Continuous associations of the brief Wisconsin Inventory of Smoking Dependence Motives with transitions between cigarette and e-cigarette use

Abstract

Introduction:

Measures of cigarette and e-cigarette dependence may predict product transitions, and they could be used to inform tailored messaging for cigarette cessation.

Methods:

We used data from a longitudinal cohort study of 353 adults who smoked cigarettes daily, with or without e-cigarettes, with product dependence assessed at baseline and year one and follow-up visits every two months (enrolled 2015–17 in Madison and Milwaukee, WI). We estimated continuous, non-linear associations between the brief Wisconsin Inventory of Smoking Dependence Motives scores for cigarettes (WISDM) and e-cigarettes (e-WISDM) with transition rates using a multistate transition model.

Results:

Lower WISDM scores were associated with lower rates of transitioning from cigarette-only to dual use; only 3.5% (95% CI: 1.4%, 9.3%) of those at the 10th WISDM percentile were predicted to transition to dual use compared with 25.0% (95% CI: 14.0%, 39.1%) at the 50th percentile. Higher e-WISDM scores were associated with lower rate of transitioning from dual use to either cigarette-only or e-cigarette-only use, with 50.7% (95% CI: 31.9%, 65.7%) of those who used both products predicted to continue to do so at the 50th percentile of e-WISDM compared with 91.1% (95% CI: 78.7%, 96.1%) predicted to continue both products at the 90th percentile. Ratios of e-WISDM:WISDM above 1 were associated with an increased likelihood of transitioning from dual to e-cigarette-only rather than cigarette-only use.

Conclusions:

Product dependence scores were non-linearly associated with transition rates between cigarette and e-cigarette use, which may help to guide the development of personalized, adaptive cigarette cessation recommendations.

Implications:

This work may inform the development of personalized and adaptive clinical cessation recommendations to promote harm-reducing product use transitions by using information about an individual’s absolute and relative dependence on cigarettes and e-cigarettes. Additionally, our work helps to develop predictors of transitions that are consistent across diverse e-cigarette device types.

INTRODUCTION

E-cigarettes, which are not harmless^1–3 but are likely less harmful than cigarettes,^4–6 have been promoted as tools for smoking cessation or harm reduction. The transition from exclusive smoking to exclusive e-cigarette use often occurs through a period of dual use of both products.⁷ A better understanding of transition patterns among cigarette-only, e-cigarette-only, and dual product use—and factors that predict such transitions—could inform public health projections of product use and health outcomes, enhance targeted health promotion messaging, and optimize clinical intervention strategies to reduce harm. One barrier to developing these predictions is the difficulty in assessing and interpreting an individual’s e-cigarette use patterns (e.g., frequency, intensity, total exposure). E-cigarette devices vary widely in their nicotine concentrations and other device characteristics, and they lack a clear, discrete unit of use analogous to cigarettes-per-day. Developing easy-to-measure predictors of transitions across diverse device types would aid researchers and practitioners, particularly in the rapidly evolving tobacco marketplace. One promising approach is to use measures of tobacco product dependence. Dependence-informed transition estimates could help to guide the development of personalized, adaptive cigarette cessation recommendations. For example, dependence measures could be used to predict whether a given individual who smokes will be more likely to quit smoking cigarettes by using e-cigarettes and cigarettes (i.e., via dual use), and, for those who take up e-cigarettes, to adaptively update predictions and recommendations as cigarette and e-cigarette dependence change.

Tobacco dependence measures are designed to assess the degree of nicotine addiction or dependence on tobacco and nicotine products. They help clarify the underlying nature and mechanisms of tobacco use motivation. Among the measures used to evaluate and predict addiction to smoking or vaping are the brief Wisconsin Inventory of Smoking Dependence Motives (WISDM)^8,9 and its e-cigarette counterpart, e-WISDM.^10,11 These measures reflect dependence by assessing multiple motivational domains through questions designed to capture various aspects of smoking behavior and attitudes. The WISDM and e-WISDM have excellent internal consistency (subscale and total score Cronbach’s alphas ranging from 0.81–0.96, exceeding the Nunnally and Bernstein 0.80 recommended cut-off for clinical instruments¹²), have a validated 11-factor structure, and capture multiple dimensions of tobacco dependence that are categorized into primary and secondary dependence motives.^8–10,13,14 Primary dependence motives (PDM) refer to core aspects of tobacco addiction (i.e., craving, tolerance, automaticity, and loss of control) and are often associated with heavy smoking. Secondary dependence motives (SDM), on the other hand, revolve around the use of tobacco for instrumental reasons, such as affective and cognitive enhancement.⁸ WISDM scores show strong correlations with other established measures of tobacco dependence, such as the Fagerström Test for Nicotine Dependence.^9,15,16 The questions for the WISDM and e-WISDM are analogous, e.g., participants are asked to rate “I smoke without deciding to” vs “I vape without deciding to” on a scale of “not true of me at all” to “extremely true of me.”

In this analysis, we estimated non-linear associations between overall WISDM and e-WISDM scores (and the PDM and SDM subscales) and transition propensities (rates) and probabilities, applying a multistate transition model to data from the Exhale study, a two-year longitudinal cohort study focusing on the use of cigarettes and e-cigarettes with bi-monthly follow-ups and two WISDM assessments one year apart. The multistate transition modeling approach has been employed in various studies of tobacco product use, estimating transition patterns as a function of an individual’s characteristics.^7,17–21 Here, we investigated how WISDM and e-WISDM scores—and their ratio—were associated with cigarette and e-cigarette cessation and initiation among those who currently smoke cigarettes only or who both smoke cigarettes and vape e-cigarettes. Our previous analysis of this cohort employed categorical dependence categories to capture crude associations between product dependence and transitions, and we identified potentially non-linear and even non-monotonic relationships. For example, we found a greater increase in transition rates for dual to cigarette-only use between those with low and medium WISDM scores than between those with medium and high scores. Additionally, medium e-WISDM scores predicted a greater rate of transitioning from dual to e-cigarette-only use than did low or high e-WISDM scores.²² These results motivated this analysis estimating non-linear associations, treating the WISDM and e-WISDM as continuous measures. This approach provides a framework for developing a more nuanced understanding of how dependence measures are associated with transitions in tobacco use and how these associations differ for those smoking cigarettes only compared to those using both cigarettes and e-cigarettes.

METHODS

This secondary analysis used data from the previously described Exhale study, a longitudinal observational cohort study of 422 adults who smoked cigarettes daily or used both cigarettes and e-cigarettes.^23,24 We hypothesize that the associations between dependence measures and product transitions in this sample are non-linear. We consider this analysis to be exploratory: while our hypothesis is consistent with the literature and prior work, the original study was not designed or powered for this present analysis, and this analysis was not pre-registered. However, the study’s longitudinal design with short-term follow-up, described below, offers a valuable opportunity to demonstrate preliminary effect sizes that could motivate and inform further investigation, study design, and hypothesis testing.

Study population and measures

Participants were recruited from the greater Madison and Milwaukee, Wisconsin areas through advertisements between October 2015 and July 2017. Participants were adults who did not intend to quit smoking or vaping in the next 30 days and who were not currently in treatment for psychosis or bipolar disorder. Participants had to either be using cigarettes only, defined as smoking more than 5 cigarettes daily in the past 6 months and not using e-cigarettes in the past 3 months, or using both cigarettes and e-cigarettes, defined as smoking daily and using nicotine-containing e-cigarettes more than once a week in the past 3 months. Enrolled participants were followed every 2 months for up to 2 years, but participants’ WISDM and e-WISDM scores were assessed only at baseline and at 1 and 2 years of follow-up. Higher values of WISDM and e-WISDM (or their PDM or SDM subscores) indicate greater dependence. We assessed whether WISDM scores were different for those using cigarettes only vs using both cigarettes and ENDS at baseline with a Wilcoxon rank sum test. We also assessed crude associations between either dependence score with frequency of cigarette and e-cigarette product use, associations between demographic variables and tertiles of each dependence score, and the correlation of the two dependence scores among people smoking and vaping.

Based on self-reported past-30-day smoking and vaping, we classified participants into one of four product-use states at each time point: non-current use, cigarette-only use, e-cigarette-only use, and dual use of cigarettes and e-cigarettes. Our analysis focused on transitions from cigarette-only to non-current use or to dual use, and from dual-use to cigarette-only or to e-cigarette-only use, considering the small sample sizes of other states (because of the study design). Transition analysis considers pairs of observations: a “from” observation and a “to” observation. Because of the nature of transitions, we excluded participants with fewer than two observations total. Additionally, since we were estimating associations between transitions and the dependence scores, we excluded transition pairs where the corresponding dependence score was missing from the first time point (the “from” observation). For example, if a participant only smoked cigarettes at baseline, we included their baseline and 2-month follow-up data only if they had a WISDM score at baseline. Similarly, if a participant used both cigarettes and e-cigarettes at 12 months, we only included their 12- and 14- month follow-up data if they had both WISDM and e-WISDM scores at 12 months. Transitions from cigarette-only or dual use at other time points (e.g. between 2- and 4-month follow-up visits) are not included in this analysis because we did not have dependence measures at the “from” observation. However, because multistate transitions models, discussed below, look at transitions between all states holistically, we included all transitions from non-current use and e-cigarette-only use regardless of WISDM or e-WISDM score or time point, as we did not assess associations between product dependence and transitions from those states. In total, 353 participants and 737 observation pairs were included (the number of participants decreased from 422 to 371 because of the requirement to have paired observations and further decreased to 353 because of missingness in the relevant dependence measures).

Transition modeling

We used a Markov multistate transition model (MSTM) to estimate how certain transition rates between the product-use states depended on the product dependence scores. Informed by prior work,⁷ we assumed that participants transitioned one product at a time. Our approach assumed transitions between product use states are governed by a continuous-time, stochastic process where transition propensities (rates) to an individual’s “to” state depend on an individual’s current “from” state and any relevant covariates but not on previous states.²⁵ Although participants were followed every 2 months, the model assumes that transitions could occur at any time between follow-up assessments. We estimated the instantaneous transition rates from the model that collectively gave rise to the transitions probabilities that best fit the observed transitions.

In brief, the transition rate matrix is the matrix $Q$, containing entries $q_{ij}$ that represent instantaneous transition rates between states $i$ and $j$. (The entries are defined to be 0 where transitions cannot occur, and the diagonal entries are given by $q_{ij}=-{\sum }_{j\ne i}{q}_{ij}$). The transition matrix for a given set of covariates defines a transition probability matrix $P=\mathit{\text{expm}}\left(Q{\Delta }_t\right)$ (where $\mathit{\text{expm}}\left(\right)$ denotes a matrix exponential), giving the modeled probability of being in each state after time ${\Delta }_t$. We calculate the likelihood of the data given $P$ and maximize the likelihood as a function of the transition rates. Further technical details of the method are given in Jackson (2011).²⁶

While categorization of continuous variables is commonly used to estimate non-linear associations, that approach does not estimate the actual form of the association and creates several statistical and inferential issues.^27–29 Here, we applied a natural cubic spline estimator to model the continuous relationship between the continuous WISDM score covariates and the transition rates; the choice of a spline estimator minimizes information loss from categorizing while allowing for a flexible, nonlinear association. Natural cubic splines are piecewise polynomial functions comprising piecewise cubic segments, defined by a set of knots.³⁰ Natural splines are also constrained to be linear at the boundaries of the covariate values. We chose knots based on one-third and two-thirds quantiles of the WISDM and e-WISDM distributions, which is a typical choice for knot position.³¹ We also calculated confidence intervals for the cubic spline associations.³⁰ To validate the cubic spline estimators, we compared their results with transition rates calculated from participants with WISDM or e-WISDM scores within each 10 point interval between 10 and 80 for WISDM (range: 11–77) and between 10 and 70 for e-WISDM (range 11 –71). Additionally, to avoid computational issues associated with negligible transition rates, we fixed the transition rate to 0 at an edge of the dependence score range if there were no observed transitions scores of ≤30 (for the low bound) or ≥60 (for the upper bound). The associations between WISDM and e-WISDM and transition rates were estimated in separate, univariable models. We repeated these analyses with WISDM and e-WISDM PDM and SDM scores (each ranging 1 to 7), separately, with a single knot at the 50% quantile because of the smaller range. Finally, we estimated transition rates from dual use as a function of the e-WISDM:WISDM ratio. Transition rates and probabilities from each model were compared to those estimated in an unadjusted model without covariates, i.e., the model just estimating mean rates between the product use categories overall.

Because we estimated continuous associations, our model can estimate transition rates and probabilities for any specific value of WISDM or e-WISDM scores. To facilitate interpretation for this analysis, we chose to estimate transition rates at the 10th, 50th, and 90th percentiles of the WISDM and e-WISDM scores and at e-WISDM:WISDM ratios of 0.5, 1.0, and 1.5, and we converted these rates into one-year transition probabilities.

The analysis was conducted using the msm package (v1.8.2) in R (v4.4.2).²⁶ This secondary analysis was not regulated as human subjects research (University of Michigan Institutional Review Board HUM00179890).

RESULTS

The 353 participants included in this study were 51% male and 49% female, with a median age of 40 years (range: 18–72; 25% aged 18–29 years, 44% aged 30–49 years, and 31% aged 50 years and older). Participants self-identified race and ethnicity: 67% non-Hispanic White, 23% non-Hispanic Black, 5% Hispanic, and 8% another non-Hispanic race or multiracial. Among the 257 participants aged 25 years and older, 61% had completed more than high school, 31% had completed high school or equivalent, and 6% had completed less than high school. About half of participants (52%) self-reported diagnosis or treatment of a psychiatric condition (which was defined broadly to include anxiety, depression, attention deficit disorder, bipolar disorder, schizophrenia, panic disorder, and post-traumatic stress disorder). A more detailed description of the broader sample is given in Piper et al.²³ A comparison of demographic variables across tertiles of WISDM and e-WISDM scores at baseline is given in Table S1. There is potential for confounding of the relationship between dependence scores and transitions by these variables that will need to be addressed in future work.

Histograms of WISDM scores and e-WISDM scores at baseline, distinguishing between cigarette-only use participants and dual use participants, are given in Fig 1. WISDM scores were approximately normally distributed, with similar distributions in the two product use groups (p=0.44 for a Wilcoxon test of WISDM scores between cigarette-only and dual use), with median values around 40. Across all smoking participants, the 10^th, 50^th, and 90^th percentiles of WISDM scores were 26, 44, and 62, respectively. In contrast, the distribution of e-WISDM scores among those in the dual use of cigarettes and e-cigarettes were right skewed, with a median score of 28, and 10^th and 90^th percentiles of 16 and 51. Associations between WISDM and e-WISDM scores and cigarettes-per-day or e-cigarette use frequency among participants using cigarettes only, e-cigarettes only, or both products are given in Fig S1. Among those smoking cigarettes only or using both cigarette and e-cigarettes, the mean WISDM scores for those smoking 2, 10, and 30 cigarettes per day were 30, 40, and 55, respectively. For participants using both products, e-WISDM scores increased on average from about 20 for those using e-cigarettes once per day to around 35 for those using 10 times per day, where it plateaued. Among participants using both products at baseline, correlation between WISDM and e-WISDM scores was low (0.253).

Fig 1. Distributions of Wisconsin Inventory of Smoking Dependence Motives (WISDM) score and e-cigarette e-WISDM scores by product use at study baseline.

Data were collected in Wisconsin, US in 2015–19. WISDM scores are given in orange and e-WISDM scores in blue.

In this sample, we observed 308 transitions pairs where the participant was only smoking cigarettes in the “from” state; of these, the participant remained only smoking cigarettes after two months in 289 pairs, transitioned to dual use in 15 pairs, transitioned to non-current in 3 pairs, and transitioned to e-cigarette-only use in 1 pair. We observed 219 transition pairs where the participant was smoking and vaping in the “from” state; of these the participant was still using both products in 196 pairs, was only smoking in 17 pairs, was only vaping in 5 pairs, and transitioned to non-current use in 1 pair. These transitions were not evenly distributed across the dependence scores (Fig S2). In the unadjusted transition model not accounting for either dependence measure, the estimated 1-year transition probabilities from cigarette-only use were 3.7% (95% CI: 3.7%, 9.7%) to non-current use, 73.7% (95% CI: 61.7%, 82.0%) remaining smoking only, 1.7% (95% CI: 0.7%, 3.9%) to e-cigarette-only use, and 20.9% (95% CI: 13.7%, 31.5%) to dual use. From dual use, the estimated 1-year transition probabilities were 1.5% (95% CI: 0.6%, 3.6%) to non-current use, 31.7% (95% CI: 21.2%, 43.9%) to cigarette-only, 8.2% (95% CI: 4.0%, 16.4%) to e-cigarette only, and 58.6% (95% CI: 45.9%, 69.0%) remaining using both products.

Continuous, nonlinear associations between WISDM and e-WISDM scores and the transition rates are given in Fig 2. There was generally considerable uncertainty in the continuous estimates, particularly for the uncommon transitions. We observed increasing transition rates for cigarette-only to dual use (Fig 2B) with increasing WISDM score (at least up to about WISDM scores of 70), indicating that those with higher cigarette dependence were more likely to adopt e-cigarettes and use both products. (Note that the shrunk confidence interval at the lower end of Fig 2B and other subfigures is a consequence of our fixing the rate to 0 to avoid computational issues from a lack of transitions, as described in the Methods). Conversely, the rate of transitioning from dual use to e-cigarette-only use decreased with increasing WISDM score (Fig 2D). We found that WISDM score was not clearly associated with either the cigarette-only to non-current use transition (Fig 2A) or the dual use to cigarette-only use transition (Fig 2C). We found that e-WISDM scores by themselves were not clearly associated with transitions away from dual use across most of the span of scores (Fig 2E, 2F), but those with the highest scores (above 60, approximately) were unlikely to transition, though the sample size was small (n=4 with e-WISDM≥60 at a “from” state). Associations between rates for PDM and SDM and their e-WISDM counterparts and transitions rates were qualitatively similar to those for overall WISDM and e-WISDM scores (Fig S3 and S4). Transition rates from dual use to cigarette-only use decreased for participants with e-WISDM:WISDM ratios above 1 (Fig 3), while transition rates to e-cigarette-only use appeared to increase, though not statistically significantly.

Fig 2. Transition rates from cigarette-only or dual use of cigarettes and e-cigarettes as a function of Wisconsin Inventory of Smoking Dependence Motives score for cigarettes (WISDM) and e-cigarettes (e-WISDM).

Rates were estimated using a multistate transition model. The colored points (orange for WISDM, blue for e-WISDM) represent transitions rates estimated on a subset of participants with WISDM or e-WISDM scores in a 10-point range (with point area proportional to sample size), the curved line and ribbon represent the best fit association and confidence intervals, and the dotted line is the estimated transition rate for that transition when not accounting for the dependence measures. WISDM scores are given in orange and e-WISDM scores in blue.

Fig 3. Transition rates from dual use of cigarettes and e-cigarettes as a function of the ratio of the Wisconsin Inventory of Smoking Dependence Motives scores for cigarettes (WISDM) and e-cigarettes (e-WISDM).

Rates were estimated using a multistate transition model. The points represent transitions rates estimated on a subset of participants with e-WISDM:WISDM ratios in a subset of the full range (with point area proportional to sample size), the curved line and ribbon represent the best fit association and confidence intervals, and the dotted line is the estimated transition rate for that transition when not accounting for the dependence measures.

We estimated the one-year cumulative transition probabilities for individuals at the 10th, 50th, and 90th percentiles of the dependence scores to better understand how varying levels of dependence might influence these transitions (Fig 4). Estimates indicated that, among participants only smoking cigarettes, as WISDM scores increased, so did the probability that participants would transition from cigarette-only use to dual use, with probabilities increasing from 3.5% (95% CI: 1.4, 9.3%) at the 10th percentile to 25.0% (95% CI: 14.0, 39.1%) at the median and to 32.7% (95% CI: 17.2, 54.0%) at the 90th percentile (Fig 4A, C, and E). In contrast, transitions for participants in the dual use group were largely stable across the distribution of WISDM scores, with approximately 60% predicted to continue using both products and 30% predicted to transition to cigarette-only use regardless of WISDM score. However, the predicted probability of transitions to e-cigarette-only use decreased from 12.6% (95% CI: 5.9, 26.5%) at the median to 0.6% (95% CI: 0.2, 2.0%) at the 90th percentile of WISDM scores.

Fig 4. One-year transition probabilities at 10th, 50th, and 90th percentiles of Wisconsin Inventory of Dependence Motives score for cigarettes (WISDM) and e-cigarettes (e-WISDM).

Score values in subfigure titles indicate the score value at that percentile. A “transition” from a state to itself (e.g., dual use to dual use) is the sum of the probability of not transitioning (persistence) and transitioning away and back again in one year.

Predicted transition probabilities from dual use were similar for the 10th and 50th percentiles of e-WISDM, with about 30–40% of participants in the dual use group predicted to transition to cigarette-only use (42.1% (95% CI: 21.9, 67.4) at the 10^th percentile and 33.8% (95% CI: 19.1, 54.3) at the 50th percentile) and about 50% predicted to remain in the dual use group (52.0% (95% CI: 25.3, 71.5) at the 10^th percentile and 50.7% (95% CI: 31.9, 65.7) at the 50th percentile). However, the predicted probabilities of these transitions shifted to 8.1% (95% CI: 3.2, 19.3%) and 91.1% (95% CI 78.7, 96.1%), respectively, at the 90th percentile of e-WISDM (Fig. 4B, D, and F). Predicted transition probabilities from dual use were similar for the e-WISDM:WISDM ratios of 0.5 and 1.0, with about 30% of participants in the dual use group predicted to transition to cigarette-only use and about 10% predicted to transition to e-cigarette-only use. The predicted probabilities of these transitions shifted to 3.8% (95% CI: 1.4, 13.5%) and 38.7% (95% CI 6.4, 80.2%), respectively, for an e-WISDM:WISDM ratio of 1.5 (Fig. 5).

Fig 5. One-year transition probabilities at three values of the ratio of the Wisconsin Inventory of Smoking Dependence Motives scores for cigarettes (WISDM) and e-cigarettes (e-WISDM).

Score values in subfigure titles indicate the score value at that percentile. A “transition” from a state to itself (i.e., dual use to dual use) is the sum of the probability of not transitioning (persistence) and the probability of transitioning away and back again in one year.

DISCUSSION

In this exploratory analysis of transitions in product use in a longitudinal cohort, we found that cigarette and e-cigarette dependence scores were informative for predicting product use transitions. Cigarette dependence was positively associated with a transition to dual use. Participants with lower dependence on cigarettes were predicted to be less likely to transition from cigarette-only use to dual use: compared to the estimated 20.9% (95% CI: 13.7%, 31.5%) transitioning in one year not accounting for cigarette dependence, an estimated 3.5% (95% CI: 1.4, 9.3%) of those in the 10th percentile of cigarette dependence would transition to dual use, compared with 25.0% (95% CI: 14.0, 39.1%) at the 50th percentile and 32.7% (95% CI: 17.2, 54.0%) at the 90th percentile. Cigarette dependence was largely not predictive of transitions among those both smoking and vaping, although higher WISDM scores were associated with a lower transition probability to e-cigarettes only. Dependence on e-cigarettes was associated with reduced transitions from dual use to smoking only at the highest levels of dependence, with the estimated probability of this transition decreasing from 33.8% (95% CI: 19.1, 54.3) at the 50th percentile to 8.1% (95% CI 3.2, 19.3%) at the 90th percentile, compared to 31.7% (95% CI: 21.2%, 43.9%) when not accounting for dependence. Among those using both products, the ratio of e-cigarette to cigarette dependence was predictive; higher e-WISDM:WISDM ratio values were associated with reduced likelihood of returning to cigarette-only use and perhaps associated with an increased likelihood of transitioning to e-cigarette-only use (which was rare in this sample).

Participants who smoked cigarettes only and who had lower cigarette dependence (WISDM) scores exhibited a reduced rate of transitioning to dual use by adopting e-cigarettes compared to those with higher dependence scores. This relationship may be because participants with lower WISDM scores may have comparatively little need or desire for additional nicotine intake via e-cigarettes. We found positive crude associations between cigarette dependence and frequency or intensity of product use in this analysis, and there is considerable evidence for such a relationship in the literature.^8,32,33 It may also be that those with lower dependence have lower perceived need for nicotine and lower motivation to quit smoking or to mitigate the associated harms by switching to e-cigarette use (via dual use). High levels of dependence appeared to sustain current use patterns. Those who used both products and had higher cigarette dependence were less likely to quit smoking, while those with higher e-cigarette dependence also showed a lower probability of transitioning to single product use (either cigarette-only use or e-cigarette-only use), a pattern that has also been suggested in previous work.³⁴ A high level of dependence on either product was associated with a reduced likelihood of transitioning from dual to e-cigarette-only use, a harm reducing transition, although the small number of such transitions reduced confidence in the specific observed patterns.

Our work may inform the development of personalized and adaptive approaches to harm reduction and cigarette cessation in clinical interventions, such as by encouraging the pathway to cigarette cessation (e.g., direct cessation to non-current use vs cessation via dual use to e-cigarette only use) most likely to succeed depending on the patient’s current level of dependence. Specifically, we found that those who smoke cigarettes only and have moderate cigarette dependence may be the most likely to successfully quit cigarettes via dual use, whereas those with low dependence are the most likely to be able to quit directly, and those with high dependence are more likely to persistent in dual use rather than quit cigarettes. These exploratory results would need to be replicated in future research, but, if so, they may have value in educating the public via public health campaigns. For instance, such campaigns might communicate that for individuals who are highly dependent on cigarettes, taking up e-cigarettes in addition to cigarettes may not help them quit smoking. For such individuals, dual use is not a pathway to being smoke-free. This message might reduce the pattern of highly dependent people being especially likely to take up e-cigarettes, given that e-cigarettes are very unlikely to help them quit. Similarly, among individuals who become very highly dependent on e-cigarettes, the majority will continue dual use and only a small percentage will stop cigarette use. Future work could shed light on the extent to which e-cigarettes serve as a complementary product rather than a substitute,²³ particularly for those with stronger overall nicotine dependence. In transitioning from cigarette-only to dual use, e-cigarette use may fulfill their nicotine needs but it is uncertain whether it contributes to reducing or maintaining cigarette dependence. Additionally, although we did not observe meaningful differences between the associations between PDM or SDM and transitions, they may be clinically useful in crafting effective messaging for behavior change.

One strength of this study lies in the Exhale study's short-term (2 month) sampling schedule, which contrasts with the annual follow-ups in studies such as Tobacco Use Supplement to the Current Population Survey (TUS-CPS) and the Population Assessment of Tobacco and Health (PATH) Study. The more frequent follow-up offers more precise data to capture short-term transitions among those using tobacco products. Further, our use of a multistate modeling framework combined with a spline-based approach to estimating associations with continuous variables allows both a holistic look at transitions while reducing the information loss associated with categorization of continuous variables.^27–29 Limitations of this study include 1) the relatively small sample size (although follow-up was every two months, dependence was only measured yearly, reducing the number of usable transition pairs), especially for the highest and lowest values of product dependence, 2) potential confounding by age, sex, race, and psychiatric history, which we were not able to account for because of low statistical power, but which should be addressed in future work, 3) the assumption of constant dependence over the 2-month follow-up period, and 4) the timeframe of the study (2015–19), which was largely prior to the marketplace transition to nicotine-salt-based pod e-cigarettes, such as JUUL. Fortunately, even though transitions may differ by device characteristics such as nicotine concentration,²² dependence may mediate this relationship (i.e., nicotine concentration affects dependence and dependence affects transitions), and thus the associations between dependence and transitions may largely be agnostic to device characteristics.¹⁴ Additionally, while they are excellent tools for an in-depth assessment of dependence, the length of the WISDM and e-WISDM assessments (the short forms include 37 items) may limit the feasibility of clinical use.

Conclusion

In this study, we found nonlinear associations between measures of tobacco dependence and transitions in the use of cigarettes and e-cigarettes. A better understanding of the association between transitions and tobacco dependence in a current sample using modern e-cigarettes is needed (alongside other metrics such as frequency of use and biomarkers of exposure) to provide important insights for shaping tobacco control strategies and developing personalized intervention strategies and public health education campaigns to support cigarette cessation.

Data availability statement

Data and code may be available by contacting the corresponding author with a reasonable request. IRB approval and a DUA may be required for data access.