Links to surveys were sent to participants’ email addresses and smartphones. Staff reminded participants to complete assessments via text message, telephone, and email. All procedures were approved by the University of California, San Diego Institutional Review Board. Demographic characteristics including age, sex, race, ethnic background, and student status were measured at baseline by self-report. Student status was collapsed into a dichotomous variable comparing full-time students to all other participants. Cigarette and other tobacco use were assessed at each of the 9 time points. At baseline and 12 and 24 months later, the Timeline Follow Back  was used to evaluate number of cigarettes smoked, as well as whether participants had used each of marijuana, alcohol, ecigarettes, hookah tobacco, and any other tobacco product , on each of the 14 days preceding the day of assessment receipt. At the 3, 6, 9, 15, 18 and 21 month assessments, participants reported whether they had used marijuana, alcohol, e-cigarettes, hookah tobacco, and OTPs in the past 24 hours on each of 9 consecutive days. Raw data for each of the days assessed were aggregated to create variables reflecting quantity of cigarettes smoked over 9 or 14 days of each assessment period , and frequency or number of days on which marijuana , cigarettes , e-cigarettes , hookah tobacco , and OTPs were used during each assessment period. We created a count variable that reflected the number of days at each time point on which participants reported using any tobacco product , and a binary variable that assessed whether or not they reported use of multiple tobacco products at each time point . The marijuana days variable was used to calculate a time-varying variable that measured cumulative number of time points,grow trays 4×4 up to and including the one being assessed, at which marijuana days was greater than 0. For example, if a participant reported 1 marijuana day at baseline, 0 at 3 months, and 4 at 6 months, his or her values for marijuana frequency at those time points would be 1 , 1 , and 2 , respectively.

The purpose of this variable was to capture cumulative marijuana use aggregated over the full two years, rather than within each assessment period. We assumed that if marijuana use is a predictor of heavier tobacco use, individuals who use marijuana more frequently over the entire study period would be most vulnerable to this association. Thus, we believed that this variable would better capture marijuana use over time relative to a variable that evaluated marijuana frequency at each assessment but did not account for previous use. Consequently, analyses included cumulative marijuana frequency as a predictor of tobacco outcomes over 2 years. Similar variables were calculated to reflect cumulative frequencies of cigarette use, overall tobacco use, poly tobacco use, and alcohol use. Because time points varied in the number of days on which use was assessed, we also created a time-varying variable that measured number of days on which use was assessed at each time point.All analyses were conducted using Stata 15.0 , with α=.05. We used bivariate tests to evaluate relationships between demographic, predictor, and outcome variables. Tests of associations between cumulative marijuana frequency and tobacco use over time were conducted by testing separate models of the association of the predictor with each time-varying outcome . Each model included cumulative alcohol frequency and assessment days as covariates, as well as terms for both linear and quadratic time and their interactions with marijuana frequency. Non-significant interaction terms that were removed and the model refit. Count outcomes were evaluated via longitudinal negative binomial regression, using Stata’s xtnbreg module, because that was a better fit to the data than linear or Poisson models. Polytobacco use, as a time-varying binary outcome, was analyzed using a longitudinal logistic regression model via the generalized estimating equations approach using xtgee in Stata. Tests of whether tobacco frequency was associated with marijuana use over time were conducted by fitting separate models of the associations of each predictor with marijuana days over time, again utilizing negative binomial models.

The proportion of the sample completing each post-baseline assessment generally decreased over time: 94% at 3 months, 88% 6 months, 85% at 9 months, 89% at 12 months, 84% at 15 months, 82% at 18 months, 78% at 21 months and 81% at 24 months. Having missing data at a specific time point was not significantly associated with predictor or outcome variables at the previous assessment. Quantity and frequency of cigarette and marijuana use over time are shown in Table 2. Bivariate assessments indicated that women, full-time students, and Asian Americans smoked fewer cigarettes than others , and therefore sex, student status, and race/ethnicity were included as covariates in subsequent hypothesis tests.The final model is shown in Table 3. All interactions were non-significant, indicating the association between marijuana frequency and total cigarettes was consistent over time; these terms were excluded from the final model. There was a significant main effect of marijuana frequency [Incidence Rate Ratio =1.11 , p<.001]. The effect size indicates that each additional time point at which recent marijuana use was reported was associated with an 11% increase in number of cigarettes. Put another way, if Participant A reported never using marijuana through the first 5 assessments, and Participant B reported recent marijuana use at each of these assessments, Participant B would be expected to report 55% more cigarettes at the 5th assessment than Participant A. The models of cigarette and tobacco frequency are shown in Table 3. Both yielded similar results as the first analysis. Cumulative marijuana frequency was a significant predictor of cigarette [IRR=1.09 , p<.001] and overall tobacco [IRR=1.09 , p<. 001] frequencies. In both cases, the association was stable over time. These analyses suggest that each additional assessment period with recently marijuana use predicted a 9% increase in both the number of cigarette days and in the number of days on which any tobacco product was used. The GEE model indicated poly tobacco use was more common among men but did not vary by race/ethnicity or student status.

There were significant interactions between marijuana use and time, suggesting the impact of marijuana frequency on poly tobacco use changed over time. More specifically, the interaction between cumulative marijuana use and time2 was a significant predictor of likelihood of concurrent use of multiple tobacco products over time . To better understand this interaction, we calculated odds ratios indicating the association between cumulative marijuana frequency and odds of poly tobacco use at each individual time point, accounting for all covariates in the original model. A plot of these odds ratios indicates the association between cumulative marijuana use and poly tobacco use was highest at baseline, when the possible values for the former were 0 and 1 . At baseline, participants who used marijuana recently were 65% more likely to report use of multiple tobacco products than those who reported no recent marijuana use. This association decreased over time as a function of the increasing range of possible values of cumulative marijuana frequency. More specifically, at each time point, the odds ratio reflects change in odds of recent multiple product use with a one-point change in the cumulative marijuana predictor. As the range of cumulative marijuana frequency increased over time, a one-point change became relatively smaller. Over the second year of observation,horticulture products each additional time point of marijuana use was associated with a 10-21% increase in the odds of poly tobacco use. The model examining cumulative frequency of cigarette smoking on marijuana frequency over time yielded a significant main effect [IRR=1.20 , p < .001; Table 4] that did not vary over time. When we modeled the association between cumulative all tobacco use and marijuana frequency, we found a significant main effect [IRR=1.22 , p<. 001] that did not vary with time. Similarly, analyses showed a significant main effect of cumulative frequency of poly tobacco use on marijuana frequency [IRR=1.19 , p<.001] but no interaction with time. These results indicate that each additional time point at which participants reported any tobacco use or poly tobacco use predicted 22% and 19% more days of marijuana use, respectively.The aim of this study was to examine whether cumulative frequency of recent marijuana use at quarterly assessments over 2 years would be associated with quantity and frequency of tobacco use among young adults who were non-daily cigarette smokers at baseline. Additionally, we sought to examine whether cumulative tobacco use over time predicted frequency of marijuana use. As expected, we found a dose-response relationship, such that participants with greater marijuana use reported greater quantity and frequency of cigarette use, and greater frequency of use of any tobacco product. Cumulative marijuana use also predicted likelihood of use of multiple tobacco products at single time points over time. Each additional time point of recent marijuana use was generally associated with a 10-20% increase in tobacco quantity/frequency. Similarly, non-daily cigarette smokers who used multiple tobacco products more frequently also reported more frequent use of marijuana. Each additional time point at which participants used cigarettes, all tobacco, or multiple tobacco products was associated with approximately 20% greater marijuana frequency.

These findings are consistent with cross-sectional studies suggesting substantial overlap between marijuana and tobacco use . However, our data also meaningfully extend previous work by demonstrating that longer-term use of marijuana is associated with greater tobacco consumption and vice versa. These associations were comparable in magnitude, suggesting a bidirectional relationship in which either may be the initial substance of interest. Given decreasing legal barriers to marijuana use, the fact that cumulative marijuana use was associated with increasing tobacco frequency in a sample of non-daily cigarette smokers is concerning, as it indicates that marijuana use may promote tobacco progression, increasing risk of poor health outcomes.For example, more frequent simultaneous use of marijuana and tobacco would lead one substance to serve as a behavioral cue for the other, and possibly to increased use of both. Additionally, learned cognitions may play a role, as demonstrated in a study examining expectancies of interactions between marijuana and tobacco effects . Higher expectations that marijuana use increases tobacco use and urges have been positively associated with tobacco and marijuana frequency, severity of marijuana use, and proportion of days of marijuana and tobacco co-use . Thus, individuals who hold these expectancies and use marijuana may experience more tobacco urges, leading to increased tobacco use over time. Further mechanisms are suggested by a recent review of neurobiological mechanisms underlying co-use . One proposed mechanism centers on synergistic effects or functional interactions, whereby use of one substance enhances the reinforcing effects of the other. Currently, the few studies that directly addressed this question have yielded conflicting findings. Some have supported the notion that nicotine enhances the effects of marijuana, while others have failed to support this relationship . As such, further study of this relationship is warranted. Another mechanism centers on compensatory effects, whereby use of one substance alleviates negative effects of the other. This hypothesis derives from evidence that marijuana withdrawal effects may be ameliorated by nicotine and vice versa . In support of this mechanism, a study of expectancies for the interactive effects of nicotine and marijuana found that higher expectations of smoking as a means to cope with marijuana urges were associated with greater marijuana cravings . Whether this influences the progression of marijuana and tobacco co-use is currently unknown and merits exploration. In all, multiple active mechanisms are likely contributing to this overlap, consistent with our finding of a bidirectional relationship. Exploration of such mechanisms and trajectories of co-use have clear clinical implications, especially in the context of smoking cessation. Evaluations of the influence of cannabis use on cessation outcomes have primarily comprised secondary analyses of cessation trials. Similarly, knowledge about the impact of tobacco on cannabis cessation is based on secondary analyses. There is preliminary evidence that pharmacotherapy and behavioral therapy may be effective treatments for cooccurring marijuana and tobacco use. Our findings converge with this evidence to encourage further systematic exploration into how marijuana-tobacco relationships impact clinical outcomes and into what may be effective at treating concurrent use. Clinically, these findings also reinforce the importance of evaluating use of both products even for intermittent users, and of incorporating evaluation outcomes into efforts to quit using one or both products. This study has some limitations.