempt may be more vulnerable to the iatrogenic effects of alcohol on smoking. In other words, the clinical recommendation that individuals reduce their drinking during a quit attempt is well supported by these data as these two behaviors are still strongly related throughout a cessation attempt. Furthermore, independent of one another, we found use of each substance is predictive of future use respectively across 6-months. This calls attention to the potential benefit of reductions in either substance at any phase of a cessation attempt to exert a meaningful impact on successive use. However, even individuals who are actively addressing their drinking and receiving pharmacotherapy for alcohol use are continuously affected by the bidirectional relationship between smoking and drinking. The present study must be interpreted in light of strengths and limitations. Strengths include study design affording multiple assessments of drinking and smoking behaviors across a span of 6 months and within subjects. Additionally, the use of percentile bootstrap confidence intervals to assess the indirect effect in our mediation models as this approach has been shown to produce inferences that are more likely to be accurate than the normal theory approach . A noteworthy limitation is that while we did not have excess attrition throughout, reductions in our sample size across the trial may have reduced our power to detect more nuanced relationships, including mediated effects, among medication, drinking, and smoking outcomes.