Abstract
Background and Aims:
Long-term frequent marijuana use is associated with significant negative outcomes, yet little is known about the longitudinal course of marijuana use among those who start frequent use during adolescence. Objectives are (a) to identify latent patterns of within-person marijuana use from ages 19–30 among 12th graders reporting frequent marijuana use, (b) to examine if membership in identified patterns has changed across historical time, and (c) to examine if key covariates differentiate class membership.
Design, Setting, Participants:
Longitudinal, national US panel data from 4,423 individuals (53.4% of the eligible sample; 2,744 [62%] males) who reported frequent marijuana use in 12th grade (modal age 18 years; senior year cohorts 1976–2006) followed biennially from age 19/20 through 29/30.
Measurements:
Self-reported past 30-day marijuana use (frequent use defined as use on 20+ occasions), demographics, college graduation, marriage, and parenthood.
Findings:
Repeated measures latent class analysis (RMLCA) identified five latent classes of past 30-day marijuana use from ages 19/20 through 29/30: Continued Frequent Users (estimated membership 23.4%); Frequent to Non-Frequent Users (15.5%); Consistent Non-Frequent Users (18.4%); Non-Frequent Users to Discontinuers (19.5%); and Discontinuers (23.2%). In multivariable models, membership in the highest-risk latent class (Continued Frequent Users) versus one or more of the lower-risk latent classes was more likely for recent cohorts (p=0.038 to <0.001), as well as those who did not marry (p=0.039 to <0.001) or become parents (p=0.001) by modal age 29/30.
Conclusions:
Nearly one in four 12th grade (modal age 18 years) frequent marijuana users in the US continues to report high frequency use through age 30; the proportion continuing high frequency use across young adulthood has increased among more recent cohorts.
Introduction
High frequency marijuana use is associated with enduring negative effects [1–4], and chronic heavy use (particularly use that starts in adolescence and continues throughout young adulthood) is associated with higher degrees of impaired functioning than adolescent-limited use [3,5–11]. Yet, studies have not examined long-term marijuana use trajectories among individuals who report frequent use by the end of high school. Little is known about possible heterogeneity in later marijuana use among those who use marijuana frequently as adolescents, the stability of such patterns across historical time, or characteristics associated with diverging patterns of use across development. Such information is essential for accurate projections of future treatment demand and potential social and individual costs, given that frequent marijuana use is associated with increased addiction/dependence, cognitive impairment, diminished life satisfaction/achievement, poor psychological/physical health, etc. [2–6,8–12].
Several studies have examined developmental patterns of marijuana use from adolescence into adulthood among the general population and have identified heavy or chronic use subgroup(s) [6–10,13–20]. Similar studies are not available focusing on individuals who are heavy users by late adolescence. In one national sample of U.S. 12th grade students in 1975, 51% of those reporting frequent use (defined in the current study as use on 20 or more occasions in the previous 30 days) in 12th grade also reported frequent use four years later, while 34% reported non-frequent use [21]. Not only are more recent data needed, but also needed is an analytical approach allowing identification of underlying use patterns across age while incorporating measurement error and providing probability-based estimates of pattern membership, and modelling of possible changes in such patterns over historical time.
Prevalence data for frequent use among both adolescents and young adults indicate there may be historical differences in developmental patterns of frequent marijuana use. Since the mid-1970s, frequent marijuana use prevalence among US 12th grade students ranged from a high of 10.7% in 1978 to a low of 1.9% in 1992; since the early 2000s, prevalence has remained near 6% [22]. National data indicate frequent marijuana use prevalence has increased significantly among young adults and adults in general [23,24]; further, the typical mid-to late 20s developmental decrease in frequent marijuana use has slowed for recent cohorts [25]. Such changes indicate the number of adolescent frequent marijuana users who continue frequent use across young adulthood may be growing, with meaningful implications for projections of future health and treatment needs.
Several key covariates may be associated with the likelihood of membership in patterns of marijuana use across young adulthood among those who initiated frequent use during adolescence. The social roles of marriage and parenthood have consistently been found to be protective against substance use [26,27]. Among the general population, lower likelihood of membership in heavy/chronic marijuana use subgroups has been found for both marriage [9,10] and parenthood [9]. Johnston found that continued frequent marijuana use four years after 12th grade was less likely for those who were married or had children [21]. General population studies also have found membership in heavy/chronic marijuana use subgroups is more likely for men [7,9,10,15,22,28,29] and individuals whose parents have higher education levels [10,15,18,22]; mixed results have been found for both race/ethnicity and college attendance/completion in regards to heavy/chronic marijuana use patterns [10,15,17,18,29]. The extent to which these covariates differentiate long-term marijuana use patterns within individuals who initiated frequent use by late adolescence is unknown.
The current study uses US national panel data from 32 cohorts of 12th grade students (1) to identify latent, heterogeneous patterns of within-person marijuana use from ages 19–30 among individuals reporting frequent marijuana use in 12th grade; (2) to examine the extent to which latent class membership probability has varied across historical time (i.e., cohorts); and (3) to examine if sex, race/ethnicity, parental education, 4-year college experience, marital status, and parental status differentiate membership likelihood for identified classes.
Method
Participants
Since 1975, MTF has surveyed annually nationally representative cross-sectional samples of approximately 15,000 12th grade students from 130 public and private schools in the coterminous US (school samples are revised annually) [22]. About 2,450 students are selected annually for longitudinal follow-up; drug users are oversampled [24]. A random half of the follow-up sample receives biennial follow-ups beginning one year after 12th grade (modal age 19); the other half begins biennial follow-up beginning two years after 12th grade (modal age 20). Mailed questionnaires collect data at six follow-up time points: modal ages 19/20, 21/22, 23/24, 25/26, 27/28, and 29/30 (hereafter, modal age is referred to simply as “age”). A University of Michigan Institutional Review Board approved the study.
Analysis was limited to 12th grade cohorts with the opportunity for age 29/30 follow-up survey participation as of the date of analysis: 1976–2006 12th grade cohorts (age 29/30 data collected during 1987–2017). Twelfth grade student response rates averaged 82.7%; absenteeism was the primary reason for nonresponse [22]. Of the 74,525 individuals selected for follow-up, 72,643 (97.5%) provided valid data on 12th grade past 30-day marijuana use; 8,279 (11.4%) reported frequent use (use on 20 or more occasions in the past 30 days). Given the analytical focus of use throughout the 20s and covariates including marriage, parenthood, etc., included cases were required to respond to at least one of the age 25/26, 27/28 or 29/30 follow-up surveys. Among 12th grade frequent marijuana users, 4,676 (56.5%) responded to at least one of these surveys; 4,423 (53.4%) provided valid data on past 30-day marijuana use and covariates (2,744 (62.0% male) (see Supplement Figure 1 for sample flow chart). The mean number of marijuana use responses per respondent was 4.9 (42% of respondents provided data at all 6; 25% 5; 18% 4; 10% 3; 4% 2, and 1% only 1). Attrition adjustments are discussed below.
Measures
Marijuana use.
At each survey, participants were asked about the number of occasions (if any) they used marijuana or hashish during the past 30 days. Given the primary interest in examining patterns of frequent marijuana use across development, as well as the desire to model non-frequent use, responses (on a 7-point scale of 0, 1–2, 3–5, 6–9, 10–19, 20–39, 40 or more occasions) were recoded into a trichotomy of frequent use (20 or more occasions), non-frequent use (1 to 19 occasions), and no use (0 occasions).
Cohort (indicating year of 12th grade survey) was coded into six non-overlapping groups: 1976–1980, 1981–1985, 1986–1990, 1991–1995, 1996–2000, and 2001–2006.
Covariates.
Sex (male, female), race/ethnicity (African American, Hispanic, White, or other [including multiracial identity]), and parental education (at least one parent had “some college” or more) were self-reported at 12th grade. Remaining covariates were measured from ages 19/20–29/30, and indicated if respondents reported the following at any follow-up survey: 4-year college experience (ever graduating; attending but not graduating; not attending); ever married (yes/no); having one or more children at any survey (yes/no).
Analysis
Descriptive analysis was conducted using SAS v.9.4; repeated measures latent class analysis (RMLCA) models were fit in Mplus v.7.4 [30], using full information maximum likelihood estimation with robust standard errors. RMLCA models used six trichotomous past 30-day marijuana use indicators (one for each follow-up survey). Because the six indicators could result in 36 (729) possible patterns (not including missingness), RMLCA provided both a data reduction technique to identify key longitudinal patterns and the ability to account explicitly for both measurement error (i.e., mismatch between observed reports and class assignment) and missing marijuana use indicator data [31]. Missing data on marijuana use indicators were assumed to be missing at random and were handled using Mplus’ full information maximum likelihood estimation procedure [30]. Cases with missing data on covariates were excluded. Maximum likelihood solution identification was confirmed using 500 initial stage random starts and 250 final stage optimizations. Because trichotomous indicators preclude use of the bootstrap likelihood ratio test as a criterion for selection of the optimal number of classes, a variety of fit criteria were relied on, of which the BIC and sample size-adjusted BIC (a-BIC) have been shown in simulations to perform particularly well for selecting “correct” latent class models [32]. Using the R3STEP command (wherein the latent class model is estimated first, the most likely class variable is then created using the latent class posterior distribution taking into account measurement error, and finally auxiliary variables are included [33]), covariates were added (first bivariately, then all simultaneously) as auxiliary variables to baseline-category, multinomial logistic regression models including all cases. Unadjusted covariate prevalence estimates across latent classes were obtained using the BCH command. By design, covariates are not permitted to affect latent class formation in either R3STEP or BCH [33]. All analyses were weighted using an attrition weight, calculated as the inverse of the probability of responding at age 29/30 based on 12th grade covariates, including the sampling weight correcting for over-sampling of 12th grade substance users [9].
Results
Descriptive statistics
Age 18 frequent marijuana users were primarily male (68%) and White (79%); most (62%) reported at least one parent had attained some college education (see Table 1). By age 29/30, 24% had completed a 4-year college degree; approximately half reported ever being married (49%) or having children (50%).
Table 1.
Sample Descriptives: 12th grade frequent marijuana users followed through age 29/30
| % | (SE) | % | (SE) | |||
|---|---|---|---|---|---|---|
| Sexa | Ever marriedb | |||||
| Female | 31.7 | (0.74) | No | 51.1 | (0.84) | |
| Male | 68.3 | (0.74) | Yes | 48.9 | (0.84) | |
| Race/ethnicitya | Ever have a childb | |||||
| White | 79.0 | (0.83) | No | 50.0 | (0.84) | |
| Black | 8.5 | (0.63) | Yes | 50.0 | (0.84) | |
| Hispanic | 5.9 | (0.51) | ||||
| Other | 6.6 | (0.45) | Cohort group | |||
| Parental educationa | 1976–1980 | 24.5 | (0.61) | |||
| No college education | 38.5 | (0.82) | 1981–1985 | 16.7 | (0.57) | |
| Any college education | 61.5 | (0.82) | 1986–1990 | 9.7 | (0.48) | |
| 4-year college experienceb | 1991–1995 | 9.3 | (0.50) | |||
| Not attend | 64.9 | (0.80) | 1996–2000 | 18.7 | (0.73) | |
| Attend only | 11.3 | (0.54) | 2001–2006 | 21.2 | (0.79) | |
| Graduate | 23.8 | (0.70) | ||||
Notes: Sample n (unwtd) = 4,423.
Reported at 12th grade survey.
Reported from ages 19/20 through 29/30.
Latent marijuana use classes
RMLCAs with one to seven classes were considered. The optimal number of classes was determined by examining model fit, interpretability, parsimony, and stability/identification. Item response probability values of 0.50 or higher were considered to indicate “high” probability of endorsing the specified marijuana use level; in the absence of probabilities ≥0.50, the next highest probability value was considered. Table 2 reports model fit/selection criteria. No improvement in BIC was seen between the 6- and 7-class model; thus, 5- and 6-class models were considered for interpretation clarity and utility. Because the 6-class model included one class for which interpretation was questionable, the 5-class model was selected as optimal.
Table 2.
Fit Information for RMLCAs Modeling Past 30-day Marijuana Use from Ages 19/20 through 29/30 with 1–7 Latent Classes
| Classes | df | AIC | BIC | a-BIC | Entropy | Stability |
|---|---|---|---|---|---|---|
| 1 | 716 | 44687.77 | 44764.51 | 44726.37 | 1.0000 | |
| 2 | 703 | 39182.36 | 39342.22 | 39262.78 | 0.78 | 1.0000 |
| 3 | 690 | 37898.43 | 38141.43 | 38020.68 | 0.71 | 1.0000 |
| 4 | 677 | 37640.65 | 37966.77 | 37804.72 | 0.67 | 0.9760 |
| 5 | 664 | 37491.95 | 37901.20 | 37697.84 | 0.65 | 0.7200 |
| 6 | 651 | 37393.67 | 37886.06 | 37641.38 | 0.63 | 0.4960 |
| 7 | 638 | 37321.58 | 37897.09 | 37611.10 | 0.63 | 0.6080 |
Notes: Sample n (unwtd) = 4,423. RMLCA = repeated measures latent class analysis; AIC = Akaike information criterion; BIC = Bayesian information criterion; a-BIC = adjusted BIC; Stability = proportion of time the maximum-likelihood solution was selected out of 250 final stage optimizations (preceded by 500 initial stage sets of random starting values). Bold font indicates selected model.
Table 3 reports class membership and item response probability parameter estimates for the 5-class solution (see Supplement Figure 2 for graphical representation). The overall probability of reporting frequent marijuana use in the past 30 days dropped from 0.483 to 0.237 between ages 19/20 and 29/30. The overall probability of reporting non-frequent use was approximately 0.334 from ages 19/20–23/24, and then decreased to 0.266 by age 29/30. The overall probability of reporting no past 30-day marijuana use increased from 0.183 to 0.497 from ages 19/20–29/30. Thus, overall, the probability of continuing frequent marijuana use dropped by approximately 50%, the probability of discontinuation more than doubled, and the probability of non-frequent use dropped by approximately 20%. Latent classes were labeled and described as follows:
Table 3.
Five-class model of past 30-day marijuana use from ages 19/20 through 29/30 among 12th grade frequent marijuana users
| Modal age | Overall probability | Continued Frequent Users | Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | Discontinuers | |
|---|---|---|---|---|---|---|---|
| Latent class membership probabilities (estimated class counts) | |||||||
| 0.234 (1,033) | 0.155 (687) | 0.184 (816) | 0.195 (863) | 0.232 (1,024) | |||
| Item response probabilities | |||||||
| Frequent 30-day use | |||||||
| 19/20 | 0.483 | 0.760 | 0.771 | 0.377 | 0.418 | 0.107 | |
| 21/22 | 0.374 | 0.753 | 0.771 | 0.166 | 0.212 | 0.016 | |
| 23/24 | 0.327 | 0.800 | 0.646 | 0.106 | 0.059 | 0.030 | |
| 25/26 | 0.277 | 0.799 | 0.368 | 0.136 | 0.014 | 0.016 | |
| 27/28 | 0.251 | 0.840 | 0.104 | 0.174 | 0.000 | 0.015 | |
| 29/30 | 0.237 | 0.772 | 0.104 | 0.171 | 0.005 | 0.022 | |
| Non-frequent 30-day use | |||||||
| 19/20 | 0.334 | 0.180 | 0.193 | 0.473 | 0.501 | 0.341 | |
| 21/22 | 0.347 | 0.171 | 0.229 | 0.658 | 0.626 | 0.119 | |
| 23/24 | 0.335 | 0.138 | 0.316 | 0.695 | 0.524 | 0.086 | |
| 25/26 | 0.322 | 0.162 | 0.426 | 0.746 | 0.303 | 0.085 | |
| 27/28 | 0.297 | 0.122 | 0.608 | 0.720 | 0.123 | 0.074 | |
| 29/30 | 0.266 | 0.170 | 0.495 | 0.613 | 0.116 | 0.054 | |
| No 30-day use | |||||||
| 19/20 | 0.183 | 0.060 | 0.035 | 0.149 | 0.081 | 0.552 | |
| 21/22 | 0.279 | 0.077 | 0.000 | 0.176 | 0.162 | 0.865 | |
| 23/24 | 0.338 | 0.062 | 0.038 | 0.198 | 0.418 | 0.884 | |
| 25/26 | 0.401 | 0.039 | 0.206 | 0.118 | 0.683 | 0.899 | |
| 27/28 | 0.453 | 0.038 | 0.288 | 0.106 | 0.877 | 0.911 | |
| 29/30 | 0.497 | 0.059 | 0.402 | 0.216 | 0.880 | 0.924 | |
Notes: Sample n (unwtd) = 4,423. Estimated class counts obtained from estimated posterior probabilities. Bold font indicates item-response probabilities of ≥ 0.50. Bold and italicized font indicates the highest observed item-response probabilities when no probabilities ≥ .50 are observed. Latent class membership probabilities sum to 1 horizontally. Item response probabilities (reflecting the probabilities of each level of use for the sample overall, as well as within latent classes), sum to 1 vertically within age. For example, at age 19/20, the overall probability of using marijuana frequently in the past 30 days was 0.483, the probability of using non-frequently was 0.334, and was 0.183 for not using in the past 30 days.
Continued Frequent Users (estimated membership probability=0.234 [estimated n=1,033], indicating 23.4% of those reporting frequent marijuana use as 12th grade students exhibited this pattern of 30-day marijuana use from ages 19/20–29/30): high probabilities (0.753–0.840) of frequent past 30-day marijuana use from ages 19/20–29/30.
Frequent to Non-Frequent Users (15.5%, n=687): high probabilities (0.646–0.771) of frequent use from ages 19/20–23/24, and high probabilities (0.426–0.608) of non-frequent use from ages 25/26–29/30.
Consistent Non-Frequent Users (18.4%, n=816): high probabilities (0.473–0.746) of non-frequent use from ages 19/20–29/30.
Non-Frequent Users to Discontinuers (19.5%, n=863): high probabilities (0.501–0.626) of non-frequent use from ages 19/20–23/24, and high probabilities (0.683–0.879) of discontinuation from ages 25/26–29/30.
Discontinuers (23.2%, n=1,024): high probabilities (0.552–0.924) of no past 30-day marijuana use from ages 19/20–29/30.
Covariate associations
Multinomial logistic regression models examined associations between covariates and membership likelihood in (a) the Continued Frequent Users class versus other classes, and (b) the Discontinuers class versus other classes. Table 4 presents unadjusted covariate prevalence estimates across latent classes and summarizes significant bivariate and multivariable analyses reported in Tables 5–8.
Table 4.
Estimated unadjusted proportions of covariates across five-class model of past 30-day marijuana use from ages 19/20 through 29/30 among 12th grade frequent marijuana users
| Class 1: | Class 2: | Class 3: | Class 4: | Class 5: | |
|---|---|---|---|---|---|
| Continued Frequent Users | Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | Discontinuers | |
| Prop. (SE) | Prop. (SE) | Prop. (SE) | Prop. (SE) | Prop. (SE) | |
| Covariates measured at 12th grade survey | |||||
| Sex | |||||
| Female (ref) | 0.288 (0.018) | 0.311 (0.030) | 0.265 (0.024) | 0.394 (0.025) | 0.327 (0.020) |
| Male | 0.712 (0.018) | 0.689 (0.030) | 0.735 (0.024) | 0.606 (0.025)1a,1b,5b | 0.673 (0.020) |
| Race/ethnicity | |||||
| White (ref) | 0.819 (0.020) | 0.777 (0.033) | 0.816 (0.028) | 0.828 (0.027) | 0.718 (0.023) |
| African American | 0.071 (0.015)5a,5b | 0.072 (0.025) | 0.101 (0.023) | 0.060 (0.019) | 0.117 (0.018)1a,1b |
| Hispanic | 0.050 (0.012)5a,5b | 0.067 (0.021) | 0.037 (0.015)5a | 0.042 (0.016) | 0.093 (0.016)1a,1b |
| Other | 0.060 (0.011) | 0.084 (0.019) | 0.047 (0.014) | 0.070 (0.016) | 0.072 (0.012) |
| Parental education | |||||
| No college education (ref) | 0.372 (0.020) | 0.407 (0.033) | 0.363 (0.027) | 0.350 (0.026) | 0.429 (0.022) |
| Any college education | 0.628 (0.020) | 0.593 (0.033) | 0.637 (0.027) | 0.650 (0.026)1b,5a,5b | 0.571 (0.022) |
| Covariates measured from ages 19/20–29/30 | |||||
| 4-Year college experience | |||||
| Not attend | 0.642 (0.020) | 0.683 (0.032) | 0.662 (0.027) | 0.597 (0.026) | 0.666 (0.021) |
| Attend only | 0.116 (0.014) | 0.108 (0.022) | 0.093 (0.016) | 0.126 (0.017) | 0.117 (0.015) |
| Graduate | 0.242 (0.018) | 0.209 (0.028) | 0.245 (0.024) | 0.277 (0.023)1b,5b | 0.217 (0.018) |
| Ever married | |||||
| No (ref) | 0.591 (0.021) | 0.499 (0.034) | 0.607 (0.027) | 0.437 (0.028) | 0.423 (0.023) |
| Yes | 0.409 (0.021)5a,5b | 0.501 (0.034)1a,5b | 0.393 (0.027)1b,5a,5b | 0.563 (0.028)1a,5b | 0.577 (0.023)1a,1b |
| Ever have a child | |||||
| No (ref) | 0.572 (0.021) | 0.495 (0.034) | 0.556 (0.028) | 0.500 (0.027) | 0.387 (0.022) |
| Yes | 0.428 (0.021)5a,5b | 0.505 (0.034)5a | 0.444 (0.028)5a | 0.500 (0.027)1a,5a | 0.613 (0.022)1a,1b |
| Cohort group | |||||
| 1976–1980 | 0.170 (0.014)5b | 0.350 (0.029)1a,1b,5a,5b | 0.282 (0.022)1a,1b,5a,5b | 0.334 (0.023)1a,1b,5a,5b | 0.146 (0.013)1b |
| 1981–1985 | 0.118 (0.013) | 0.149 (0.022)1a | 0.209 (0.021)1a,1b,5a,5b | 0.200 (0.020)1a,1b,5a,5b | 0.166 (0.015) |
| 1986–1990 | 0.107 (0.012) | 0.081 (0.019) | 0.074 (0.015) | 0.118 (0.017)1a,1b,5a,5b | 0.097 (0.013) |
| 1991–1995 | 0.103 (0.013) | 0.072 (0.019) | 0.097 (0.017)5b | 0.094 (0.017)1a,1b,5a,5b | 0.090 (0.013) |
| 1996–2000 | 0.223 (0.019) | 0.160 (0.030) | 0.156 (0.024) | 0.146 (0.022) | 0.230 (0.020) |
| 2001–2006 (ref) | 0.279 (0.020) | 0.186 (0.031) | 0.183 (0.026) | 0.107 (0.024) | 0.271 (0.022) |
Notes: Sample n (unwtd) = 4,423.
Significantly different (p < 0.05 or stronger) odds than referent category vs. Class 1 (Continued Frequent Users) in bivariate models; exact p-values reported in Table 5.
Significantly different (p < 0.05 or stronger) odds than referent category vs. Class 1 (Continued Frequent Users) in multivariable models; exact p-values reported in Table 6.
Significantly different (p < 0.05 or stronger) odds than referent category vs. Class 5 (Discontinuers) in bivariate models; exact p-values reported in Table 7.
Significantly different (p < 0.05 or stronger) odds than referent category vs. Class 5 (Discontinuers) in multivariable models; exact p-values reported in Table 8.
Table 5.
Unadjusted relative risk ratios from bivariate multinomial logistic regressions of past 30-day marijuana use latent class membership on covariates: Referent class = Continued Frequent Users
| Class 2: | Class 3: | Class 4: | Class 5: | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | Discontinuers | |||||||||
| [v. CFU] | [v. CFU] | [v. CFU] | [v. CFU] | |||||||||
| RR | (95% CI) | p | RR | (95% CI) | p | RR | (95% CI) | p | RR | (95% CI) | p | |
| Covariates measured at 12th grade survey | ||||||||||||
| Sex | ||||||||||||
| Female | (ref) | (ref) | (ref) | (ref) | ||||||||
| Male | 0.90 | (0.63, 1.28) | 0.555 | 1.12 | (0.82, 1.52) | 0.469 | 0.62 | (0.48, 0.81) | <0.001 | 0.83 | (0.65, 1.07) | 0.151 |
| Race/ethnicity | ||||||||||||
| White | (ref) | (ref) | (ref) | (ref) | ||||||||
| African American | 1.08 | (0.42, 2.79) | 0.875 | 1.43 | (0.72, 2.85) | 0.308 | 0.84 | (0.39, 1.84) | 0.670 | 1.90 | (1.08, 3.34) | 0.026 |
| Hispanic | 1.41 | (0.56, 3.54) | 0.466 | 0.74 | (0.28, 1.94) | 0.537 | 0.83 | (0.34, 2.03) | 0.682 | 2.12 | (1.15, 3.89) | 0.016 |
| Other | 1.47 | (0.74, 2.94) | 0.276 | 0.78 | (0.37, 1.62) | 0.507 | 1.15 | (0.64, 2.08) | 0.633 | 1.37 | (0.81, 2.31) | 0.238 |
| Parental education | ||||||||||||
| No college education | (ref) | (ref) | (ref) | (ref) | ||||||||
| Any college education | 0.86 | (0.60, 1.23) | 0.416 | 1.04 | (0.77, 1.40) | 0.809 | 1.10 | (0.83, 1.44) | 0.513 | 0.79 | (0.61, 1.01) | 0.057 |
| Covariates measured from ages 19/20–29/30 | ||||||||||||
| 4-Year college experience | ||||||||||||
| Not attend | (ref) | (ref) | (ref) | (ref) | ||||||||
| Attend only | 0.87 | (0.48, 1.57) | 0.649 | 0.77 | (0.47, 1.25) | 0.296 | 1.16 | (0.77, 1.74) | 0.469 | 0.97 | (0.65, 1.44) | 0.873 |
| Graduate | 0.81 | (0.53, 1.24) | 0.336 | 0.98 | (0.70, 1.37) | 0.906 | 1.23 | (0.91, 1.66) | 0.172 | 0.86 | (0.64, 1.15) | 0.318 |
| Ever married | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 1.45 | (1.03, 2.06) | 0.036 | 0.93 | (0.70, 1.25) | 0.644 | 1.86 | (1.42, 2.44) | <0.001 | 1.97 | (1.54, 2.53) | <0.001 |
| Ever have a child | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 1.37 | (0.96, 1.94) | 0.081 | 1.07 | (0.80, 1.43) | 0.658 | 1.34 | (1.03, 1.75) | 0.031 | 2.12 | (1.65, 2.72) | <0.001 |
| Cohort group | ||||||||||||
| 1976–1980 | 3.10 | (1.84, 5.23) | <0.001 | 2.54 | (1.62, 3.97) | <0.001 | 5.14 | (3.00, 8.82) | <0.001 | 0.89 | (0.62, 1.28) | 0.527 |
| 1981–1985 | 1.90 | (1.03, 3.50) | 0.040 | 2.70 | (1.65, 4.42) | <0.001 | 4.43 | (2.49, 7.90) | <0.001 | 1.45 | (0.98, 2.14) | 0.064 |
| 1986–1990 | 1.14 | (0.54, 2.42) | 0.723 | 1.07 | (0.58, 1.96) | 0.840 | 2.88 | (1.54, 5.39) | 0.001 | 0.94 | (0.60, 1.47) | 0.791 |
| 1991–1995 | 1.05 | (0.47, 2.33) | 0.904 | 1.43 | (0.79, 2.60) | 0.239 | 2.38 | (1.22, 4.62) | 0.011 | 0.90 | (0.56, 1.44) | 0.649 |
| 1996–2000 | 1.08 | (0.55, 2.12) | 0.833 | 1.07 | (0.61, 1.86) | 0.815 | 1.71 | (0.92, 3.18) | 0.091 | 1.06 | (0.73, 1.55) | 0.748 |
| 2001–2006 | (ref) | (ref) | (ref) | (ref) | ||||||||
Notes: Sample n (unwtd) = 4,423. CFU = Consistent Frequent Users; RR = Relative risk ratio (obtained from models regressing past 30-day marijuana use latent class membership on one covariate at a time).
Referent=Continued Frequent Users.
In Table 4, significantly (p<0.05) different membership likelihoods in the noted class versus Continued Frequent Users (or CFU, Class 1) by covariates in bivariate and multivariate models are indicated by the superscripts “1a” and “1b”, respectively. For example, the “0.6061a,1b” for males in the Non-Frequent Users to Discontinuers class (Class 4) column indicates that membership likelihood for this class versus CFU differed significantly by sex in bivariate and multivariable models. The estimated proportion of the class that was male was 0.606 for Non-Frequent Users to Discontinuers versus 0.712 for CFU. Tables 5 and 6 provide unadjusted and adjusted relative risk ratios (RRs and ARRs, respectively), 95% confidence intervals, and exact p-values for all associations. To continue the example above, the likelihood of membership in the Non-Frequent Users to Discontinuers class versus CFU was significantly lower for males than females (Table 5 RR=0.62, p<0.001; Table 6 ARR=0.62, p=0.001). Below, we discuss only significant multivariable associations.
Table 6.
Adjusted relative risk ratios from multivariable multinomial logistic regressions of past 30-day marijuana use latent class membership on covariates: Referent class = Continued Frequent Users
| Class 2: | Class 3: | Class 4: | Class 5: | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | Discontinuers | |||||||||
| [v. CFU] | [v. CFU] | [v. CFU] | [v. CFU] | |||||||||
| ARR | (95% CI) | p | ARR | (95% CI) | p | ARR | (95% CI) | p | ARR | (95% CI) | p | |
| Covariates measured at 12th grade survey | ||||||||||||
| Sex | ||||||||||||
| Female | (ref) | (ref) | (ref) | (ref) | ||||||||
| Male | 0.98 | (0.66, 1.45) | 0.920 | 1.14 | (0.83, 1.58) | 0.423 | 0.62 | (0.47, 0.81) | 0.001 | 0.98 | (0.75, 1.29) | 0.893 |
| Race/ethnicity | ||||||||||||
| White | (ref) | (ref) | (ref) | (ref) | ||||||||
| African American | 1.05 | (0.36, 3.04) | 0.929 | 1.38 | (0.68, 2.78) | 0.373 | 0.94 | (0.43, 2.06) | 0.884 | 2.00 | (1.09, 3.67) | 0.025 |
| Hispanic | 2.00 | (0.76, 5.29) | 0.163 | 0.94 | (0.33, 2.68) | 0.911 | 1.29 | (0.52, 3.18) | 0.583 | 2.23 | (1.16, 4.26) | 0.016 |
| Other | 1.59 | (0.75, 3.40) | 0.230 | 0.88 | (0.42, 1.85) | 0.733 | 1.26 | (0.65, 2.45) | 0.503 | 1.51 | (0.86, 2.66) | 0.153 |
| Parental education | ||||||||||||
| No college education | (ref) | (ref) | (ref) | (ref) | ||||||||
| Any college education | 1.16 | (0.75, 1.79) | 0.496 | 1.24 | (0.89, 1.72) | 0.208 | 1.36 | (1.00, 1.83) | 0.048 | 0.89 | (0.68, 1.18) | 0.427 |
| Covariates measured from ages 19/20–29/30 | ||||||||||||
| 4-Year college experience | ||||||||||||
| Not attend | (ref) | (ref) | (ref) | (ref) | ||||||||
| Attend only | 0.96 | (0.51, 1.78) | 0.884 | 0.77 | (0.45, 1.31) | 0.332 | 1.42 | (0.93, 2.16) | 0.100 | 1.06 | (0.70, 1.60) | 0.785 |
| Graduate | 0.94 | (0.59, 1.52) | 0.817 | 1.14 | (0.79, 1.64) | 0.491 | 1.57 | (1.12, 2.18) | 0.008 | 1.01 | (0.72, 1.41) | 0.978 |
| Ever married | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 1.04 | (0.66, 1.63) | 0.875 | 0.69 | (0.48, 0.98) | 0.039 | 1.26 | (0.92, 1.72) | 0.152 | 1.84 | (1.36, 2.48) | <0.001 |
| Ever have a child | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 1.28 | (0.79, 2.06) | 0.319 | 1.24 | (0.85, 1.81) | 0.269 | 1.19 | (0.87, 1.62) | 0.282 | 1.63 | (1.21, 2.21) | 0.001 |
| Cohort group | ||||||||||||
| 1976–1980 | 3.18 | (1.76, 5.75) | <0.001 | 2.98 | (1.85, 4.80) | <0.001 | 5.82 | (3.33, 10.18) | <0.001 | 0.65 | (0.43, 0.98) | 0.038 |
| 1981–1985 | 1.88 | (0.96, 3.66) | 0.065 | 3.00 | (1.79, 5.02) | <0.001 | 5.16 | (2.85, 9.32) | <0.001 | 1.15 | (0.76, 1.74) | 0.517 |
| 1986–1990 | 1.18 | (0.55, 2.54) | 0.673 | 1.19 | (0.64, 2.22) | 0.585 | 3.06 | (1.62, 5.77) | 0.001 | 0.85 | (0.54, 1.33) | 0.475 |
| 1991–1995 | 1.01 | (0.44, 2.28) | 0.990 | 1.59 | (0.87, 2.89) | 0.133 | 2.44 | (1.26, 4.72) | 0.008 | 0.76 | (0.47, 1.24) | 0.272 |
| 1996–2000 | 1.04 | (0.52, 2.09) | 0.916 | 1.11 | (0.63, 1.94) | 0.720 | 1.82 | (0.98, 3.39) | 0.057 | 0.92 | (0.62, 1.37) | 0.677 |
| 2001–2006 | (ref) | (ref) | (ref) | (ref) | ||||||||
Notes: Sample n (unwtd) = 4,423. CFU = Consistent Frequent Users; ARR = Adjusted relative risk ratio (obtained from models regressing past 30-day marijuana use latent class membership on all listed covariates simultaneously).
Membership likelihood in the Frequent to Non-Frequent Users class (Class 2) versus CFU was significantly higher for the 1976–1980 (vs. 2001–2006) cohorts. Membership likelihood in the Consistent Non-Frequent Users class (Class 3) versus CFU was significantly higher for those never married and the 1976–1980 and 1981–1985 (vs. 2001–2006) cohorts. Membership in the Non-Frequent Users to Discontinuers class (Class 4) versus CFU was significantly more likely for females (as presented above), those whose parents had any college education, those graduating from a 4-year college, and the 1976–1980, 1981–1985, 1986–1990, and 1991–1995 (vs. 2001–2006) cohorts. Membership in the Discontinuers class (Class 5) versus CFU was significantly more likely for African American and Hispanic than White individuals, for those ever married, and those having at least one child; membership was significantly less likely for the 1976–1980 (vs. 2001–2006) cohorts.
Referent=Discontinuers.
In Table 4, significantly different likelihood of membership in the noted class versus the Discontinuers class in bivariate and multivariable models is indicated by superscripts “5a” and “5b”, respectively. Tables 7 and 8 provide RRs and ARRs, 95% confidence intervals, and exact p-values. Covariate associations differentiating membership in the CFU class versus Discontinuers were discussed above; the following focuses on significant multivariable covariate associations with membership in the Frequent to Non-Frequent Users, Consistent Non-Frequent Users, and Non-Frequent Users to Discontinuers classes versus Discontinuers.
Table 7.
Unadjusted relative risk ratios from bivariate multinomial logistic regressions of past 30-day marijuana use latent class membership on covariates: Referent class = Discontinuers
| Class 1: | Class 2: | Class 3: | Class 4: | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Continued Frequent Users | Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | |||||||||
| [v. Discontinuers] | [v. Discontinuers] | [v. Discontinuers] | [v. Discontinuers] | |||||||||
| RR | (95% CI) | p | RR | (95% CI) | p | RR | (95% CI) | p | RR | (95% CI) | p | |
| Covariates measured at 12th grade survey | ||||||||||||
| Sex | ||||||||||||
| Female | (ref) | (ref) | (ref) | (ref) | ||||||||
| Male | 1.20 | (0.94, 1.54) | 0.151 | 1.08 | (0.78, 1.48) | 0.645 | 1.34 | (1.00, 1.81) | 0.052 | 0.75 | (0.55, 1.02) | 0.065 |
| Race/ethnicity | ||||||||||||
| White | (ref) | (ref) | (ref) | (ref) | ||||||||
| African American | 0.53 | (0.30, 0.93) | 0.026 | 0.57 | (0.26, 1.25) | 0.161 | 0.75 | (0.41, 1.37) | 0.355 | 0.44 | (0.19, 1.02) | 0.055 |
| Hispanic | 0.47 | (0.26, 0.87) | 0.016 | 0.67 | (0.32, 1.40) | 0.285 | 0.35 | (0.14, 0.86) | 0.021 | 0.39 | (0.15, 1.02) | 0.055 |
| Other | 0.73 | (0.43, 1.23) | 0.238 | 1.08 | (0.59, 1.96) | 0.812 | 0.57 | (0.28, 1.16) | 0.122 | 0.84 | (0.43, 1.67) | 0.627 |
| Parental education | ||||||||||||
| No college education | (ref) | (ref) | (ref) | (ref) | ||||||||
| Any college education | 1.27 | (0.99, 1.63) | 0.057 | 1.10 | (0.80, 1.51) | 0.567 | 1.32 | (0.99, 1.77) | 0.063 | 1.39 | (1.01, 1.93) | 0.045 |
| Covariates measured from ages 19/20–29/30 | ||||||||||||
| 4-Year college experience | ||||||||||||
| Not attend | (ref) | (ref) | (ref) | (ref) | ||||||||
| Attend only | 1.03 | (0.69, 1.54) | 0.873 | 0.90 | (0.53, 1.54) | 0.702 | 0.80 | (0.49, 1.30) | 0.365 | 1.20 | (0.73, 1.97) | 0.469 |
| Graduate | 1.16 | (0.87, 1.55) | 0.318 | 0.94 | (0.64, 1.39) | 0.756 | 1.14 | (0.81, 1.60) | 0.459 | 1.43 | (0.99, 2.05) | 0.054 |
| Ever married | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 0.51 | (0.39, 0.65) | <0.001 | 0.74 | (0.54, 1.01) | 0.059 | 0.47 | (0.35, 0.63) | <0.001 | 0.94 | (0.68, 1.31) | 0.727 |
| Ever have a child | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 0.47 | (0.37, 0.61) | <0.001 | 0.64 | (0.47, 0.89) | 0.007 | 0.50 | (0.38, 0.68) | <0.001 | 0.63 | (0.46, 0.87) | 0.005 |
| Cohort group | ||||||||||||
| 1976–1980 | 1.12 | (0.78, 1.61) | 0.527 | 3.48 | (2.16, 5.61) | <0.001 | 2.85 | (1.81, 4.49) | <0.001 | 5.78 | (3.10, 10.79) | <0.001 |
| 1981–1985 | 0.69 | (0.47, 1.02) | 0.064 | 1.31 | (0.76, 2.25) | 0.326 | 1.87 | (1.16, 3.01) | 0.010 | 3.06 | (1.60, 5.87) | 0.001 |
| 1986–1990 | 1.06 | (0.68, 1.65) | 0.791 | 1.22 | (0.61, 2.40) | 0.576 | 1.13 | (0.61, 2.08) | 0.694 | 3.06 | (1.47, 6.36) | 0.003 |
| 1991–1995 | 1.12 | (0.69, 1.79) | 0.649 | 1.17 | (0.57, 2.43) | 0.668 | 1.60 | (0.88, 2.92) | 0.127 | 2.65 | (1.22, 5.77) | 0.014 |
| 1996–2000 | 0.94 | (0.65, 1.37) | 0.748 | 1.01 | (0.55, 1.87) | 0.971 | 1.00 | (0.58, 1.74) | 0.988 | 1.61 | (0.79, 3.27) | 0.190 |
| 2001–2006 | (ref) | (ref) | (ref) | (ref) | ||||||||
Notes: Sample n (unwtd) = 4,423. RR = Relative risk ratio (obtained from models regressing past 30-day marijuana use latent class membership on one covariate at a time).
Table 8.
Adjusted relative risk ratios from multivariable multinomial logistic regressions of past 30-day marijuana use latent class membership on covariates: Referent class = Discontinuers
| Class 1: | Class 2: | Class 3: | Class 4: | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Continued Frequent Users | Frequent to Non-Frequent Users | Consistent Non-Frequent Users | Non-Frequent Users to Discontinuers | |||||||||
| [v. Discontinuers] | [v. Discontinuers] | [v. Discontinuers] | [v. Discontinuers] | |||||||||
| ARR | (95% CI) | p | ARR | (95% CI) | p | ARR | (95% CI) | p | ARR | (95% CI) | p | |
| Covariates measured at 12th grade survey | ||||||||||||
| Sex | ||||||||||||
| Female | (ref) | (ref) | (ref) | (ref) | ||||||||
| Male | 1.02 | (0.78, 1.33) | 0.893 | 1.00 | (0.70, 1.42) | 0.993 | 1.16 | (0.84, 1.62) | 0.367 | 0.63 | (0.45, 0.88) | 0.006 |
| Race/ethnicity | ||||||||||||
| White | (ref) | (ref) | (ref) | (ref) | ||||||||
| African American | 0.50 | (0.27, 0.92) | 0.025 | 0.53 | (0.22, 1.24) | 0.140 | 0.69 | (0.38, 1.26) | 0.226 | 0.47 | (0.21, 1.07) | 0.073 |
| Hispanic | 0.45 | (0.23, 0.86) | 0.016 | 0.90 | (0.41, 1.97) | 0.787 | 0.42 | (0.16, 1.15) | 0.091 | 0.58 | (0.24, 1.42) | 0.233 |
| Other | 0.66 | (0.38, 1.16) | 0.153 | 1.05 | (0.55, 2.01) | 0.873 | 0.58 | (0.29, 1.18) | 0.135 | 0.83 | (0.39, 1.79) | 0.638 |
| Parental education | ||||||||||||
| No college education | (ref) | (ref) | (ref) | (ref) | ||||||||
| Any college education | 1.12 | (0.85, 1.47) | 0.427 | 1.30 | (0.88, 1.92) | 0.185 | 1.38 | (0.99, 1.93) | 0.056 | 1.52 | (1.07, 2.16) | 0.021 |
| Covariates measured from ages 19/20–29/30 | ||||||||||||
| 4-Year college experience | ||||||||||||
| Not attend | (ref) | (ref) | (ref) | (ref) | ||||||||
| Attend only | 0.94 | (0.62, 1.43) | 0.785 | 0.90 | (0.51, 1.59) | 0.721 | 0.72 | (0.42, 1.26) | 0.250 | 1.34 | (0.80, 2.24) | 0.260 |
| Graduate | 1.00 | (0.71, 1.40) | 0.978 | 0.94 | (0.60, 1.47) | 0.788 | 1.13 | (0.77, 1.66) | 0.527 | 1.56 | (1.02, 2.38) | 0.040 |
| Ever married | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 0.54 | (0.40, 0.74) | <0.001 | 0.56 | (0.38, 0.85) | 0.006 | 0.37 | (0.26, 0.53) | <0.001 | 0.68 | (0.47, 0.99) | 0.047 |
| Ever have a child | ||||||||||||
| No | (ref) | (ref) | (ref) | (ref) | ||||||||
| Yes | 0.61 | (0.45, 0.83) | 0.001 | 0.78 | (0.51, 1.20) | 0.255 | 0.76 | (0.52, 1.11) | 0.150 | 0.73 | (0.50, 1.05) | 0.089 |
| Cohort group | ||||||||||||
| 1976–1980 | 1.54 | (1.02, 2.33) | 0.038 | 4.91 | (2.82, 8.53) | <0.001 | 4.60 | (2.77, 7.65) | <0.001 | 9.00 | (4.60, 17.59) | <0.001 |
| 1981–1985 | 0.87 | (0.57, 1.32) | 0.517 | 1.64 | (0.90, 2.96) | 0.103 | 2.61 | (1.56, 4.37) | <0.001 | 4.49 | (2.27, 8.87) | <0.001 |
| 1986–1990 | 1.18 | (0.75, 1.85) | 0.475 | 1.39 | (0.69, 2.81) | 0.360 | 1.40 | (0.75, 2.63) | 0.293 | 3.60 | (1.70, 7.61) | 0.001 |
| 1991–1995 | 1.32 | (0.81, 2.15) | 0.272 | 1.32 | (0.62, 2.81) | 0.466 | 2.09 | (1.12, 3.88) | 0.020 | 3.20 | (1.47, 6.99) | 0.003 |
| 1996–2000 | 1.09 | (0.73, 1.62) | 0.677 | 1.13 | (0.60, 2.14) | 0.709 | 1.20 | (0.68, 2.13) | 0.521 | 1.98 | (0.96, 4.09) | 0.063 |
| 2001–2006 | (ref) | (ref) | (ref) | (ref) | ||||||||
Notes: Sample n (unwtd) = 4,423. ARR = Adjusted relative risk ratio (obtained from models regressing past 30-day marijuana use latent class membership on all listed covariates simultaneously).
Membership likelihood in the Frequent to Non-Frequent Users class (Class 2) versus Discontinuers was significantly higher for those never married and the 1976–1980 (vs. 2001–2006) cohorts. Membership likelihood in the Consistent Non-Frequent Users class (Class 3) versus Discontinuers was significantly higher for those never married and the 1976–1980, 1981–1985, and 1991–1995 (vs. 2001–2006) cohorts. The likelihood of membership in the Non-Frequent Users to Discontinuers class (Class 4) versus Discontinuers was significantly higher for females, those whose parents had some college education, those who graduated from a 4-year college, those who never married, and the 1976–1980, 1981–1985, 1986–1990, and 1991–1995 (vs. 2001–2006) cohorts.
Discussion
Using data from multiple cohorts of US nationally representative samples of 12th grade students who were followed longitudinally on a biennial basis, we found that those who reported frequent marijuana use as 12th grade students (approximately 6% of recent 12th grade samples [22]) could be meaningfully separated into five latent classes based on longitudinal patterns of past 30-day marijuana use from ages 19/20–29/30. Latent class membership was associated significantly with sociodemographic covariates and varied across cohorts. The highest risk for continued frequent marijuana use belonged to individuals in more recent cohorts and those who did not enter roles of marriage and parenthood.
These results support findings from prior studies showing strong connections between adolescent marijuana use and continued use during adulthood [34], particularly among adolescents reporting frequent use [8]. Almost 60% of 12th grade frequent marijuana users were estimated to be members of latent classes involving some level of past 30-day marijuana use through age 29/30. Yet, approximately 40% of 12th grade frequent users ceased reporting past 30-day marijuana use by age 29/30—slightly more than half of these individuals discontinued from age 19/20 onwards. Observed cohort differences indicated that the likelihood of being in either “extreme” class (CFU or Discontinuers) compared with other classes was significantly higher for more recent cohorts (2001–2006). What might be behind increasing membership likelihood in these very different classes?
Growing CFU class membership indicates that an increasing number of individuals who were frequent marijuana users as 12th graders have continued frequent use through age 29/30 versus exhibiting age-related developmental decreases in use. Such increases would be consistent with recent MTF research finding (a) historically recent high levels of frequent marijuana use among young adults aged 27–30 in general [24]; and (b) developmental patterns of frequent marijuana use that do not reflect convergence to lower use levels by age 29/30 among recent cohorts [25].
Increasing CFU class membership may be associated with changes in marijuana potency and perceived risk, as well as changing trends related to adult social roles. US marijuana potency has increased notably since the 1980s [35,36]; increasing potency may be associated with higher dependence among long-term frequent marijuana users [37,38], and decreased likelihood of reducing use with age. The protective association between perceived risk of marijuana use and actual use likelihood also has weakened among those aged 26–30 in recent cohorts [39]. A weakened perceived risk/use association in the late 20s may be contributing to decreased likelihood of age-related use reduction. A third possibility for CFU class growth may be changing involvement in social roles of marriage and parenthood [26]. Both marriage and parenthood are associated with increased likelihood of decreasing/stopping marijuana use [26,40–42]. A growing percentage of US young adults are delaying marriage [43] and parenthood [44–46]. To the extent that the analytic sample reflects decreasing rates of marriage and parenthood, recent cohorts would experience fewer reasons to move towards lower levels of marijuana use across age. Additional analyses confirmed that this was, indeed, the case. The percentage of the analytic sample reporting ever marrying by age 29/30 decreased linearly from 66.6% among the 1976–1980 cohort group to 29.0% among the 2001–2006 cohort group, and the percentage who reported parenthood decreased linearly from 54.6% to 44.0% (Mantel-Haenszel chi square p=<0.001 for both). Whatever the contributing factors may be, growing CFU class membership indicates increased health risks associated with chronic frequent marijuana use [2–6,8,10,11] for US adults entering their 30s.
Given the potential factors behind increasing CFU class membership discussed above, a co-occurring historical increase in the Discontinuers class appeared all the more surprising. Prior research examining desistance versus continued use among young adult heavy marijuana users wishing to stop use found that those who successfully desisted pointed to a key life event that induced quitting: either experiencing acute negative effects of use, or an event that resulted in marijuana no longer being a suitable part of their lives [47]. From this perspective, both increasing marijuana potency and historical change in social roles may be contributing to growing Discontinuers class membership. Increasing potency is linked with negative experiences such as psychotic outcomes [48]. More recent 12th grade cohorts who report frequent use are experiencing higher-potency marijuana than prior cohorts, and may be at higher risk for negative acute effects. Historical change in social roles may also be a factor, in that while marriage and parenthood are decreasing, college attendance has increased markedly [49]. Marijuana use, particularly frequent marijuana use, among full-time college students is markedly lower than among others [24], and the percentage of the analytic sample who reported ever graduating with a 4-year college degree by age 29/30 increased linearly from 19.1% among the 1976–1980 cohort group to 31.5% among the 2001–2006 cohort group (Mantel-Haenszel chi square p<0.001). To the extent that a 12th grade frequent marijuana user considers such use to be incongruent with a desired transition to college, the end of high school may provide a key life event precipitating discontinuation from ages 19/20 onward. More generally, our findings regarding cohort variation in the trajectories indicate that the etiology of frequent marijuana use has shifted in important ways over the past three decades.
With respect to covariates other than cohort, the strongest associations with class membership were related to social roles of marriage and parenthood, with slightly weaker associations related to college attendance. These findings are consistent with prior substance use trajectory research among the general population [27], and may help identify individuals at higher risk of continuing frequent marijuana use. To the extent that national trends in marriage, parenthood, and college attendance continue, continued increases in the percentage of individuals who continue frequent marijuana use across young adulthood may be likely.
Strengths and limitations
All data were self-report, based on individuals who reported frequent marijuana use in 12th grade. School dropout is associated with higher marijuana use [50]; results may or may not generalize to frequent marijuana users who drop out prior to 12th grade. The sample was also subject to attrition. While use of weighting adjustments addressed attrition in part, frequent marijuana use is likely associated with missingness; estimated membership prevalence for the CFU class may thus be somewhat underestimated. Future research should examine if the associations between the social roles of marriage and parenthood and class membership vary between men and women. Such limitations notwithstanding, the current study’s strengths include multi-cohort longitudinal data based on nationally representative samples of 12th grade students, collected using consistent measurement over four decades (1976 to 2017).
Conclusions
Across the past three decades, almost one in four US 12th grade frequent marijuana users continued high frequency use across young adulthood. The proportion of continuing frequent marijuana users has increased among more recent cohorts, with meaningful implications for future treatment demand and potential social and individual costs.
Supplementary Material
Acknowledgments
This work was supported by the National Institute on Drug Abuse (DA001411 and DA016575). The National Institute on Drug Abuse had no further role in the study design; in the collection, management, analysis, and interpretation of the data; in the writing of the manuscript; nor in the decision to submit the paper for publication. The views expressed in this article are those of the authors and do not necessarily reflect the views of the funders.
Footnotes
Declarations of competing interest: None
References
- 1.Becker MP, Collins PF, Luciana M Neurocognition in college-aged daily marijuana users. J Clin Exp Neuropsychol 2014; 36: 379–398. doi: 10.1080/13803395.2014.893996 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Bolla KI, Brown K, Eldreth D, Tate K, Cadet JL Dose-related neurocognitive effects of marijuana use. Neurol 2002; 59: 1337–1343. 10.1212/01.WNL.0000031422.66442.49 [DOI] [PubMed] [Google Scholar]
- 3.Crean RD, Crane NA, Mason BJ An evidence based review of acute and long-term effects of cannabis use on executive cognitive functions. J Addict Med 2011; 5: 1–8. doi: 10.1097/ADM.0b013e31820c23fa [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hall W, Degenhardt L Adverse health effects of non-medical cannabis use. Lancet 2009; 347: 1383–1391. 10.1016/S0140-6736(09)61037-0 [DOI] [PubMed] [Google Scholar]
- 5.Auer R, Vittinghoff E, Yaffe K, Künzi A, Kertesz SG, Levine DA et al. Association between lifetime marijuana use and cognitive function in middle age: the Coronary Artery Risk Development in Young Adults (CARDIA) study. JAMA Intern Med 2016; 176: 352–361. doi: 10.1001/jamainternmed.2015.7841 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Caldeira KM, O’Grady KE, Vincent KB, Arria AM Marijuana use trajectories during the post-college transition: health outcomes in young adulthood. Drug Alcohol Depend 2012; 125: 267–275. doi: 10.1016/j.drugalcdep.2012.02.022 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Epstein M, Hill KG, Nevell AM, Guttmannova K, Bailey JA, Abbott RD et al. Trajectories of marijuana use from adolescence into adulthood: environmental and individual correlates. Dev Psychol 2015; 51: 1650–1663. doi: 10.1037/dev0000054 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Patton GC, Coffey C, Lynskey MT, Reid S, Hemphill S, Carlin JB, et al. Trajectories of adolescent alcohol and cannabis use into young adulthood. Addiction 2007; 102: 607–615. doi: 10.1111/j.1360-0443.2006.01728.x [DOI] [PubMed] [Google Scholar]
- 9.Schulenberg JE, Merline AC, Johnston LD, O’Malley PM, Bachman JG, Laetz VB Trajectories of marijuana use during the transition to adulthood: the big picture based on national panel data. J Drug Issues 2005; 35: 255–279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Terry-McElrath YM, O’Malley PM, Johnston LD, Bray BC, Patrick ME, Schulenberg JE Longitudinal patterns of marijuana use across ages 18–50 in a US national sample: a descriptive examination of predictors and health correlates of repeated measures latent class membership. Drug Alcohol Depend 2017; 171: 70–83. 10.1016/j.drugalcdep.2016.11.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Volkow ND, Baler RD, Compton WM, Weiss SRB Adverse health effects of marijuana use. N Engl J Med 2014; 370: 2219–2227. doi: 10.1056/NEJMra1402309 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Maggs JL, Staff J, Kloska DD, Patrick ME, O’Malley PM, Schulenberg JE Predicting young adult degree attainment by late adolescent marijuana use. J Adolesc Health 2015; 57: 205–211. doi: 10.1016/j.jadohealth.2015.04.028 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Brook JS, Zhang C, Brook DW Antisocial behavior at age 37: developmental trajectories of marijuana use extending from adolescence to adulthood. Am J Addict 2011; 20: 509–515. doi: 10.1111/j.1521-0391.2011.00179.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Brown TL, Flory K, Lynam DR, Leukefeld C, Clayton RR Comparing the developmental trajectories of marijuana use of African American and Caucasian adolescents: patterns, antecedents, and consequences. Exp Clin Psychopharmacol 2004; 12: 47–56. doi: 10.1037/1064-1297.12.1.47 [DOI] [PubMed] [Google Scholar]
- 15.Ellickson PL, Martino SC, Collins RL Marijuana use from adolescence to young adulthood: multiple developmental trajectories and their associated outcomes. Health Psychol 2004; 23: 299–307. doi: 10.1037/0278-6133.23.3.299 [DOI] [PubMed] [Google Scholar]
- 16.Jackon KM, Sher KJ, Schulenberg JE Conjoint developmental trajectories of young adult substance use. Alcohol Clin Exp Res 2008; 32: 723–737. doi: 10.1111/j.1530-0277.2008.00643.x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Nelson SE, Van Ryzin MJ, Dishion TJ Alcohol, marijuana, and tobacco use trajectories from age 12 to 24 years: demographic correlates and young adult substance use problems. Dev Psychopathol 2015; 27: 253–277. doi: 10.1017/S0954579414000650 [DOI] [PubMed] [Google Scholar]
- 18.Passarotti AM, Crane NA, Hedeker D, Mermelstein RJ Longitudinal trajectories of marijuana use from adolescence to young adulthood. Addict Behav 2015; 45: 301–308. 10.1016/j.addbeh.2015.02.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Tait RJ, Mackinnon A, Christensen H Cannabis use and cognitive function: 8-year trajectory in a young adult cohort. Addiction 2011; 106: 2195–2203. doi: 10.1111/j.1360-0443.2011.03574.x [DOI] [PubMed] [Google Scholar]
- 20.White HR, Bechtold J, Loeber R, Pardini D Divergent marijuana trajectories among men: socioeconomic, relationship, and life-satisfaction outcomes in the mid-30s. Drug Alcohol Depend 2015; 156: 62–69. 10.1016/j.drugalcdep.2015.08.031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Johnston LD Characteristics of the daily marijuana user In: de Silva R, DuPont RL, Russell GK, editors, Treating the Marijuana-Dependent Person. New York: The American Council on Marijuana; 1981, pp. 12–15. [Google Scholar]
- 22.Miech RA, Johnston LD, O’Malley PM, Bachman JG, Schulenberg JE, Patrick ME Monitoring the Future National Survey Results on Drug Use, 1975–2017: Volume I, Secondary School Students. Ann Arbor: Institute for Social Research, The University of Michigan; 2018. [Google Scholar]
- 23.Compton WM, Han B, Jones CM, Blanco C, Hughes A Marijuana use and use disorders in adults in the USA, 2002–14: analysis of annual cross-sectional surveys. Lancet Psychiatry 2016; 3: 954–964. 10.1016/S2215-0366(16)30208-5 [DOI] [PubMed] [Google Scholar]
- 24.Schulenberg JE, Johnston LD, O’Malley PM, Bachman JG, Miech RA, Patrick ME Monitoring the Future National Survey Results on Drug Use, 1975–2017: Volume II, College Students and Adults Ages 19–55. Ann Arbor, MI: Institute for Social Research, The University of Michigan; 2018. [Google Scholar]
- 25.Terry-McElrath YM, Patrick ME, O’Malley PM, Johnston LD The end of convergence in developmental patterns of frequent marijuana use from ages 18 to 30: an analysis of cohort change from 1976–2016. Drug Alcohol Depend 2018; 191: 203–209. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Bachman JG, O’Malley PM, Schulenberg JE, Johnston LD, Bryant AL, Merline AC The decline of substance use in young adulthood: changes in social activities, roles, and beliefs. Mahwah, NJ: Lawrence Erlbaum Associates, Publishers; 2002. [Google Scholar]
- 27.Staff J, Schulenberg JE, Maslowsky J, Bachman JG, O’Malley PM, Maggs JL, Johnston LD Substance use changes and social role transitions: proximal developmental effects on ongoing trajectories from late adolescence through early adulthood. Dev Psychopathol 2010; 22: 917–932. doi: 10.1017/S0954579410000544 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Juon H-S, Fothergill KE, Green KM, Doherty EE, Ensminger ME Antecedents and consequences of marijuana use trajectories over the life course in an African American population. Drug Alcohol Depend 2011; 118: 216–223. doi: 10.1016/j.drugalcdep.2011.03.027 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Silins E, Hutchinson D, Swift W, Slade T, Toson B, Rodgers B Factors associated with variability and stability of cannabis use in young adulthood. Drug Alcohol Depend 2013; 133: 452–458. 10.1016/j.drugalcdep.2013.07.003 [DOI] [PubMed] [Google Scholar]
- 30.Muthén LK, Muthén BO Mplus users’s guide. Eighth edition. Los Angeles, CA: Muthén & Muthén; 1998–2017. [Google Scholar]
- 31.Collins LM, Lanza ST Latent class and latent transition analysis: with applications in the social, behavioral, and health sciences. Hoboken, NJ: John Wiley & Sons, Inc.; 2010. [Google Scholar]
- 32.Dziak JJ, Lanza ST, Tan X Effect size, statistical power, and sample size requirements for the bootstrap likelihood ratio test in latent class analysis. Struct Equ Modeling 2014; 21, 1–19. doi: 10.1080/10705511.2014.919819 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Asparouhov T, Muthén B Auxiliary variables in mixture modeling: using the BCH method in Mplus to estimate a distal outcome model and an arbitrary secondary model. Mplus Web Notes #21, Version 2; May 14 2015. Available at: http://www.statmodel.com/examples/webnotes/webnote21.pdf (accessed 13 August 2018) (archived at http://www.webcitation.org/71eYN6Bg9 on 13 August 2018).
- 34.Merline AC, O’Malley PM, Schulenberg JE, Bachman JG, Johnston LD Substance use among adults 35 years of age: prevalence, adulthood predictors, and impact of adolescent substance use. Am J Public Health 2004; 94: 96–102. https://ajph.aphapublications.org/doi/pdf/10.2105/AJPH.94.1.96 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.ElSohly MA, Mehmedic Z, Foster S, Gon C, Chandra S, Church JC, Changes in cannabis potency over the last 2 decades (1995–2014): analysis of current data in the United States. Biol Psychiatry 2016; 79: 613–619. 10.1016/j.pbipsych.2016.01.004 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.ElSohly MA, Ross SA, Mehmedic A, Arafat R, Yi B, Banahan BF III., Potency trends of Δ9-THC and other cannabinoids in confiscated marijuana from 180–1997. J Forensic Sci 2000; 43: 24–30. [PubMed] [Google Scholar]
- 37.Freeman TP, van der Pol P, Kuijpers W, Wisselink J, Das RK, Rigter S, et al. Changes in cannabis potency and first-time admissions to drug treatment: a 16-year study in the Netherlands. Psycholol Medicine 2018. 10.1017/S0033291717003877 [DOI] [PubMed] [Google Scholar]
- 38.Freeman TP, Winstock AR Examining the profile of high-potency cannabis and its association with severity of cannabis dependence. Psychol Medicine 2015; 45: 3181–3189. doi: 10.1017/S0033291715001178 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Terry-McElrath YM, Patrick ME, O’Malley PM Stability and change in perceived risk associations with binge drinking and marijuana use among US young adults: a national study, 1990–2016 In: Terry-McElrath YM, VanderWaal CJ, Baltazar AM, Trim DJB, editors, Promoting the Public Good: Policy in the Public Square and the Church. Cooranbong, NSW: Avondale Academic Press; Silver Spring, MD: Office of Archives, Statistics and Research, General Conference of Seventh-day Adventists; [Google Scholar]
- 40.Brook JS, Richter L, Whiteman M, Cohen P Consequences of adolescent marijuana use: incompatibility with the assumption of adult roles. Genet Soc Gen Psychol Monogr 1999; 125: 193–207. [PubMed] [Google Scholar]
- 41.Maume MO, Ousey GC, Eaver K Cutting the grass: a reexamination of the link between marital attachment, delinquent peers and desistance from marijuana use. J Quant Criminol 2005; 21: 27–53. doi: 10.1007/s10940-004-1786-3 [DOI] [Google Scholar]
- 42.Yamaguchi K, Kandel DB On the resolution of role incompatibility: a life event history analysis of family roles and marijuana use. Am J Sociology 1985; 90: 1284–1325. [Google Scholar]
- 43.United States Census Bureau. Median Age at First Marriage: 1890 to Present. Suitland, MD: US Department of Commerce, Economics and Statistics Administration, US Census Bureau; 2017. Available at: https://www.census.gov/content/dam/Census/library/visualizations/time-series/demo/families-and-households/ms-2.pdf (accessed 13 August 2018). [Google Scholar]
- 44.Khandwala YS, Zhang CA, Lu Y, Eisenberg ML The age of fathers in the USA is rising: an analysis of 168,867,480 births from 1972 to 2015. Hum Reprod 2017; 32: 2110–2116. 10.1093/humrep/dex267 [DOI] [PubMed] [Google Scholar]
- 45.Mathews TJ, Hamilton BE Delayed Childbearing: More Women Are Having Their First Child Later in Life NCHS Data Brief N. 21. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics; 2009. Available at: https://www.cdc.gov/nchs/data/databriefs/db21.pdf (accessed 13 August 2018) (archived at http://www.webcitation.org/71eYbftPv 13 August 2018). [Google Scholar]
- 46.Mathews TJ, Hamilton BE Mean Age of Mothers is On the Rise: United States, 2000–2014 NCHS Data Brief N. 232. Hyattsville, MD: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics; 2016. Available at: https://www.cdc.gov/nchs/data/databriefs/db232.pdf (accessed 13 August 2018) (archived at http://www.webcitation.org/71eYg8Fhb 13 August 2018). [Google Scholar]
- 47.Liebregts N, van der Pol P, de Graaf R, van Laar M, van den Brink W, Korf DJ Persistence and desistance in heavy cannabis use: the role of identity, agency, and life events. J Youth Stud 2015; 18: 617–533. 10.1080/13676261.2014.992320 [DOI] [Google Scholar]
- 48.Murray RM, Quigley H, Quattrone D, Englund A, Di Forti M Traditional marijuana, high-potency cannabis and synthetic cannabinoids: increasing risk for psychosis. World Psychiatry 2016; 15: 195–204. doi: 10.1002/wps.20341 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.National Center for Education Statistics. US college enrollment statistics for public and private colleges from 1965 to 2016 and projections up to 2017 (in millions). In Statista: The Statistics Portal; n.d. Available at: https://www.statista.com/statistics/183995/us-college-enrollment-and-projections-in-public-and-private-institutions/ (accessed 13 August 2018) (archived at http://www.webcitation.org/71eYkFoiE 13 August 2018). [Google Scholar]
- 50.Tice P, Lipari RN, Van Horn SL Substance Use Among 12th Grade Aged Youths, by Dropout Status The CHBSQ Report: August 15. Rockville, MD: Center for Behavioral Health Statistics and Quality, Substance Abuse and Mental Health Services Administration; 2017. Available at: https://www.samhsa.gov/data/sites/default/files/report_3196/ShortReport-3196.pdf (accessed 13 August 2018). [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.