Longitudinal Tobacco Use Transitions Among Adolescents and Young Adults: 2014–2016

Longitudinal Tobacco Use Transitions Among Adolescents and Young Adults: 2014–2016 Abstract Introduction Among youth, the frequency and prevalence of using more than one tobacco (small cigar, cigarette, and hookah) or nicotine-containing product (e-cigarettes-ENDS) are changing. These shifts pose challenges for regulation, intervention, and prevention campaigns because of scant longitudinal data on the stability of use patterns in this changing product landscape. Methods A nationally representative longitudinal survey of 15- to 21-year olds (n = 15,275) was used to describe transitions between never use, noncurrent use, and past 30-day use of combustible tobacco, e-cigarettes (ENDS), and dual use of both kinds of products. A multistate model was fit to observations collected every 6 months across 2.5 years to estimate the probability of transitions between states (TPs), the average time in state (sojourn time), and the effect of age on transitions. Results Current state strongly predicted future state over time intervals of 1 year or less, but only weakly predicted future state at longer intervals: TP to noncurrent use was higher for ENDS-only than combustible-only users over a 6-month interval but was similar for both groups over a 2-year interval. Sojourn time was significantly longer for combustible-only (0.52 years) and dual use (0.55 years) than ENDS-only use (0.27 years); older youth were more likely than younger youth to stay combustible tobacco users or noncurrent users. Conclusions The dynamics of transitions between combustible tobacco products and ENDS in a population of youth and young adults suggest that policy and prevention efforts must consider the frequent changes and instability over a 1-year or less time period in use patterns among young people. Implications The study addresses an urgent need in public health for timely information on how youth and young adults use tobacco and nicotine products. We found that youth, particularly adolescents, moved frequently between using ENDS and combustible tobacco products either alone or together. Importantly, the utility of current-use states for predicting future use states declined for time horizons longer than 1 year. Our results demonstrate a need for caution in interpreting product transitions. Longitudinal data with frequent observations and coverage of a wide range of possible product types is required to fully characterize usage patterns in youth. Introduction The prevalence of cigarette use has declined to record lows in the past 5 years. Over that time frame, prevalence of past 30-day cigarette use fell from 19.8% to 4.9% in 10th graders and from 17.1% to 10.5% in 12th graders.1 Contemporaneous with that decline has been an increasingly diverse tobacco and nicotine product landscape. As youth and young adults engage in trial and intermittent use, particularly before the onset of established smoking, use of products other than cigarettes has become more prevalent.2,3 Among high-school students, from 2011 to 2016, prevalence of hookah use increased from 4.1% to 7.2% and e-cigarette (ENDS) use increased from 1.5% to 11.3%.4 Among adults, 18- to 24-year olds represent a significant portion of other product users, making up 55.8% of hookah and 24.3% of ENDS users in 2014.5 Changes in the tobacco and nicotine product landscape have also influenced an increase in the percent of youth and young adults who use two or more products (ie, dual use or poly use).3,4,6–8 In 2016, 20.2% of US high schoolers reported any past 30-day product use, and of those, 47.2% were poly users, representing 9.6% of high schoolers.4 Even youth who are not past 30-day poly users are likely to move between different kinds of products. In 2015, while 4.0% of middle and high-school students reported exclusive past 30-day e-cigarette use, 61% of them had previously used at least one other tobacco product.8 Understanding the patterns of different product use among youth and young adults, and the temporal dynamics of how youth transition between different products, is critical to identifying priorities in tobacco control policies.3,9–13 To date, most longitudinal studies have focused on transitions into a single product, typically cigarettes. For example, several studies have looked at the relative risk of initiating cigarette smoking for individuals who have never used any product compared with individuals who have ever used ENDS but not cigarettes. These studies have found that ever ENDS use is associated with ever cigarette use.13–19 However, these studies are limited to specific populations observed over periods of a year or less. This means the stability of these relative risk patterns over longer periods of time is not known while other relevant transitions (eg, into and out of nonuse) are not modeled. Few studies have investigated product transitions using a framework that takes into account how participants might start and stop use of multiple different products across time.3,8 In a study of California high-school freshman, Huh and Leventhal used latent transition analysis to identify three classes of students based on past 6-month product use.10 At the follow-up 6 months later, between 13% and 19% of students had changed classes, highlighting the utility of methods that account for transitions into and out of use. Latent class models allow for data-driven classification of participants into groups. A multistate model is an alternative approach that allows for a priori specification of the categories of interest. Kaufman and colleagues used a multistate model to estimate transition probabilities into and out of single and dual use of smokeless tobacco and cigarettes using Add Health data from 1995 to 2009.11 They found that, over a 1-year interval, cigarette-only users and noncurrent users of any products were about equally likely to transition to using smokeless tobacco. Further, they found noncurrent use and cigarette-only use were the most stable states. The authors, along with several other groups, have called for greater application of these models to more product types, including ENDS.9,11,12 The goal of this study was to examine transitions into and out of tobacco and nicotine product use states using a large, nationally representative sample of youth and young adults with frequent observations over 2.5 years. We use a multistate modeling methodology and focus on transitions between single and dual use of combustible tobacco products and ENDS.20 The model is used to address two primary research questions: (1) What does an individual’s current use state say about the types of products they are likely to be using (or not using) 6 months, 1 year, or 2 years in the future? and (2) How long do individuals tend to stay in each use state before adding, dropping, or switching products, or stopping use for at least 30 days? We also examine whether age may influence transitions between tobacco products. Method Sample and Survey Procedures Data for this study were drawn from the Truth Longitudinal Cohort (TLC), a probability-based, nationally representative survey of youth and young adults. TLC participants were primarily recruited via address-based sampling, with subsamples recruited through random digit dialing and from GfK’s KnowledgePanel. Retention rates across all waves ranged from 61% to 71%. Additional details regarding the methodology of the TLC, including details about response rates, have been published elsewhere.21 Participants who completed two or more surveys across six waves of data collection were included in the analyses (n = 15,275). The majority of participants (n = 13,892) entered the study at Wave 1 between April and August 2014, and were followed for five additional waves at 6-month intervals, with Wave 6 data collected from January to April 2017. At Waves 2 through 5, refreshment samples of ~1000 additional participants, drawn from the same population as the original sample, were added to the total sample and followed for all remaining waves (eg, the refreshment participants who entered at data collection period Wave 3 were invited to participate in follow-up surveys for Waves 4 through 6). Because participants entered the study at different times, the term Time 0 in this report refers to the first observation for each individual, regardless of when they joined the study. The study protocol was approved by Chesapeake Institutional Review Board. Measures Products included in the combustible tobacco category were cigarettes, hookah, and all types of cigars. Products included in the ENDS category were e-cigarettes and e-hookah (see Appendix for survey questions). At each wave, all participants with data for that wave were classified into one of five mutually exclusive product use states: (1) Never use: Those who reported never having used any ENDS or combustible tobacco product; (2) Noncurrent use: Those who reported having used an ENDS or combustible product, but not in the past 30 days; (3) ENDS-only use: Those who reported having used an ENDS product in the past 30 days, but reported no past 30-day use of a combustible tobacco product; (4) Combustible-only use: Those who reported having used a combustible product in the past 30 days, but reported no past 30-day use of ENDS products; (5) Dual use: Those who reported having used both a combustible and an ENDS product in the past 30 days. Observation times for each participant were calculated for each wave as the difference in years between the time stamp for when they submitted the survey for that wave and the submission time stamp for their survey at Time 0 (ie, study entry). This yielded a continuous variable with peaks around each of the planned survey distribution times (ie, 0, 0.5, 1, 1.5, 2, and 2.5 years). Participant age at Time 0 was included as a covariate, and its effects were estimated and analyzed for all transitions in the model (see below). Three other individual characteristics were included as control variables, but their effects were not specifically analyzed: gender, race (white/not-white), and parent education (more than high school/high-school degree or less) were entered in the model as two-level categorical variables. Inclusion of more covariates, or additional levels of the included covariates, led to model nonconvergence because of small cell sizes within the transition matrix. Model Specification Continuous-time, multistate Markov models were fit to the data using the msm package in R.22 The continuous time model allows transitions between states to occur at unknown times between the observation timepoints, including multiple transitions between observations. The model uses all data available (ie, from respondents with two or more observations) to estimate the transition intensity matrix, which represents the instantaneous risk of transitions between states. The transition intensity matrix is then used to compute the probability of transitioning between all pairs of states for specified time intervals, and the mean duration of time spent in each state (ie, the sojourn time). The instantaneous transitions between states are shown in Figure 1, and initialization and intensity matrices are provided in Appendix Tables 1–5. One-way transitions were allowed from Never use into ENDS-only and Combustible-only: Once someone has tried a product, they cannot go back to never having tried a product. A direct transition from Never use to Noncurrent use was not allowed, as Never users would need to have tried one or more products before becoming a Noncurrent user. In other words, it would be implausible for concurrent reporting of never having used a product and having used a product but not in the last 30 days (Noncurrent use). Rather, the participant would need to transition from Never use to Noncurrent use via one (or more) of the current use states. Because our data were collected at 6-month intervals, it is possible for participants to report Never use at one observation and Noncurrent use at the next observation. Figure 1. View largeDownload slide Structure of allowed transitions for the multistate model. Unidirectional paths between states are shown with dashed lines and single-headed arrows; bidirectional paths are shown with solid lines and double-headed arrows (see text). Figure 1. View largeDownload slide Structure of allowed transitions for the multistate model. Unidirectional paths between states are shown with dashed lines and single-headed arrows; bidirectional paths are shown with solid lines and double-headed arrows (see text). The instantaneous transition matrix also excluded the transitions Never use to Dual use and Dual use to Noncurrent use. While these transitions are theoretically possible, they were extremely rare in our data set and so were set to 0 to improve model convergence and allow for inclusion of covariates. Missing data for use state and covariates were left as missing for model maximum likelihood estimation. A baseline model without any covariates was first fit to the data, followed by a series of sequential models adding one covariate at a time with the covariate effect estimated for all allowed transitions without constraint. The specific command used for model fitting are in the Appendix. Log-likelihood ratio goodness-of-fit tests revealed that each covariate improved model fit and that the model with all covariates was the best fit to the data (all p < .001, full test results in Appendix Table 3). Results are reported from the full model. Results Sample Description The sample covered the targeted ages well, with a fairly uniform distribution across the 15- to-21-year-old range at Time 0 (M = 18.61, SD = 2.11, range = 12.59 - 23.77). The distributions of the control variables were as follows: Gender: 48.7% female, 51.3% male (0 missing gender information); Race: 66.6% white, 33.1% nonwhite, 0.003% missing; Parent education: 84.0% greater than high school, 14.6% high-school graduate or less, and 1.4% missing. Participants were unevenly distributed across the five possible use states. At Time 0, the sample (n = 15,275) was 60.5% Never users, 21.3% Noncurrent users, 3.1% ENDS-only users, 8.9% Combustible-only users, and 6.1% Dual users. Over time, the largest changes in sample prevalence were seen in the Never use and Noncurrent use states, with the relative proportion of participants who reported using a combustible or ENDS product in the past 30 days remaining fairly constant. The Time 2.5 sample (n = 8571) was 45.1% Never, 35.7% Noncurrent, 3.3% ENDS-only, 11.1% Combustible-only, and 4.9% Dual. Probability of Transitioning Between States Over Time To examine use patterns over time, we estimated the transition probability between all pairs of states across time intervals between 0.5 and 2.0 years (Figure 2). The transition probability from State X to State Y represents the conditional probability that a participant will be in State Y at the end of the time interval, given that they were in State X at the beginning of the interval. Because states are mutually exclusive, the transition probabilities from any state for a given time interval must sum to 1. When estimating the transition probabilities, the model takes into account that a participant may make multiple transitions during the time interval. Figure 2. View largeDownload slide Transition probabilities across intervals of increasing duration. Each panel represents a different start state, and the series represent the probability of being in the each of the five states at a certain time in the future conditioned on that start state. Probabilities were estimated at the mean of all covariates. Matrix of model estimates is also provided in Appendix Table 6. Figure 2. View largeDownload slide Transition probabilities across intervals of increasing duration. Each panel represents a different start state, and the series represent the probability of being in the each of the five states at a certain time in the future conditioned on that start state. Probabilities were estimated at the mean of all covariates. Matrix of model estimates is also provided in Appendix Table 6. Transition probabilities estimated at the mean of all three covariates are illustrated in Figure 2. Never users pattern quite differently than all current and noncurrent users. The probability of a Never user moving into a current use state is low across all time windows, with a maximum of 6.2% probability of moving to Combustible-only use after 2 years. Setting aside the Never users, the most probable future state is highly variable across the other use states for 6-month intervals but diminishes for intervals of 1 year or more. For example, the probability of being a noncurrent user across any 6-month interval has a range of 61.6% (min = 18.2% for Dual users, max = 79.8% for Noncurrent users). However, over a 2-year interval, the range in probabilities is reduced to 9.3% (min = 56.1% for Dual users, max = 65.4% for Noncurrent users). The probability of transitioning to the Combustible-only state is similarly reduced over time, with a range of 35% for a 6-month interval but only 4% for a 2-year interval. The convergence of the transition probabilities over time across the use states indicates that individuals’ current state is much more informative for predicting their product use 6 months in the future compared with 2 years in the future. Regardless of whether participants are current or noncurrent users of combustibles, ENDS, or both, the most likely state 2 years in the future is noncurrent use (ie, no use in the past 30 days). For individuals who do remain or become active users 2 years in the future, the most likely use state is Combustible-only. Figure 2 also highlights the differences between current use states at intervals shorter than 12 months. During these shorter intervals, there is a gradient of risk across the ENDS-only, Combustible-only, and Dual use states. Across those three states, ENDS-only users are most likely to transition to noncurrent use, and Dual users are least likely, with Combustible-only users falling in between. Additionally, over those same intervals, ENDS-only users are least likely to be using combustibles (~23% probability) at the later time point, Dual users are most likely (~73%), and Combustible-only users fall in between (~60%), with percentages representing combined transition probabilities into Combustible-only and Dual use. Average Time in State Time-in-state, or sojourn time, was estimated for each use state at the mean of all covariates. Estimated sojourn time was approximately 3 months for the ENDS-only state (0.27 years, 95% CI: 0.25–0.29); 6 months for Combustible-only (0.52 years, 95% CI: 0.49–0.55) and Dual use (0.55 years, 95% CI: 0.52–0.58); and 18 months for Noncurrent use (1.50 years, 95% CI: 1.43–1.58). The relatively short sojourn times in the current use states suggest that 15- to 21-year olds are moving frequently between types and combinations of products. The longer sojourn time for Noncurrent use implies that many people in this age group were experimenting with combustible tobacco or ENDS products infrequently, or were going long stretches between use episodes. Finally, the estimated sojourn time for the Never state (7.71 years, 95% CI: 7.34–8.07) was much longer than any of the current or noncurrent use states, reflecting the fact that most people who have not tried a tobacco product by mid-adolescence will not go on to use combustible tobacco products or ENDS (the actual estimate is constrained by our data and so should not be interpreted to mean that those individuals will try a combustible tobacco or ENDS product in 7.7 years). Effects of Age The effect of Age at Time 0 on transition intensity is shown in Table 1 as the hazard ratio of 1 year of age for each allowed instantaneous transition. The hazard ratios indicate that older participants had lower risk of making instantaneous transitions out of Noncurrent use and Combustible-only use. Each year older a participant was at Time 0 was associated with a significantly decreased risk of transitioning from Noncurrent use to each of the three current use states. Each additional year of age was also associated with a significantly decreased risk of transitioning from Combustible-only to Noncurrent use and to Dual use. The only transition for which age increased risk was from Dual use to Combustible-only use. For all other transitions, the effect of age was not significant. Table 1. Hazard Ratios [95% CI] Depicting the Effect of Age at Time 0 on Transition Intensities Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Ratios significantly different than 1 are bolded. Cells without numbers represent instantaneous transitions that were not allowed in the model (see Method). View Large Table 1. Hazard Ratios [95% CI] Depicting the Effect of Age at Time 0 on Transition Intensities Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Ratios significantly different than 1 are bolded. Cells without numbers represent instantaneous transitions that were not allowed in the model (see Method). View Large Another way to understand the effect of age is by examining how sojourn times in each state vary with age (Figure 3). As implied by the decreased risk of transitioning out of Noncurrent use with age, 21-year olds were estimated to spend almost twice as long as 15-year olds in Noncurrent use before transitioning into a current use state. Sojourn time for Combustible-only use also increased with age, with 15-year olds spending an average of 4.60 months (0.383 years) as Combustible-only users before transitioning to another state and 21-year olds spending 7.61 months (0.634 years). Together, these patterns suggest that as youth move through adolescence and into early adulthood, their combustible use (or nonuse) becomes more stable. Figure 3. View largeDownload slide Estimated average time in each state (sojourn time) as a function of participants’ age with 95% confidence intervals. The left panel includes all five states. The right panel removes the Never state in order to better show the differences between the other four states. Matrix of model estimates is also provided in Appendix Table 7. Figure 3. View largeDownload slide Estimated average time in each state (sojourn time) as a function of participants’ age with 95% confidence intervals. The left panel includes all five states. The right panel removes the Never state in order to better show the differences between the other four states. Matrix of model estimates is also provided in Appendix Table 7. Discussion Our findings provide critical and current information about adolescents’ and young adults’ movement between different combustible and ENDS use patterns in a large, nationally representative sample. Results indicate those who used only combustible products or who used both ENDS and combustibles did so for about 6 months before changing to a different use state, while those who used only ENDS did so for approximately 3 months before transitioning. These relatively short durations suggest youth change their use patterns frequently. In contrast, nonuse states were more stable, spanning an estimated 1.5 years for those who reported having used a product in the past, but not in the last 30 days, and 7.6 years for those who reported never having used a product. We found that the relative probability of possible future states depended on both the current state and the length of the time interval of interest. At time intervals of 1 year or less, the prior state had a large influence on predictions for future states. At these short intervals, combustible-only and dual users were more likely to remain users than to switch to noncurrent use, while ENDS-only users were more likely to switch to noncurrent use than any other state. At intervals of longer than a year, however, the prior use state was less informative about the future state. For an interval of 2 years, all past 30-day users and noncurrent users had an estimated 60% probability of being a noncurrent user and a 20% probability of being a combustible-only user. When considering the relative probabilities for different state transitions, one should keep in mind that the use states represent very different percentages of the population of youth and young adults. As reported above, at study entry, our sample was 21.3% noncurrent users, 3.1% ENDS-only users, 8.9% combustible-only users, and 6.1% dual users. The approximately 20% of each of these groups that were estimated to be combustible-only users after 2 years therefore represent 4.3%, 0.6%, 1.8%, and 1.2% of the population. It is important to note that our unweighted sample data had somewhat lower prevalence of ENDS use than reported in other national surveillance data sets. The 2015 National Youth Tobacco survey of high-school seniors reported 5.9% e-cigarette only, 6.8% combustible-only, and 6.6% dual combustible/e-cigarette users, while the 2013–2014 National Adult Tobacco survey of 18- to 24-year olds reported 6.1% e-cigarette only, 9.6% cigarette only, and 7.5% dual cigarettes/e-cigarettes.23 Age significantly influenced transition probabilities. Consistent with other research, we found older participants had significantly more stable use patterns than younger participants.24–26 For example, a noncurrent user who had tried a tobacco product in the past was likely to remain a nonuser for 1 year if they were 15 years old, but 2 years if they were 21 years old. Alternatively, a 21-year old who used only combustible products in the past 30 days was estimated to spend 7.6 months as a combustible-only user, whereas a 15-year old was estimated to spend 4.6 months. These findings suggest significant variation in use patterns across late adolescence and early adulthood, which should be considered when designing interventions aimed at reducing tobacco and nicotine product use in this age group. A novel contribution of our analysis is the ability to compare predicted outcomes for different time intervals, providing a sense of how generalizable patterns across shorter time intervals are to longer intervals. Across intervals of 1 year or less, our findings are very consistent with those of other studies working in that time frame. Recent research suggests that dual use of combustibles and ENDS products may be a marker of a high-risk state for some youth,10,27 and our model transition estimates offer qualified support for that interpretation: Over a 1-year interval, the dual users were less likely than combustible-only (or ENDS-only) users to transition to noncurrent use. However, the findings were quite different when transition probabilities were estimated for a 2-year interval: Dual users, combustible-only users, and ENDS-only users had very similar probabilities of transitioning to noncurrent use. Our findings suggest that for youth navigating the current product landscape, relative risk may depend critically on the time frame examined. Our findings provide some evidence that ENDS users are not particularly vulnerable to transitioning to use of combustible products. Over a 1 year interval, ENDS-only users had a higher probability of transitioning to combustible use than those who had never used any tobacco products. ENDS-only users, however, were far more likely to stop using any tobacco products during this interval. The very short sojourn time for ENDS-only use (3 months) suggests that ENDS-only use is an extremely transient state. Further, as mentioned above, the probability of being a combustibles user was very similar across all use states (other than never users) when estimated over a 2-year interval. Consistent with prior work on transitions between cigarettes and smokeless tobacco,11 we found that combustible-only use was the most likely past 30-day use end state for intervals 1.5 years and longer for participants starting in all the use states. Overall, our analysis suggests that as participants age into early adulthood, those who are continuing to use tobacco and nicotine products are most likely to use combustibles. When interpreting time intervals from the current study, one should keep in mind the ways in which this study deviates from a typical “cohort” design. First, while the majority of participants joined our study at Wave 1 of data collection, 9% joined the study at later time points. Second, the model uses all available data when estimating the instantaneous transition intensity matrix from which the transition probabilities and sojourn times are derived.22 Therefore, when transition probabilities are estimated for a 6-month interval, that interval represents any 6-month interval within the study bounds, not only the 6 months after Time 0. This approach, especially coupled with the relatively wide age range of the sample, makes these findings more generalizable than studies that focus on a 6- or 12-month follow-up with a specific cohort of youth. Several limitations should be noted. First, it was not possible to account for all types of product use. Smokeless tobacco use was omitted from the analysis because of prevalence of less than 0.05% within our sample. Smokeless tobacco use may influence transitions between the products studied, and future work should investigate that specifically. Further, the ENDS products used as examples in the survey were those that were on the market in 2014 when the survey began. Given subsequent changes in product availability, it is possible that ENDS use was under-reported because of mismatch between the examples and participants’ experiences. Second, the results depend on how well the Markov assumption holds: the conditional probability distribution of future states depends only upon the present state, not on the preceding sequence of events. The extent to which this modeling framework is a good fit for longitudinal changes in tobacco and nicotine use should be further explored. Third, the modeling approach was driven by an interest in states defined a priori to answer questions about specific product transitions. This means that use patterns that fall within the combustibles category, such as transitions between hookah, small cigar use, and cigarettes, were outside the scope of the current analysis. Fourth, covariate inclusion was highly constrained in this modeling framework. Future work should more fully investigate factors that moderate the likelihood of transitions between products. Finally, our analysis was limited to past 30-day use; future work should examine how transitions between products are related to the progression from experimentation to daily use as youth become young adults.25,28 In conclusion, the TLC survey represents a unique opportunity to examine transitions in a large nationally representative sample spanning a wide age range from adolescence to young adulthood, with multiple follow-ups across 2.5 years. The current study addressed a significant gap in the literature by employing the multistate modeling framework to simultaneously estimate transitions between multiple product use states. Our results demonstrate a need for caution in interpreting transitions, especially at younger ages and over intervals of 1 year or less. That more stable combustible and noncurrent use states begin to emerge in young adulthood suggests that unstable states may reflect typical adolescent experimentation.25,29 Prevention and policy must consider how to use this precision in analysis to better tailor interventions, policy, and regulations to prevent youth and young adults from using any and all forms of nicotine or tobacco-containing products. Funding This study was funded by Truth Initiative. Declaration of Interests None declared. References 1. Johnston LD, O’Malley PM, Miech RA, Bachman J, Schulenberg J. Monitoring the Future National Survey Results on Drug Use, 1975–2016: Overview, Key Findings on Adolescent Drug Use . Ann Arbor, MI: Institute for Social Research, The University of Michigan; 2017. 2. Chaffee BW, Couch ET, Gansky SA. Trends in characteristics and multi-product use among adolescents who use electronic cigarettes, United States 2011–2015. PLoS One . 2017; 12( 5): e0177073. Google Scholar CrossRef Search ADS PubMed  3. Villanti AC, Pearson JL, Glasser AMet al.   Frequency of youth E-cigarette and tobacco use patterns in the United States: measurement precision is critical to inform public health. Nicotine Tob Res . 2017;19(11):1345–1350. 4. Jamal A, Gentzke A, Hu SS,et al.   Tobacco use among middle and high school students—United States, 2011–2016. MMWR Morb Mortal Wkly Rep . 2017; 66: 597– 603. Google Scholar CrossRef Search ADS PubMed  5. Hu SS, Neff L, Agaku IT,et al.   Tobacco Product Use Among Adults—United States, 2013–2014. MMWR Morb Mortal Wkly Rep  2016; 65: 685– 691. Google Scholar CrossRef Search ADS PubMed  6. Rath JM, Villanti AC, Abrams DB, Vallone DM. Patterns of tobacco use and dual use in US young adults: the missing link between youth prevention and adult cessation. J Environ Public Health . 2012; 2012: 679134. Google Scholar CrossRef Search ADS PubMed  7. Richardson A, Williams V, Rath J, Villanti AC, Vallone D. The next generation of users: prevalence and longitudinal patterns of tobacco use among US young adults. Am J Public Health . 2014; 104( 8): 1429– 1436. Google Scholar CrossRef Search ADS PubMed  8. Collins LK, Villanti AC, Pearson JLet al.   Frequency of youth E-cigarette, tobacco, and poly-use in the United States, 2015: update to Villanti et al., “frequency of youth E-cigarette and tobacco use patterns in the United States: measurement precision is critical to inform public health”. Nicotine Tob Res . 2017;19(10), 1 October 2017; 1253–1254. 9. Cobb CO, Villanti AC, Graham ALet al.   Markov modeling to estimate the population impact of emerging tobacco products: a proof-of-concept study. Tob Regul Sci . 2015; 1( 2): 129– 141. Google Scholar CrossRef Search ADS   10. Huh J, Leventhal AM. Progression of poly-tobacco product use patterns in adolescents. Am J Prev Med . 2016; 51( 4): 513– 517. Google Scholar CrossRef Search ADS PubMed  11. Kaufman AR, Land S, Parascandola M, Augustson E, Backinger CL. Tobacco use transitions in the United States: the National Longitudinal Study of Adolescent Health. Prev Med . 2015; 81: 251– 257. Google Scholar CrossRef Search ADS PubMed  12. Tam J, Day HR, Rostron BL, Apelberg BJ. A systematic review of transitions between cigarette and smokeless tobacco product use in the United States. BMC Public Health . 2015; 15: 258. Google Scholar CrossRef Search ADS PubMed  13. Kozlowski LT, Warner KE. Adolescents and e-cigarettes: objects of concern may appear larger than they are. Drug Alcohol Depen . 2017; 174: 209– 214. Google Scholar CrossRef Search ADS   14. Leventhal AM, Strong DR, Kirkpatrick MGet al.   Association of electronic cigarette use with initiation of combustible tobacco product smoking in early adolescence. JAMA . 2015; 314( 7): 700– 707. Google Scholar CrossRef Search ADS PubMed  15. Park JY, Seo DC, Lin HC. E-Cigarette use and intention to initiate or quit smoking among US youths. Am J Public Health . 2016; 106( 4): 672– 678. Google Scholar CrossRef Search ADS PubMed  16. Wills TA, Knight R, Sargent JD, Gibbons FX, Pagano I, Williams RJ. Longitudinal study of e-cigarette use and onset of cigarette smoking among high school students in Hawaii. Tob Control . 2017;26:34–39. 17. Barrington-Trimis JL, Urman R, Berhane Ket al.   E-cigarettes and future cigarette use. Pediatrics . 2016; 138( 1): e20160379. Google Scholar CrossRef Search ADS PubMed  18. Soneji S, Barrington-Trimis JL, Wills TAet al.   Association between initial use of e-cigarettes and subsequent cigarette smoking among adolescents and young adults: a systematic review and meta-analysis. JAMA Pediatr . 2017; 171( 8): 788– 797. Google Scholar CrossRef Search ADS PubMed  19. Warner KE. Frequency of E-cigarette use and cigarette smoking by American students in 2014. Am J Prev Med . 2016; 51( 2): 179– 184. Google Scholar CrossRef Search ADS PubMed  20. Walton KM, Abrams DB, Bailey WCet al.   NIH electronic cigarette workshop: developing a research agenda. Nicotine Tob Res . 2015; 17( 2): 259– 269. Google Scholar CrossRef Search ADS PubMed  21. Cantrell J, Hair EC, Smith A,et al.   Recruiting and retaining youth and young adults: challenges and opportunities in survey research for tobacco control Tobacco Control Published Online First: 21 April 2017. doi:10.1136/tobaccocontrol-2016-053504 22. Jackson CH. Multi-state models for panel data: the msm package for R. J Stat Softw . 2011; 38( 8): 1– 29. Google Scholar CrossRef Search ADS   23. U.S. Department of Health and Human Services. E-Cigarette Use Among Youth and Young Adults: A Report of the Surgeon General—Executive Summary . Atlanta, GA: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2016. 24. Chassin L, Presson CC, Pitts SC, Sherman SJ. The natural history of cigarette smoking from adolescence to adulthood in a midwestern community sample: multiple trajectories and their psychosocial correlates. Health Psychol . 2000; 19( 3): 223– 231. Google Scholar CrossRef Search ADS PubMed  25. Hair E, Bennett M, Williams Vet al.   Progression to established patterns of cigarette smoking among young adults. Drug Alcohol Depend . 2017; 177: 77– 83. Google Scholar CrossRef Search ADS PubMed  26. Chen K, Kandel DB. The natural history of drug use from adolescence to the mid-thirties in a general population sample. Am J Public Health . 1995; 85( 1): 41– 47. Google Scholar CrossRef Search ADS PubMed  27. Doran N, Brikmanis K, Petersen Aet al.   Does e-cigarette use predict cigarette escalation? A longitudinal study of young adult non-daily smokers. Prev Med . 2017; 100: 279– 284. Google Scholar CrossRef Search ADS PubMed  28. Terry-McElrath YM, O’Malley PM, Johnston LD. Discontinuous patterns of cigarette smoking from ages 18 to 50 in the United States: a repeated-measures latent class analysis. Nicotine Tob Res . 2017; 20(1): 108–116. 29. Vanyukov MM, Tarter RE, Kirillova GPet al.   Common liability to addiction and “gateway hypothesis”: theoretical, empirical and evolutionary perspective. Drug Alcohol Depend . 2012; 123( suppl 1): S3– S17. Google Scholar CrossRef Search ADS PubMed  Appendix Survey Questions for Combustible/ENDs Use 1. Have you ever tried cigarette smoking (even 1 or 2 puffs)? (yes/no) 2. During the past 30 days, on how many days did you smoke cigarettes (even 1 or 2 puffs)? (number of days in the past 30) 3. Have you ever tried smoking large cigars, little cigars or cigarillos (like Winchester, Swisher Sweets, Phillies Blunts, Dutch Masters, Black & Mild, or White Owl) (even 1 or 2 puffs)? (yes/no) 4. During the past 30 days, on how many days did you smoke large cigars, little cigars, or cigarillos (like Winchester, Swisher Sweets, Phillies Blunts, Dutch Masters, Black & Mild, or White Owl) (even 1 or 2 puffs)? (number of days in the past 30) 5. Have you ever used or tried any of the following tobacco products? (yes/no)  a. Hookah/shisa/waterpipe (with tobacco) (even 1 or 2 puffs)  b. E-cigarettes (like blu, NJOY, Smoking Everywhere)  c. E-Hookah/E-Cigars/Vape Pens/Hookah pens/Vape Pipes 6. During the past 30 days, on how many days did you use each of the following products? (number of days in the past 30)  a. Hookah/shisa/waterpipe (with tobacco) (even 1 or 2 puffs)  b. E-cigarettes (like blu, NJOY, Smoking Everywhere)  c. E-Hookah/E-Cigars/Vape Pens/Hookah pens/Vape Pipe Model Specification Details Table 1. The Matrix of Allowed Transitions Used to Begin the Weight Initialization Process   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2  View Large Table 1. The Matrix of Allowed Transitions Used to Begin the Weight Initialization Process   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2  View Large Table 2. The Crude Initial Intensity Matrix Generated from the Data, Used in All Models: crudeQ <−crudeinits.msm(TobUsePttr~surveyYear, chatid, qmatrix = transAllow, data = d)   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054  Call to fit the final model msm(formula = TobUsePttr ~ surveyYear, subject = chatid, data = d, qmatrix = crudeQ, covariates = ~AgeBaseline + Gender_0 + Race2Cat_0 + ParentEdu_0, hessian = TRUE, use.expm = TRUE, control = list(fnscale = 60000, maxit = 1000)) View Large Table 2. The Crude Initial Intensity Matrix Generated from the Data, Used in All Models: crudeQ <−crudeinits.msm(TobUsePttr~surveyYear, chatid, qmatrix = transAllow, data = d)   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054  Call to fit the final model msm(formula = TobUsePttr ~ surveyYear, subject = chatid, data = d, qmatrix = crudeQ, covariates = ~AgeBaseline + Gender_0 + Race2Cat_0 + ParentEdu_0, hessian = TRUE, use.expm = TRUE, control = list(fnscale = 60000, maxit = 1000)) View Large Table 3. Log-likelihood Ratio Tests for Model Selection Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  The numbers in each column represent the log-likelihood ratio test against the model in the row immediately above. View Large Table 3. Log-likelihood Ratio Tests for Model Selection Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  The numbers in each column represent the log-likelihood ratio test against the model in the row immediately above. View Large Intensity Matrices Table 4. Transition Intensity Matrix for the Final Model with 95% Confidence Intervals, Estimated at the Mean of All Covariates Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  View Large Table 4. Transition Intensity Matrix for the Final Model with 95% Confidence Intervals, Estimated at the Mean of All Covariates Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  View Large Table 5. Transition Intensity Matrices for the Final Model with 95% Confidence Intervals, Estimated at Ages = 15, 18, 21, and at the Mean of All Other Covariates Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  View Large Table 5. Transition Intensity Matrices for the Final Model with 95% Confidence Intervals, Estimated at Ages = 15, 18, 21, and at the Mean of All Other Covariates Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  View Large Transition Probabilities Table 6. Transition Probabilities Across Multiple Time Intervals Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Estimated at mean of all covariates. These data are plotted in Figure 2. View Large Table 6. Transition Probabilities Across Multiple Time Intervals Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Estimated at mean of all covariates. These data are plotted in Figure 2. View Large Sojourn Times Table 7. Sojourn Times Estimated at Ages = 15, 18, 21 and the Mean of All Other Covariates Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  These data are plotted in Figure 3. View Large Table 7. Sojourn Times Estimated at Ages = 15, 18, 21 and the Mean of All Other Covariates Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  These data are plotted in Figure 3. View Large © The Author 2018. Published by Oxford University Press on behalf of the Society for Research on Nicotine and Tobacco. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nicotine and Tobacco Research Oxford University Press

Loading next page...
 
/lp/ou_press/longitudinal-tobacco-use-transitions-among-adolescents-and-young-78Ov0mX0dt
Publisher
Oxford University Press
Copyright
© The Author 2018. Published by Oxford University Press on behalf of the Society for Research on Nicotine and Tobacco. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
1462-2203
eISSN
1469-994X
D.O.I.
10.1093/ntr/ntx285
Publisher site
See Article on Publisher Site

Abstract

Abstract Introduction Among youth, the frequency and prevalence of using more than one tobacco (small cigar, cigarette, and hookah) or nicotine-containing product (e-cigarettes-ENDS) are changing. These shifts pose challenges for regulation, intervention, and prevention campaigns because of scant longitudinal data on the stability of use patterns in this changing product landscape. Methods A nationally representative longitudinal survey of 15- to 21-year olds (n = 15,275) was used to describe transitions between never use, noncurrent use, and past 30-day use of combustible tobacco, e-cigarettes (ENDS), and dual use of both kinds of products. A multistate model was fit to observations collected every 6 months across 2.5 years to estimate the probability of transitions between states (TPs), the average time in state (sojourn time), and the effect of age on transitions. Results Current state strongly predicted future state over time intervals of 1 year or less, but only weakly predicted future state at longer intervals: TP to noncurrent use was higher for ENDS-only than combustible-only users over a 6-month interval but was similar for both groups over a 2-year interval. Sojourn time was significantly longer for combustible-only (0.52 years) and dual use (0.55 years) than ENDS-only use (0.27 years); older youth were more likely than younger youth to stay combustible tobacco users or noncurrent users. Conclusions The dynamics of transitions between combustible tobacco products and ENDS in a population of youth and young adults suggest that policy and prevention efforts must consider the frequent changes and instability over a 1-year or less time period in use patterns among young people. Implications The study addresses an urgent need in public health for timely information on how youth and young adults use tobacco and nicotine products. We found that youth, particularly adolescents, moved frequently between using ENDS and combustible tobacco products either alone or together. Importantly, the utility of current-use states for predicting future use states declined for time horizons longer than 1 year. Our results demonstrate a need for caution in interpreting product transitions. Longitudinal data with frequent observations and coverage of a wide range of possible product types is required to fully characterize usage patterns in youth. Introduction The prevalence of cigarette use has declined to record lows in the past 5 years. Over that time frame, prevalence of past 30-day cigarette use fell from 19.8% to 4.9% in 10th graders and from 17.1% to 10.5% in 12th graders.1 Contemporaneous with that decline has been an increasingly diverse tobacco and nicotine product landscape. As youth and young adults engage in trial and intermittent use, particularly before the onset of established smoking, use of products other than cigarettes has become more prevalent.2,3 Among high-school students, from 2011 to 2016, prevalence of hookah use increased from 4.1% to 7.2% and e-cigarette (ENDS) use increased from 1.5% to 11.3%.4 Among adults, 18- to 24-year olds represent a significant portion of other product users, making up 55.8% of hookah and 24.3% of ENDS users in 2014.5 Changes in the tobacco and nicotine product landscape have also influenced an increase in the percent of youth and young adults who use two or more products (ie, dual use or poly use).3,4,6–8 In 2016, 20.2% of US high schoolers reported any past 30-day product use, and of those, 47.2% were poly users, representing 9.6% of high schoolers.4 Even youth who are not past 30-day poly users are likely to move between different kinds of products. In 2015, while 4.0% of middle and high-school students reported exclusive past 30-day e-cigarette use, 61% of them had previously used at least one other tobacco product.8 Understanding the patterns of different product use among youth and young adults, and the temporal dynamics of how youth transition between different products, is critical to identifying priorities in tobacco control policies.3,9–13 To date, most longitudinal studies have focused on transitions into a single product, typically cigarettes. For example, several studies have looked at the relative risk of initiating cigarette smoking for individuals who have never used any product compared with individuals who have ever used ENDS but not cigarettes. These studies have found that ever ENDS use is associated with ever cigarette use.13–19 However, these studies are limited to specific populations observed over periods of a year or less. This means the stability of these relative risk patterns over longer periods of time is not known while other relevant transitions (eg, into and out of nonuse) are not modeled. Few studies have investigated product transitions using a framework that takes into account how participants might start and stop use of multiple different products across time.3,8 In a study of California high-school freshman, Huh and Leventhal used latent transition analysis to identify three classes of students based on past 6-month product use.10 At the follow-up 6 months later, between 13% and 19% of students had changed classes, highlighting the utility of methods that account for transitions into and out of use. Latent class models allow for data-driven classification of participants into groups. A multistate model is an alternative approach that allows for a priori specification of the categories of interest. Kaufman and colleagues used a multistate model to estimate transition probabilities into and out of single and dual use of smokeless tobacco and cigarettes using Add Health data from 1995 to 2009.11 They found that, over a 1-year interval, cigarette-only users and noncurrent users of any products were about equally likely to transition to using smokeless tobacco. Further, they found noncurrent use and cigarette-only use were the most stable states. The authors, along with several other groups, have called for greater application of these models to more product types, including ENDS.9,11,12 The goal of this study was to examine transitions into and out of tobacco and nicotine product use states using a large, nationally representative sample of youth and young adults with frequent observations over 2.5 years. We use a multistate modeling methodology and focus on transitions between single and dual use of combustible tobacco products and ENDS.20 The model is used to address two primary research questions: (1) What does an individual’s current use state say about the types of products they are likely to be using (or not using) 6 months, 1 year, or 2 years in the future? and (2) How long do individuals tend to stay in each use state before adding, dropping, or switching products, or stopping use for at least 30 days? We also examine whether age may influence transitions between tobacco products. Method Sample and Survey Procedures Data for this study were drawn from the Truth Longitudinal Cohort (TLC), a probability-based, nationally representative survey of youth and young adults. TLC participants were primarily recruited via address-based sampling, with subsamples recruited through random digit dialing and from GfK’s KnowledgePanel. Retention rates across all waves ranged from 61% to 71%. Additional details regarding the methodology of the TLC, including details about response rates, have been published elsewhere.21 Participants who completed two or more surveys across six waves of data collection were included in the analyses (n = 15,275). The majority of participants (n = 13,892) entered the study at Wave 1 between April and August 2014, and were followed for five additional waves at 6-month intervals, with Wave 6 data collected from January to April 2017. At Waves 2 through 5, refreshment samples of ~1000 additional participants, drawn from the same population as the original sample, were added to the total sample and followed for all remaining waves (eg, the refreshment participants who entered at data collection period Wave 3 were invited to participate in follow-up surveys for Waves 4 through 6). Because participants entered the study at different times, the term Time 0 in this report refers to the first observation for each individual, regardless of when they joined the study. The study protocol was approved by Chesapeake Institutional Review Board. Measures Products included in the combustible tobacco category were cigarettes, hookah, and all types of cigars. Products included in the ENDS category were e-cigarettes and e-hookah (see Appendix for survey questions). At each wave, all participants with data for that wave were classified into one of five mutually exclusive product use states: (1) Never use: Those who reported never having used any ENDS or combustible tobacco product; (2) Noncurrent use: Those who reported having used an ENDS or combustible product, but not in the past 30 days; (3) ENDS-only use: Those who reported having used an ENDS product in the past 30 days, but reported no past 30-day use of a combustible tobacco product; (4) Combustible-only use: Those who reported having used a combustible product in the past 30 days, but reported no past 30-day use of ENDS products; (5) Dual use: Those who reported having used both a combustible and an ENDS product in the past 30 days. Observation times for each participant were calculated for each wave as the difference in years between the time stamp for when they submitted the survey for that wave and the submission time stamp for their survey at Time 0 (ie, study entry). This yielded a continuous variable with peaks around each of the planned survey distribution times (ie, 0, 0.5, 1, 1.5, 2, and 2.5 years). Participant age at Time 0 was included as a covariate, and its effects were estimated and analyzed for all transitions in the model (see below). Three other individual characteristics were included as control variables, but their effects were not specifically analyzed: gender, race (white/not-white), and parent education (more than high school/high-school degree or less) were entered in the model as two-level categorical variables. Inclusion of more covariates, or additional levels of the included covariates, led to model nonconvergence because of small cell sizes within the transition matrix. Model Specification Continuous-time, multistate Markov models were fit to the data using the msm package in R.22 The continuous time model allows transitions between states to occur at unknown times between the observation timepoints, including multiple transitions between observations. The model uses all data available (ie, from respondents with two or more observations) to estimate the transition intensity matrix, which represents the instantaneous risk of transitions between states. The transition intensity matrix is then used to compute the probability of transitioning between all pairs of states for specified time intervals, and the mean duration of time spent in each state (ie, the sojourn time). The instantaneous transitions between states are shown in Figure 1, and initialization and intensity matrices are provided in Appendix Tables 1–5. One-way transitions were allowed from Never use into ENDS-only and Combustible-only: Once someone has tried a product, they cannot go back to never having tried a product. A direct transition from Never use to Noncurrent use was not allowed, as Never users would need to have tried one or more products before becoming a Noncurrent user. In other words, it would be implausible for concurrent reporting of never having used a product and having used a product but not in the last 30 days (Noncurrent use). Rather, the participant would need to transition from Never use to Noncurrent use via one (or more) of the current use states. Because our data were collected at 6-month intervals, it is possible for participants to report Never use at one observation and Noncurrent use at the next observation. Figure 1. View largeDownload slide Structure of allowed transitions for the multistate model. Unidirectional paths between states are shown with dashed lines and single-headed arrows; bidirectional paths are shown with solid lines and double-headed arrows (see text). Figure 1. View largeDownload slide Structure of allowed transitions for the multistate model. Unidirectional paths between states are shown with dashed lines and single-headed arrows; bidirectional paths are shown with solid lines and double-headed arrows (see text). The instantaneous transition matrix also excluded the transitions Never use to Dual use and Dual use to Noncurrent use. While these transitions are theoretically possible, they were extremely rare in our data set and so were set to 0 to improve model convergence and allow for inclusion of covariates. Missing data for use state and covariates were left as missing for model maximum likelihood estimation. A baseline model without any covariates was first fit to the data, followed by a series of sequential models adding one covariate at a time with the covariate effect estimated for all allowed transitions without constraint. The specific command used for model fitting are in the Appendix. Log-likelihood ratio goodness-of-fit tests revealed that each covariate improved model fit and that the model with all covariates was the best fit to the data (all p < .001, full test results in Appendix Table 3). Results are reported from the full model. Results Sample Description The sample covered the targeted ages well, with a fairly uniform distribution across the 15- to-21-year-old range at Time 0 (M = 18.61, SD = 2.11, range = 12.59 - 23.77). The distributions of the control variables were as follows: Gender: 48.7% female, 51.3% male (0 missing gender information); Race: 66.6% white, 33.1% nonwhite, 0.003% missing; Parent education: 84.0% greater than high school, 14.6% high-school graduate or less, and 1.4% missing. Participants were unevenly distributed across the five possible use states. At Time 0, the sample (n = 15,275) was 60.5% Never users, 21.3% Noncurrent users, 3.1% ENDS-only users, 8.9% Combustible-only users, and 6.1% Dual users. Over time, the largest changes in sample prevalence were seen in the Never use and Noncurrent use states, with the relative proportion of participants who reported using a combustible or ENDS product in the past 30 days remaining fairly constant. The Time 2.5 sample (n = 8571) was 45.1% Never, 35.7% Noncurrent, 3.3% ENDS-only, 11.1% Combustible-only, and 4.9% Dual. Probability of Transitioning Between States Over Time To examine use patterns over time, we estimated the transition probability between all pairs of states across time intervals between 0.5 and 2.0 years (Figure 2). The transition probability from State X to State Y represents the conditional probability that a participant will be in State Y at the end of the time interval, given that they were in State X at the beginning of the interval. Because states are mutually exclusive, the transition probabilities from any state for a given time interval must sum to 1. When estimating the transition probabilities, the model takes into account that a participant may make multiple transitions during the time interval. Figure 2. View largeDownload slide Transition probabilities across intervals of increasing duration. Each panel represents a different start state, and the series represent the probability of being in the each of the five states at a certain time in the future conditioned on that start state. Probabilities were estimated at the mean of all covariates. Matrix of model estimates is also provided in Appendix Table 6. Figure 2. View largeDownload slide Transition probabilities across intervals of increasing duration. Each panel represents a different start state, and the series represent the probability of being in the each of the five states at a certain time in the future conditioned on that start state. Probabilities were estimated at the mean of all covariates. Matrix of model estimates is also provided in Appendix Table 6. Transition probabilities estimated at the mean of all three covariates are illustrated in Figure 2. Never users pattern quite differently than all current and noncurrent users. The probability of a Never user moving into a current use state is low across all time windows, with a maximum of 6.2% probability of moving to Combustible-only use after 2 years. Setting aside the Never users, the most probable future state is highly variable across the other use states for 6-month intervals but diminishes for intervals of 1 year or more. For example, the probability of being a noncurrent user across any 6-month interval has a range of 61.6% (min = 18.2% for Dual users, max = 79.8% for Noncurrent users). However, over a 2-year interval, the range in probabilities is reduced to 9.3% (min = 56.1% for Dual users, max = 65.4% for Noncurrent users). The probability of transitioning to the Combustible-only state is similarly reduced over time, with a range of 35% for a 6-month interval but only 4% for a 2-year interval. The convergence of the transition probabilities over time across the use states indicates that individuals’ current state is much more informative for predicting their product use 6 months in the future compared with 2 years in the future. Regardless of whether participants are current or noncurrent users of combustibles, ENDS, or both, the most likely state 2 years in the future is noncurrent use (ie, no use in the past 30 days). For individuals who do remain or become active users 2 years in the future, the most likely use state is Combustible-only. Figure 2 also highlights the differences between current use states at intervals shorter than 12 months. During these shorter intervals, there is a gradient of risk across the ENDS-only, Combustible-only, and Dual use states. Across those three states, ENDS-only users are most likely to transition to noncurrent use, and Dual users are least likely, with Combustible-only users falling in between. Additionally, over those same intervals, ENDS-only users are least likely to be using combustibles (~23% probability) at the later time point, Dual users are most likely (~73%), and Combustible-only users fall in between (~60%), with percentages representing combined transition probabilities into Combustible-only and Dual use. Average Time in State Time-in-state, or sojourn time, was estimated for each use state at the mean of all covariates. Estimated sojourn time was approximately 3 months for the ENDS-only state (0.27 years, 95% CI: 0.25–0.29); 6 months for Combustible-only (0.52 years, 95% CI: 0.49–0.55) and Dual use (0.55 years, 95% CI: 0.52–0.58); and 18 months for Noncurrent use (1.50 years, 95% CI: 1.43–1.58). The relatively short sojourn times in the current use states suggest that 15- to 21-year olds are moving frequently between types and combinations of products. The longer sojourn time for Noncurrent use implies that many people in this age group were experimenting with combustible tobacco or ENDS products infrequently, or were going long stretches between use episodes. Finally, the estimated sojourn time for the Never state (7.71 years, 95% CI: 7.34–8.07) was much longer than any of the current or noncurrent use states, reflecting the fact that most people who have not tried a tobacco product by mid-adolescence will not go on to use combustible tobacco products or ENDS (the actual estimate is constrained by our data and so should not be interpreted to mean that those individuals will try a combustible tobacco or ENDS product in 7.7 years). Effects of Age The effect of Age at Time 0 on transition intensity is shown in Table 1 as the hazard ratio of 1 year of age for each allowed instantaneous transition. The hazard ratios indicate that older participants had lower risk of making instantaneous transitions out of Noncurrent use and Combustible-only use. Each year older a participant was at Time 0 was associated with a significantly decreased risk of transitioning from Noncurrent use to each of the three current use states. Each additional year of age was also associated with a significantly decreased risk of transitioning from Combustible-only to Noncurrent use and to Dual use. The only transition for which age increased risk was from Dual use to Combustible-only use. For all other transitions, the effect of age was not significant. Table 1. Hazard Ratios [95% CI] Depicting the Effect of Age at Time 0 on Transition Intensities Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Ratios significantly different than 1 are bolded. Cells without numbers represent instantaneous transitions that were not allowed in the model (see Method). View Large Table 1. Hazard Ratios [95% CI] Depicting the Effect of Age at Time 0 on Transition Intensities Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Dual  –  1.063 [1.022–1.106]  0.944 [0.891–1.001]  —  Combust-only  0.945 [0.9–0.991]  –  0.725 [0.446–1.177]  0.909 [0.882–0.937]  Ends-only  1.048 [0.977–1.125]  1.027 [0.787–1.341]  –  1.024 [0.987–1.063]  Noncurrent  0.813 [0.717–0.923]  0.957 [0.927–0.988]  0.832 [0.791–0.874]  –  Never  –  1.01 [0.965–1.058]  0.971 [0.938–1.005]  –  Ratios significantly different than 1 are bolded. Cells without numbers represent instantaneous transitions that were not allowed in the model (see Method). View Large Another way to understand the effect of age is by examining how sojourn times in each state vary with age (Figure 3). As implied by the decreased risk of transitioning out of Noncurrent use with age, 21-year olds were estimated to spend almost twice as long as 15-year olds in Noncurrent use before transitioning into a current use state. Sojourn time for Combustible-only use also increased with age, with 15-year olds spending an average of 4.60 months (0.383 years) as Combustible-only users before transitioning to another state and 21-year olds spending 7.61 months (0.634 years). Together, these patterns suggest that as youth move through adolescence and into early adulthood, their combustible use (or nonuse) becomes more stable. Figure 3. View largeDownload slide Estimated average time in each state (sojourn time) as a function of participants’ age with 95% confidence intervals. The left panel includes all five states. The right panel removes the Never state in order to better show the differences between the other four states. Matrix of model estimates is also provided in Appendix Table 7. Figure 3. View largeDownload slide Estimated average time in each state (sojourn time) as a function of participants’ age with 95% confidence intervals. The left panel includes all five states. The right panel removes the Never state in order to better show the differences between the other four states. Matrix of model estimates is also provided in Appendix Table 7. Discussion Our findings provide critical and current information about adolescents’ and young adults’ movement between different combustible and ENDS use patterns in a large, nationally representative sample. Results indicate those who used only combustible products or who used both ENDS and combustibles did so for about 6 months before changing to a different use state, while those who used only ENDS did so for approximately 3 months before transitioning. These relatively short durations suggest youth change their use patterns frequently. In contrast, nonuse states were more stable, spanning an estimated 1.5 years for those who reported having used a product in the past, but not in the last 30 days, and 7.6 years for those who reported never having used a product. We found that the relative probability of possible future states depended on both the current state and the length of the time interval of interest. At time intervals of 1 year or less, the prior state had a large influence on predictions for future states. At these short intervals, combustible-only and dual users were more likely to remain users than to switch to noncurrent use, while ENDS-only users were more likely to switch to noncurrent use than any other state. At intervals of longer than a year, however, the prior use state was less informative about the future state. For an interval of 2 years, all past 30-day users and noncurrent users had an estimated 60% probability of being a noncurrent user and a 20% probability of being a combustible-only user. When considering the relative probabilities for different state transitions, one should keep in mind that the use states represent very different percentages of the population of youth and young adults. As reported above, at study entry, our sample was 21.3% noncurrent users, 3.1% ENDS-only users, 8.9% combustible-only users, and 6.1% dual users. The approximately 20% of each of these groups that were estimated to be combustible-only users after 2 years therefore represent 4.3%, 0.6%, 1.8%, and 1.2% of the population. It is important to note that our unweighted sample data had somewhat lower prevalence of ENDS use than reported in other national surveillance data sets. The 2015 National Youth Tobacco survey of high-school seniors reported 5.9% e-cigarette only, 6.8% combustible-only, and 6.6% dual combustible/e-cigarette users, while the 2013–2014 National Adult Tobacco survey of 18- to 24-year olds reported 6.1% e-cigarette only, 9.6% cigarette only, and 7.5% dual cigarettes/e-cigarettes.23 Age significantly influenced transition probabilities. Consistent with other research, we found older participants had significantly more stable use patterns than younger participants.24–26 For example, a noncurrent user who had tried a tobacco product in the past was likely to remain a nonuser for 1 year if they were 15 years old, but 2 years if they were 21 years old. Alternatively, a 21-year old who used only combustible products in the past 30 days was estimated to spend 7.6 months as a combustible-only user, whereas a 15-year old was estimated to spend 4.6 months. These findings suggest significant variation in use patterns across late adolescence and early adulthood, which should be considered when designing interventions aimed at reducing tobacco and nicotine product use in this age group. A novel contribution of our analysis is the ability to compare predicted outcomes for different time intervals, providing a sense of how generalizable patterns across shorter time intervals are to longer intervals. Across intervals of 1 year or less, our findings are very consistent with those of other studies working in that time frame. Recent research suggests that dual use of combustibles and ENDS products may be a marker of a high-risk state for some youth,10,27 and our model transition estimates offer qualified support for that interpretation: Over a 1-year interval, the dual users were less likely than combustible-only (or ENDS-only) users to transition to noncurrent use. However, the findings were quite different when transition probabilities were estimated for a 2-year interval: Dual users, combustible-only users, and ENDS-only users had very similar probabilities of transitioning to noncurrent use. Our findings suggest that for youth navigating the current product landscape, relative risk may depend critically on the time frame examined. Our findings provide some evidence that ENDS users are not particularly vulnerable to transitioning to use of combustible products. Over a 1 year interval, ENDS-only users had a higher probability of transitioning to combustible use than those who had never used any tobacco products. ENDS-only users, however, were far more likely to stop using any tobacco products during this interval. The very short sojourn time for ENDS-only use (3 months) suggests that ENDS-only use is an extremely transient state. Further, as mentioned above, the probability of being a combustibles user was very similar across all use states (other than never users) when estimated over a 2-year interval. Consistent with prior work on transitions between cigarettes and smokeless tobacco,11 we found that combustible-only use was the most likely past 30-day use end state for intervals 1.5 years and longer for participants starting in all the use states. Overall, our analysis suggests that as participants age into early adulthood, those who are continuing to use tobacco and nicotine products are most likely to use combustibles. When interpreting time intervals from the current study, one should keep in mind the ways in which this study deviates from a typical “cohort” design. First, while the majority of participants joined our study at Wave 1 of data collection, 9% joined the study at later time points. Second, the model uses all available data when estimating the instantaneous transition intensity matrix from which the transition probabilities and sojourn times are derived.22 Therefore, when transition probabilities are estimated for a 6-month interval, that interval represents any 6-month interval within the study bounds, not only the 6 months after Time 0. This approach, especially coupled with the relatively wide age range of the sample, makes these findings more generalizable than studies that focus on a 6- or 12-month follow-up with a specific cohort of youth. Several limitations should be noted. First, it was not possible to account for all types of product use. Smokeless tobacco use was omitted from the analysis because of prevalence of less than 0.05% within our sample. Smokeless tobacco use may influence transitions between the products studied, and future work should investigate that specifically. Further, the ENDS products used as examples in the survey were those that were on the market in 2014 when the survey began. Given subsequent changes in product availability, it is possible that ENDS use was under-reported because of mismatch between the examples and participants’ experiences. Second, the results depend on how well the Markov assumption holds: the conditional probability distribution of future states depends only upon the present state, not on the preceding sequence of events. The extent to which this modeling framework is a good fit for longitudinal changes in tobacco and nicotine use should be further explored. Third, the modeling approach was driven by an interest in states defined a priori to answer questions about specific product transitions. This means that use patterns that fall within the combustibles category, such as transitions between hookah, small cigar use, and cigarettes, were outside the scope of the current analysis. Fourth, covariate inclusion was highly constrained in this modeling framework. Future work should more fully investigate factors that moderate the likelihood of transitions between products. Finally, our analysis was limited to past 30-day use; future work should examine how transitions between products are related to the progression from experimentation to daily use as youth become young adults.25,28 In conclusion, the TLC survey represents a unique opportunity to examine transitions in a large nationally representative sample spanning a wide age range from adolescence to young adulthood, with multiple follow-ups across 2.5 years. The current study addressed a significant gap in the literature by employing the multistate modeling framework to simultaneously estimate transitions between multiple product use states. Our results demonstrate a need for caution in interpreting transitions, especially at younger ages and over intervals of 1 year or less. That more stable combustible and noncurrent use states begin to emerge in young adulthood suggests that unstable states may reflect typical adolescent experimentation.25,29 Prevention and policy must consider how to use this precision in analysis to better tailor interventions, policy, and regulations to prevent youth and young adults from using any and all forms of nicotine or tobacco-containing products. Funding This study was funded by Truth Initiative. Declaration of Interests None declared. References 1. Johnston LD, O’Malley PM, Miech RA, Bachman J, Schulenberg J. Monitoring the Future National Survey Results on Drug Use, 1975–2016: Overview, Key Findings on Adolescent Drug Use . Ann Arbor, MI: Institute for Social Research, The University of Michigan; 2017. 2. Chaffee BW, Couch ET, Gansky SA. Trends in characteristics and multi-product use among adolescents who use electronic cigarettes, United States 2011–2015. PLoS One . 2017; 12( 5): e0177073. Google Scholar CrossRef Search ADS PubMed  3. Villanti AC, Pearson JL, Glasser AMet al.   Frequency of youth E-cigarette and tobacco use patterns in the United States: measurement precision is critical to inform public health. Nicotine Tob Res . 2017;19(11):1345–1350. 4. Jamal A, Gentzke A, Hu SS,et al.   Tobacco use among middle and high school students—United States, 2011–2016. MMWR Morb Mortal Wkly Rep . 2017; 66: 597– 603. Google Scholar CrossRef Search ADS PubMed  5. Hu SS, Neff L, Agaku IT,et al.   Tobacco Product Use Among Adults—United States, 2013–2014. MMWR Morb Mortal Wkly Rep  2016; 65: 685– 691. Google Scholar CrossRef Search ADS PubMed  6. Rath JM, Villanti AC, Abrams DB, Vallone DM. Patterns of tobacco use and dual use in US young adults: the missing link between youth prevention and adult cessation. J Environ Public Health . 2012; 2012: 679134. Google Scholar CrossRef Search ADS PubMed  7. Richardson A, Williams V, Rath J, Villanti AC, Vallone D. The next generation of users: prevalence and longitudinal patterns of tobacco use among US young adults. Am J Public Health . 2014; 104( 8): 1429– 1436. Google Scholar CrossRef Search ADS PubMed  8. Collins LK, Villanti AC, Pearson JLet al.   Frequency of youth E-cigarette, tobacco, and poly-use in the United States, 2015: update to Villanti et al., “frequency of youth E-cigarette and tobacco use patterns in the United States: measurement precision is critical to inform public health”. Nicotine Tob Res . 2017;19(10), 1 October 2017; 1253–1254. 9. Cobb CO, Villanti AC, Graham ALet al.   Markov modeling to estimate the population impact of emerging tobacco products: a proof-of-concept study. Tob Regul Sci . 2015; 1( 2): 129– 141. Google Scholar CrossRef Search ADS   10. Huh J, Leventhal AM. Progression of poly-tobacco product use patterns in adolescents. Am J Prev Med . 2016; 51( 4): 513– 517. Google Scholar CrossRef Search ADS PubMed  11. Kaufman AR, Land S, Parascandola M, Augustson E, Backinger CL. Tobacco use transitions in the United States: the National Longitudinal Study of Adolescent Health. Prev Med . 2015; 81: 251– 257. Google Scholar CrossRef Search ADS PubMed  12. Tam J, Day HR, Rostron BL, Apelberg BJ. A systematic review of transitions between cigarette and smokeless tobacco product use in the United States. BMC Public Health . 2015; 15: 258. Google Scholar CrossRef Search ADS PubMed  13. Kozlowski LT, Warner KE. Adolescents and e-cigarettes: objects of concern may appear larger than they are. Drug Alcohol Depen . 2017; 174: 209– 214. Google Scholar CrossRef Search ADS   14. Leventhal AM, Strong DR, Kirkpatrick MGet al.   Association of electronic cigarette use with initiation of combustible tobacco product smoking in early adolescence. JAMA . 2015; 314( 7): 700– 707. Google Scholar CrossRef Search ADS PubMed  15. Park JY, Seo DC, Lin HC. E-Cigarette use and intention to initiate or quit smoking among US youths. Am J Public Health . 2016; 106( 4): 672– 678. Google Scholar CrossRef Search ADS PubMed  16. Wills TA, Knight R, Sargent JD, Gibbons FX, Pagano I, Williams RJ. Longitudinal study of e-cigarette use and onset of cigarette smoking among high school students in Hawaii. Tob Control . 2017;26:34–39. 17. Barrington-Trimis JL, Urman R, Berhane Ket al.   E-cigarettes and future cigarette use. Pediatrics . 2016; 138( 1): e20160379. Google Scholar CrossRef Search ADS PubMed  18. Soneji S, Barrington-Trimis JL, Wills TAet al.   Association between initial use of e-cigarettes and subsequent cigarette smoking among adolescents and young adults: a systematic review and meta-analysis. JAMA Pediatr . 2017; 171( 8): 788– 797. Google Scholar CrossRef Search ADS PubMed  19. Warner KE. Frequency of E-cigarette use and cigarette smoking by American students in 2014. Am J Prev Med . 2016; 51( 2): 179– 184. Google Scholar CrossRef Search ADS PubMed  20. Walton KM, Abrams DB, Bailey WCet al.   NIH electronic cigarette workshop: developing a research agenda. Nicotine Tob Res . 2015; 17( 2): 259– 269. Google Scholar CrossRef Search ADS PubMed  21. Cantrell J, Hair EC, Smith A,et al.   Recruiting and retaining youth and young adults: challenges and opportunities in survey research for tobacco control Tobacco Control Published Online First: 21 April 2017. doi:10.1136/tobaccocontrol-2016-053504 22. Jackson CH. Multi-state models for panel data: the msm package for R. J Stat Softw . 2011; 38( 8): 1– 29. Google Scholar CrossRef Search ADS   23. U.S. Department of Health and Human Services. E-Cigarette Use Among Youth and Young Adults: A Report of the Surgeon General—Executive Summary . Atlanta, GA: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2016. 24. Chassin L, Presson CC, Pitts SC, Sherman SJ. The natural history of cigarette smoking from adolescence to adulthood in a midwestern community sample: multiple trajectories and their psychosocial correlates. Health Psychol . 2000; 19( 3): 223– 231. Google Scholar CrossRef Search ADS PubMed  25. Hair E, Bennett M, Williams Vet al.   Progression to established patterns of cigarette smoking among young adults. Drug Alcohol Depend . 2017; 177: 77– 83. Google Scholar CrossRef Search ADS PubMed  26. Chen K, Kandel DB. The natural history of drug use from adolescence to the mid-thirties in a general population sample. Am J Public Health . 1995; 85( 1): 41– 47. Google Scholar CrossRef Search ADS PubMed  27. Doran N, Brikmanis K, Petersen Aet al.   Does e-cigarette use predict cigarette escalation? A longitudinal study of young adult non-daily smokers. Prev Med . 2017; 100: 279– 284. Google Scholar CrossRef Search ADS PubMed  28. Terry-McElrath YM, O’Malley PM, Johnston LD. Discontinuous patterns of cigarette smoking from ages 18 to 50 in the United States: a repeated-measures latent class analysis. Nicotine Tob Res . 2017; 20(1): 108–116. 29. Vanyukov MM, Tarter RE, Kirillova GPet al.   Common liability to addiction and “gateway hypothesis”: theoretical, empirical and evolutionary perspective. Drug Alcohol Depend . 2012; 123( suppl 1): S3– S17. Google Scholar CrossRef Search ADS PubMed  Appendix Survey Questions for Combustible/ENDs Use 1. Have you ever tried cigarette smoking (even 1 or 2 puffs)? (yes/no) 2. During the past 30 days, on how many days did you smoke cigarettes (even 1 or 2 puffs)? (number of days in the past 30) 3. Have you ever tried smoking large cigars, little cigars or cigarillos (like Winchester, Swisher Sweets, Phillies Blunts, Dutch Masters, Black & Mild, or White Owl) (even 1 or 2 puffs)? (yes/no) 4. During the past 30 days, on how many days did you smoke large cigars, little cigars, or cigarillos (like Winchester, Swisher Sweets, Phillies Blunts, Dutch Masters, Black & Mild, or White Owl) (even 1 or 2 puffs)? (number of days in the past 30) 5. Have you ever used or tried any of the following tobacco products? (yes/no)  a. Hookah/shisa/waterpipe (with tobacco) (even 1 or 2 puffs)  b. E-cigarettes (like blu, NJOY, Smoking Everywhere)  c. E-Hookah/E-Cigars/Vape Pens/Hookah pens/Vape Pipes 6. During the past 30 days, on how many days did you use each of the following products? (number of days in the past 30)  a. Hookah/shisa/waterpipe (with tobacco) (even 1 or 2 puffs)  b. E-cigarettes (like blu, NJOY, Smoking Everywhere)  c. E-Hookah/E-Cigars/Vape Pens/Hookah pens/Vape Pipe Model Specification Details Table 1. The Matrix of Allowed Transitions Used to Begin the Weight Initialization Process   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2  View Large Table 1. The Matrix of Allowed Transitions Used to Begin the Weight Initialization Process   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.3  0.1  0.1  0.1  0  Combust-only  0.1  −0.3  0.1  0.1  0  ENDS-only  0.1  0.1  −0.3  0.1  0  Noncurrent  0.05  0.1  0.1  −0.25  0  Never  0  0.1  0.1  0  −0.2  View Large Table 2. The Crude Initial Intensity Matrix Generated from the Data, Used in All Models: crudeQ <−crudeinits.msm(TobUsePttr~surveyYear, chatid, qmatrix = transAllow, data = d)   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054  Call to fit the final model msm(formula = TobUsePttr ~ surveyYear, subject = chatid, data = d, qmatrix = crudeQ, covariates = ~AgeBaseline + Gender_0 + Race2Cat_0 + ParentEdu_0, hessian = TRUE, use.expm = TRUE, control = list(fnscale = 60000, maxit = 1000)) View Large Table 2. The Crude Initial Intensity Matrix Generated from the Data, Used in All Models: crudeQ <−crudeinits.msm(TobUsePttr~surveyYear, chatid, qmatrix = transAllow, data = d)   Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054    Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −0.61  0.468  0.143  0  0  Combust-only  0.24  −0.891  0.051  0.599  0  ENDS-only  0.253  0.201  −1.27  0.815  0  Noncurrent  0.071  0.2  0.081  −0.353  0  Never  0  0.031  0.022  0  −0.054  Call to fit the final model msm(formula = TobUsePttr ~ surveyYear, subject = chatid, data = d, qmatrix = crudeQ, covariates = ~AgeBaseline + Gender_0 + Race2Cat_0 + ParentEdu_0, hessian = TRUE, use.expm = TRUE, control = list(fnscale = 60000, maxit = 1000)) View Large Table 3. Log-likelihood Ratio Tests for Model Selection Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  The numbers in each column represent the log-likelihood ratio test against the model in the row immediately above. View Large Table 3. Log-likelihood Ratio Tests for Model Selection Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  Model  −2 logLR  df  p  No covariates  64842.96  –  –  Age  329.12  13  <.001  Age, Gender  76.32  13  <.001  Age, Gender, Race  320.35  13  <.001  Age, Gender, Race, Parent Education  795.92  13  <.001  The numbers in each column represent the log-likelihood ratio test against the model in the row immediately above. View Large Intensity Matrices Table 4. Transition Intensity Matrix for the Final Model with 95% Confidence Intervals, Estimated at the Mean of All Covariates Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  View Large Table 4. Transition Intensity Matrix for the Final Model with 95% Confidence Intervals, Estimated at the Mean of All Covariates Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  Dual  −1.825 [−1.939 to −1.719]  1.232 [1.13 to 1.342]  0.594 [0.523 to 0.674]  –  –  Combust-only  0.658 [0.594 to 0.728]  −1.923 [−2.03 to −1.822]  0.011 [0.001 to 0.136]  1.255 [1.179 to 1.335]  -–  ENDS-only  0.799 [0.681 to 0.938]  0.065 [0.011 to 0.405]  −3.751 [−4.035 to −3.487]  2.886 [2.671 to 3.118]  –  Noncurrent  0.046 [0.03 to 0.072]  0.369 [0.343 to 0.397]  0.249 [0.223 to 0.277]  −0.664 [−0.698 to −0.632]  -–  Never  –  0.053 [0.047 to 0.059]  0.077 [0.071 to 0.084]  –  −0.13 [−0.136 to −0.124]  View Large Table 5. Transition Intensity Matrices for the Final Model with 95% Confidence Intervals, Estimated at Ages = 15, 18, 21, and at the Mean of All Other Covariates Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  View Large Table 5. Transition Intensity Matrices for the Final Model with 95% Confidence Intervals, Estimated at Ages = 15, 18, 21, and at the Mean of All Other Covariates Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  Age  Start state  End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  15 years  Dual  −1.713 [−1.949 to −1.506]  0.984 [0.814 to 1.189]  0.73 [0.566 to 0.941]  –  –  Combust-only  0.81 [0.642 to 1.022]  −2.612 [−2.958 to −2.307]  0.035 [0.002 to 0.665]  1.768 [1.524 to 2.05]  –  ENDS-only  0.674 [0.503 to 0.904]  0.061 [0.011 to 0.32]  −3.374 [−3.886 to −2.929]  2.639 [2.263 to 3.077]  –  Noncurrent  0.098 [0.05 to 0.191]  0.435 [0.37 to 0.51]  0.486 [0.393 to 0.6]  −1.018 [−1.134 to −0.914]  –  Never  –  0.051 [0.042 to 0.061]  0.086 [0.075 to 0.098]  –  −0.137 [−0.147 to −0.127]  18 years  Dual  −1.797 [−1.919 to −1.683]  1.183 [1.074 to 1.303]  0.614 [0.538 to 0.702]  –  –  Combust-only  0.683 [0.608 – 0.766]  −2.023 [−2.153 to −1.901]  0.013 [0.001 to 0.164]  1.327 [1.234 to 1.427]  –  ENDS-only  0.776 [0.659 to 0.914]  0.066 [0.012 to 0.371]  −3.677 [−3.967 to −3.408]  2.835[2.617 to 3.071]  –  Noncurrent  0.053 [0.033 to 0.083]  0.381 [0.35 to 0.414]  0.279 [0.249 to 0.313]  −0.713 [−-0.752 to −0.675]  –  Never  –  0.052 [0.046 to 0.059]  0.079 [0.073 to 0.085]  –  −0.131 [−0.137 to −0.125]  21 years  Dual  −1.94 [−2.102 to −1.79]  1.422 [1.282 to 1.578]  0.517 [0.432 to 0.619]  –  –  Combust-only  0.576 [0.511 to 0.649]  −1.577 [−1.679 to −1.481]  0.005 [0 to 0.088]  0.996 [0.928 to 1.07]  –  ENDS-only  0.894 [0.705 to 1.134]  0.071 [0.009 to 0.596]  −4.011 [−4.47 to −3.6]  3.046 [2.702 to 3.434]  –  Noncurrent  0.028 [0.017 to 0.047]  0.333 [0.308 to 0.361]  0.161 [0.137 to 0.189]  −0.522 [−0.557 to −0.49]  –  Never  –  0.054 [0.046 to 0.063]  0.072 [0.063 to 0.082]  –  −0.126 [−0.135 to −0.117]  View Large Transition Probabilities Table 6. Transition Probabilities Across Multiple Time Intervals Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Estimated at mean of all covariates. These data are plotted in Figure 2. View Large Table 6. Transition Probabilities Across Multiple Time Intervals Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Start state    End state  Dual  Combust-only  ENDS-only  Noncurrent  Never  0.5-year interval  Dual  0.464[0.444 to 0.482]  0.266 [0.25 to 0.282]  0.088 [0.079 to 0.097]  0.182 [0.174 to 0.192]  –  Combust-only  0.145 [0.135 to 0.156]  0.454 [0.429 to 0.468]  0.032 [0.029 to 0.047]  0.37 [0.355 to 0.387]  –  ENDS-only  0.127 [0.115 to 0.141]  0.1 [0.092 to 0.138]  0.194 [0.168 to 0.213]  0.579 [0.554 to 0.597]  –  Noncurrent  0.039 [0.035 to 0.045]  0.112 [0.106 to 0.119]  0.051 [0.047 to 0.055]  0.798 [0.789 to 0.805]  –  Never  0.006 [0.005 to 0.006]  0.019 [0.017 to 0.021]  0.018 [0.016 to 0.019]  0.02 [0.019 to 0.022]  0.937 [0.934 to 0.94]  0.75-year interval  Dual  0.346 [0.33 to 0.363]  0.28 [0.265 to 0.295]  0.083 [0.074 to 0.091]  0.291 [0.279 to 0.303]  –  Combust-only  0.154 [0.144 to 0.165]  0.349 [0.328 to 0.363]  0.045 [0.041 to 0.056]  0.453 [0.438 to 0.47]  –  ENDS-only  0.125 [0.116 to 0.138]  0.139 [0.131 to 0.169]  0.113 [0.098 to 0.125]  0.622 [0.602 to 0.638]  –  Noncurrent  0.057[0.053 to 0.064]  0.138 [0.131 to 0.146]  0.056 [0.052 to 0.061]  0.748 [0.738 to 0.757]  –  Never  0.01 [0.009 to 0.011]  0.026 [0.024 to 0.028]  0.021 [0.019 to 0.022]  0.036 [0.035 to 0.039]  0.907 [0.903 to 0.911]  1-year interval  Dual  0.272 [0.257 to 0.287]  0.273 [0.258 to 0.287]  0.076 [0.069 to 0.085]  0.379 [0.366 to 0.393]  –  Combust-only  0.151 [0.143 to 0.161]  0.289 [0.269 to 0.302]  0.052 [0.049 to 0.063]  0.508 [0.493 to 0.524]  –  ENDS-only  0.121 [0.113 to 0.13]  0.163 [0.154 to 0.185]  0.082 [0.071 to 0.089]  0.634 [0.615 to 0.648]  –  Noncurrent  0.072 [0.067 to 0.079]  0.156 [0.147 to 0.164]  0.058 [0.053 to 0.063]  0.715 [0.703 to 0.724]  –  Never  0.014 [0.013 to 0.015]  0.032 [0.029 to 0.035]  0.022 [0.02 to 0.024]  0.053 [0.051 to 0.056]  0.878 [0.873 to 0.883]  1.5-year interval  Dual  0.19 [0.178 to 0.203]  0.246 [0.232 to 0.258]  0.067 [0.061 to 0.075]  0.497[0.483 to 0.511]  –  Combust-only  0.138 [0.13 to 0.148]  0.233 [0.218 to 0.244]  0.059 [0.054 to 0.067]  0.57 [0.557 to 0.582]  –  ENDS-only  0.115 [0.108 to 0.123]  0.185 [0.177 to 0.199]  0.064 [0.059 to 0.07]  0.636 [0.621 to 0.645]  –  Noncurrent  0.091 [0.086 to 0.099]  0.176 [0.167 to 0.185]  0.059 [0.055 to 0.065]  0.674 [0.662 to 0.683]  –  Never  0.021 [0.02 to 0.023]  0.043 [0.04 to 0.047]  0.025 [0.023 to 0.027]  0.088 [0.084 to 0.092]  0.823 [0.816 to 0.83]  2-year Interval  Dual  0.152 [0.143 to 0.162]  0.225 [0.213 to 0.236]  0.063 [0.058 to 0.07]  0.561 [0.547 to 0.573]  -–  Combust-only  0.128 [0.12 to 0.136]  0.212 [0.2 to 0.223]  0.06 [0.055 to 0.067]  0.6 [0.587 to 0.612]  –  ENDS-only  0.113 [0.107 to 0.121]  0.192 [0.183 to 0.204]  0.061 [0.056 to 0.067]  0.634 [0.62 to 0.643]  –  Noncurrent  0.101 [0.096 to 0.109]  0.185 [0.176 to 0.196]  0.06 [0.054 to 0.066]  0.654 [0.641 to 0.663]  –  Never  0.028 [0.026 to 0.03]  0.053 [0.05 to 0.057]  0.027 [0.025 to 0.029]  0.121 [0.115 to 0.126]  0.772 [0.762 to 0.78]  2.5-year interval  Dual  0.133[0.125 to 0.142]  0.211 [0.198 to 0.222]  0.062 [0.056 to 0.069]  0.594 [0.582 to 0.607]  –  Combust-only  0.121 [0.114 to 0.129]  0.203 [0.19 to 0.214]  0.061 [0.055 to 0.068]  0.615[0.602 to 0.628]  –  ENDS-only  0.113 [0.106 to 0.121]  0.194 [0.183 to 0.205]  0.06 [0.055 to 0.067]  0.633 [0.619 to 0.644]  –  Noncurrent  0.107 [0.101 to 0.115]  0.19 [0.179 to 0.2]  0.06 [0.055 to 0.067]  0.643 [0.63 to 0.654]  –  Never  0.033 [0.031 to 0.035]  0.062 [0.058 -to 0.067]  0.029 [0.027 to 0.032]  0.152 [0.146 -to 0.159]  0.723 [0.712 to 0.733]  Estimated at mean of all covariates. These data are plotted in Figure 2. View Large Sojourn Times Table 7. Sojourn Times Estimated at Ages = 15, 18, 21 and the Mean of All Other Covariates Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  These data are plotted in Figure 3. View Large Table 7. Sojourn Times Estimated at Ages = 15, 18, 21 and the Mean of All Other Covariates Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  Age  State  Estimates  SE  L  U  15  Dual  0.584  0.038  0.513  0.664  Combust-only  0.383  0.024  0.338  0.433  ENDS-only  0.296  0.021  0.257  0.341  Noncurrent  0.982  0.054  0.882  1.094  Never  7.318  0.284  6.781  7.897  18  Dual  0.556  0.019  0.521  0.594  Combust-only  0.494  0.016  0.464  0.526  ENDS-only  0.272  0.011  0.252  0.293  Noncurrent  1.403  0.038  1.330  1.481  Never  7.641  0.169  7.316  7.980  21  Dual  0.516  0.021  0.476  0.559  Combust-only  0.634  0.020  0.595  0.675  ENDS-only  0.249  0.014  0.224  0.278  Noncurrent  1.914  0.062  1.796  2.040  Never  7.950  0.298  7.387  8.557  These data are plotted in Figure 3. View Large © The Author 2018. Published by Oxford University Press on behalf of the Society for Research on Nicotine and Tobacco. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

Journal

Nicotine and Tobacco ResearchOxford University Press

Published: Feb 13, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off