Longitudinal associations of sleep problems with alcohol and cannabis use from adolescence to emerging adulthood

Abstract

Study Objectives

This study examined longitudinal associations of sleep problems with alcohol and cannabis use across six annual waves of data from adolescence to emerging adulthood.

Methods

Participants were 3,265 youth from California (ages 16–22 across waves). At each wave, past-month alcohol use and cannabis use, mental health, and several dimensions of sleep health (i.e. social jetlag, bedtimes, time in bed, trouble sleeping) were assessed via questionnaire. Parallel process latent growth models examined the association between sleep and alcohol or cannabis use trajectories and the role of mental health in contributing to such trajectories.

Results

Smaller declines in social jetlag (r = 0.11, p = 0.04), increases in trouble sleeping (r = 0.18, p < 0.01), and later weekday (r = 0.16, p < 0.01) and weekend bedtimes (r = 0.25, p < 0.01) were associated with increases in likelihood of alcohol use over time. Declines in weekend TIB (r = −0.13, p = 0.03), as well as increases in weekday TIB (r = 0.11, p = 0.04) and later weekday (r = 0.18, p < 0.01) and weekend bedtime (r = 0.24, p < 0.01), were associated with increases in likelihood of cannabis use over time. Most associations remained significant after controlling for time-varying mental health symptoms.

Conclusions

Trajectories of sleep health were associated with trajectories of alcohol and cannabis use during late adolescence to emerging adulthood. Improving sleep is an important target for intervention efforts to reduce the risk of substance use during this critical developmental transition.

Statement of Significance.

Using six waves of data from a large sample of older adolescents and emerging adults, we demonstrated longitudinal associations between trajectories of sleep health and trajectories of alcohol or cannabis use. Importantly, worsening of sleep health across several dimensions (e.g. bedtime, trouble sleeping, weekend time in bed) was associated with an increased likelihood of alcohol and cannabis use during this critical transition period for youth. Our findings have important implications for future intervention research, including the critical need to consider both policy changes and developmentally appropriate interventions to improve sleep health and prevent substance use in adolescents and emerging adults.

Introduction

Adolescence and emerging adulthood (ages 12–25) [1] are vulnerable developmental periods characterized by crucial transitions, such as increased autonomy [2] and changes in circadian function [3, 4], reward function [5, 6], and affect. Such changes have direct implications for the development of sleep problems and disorders [7, 8], as well as risk for the development of alcohol and cannabis use and other drug use disorders [9, 10]. Both sleep problems and alcohol and cannabis use during these critical developmental periods can set the stage for poor mental and physical health outcomes that persist through adulthood [11–16]. Among adolescents and emerging adults, the two most widely used drugs are alcohol and cannabis [10]. In recent years, cannabis has shown a significant spike in use [17], perhaps at least partially due to the legalization of recreational and medical cannabis outlets in certain states [18, 19] such as California.

Due to the confluence of biological changes in sleep timing (circadian phase delay) which begins during adolescence and persists into the early 20s, societal obligations (e.g. early school start times) and social pressures, adolescents and emerging adults are at high risk for sleep problems and circadian misalignment [20, 21]. For example, national data show that 70% of adolescents average less than 8 hours of sleep per night, although 8–10 h is the recommended sleep duration for this age group [22, 23]. However, sleep problems among adolescents extend beyond insufficient sleep, and they do not disappear at the conclusion of high school. Of particular concern is social jetlag, which refers to large discrepancies in weekend/weekday sleep timing [8, 24] or otherwise erratic sleep schedules. Social jetlag is common among adolescents and emerging adults [7, 25], owing in part to the delayed sleep phase characteristic of these developmental stages in conjunction with societal demands, such as early school start times. For example, in a study of over 1,000 undergraduates, Lund et al. [26] found high levels of insufficient sleep, as well as erratic sleep patterns and clinically significant poor sleep quality.

Among adults, sleep and circadian disruptions longitudinally predict the development of substance use and substance use disorders (SUD) [27, 28]. In addition, sleep problems are associated with a 2-fold increased risk of depression [29], which may indirectly lead to increased risk of substance use and disorders, given known comorbidities between poor mental health and SUD [30]. Several studies have demonstrated cross-sectional associations between sleep problems (including insomnia, short sleep duration, and social jetlag) and increased use of alcohol, cigarette, or other drugs in adolescent and emerging adult samples [31–37]. However, cross-sectional studies cannot elucidate causal mechanisms or the temporal ordering of associations, which is particularly important given evidence that associations between sleep problems and substance use may be bidirectional [38].

There are few longitudinal studies on sleep and any type of substance use in adolescence [38–48], and even fewer studies that examine this association during the critical developmental transition from adolescence into young adulthood [40, 46–48], when sleep problems remain prevalent [49] and alcohol and cannabis use increase [50]. Furthermore, there are several limitations of the existing longitudinal literature that affect generalizability of results and the ability to infer causal associations. First, most of the extant literature has focused on past-year alcohol or other drug use [51, 52], or has used a composite measure of any substance use [48], which limits understanding of the specific association between sleep problems and the two most commonly used substances among adolescents: alcohol and cannabis. Understanding the role of sleep problems in relation to cannabis use is particularly salient given increasing legalization of cannabis use in the United States, including states such as California [17], where recreational and medical marijuana use is legal. Second, existing studies have generally focused on one or two dimensions of sleep (e.g. insomnia [40] or social jetlag [48]). However, sleep is a multidimensional state and evidence suggests that multiple dimensions of sleep (e.g. quality, duration, and timing) may be implicated in substance use risk [53]. Finally, most existing studies measure sleep problems in childhood or early adolescence and predict alcohol or other drug use several years later or at a single, later time point [47, 54]. Such limited frequency of follow-up may miss temporal associations between sleep health and alcohol or other drug use. This is particularly relevant for identifying opportunities for intervention during the transition from adolescence into young adulthood, a developmental period characterized by increased autonomy from parents, and dramatic changes in behavior and biology [1]. Furthermore, given that mental health symptoms systematically covary with sleep problems and substance use across development [29, 30], a more fine-grained analysis with sufficient frequency of assessments is critical to elucidate the role of mental health symptoms in contributing to sleep and substance use trajectories, across this dynamic period of development.

To our knowledge, the current study is the first to examine longitudinal trajectories of sleep health and alcohol and cannabis use over six waves of annual data and spanning the ages of 16−22. With such repeated assessments across this dynamic developmental transition, we can address key gaps in the literature concerning the role of sleep in contributing to alcohol and cannabis use over time and evaluate the degree to which mental health symptoms may contribute to these trajectories. We focus on several key dimensions of sleep that have previously been associated with alcohol and cannabis use [39, 51, 52, 55], and hypothesize that reductions in time in bed (as an indicator for sleep duration), later sleep timing, poorer sleep quality, and greater social jetlag over time would be associated with increases in alcohol and cannabis use during the transition from adolescence to emerging adulthood. We utilize parallel process latent growth curve models to examine over-time associations between sleep measure and alcohol and cannabis use trajectories from ages 16 to 22 years. Recognizing the important role of mental health symptoms in contributing to both sleep problems and substance use, we include mental health as a time-varying covariate in parallel-process models to determine whether associations persist after statistically adjusting for wave-specific effects of mental health on sleep measures and alcohol and cannabis use. Finally, given that our cohort is from California, and is comprised of a diverse sample of youth, we have a unique opportunity to examine sleep and alcohol and cannabis use trajectories in a state with legalized marijuana.

Methods

Sample and procedures

The current study utilized six waves of data from wave 6 (May 2013 to April 2014) to wave 11 (July 2018 to June 2019) from a large multi-wave study of adolescents and emerging adults in southern California. The current analytic sample included 3,265 youth (mean age 16.2 at Wave 6 and 21.6 at Wave 11). The original sample included participants recruited in sixth and seventh grades (ages 11−13) in 2008 for an alcohol and other drug prevention program, CHOICE, and were representative of the 16 middle schools in southern California from which they were recruited [56]. The study has a Certificate of Confidentiality and all procedures were approved by the [BLIND] Human Subjects Protection Committee.

Participants completed waves 1 through 5 in middle school with follow-up rates ranging from 74% to 90%, excluding new youth that could have come in at a subsequent wave. Adolescents transitioned to over 200 different high schools and were re-contacted and re-consented to complete annual web-based surveys. At wave 6, 61% of adolescents participated in the follow-up survey. At subsequent annual assessments, retention rates ranged from 80% to 92%. If a participant did not complete a wave of data collection, they were still invited and eligible to complete subsequent waves. They did not “dropout” of the study once they missed a survey wave; rather, we fielded the full sample every wave so all participants were invited to complete each individual survey. Participants were paid $50 for completing each web-based survey. Substance use at wave 10 did not significantly predict retention at wave 11, similar to previous waves [57]; however, compared to those who did not complete wave 11, retained participants were slightly more likely to be female (94% vs. 91%) and tended to be slightly younger at wave 10 (mean = 20.6 years vs. 20.9 years). We did not find a significant difference in retention by race/ethnicity.

Measures

Covariates

Variables included as covariates were: self-reported age, gender, race/ethnicity, mother’s education, and CHOICE intervention status. (The CHOICE program, conducted over 10 years ago, showed effects on youths’ alcohol use at one year after the program; however, no effects were observed beyond one year for any substance use outcomes, and intervention status at wave 1 is unrelated to substance use or retention across study waves.) Participants were classified into one of five racial/ethnic groups: non-Hispanic White (reference group), non-Hispanic Black, Hispanic, Asian, and other (e.g. Native American, Native Hawaiian, multi-racial). Mental health at waves 6−11 was assessed using the Mental Health Inventory (MHI-5) [58], which comprises five items focused on past month anxiety and depression symptoms (e.g. felt downhearted and blue, been a very nervous person). Scores on the MHI-5 were scaled such that they ranged from 0 to 100, with higher scores indicating better mental health (α = 0.75 at initial Wave, i.e. wave 6). Mental health was included as a time-varying covariate given the possible fluctuations in mental health across waves.

Past month alcohol use (wave 6−11)

Use of alcohol was assessed by asking: “During the past month, how many days did you have at least one drink of alcohol?” Responses ranged from 0 days to 20–30 days. Given that the focus was on any use given low base rates of substance use, particularly at earlier waves, responses were dichotomized to indicate any (1) versus no (0) use, as in prior research [50].

Past month cannabis use (wave 6−11)

Cannabis use was assessed by asking: “During the past month, how many days did you use marijuana (e.g. pot, weed, grass, hash, bud, sins)” and was also coded as a binary variable to indicate any (1) versus no past month use (0) given lower rates of use at younger ages [57].

Sleep measures

Self-reported sleep items were added to the ongoing survey in 2013 (wave 6) and continued through 2019 (wave 11). The primary sleep measures for the current study were: weekday and weekend bedtime (separate items); weekday and weekend sleep time in bed (TIB based on separate items; derived from reported bedtimes and waketimes); trouble sleeping (as an indicator of sleep quality); and social jetlag. Similar measures have been used in several prior studies with adolescents [59, 60], and have been shown to significantly correlate with diary and actigraphy assessments of sleep [61].

Specifically, for weekday and weekend TIB, respectively, youth were asked “What time do you usually go to bed on school days (or weekend days)?” and “What time do you usually wake-up on school days (or weekends)?.” The difference between bedtime and waketime was computed and used in analyses for weekday and weekend TIB. Consistent with prior research, we consider TIB as an indicator of sleep duration, though it may overestimate sleep duration, as it does not account for sleep latency or wakefulness after sleep onset [62, 63]. Weekday and weekday TIB were calculated separately given known variability between weekday and weekend sleep patterns for adolescents [64].

Trouble sleeping was assessed using a single item taken from the 15-item Patient Health Questionnaire (PHQ) Somatic Symptom Severity Scale measure [65]. Youth were asked “During the past 4 weeks, how much have you been bothered by trouble sleeping?” Responses included 1 = not bothered, 2 = bothered a little, and 3 = bothered a lot.

Per recent recommendations for the calculation of social jetlag [66], this measure was calculated as the actual difference between the midpoint of sleep (based on bedtimes and waketimes) for weekdays and weekends, expressed in hours. Higher (positive) values indicate a greater degree of phase “delay” (later weekend relative to weekday) and negative values indicate an “advance” (“earlier weekend relative to weekday”).with the former being more relevant for adolescent substance use [67]. In the present sample, this measure captures the degree of social jetlag delay for nearly all participants in that only 2–6% of participants (depending on wave) reported a social jetlag advance.

Statistical Analysis

Latent growth models (LGMs) were first estimated for each sleep (e.g. weekend bedtime) and substance use measure (e.g. cannabis) to examine individual trajectories (i.e. changes over time) before investigating how sleep and substance use covary over time. Then for each sleep and substance use measure combination from wave 6 to wave 11, we estimated parallel process LGMs in a structural equation modeling framework using Mplus v8 (see user manual for example model specification syntax) [68]. This framework extends the standard LGM [69] used in initial analyses by simultaneously modeling multiple longitudinal processes [70]. In LGM, one obtains, at a minimum, two growth factors: one intercept and one slope. For the present models, we also allowed residual variances to freely estimate. The model intercept represents the predicted value of the outcome when the predictor is equal to zero, and thus represents a baseline level or probability. Time was coded in 1-unit increments from 0 to 5, with wave 6 coded as zero. Thus, the intercept reflects the initial status at wave 6. The slope represents the change in level or in the probability over time. Further, initial status (intercept) and change (slope), itself, can serve as both an outcome and a predictor. Given that multiple longitudinal processes were modeled simultaneously, each process has its own growth factors, and, more importantly, cross-process growth factors were also allowed to correlate (e.g., slope of weekend bedtime correlating with a slope of cannabis use; see Supplemental Figure 1). Importantly, to identify the unique association between cross-process slopes, all models regressed slopes on intercepts within-processes (e.g. sleep measure slope on sleep measure intercept) in addition to regressing cross-process slopes on intercepts (e.g. sleep measure slope on substance use intercept). This partitioning of variance permits modeling the unique association between slopes after accounting for the potential effects of within and cross-process associations. That is, slope associations were examined after controlling for the effects of both intercepts on slopes.

The first set of parallel process models that were estimated included the time-invariant covariates described above (e.g. race/ethnicity, gender). For models with significant longitudinal associations, we followed up models by incorporating mental health as a time-varying covariate. That is, we modeled longitudinal processes accounting for the wave by wave effect of mental health which could vary across each measurement wave (see Supplemental Figure 2). Each model was evaluated for model fit using conventional fit indices and criteria: root mean square error of approximation (RMSEA), comparative fit index (CFI), and Standardized Root Mean Square Residual (SRMR). Note χ ² is not included given that it is extremely sensitive to sample size and in large samples, as the one used here, even trivial differences are significant thus rendering this fit index uninformative. We used the weighted least squares with mean and variance adjusted estimator (WLSMV), which can accommodate categorical and ordinal data, missing data, and provide unbiased and consistent estimates [71]. Although prior literature [72] suggests that there may be shifts in sleep timing coinciding with post-high school education, our wave-to-wave descriptive analyses did not find evidence for non-linear patterns over time. Nevertheless, we tested for quadratic terms, and these were not statistically significant. Lastly, for all final models with significant associations between slopes after accounting for time-varying mental health, we conducted a set of exploratory models stratified by gender to evaluate potential gender differences in observed associations. A significance level of 0.05 was used throughout.

Results

As shown in Table 1, participants were 16.21 (SD = 0.72) years old on average at wave 6 (first wave of data collection for present analyses); 47% male and 45% Hispanic. Wave to wave descriptive statistics for sleep, alcohol and cannabis use, and mental health symptoms are reported in Table 2.

Table 1.

Sample descriptives (N = 3,265)

Covariates
Age (wave 6), mean (SD), years	16.2 (0.7)
Male gender, n (%)	1,526 (46.8%)
Race/ethnicity, n(%)
Non-Hispanic White	644 (19.7%)
Non-Hispanic Black	71 (2.2%)
Hispanic	1,516 (46.5%)
Non-Hispanic Asian	648 (19.9%)
Non-Hispanic other/multi-racial	385 (11.8%)
Mother’s education, n (%)
Did not finish high school	442 (13.5%)
High school	540 (16.5%)
Some college	423 (13.0%)
College	1,621 (49.7%)
Don’t know	238 (7.3%)
CHOICE intervention, n (%)	1,638 (50.3%)

Table 2.

Mental health symptoms, alcohol and cannabis use, and sleep indicators from Wave 6 to Wave 11

	Wave 6	Wave 7	Wave 8	Wave 9	Wave 10	Wave 11
MHI-5, mean (SD)	66.0 (19.6)	65.7 (20.6)	65.3 (20.4)	64.1 (20.9)	62.2 (20.3)	62.0 (20.3)
Any alcohol use in past month, n (%)	433 (17.1%)	696 (28.0%)	962 (38.5%)	1123 (46.4%)	1490 (61.5%)	1680 (67.6%)
Any cannabis use in past month, n (%)	280 (11.0%)	416 (16.7%)	593 (23.7%)	693 (28.6%)	772 (31.9%)	798 (32.1%)
Weekday TIB (h), mean (SD)	7.6 (1.1)	7.4 (1.3)	7.8 (1.4)	7.8 (1.5)	7.6 (1.5)	7.7 (1.5)
Weekend TIB (h), mean (SD)	9.3 (1.5)	9.1 (1.5)	8.9 (1.5)	8.7 (1.5)	8.6 (1.6)	8.5 (1.6)
Weekday waketime, time (SD in min)	6:34 am (42.2)	6:46 am (59.0)	7:57 am (103.7)	7:53 am (106.2)	7:41 am (110.6)	7:39 am (126.8)
Weekday bedtime, time (SD in min)	10:56 PM (68.8)	11:21 PM (77.0)	11:49 PM (86.5)	12:03 am (91.1)	12:04 am (86.7)	12:00 am (90.9)
Weekend waketime, time (SD in min)	9:27 am (105.8)	9:39 am (101.2)	9:51 am (116.1)	9:51 am (112.8)	9:42 am (128.8)	9:34 am (129.2)
Weekend bedtime, time (SD in min)	12:07 am (84.3)	12:33 am (89.2)	12:53 am (95.3)	1:04 am (94.4)	1:02 am (93.8)	1:02 am (92.5)
Social jetlag (h), mean (SD)	2.0 (1.1)	2.1 (1.2)	1.7 (1.2)	1.5 (1.2)	1.5 (1.2)	1.5 (1.2)
How bothered by trouble sleeping in past 4 weeks*, mean (SD)	1.6 (0.7)	1.6 (0.7)	1.6 (0.7)	1.6 (0.7)	1.6 (0.7)	1.7 (0.7)
Age, mean (SD)	16.2 (0.7)	17.3 (0.7)	18.3 (0.8)	19.4 (0.8)	20.7 (0.7)	21.6 (0.8)

Note. *1 = not bothered at all, 3 = bothered a lot. Weekend waketimes are included for descriptive purposes. They are not modeled as outcomes in parallel process models.

LGMs for sleep and substance use measures

Individual trajectories for each sleep measure and substance use measure are presented first, followed by parallel process models organized by substance. Model fit indices for all models are presented in Supplemental Table 1.

Across all models, intercepts were significantly different from zero. With regard to slopes, both social jetlag (b = −0.17, 95% CI: −0.23, −0.11, p < 0.01) and weekend TIB (b = −0.16, CI: −0.23, −0.09, p < 0.01) significantly decreased over time. Weekday TIB did not significantly change over time (b = 0.03, CI: −0.03, 0.09, p = 0.37). Trouble sleeping significantly increased over time (b = 0.02, CI: 0.01, 0.03, p < 0.01), as did weekday (b = 9.55, CI: 5.97, 13.12, p < 0.01) and weekend bedtime (b = 9.30, CI: 5.46, 13.13, p < 0.01), such that respondents reported going to bed later over time. Lastly, likelihood of using alcohol (b = 0.52, CI: 0.40, 0.64, p < 0.01) and cannabis (b = 0.11, CI: 0.05, 0.18, p < 0.01) increased over time. Additional parameter estimates (e.g. slope variance) can be found in Supplemental Table 2.

Parallel process latent growth models (PPLGMs)

For all PPLGMs, we allowed within-process (e.g. baseline probability of alcohol use predicted rate of change in alcohol use over time) and cross-process (e.g. baseline weekday TIB predicting the rate of change in alcohol use over time) intercepts to predict slopes. All within-process intercepts significantly predicted within-process slopes. With only a few exceptions, cross-process intercepts did not significantly predict cross-process slopes. Specifically, both baseline alcohol (b = 0.11, CI: 0.03, 0.18, p < 0.01) and cannabis (b = 0.08, CI: 0.01, 0.16, p = 0.03) use significantly predicted the slope for weekday TIB such that greater initial probability of use predicted a greater increase in weekday TIB over time. Baseline social jetlag (b = 0.09, CI: 0.01, 0.17, p = 0.04) and weekend bedtime (b = 0.09, CI: 0.02, 0.17, p = 0.04) significantly predicted cannabis use slope such that greater social jetlag and later bedtimes at baseline predicted greater probability of use over time. In the following models, we discuss associations between intercepts and slopes, after controlling for all within- and cross-process regressions mentioned above. That is, all associations presented reflect unique associations (e.g. between slopes) over and above variance explained by within- and cross-process regressions. Further, for all final models with significant associations between slopes, we explored gender-stratified models.

Alcohol use models

Intercepts:

First, we examined the association between intercepts (i.e. initial status at baseline) for alcohol and each sleep measure. Sleep measures at baseline were all significantly associated with baseline alcohol use. Specifically, at baseline, greater social jetlag (r = 0.22, CI: 0.16, 0.28, p < 0.01), greater trouble sleeping (r = 0.18, CI: 0.10, 0.25, p < 0.01), and going to bed later on weekdays (r = 0.27, CI: 0.21, 0.33, p < 0.01) and weekends (r = 0.43, CI: 0.37, 0.49, p < 0.01) were significantly associated with greater initial likelihood of alcohol use. Further, shorter TIB on weekdays (r = −0.16, CI: −0.23, −0.90, p < 0.01) and weekends (r = −0.19, CI: −0.26, −0.13, p < 0.01) were associated with greater initial likelihood of alcohol use.

Slopes:

With the exception of weekday (r = 0.06, CI: −0.05, 0.17, p = 0.28) and weekend (r = −0.05, CI: −0.16, 0.07, p = 0.43) TIB, an increase in the likelihood of alcohol use over time was associated with all other sleep measure slopes. Social jetlag was significantly associated with likelihood of alcohol use (r = 0.11, CI: 0.01, 0.22, p = 0.04) such that a smaller reduction in social jetlag was associated with greater likelihood of alcohol use over time. However, after controlling for mental health symptoms, this association was no longer significant (r = 0.10, CI: −0.03, 0.22, p = 0.13). Trouble sleeping was positively associated with alcohol use over time (r = 0.18, CI: 0.05, 0.31, p < 0.01) such that greater trouble sleeping was associated with greater likelihood of alcohol consumption over time. After controlling for mental health symptoms this association was no longer (r = 0.14, CI: −0.01, 0.29, p = 0.06). Weekday bedtime was significantly associated with alcohol use across waves (r = 0.16, CI: 0.07, 0.25, p < 0.01) such that, over time, going to bed later on weekdays was associated with greater likelihood of alcohol consumption. After accounting for mental health symptoms, the association remained statistically significant (r = 0.11, CI: 0.002, 0.22, p = 0.047. However, when stratified by gender, this effect was found in males (r = 0.21, CI: 0.06, 0.36, p < 0.01) but not females (r = 0.04, CI: −0.12, 0.20, p = 0.63). Weekend bedtime was significantly associated with alcohol use across waves (r = 0.25, CI: 0.16, 0.35, p < 0.01) such that, over time, going to bed later on weekends was associated with greater likelihood of alcohol consumption. The association remained after accounting for mental health symptoms (r = 0.19, CI: 0.09, 0.30, p < 0.01). When stratified by gender, this association was stronger for males (r = 0.25, CI: 0.21, 0.39, p < 0.01) than females (r = 0.15, CI: −0.003, 0.30, p = 0.05)

Cannabis use models

Intercepts:

First we examined the association between intercepts for cannabis use and each sleep measure, that is, initial status at baseline. All baseline sleep measures were significantly associated with likelihood of baseline cannabis use, with the exception of weekday TIB (r = −0.06, CI: −0.13, 0.01, p = 0.08). Specifically, at baseline, greater social jetlag (r = 0.21, CI: 0.15, 0.27, p < 0.01), greater trouble sleeping (r = 0.15, CI: 0.07, 0.23, p < 0.01), and going to bed later on weekdays (r = 0.19, CI: 0.13, 0.25, p < 0.01) and weekends (r = 0.38, CI: 0.32, 0.45, p < 0.01) were significantly associated with a greater likelihood of cannabis use. Further, shorter TIB on weekends (r = −0.19, CI: −0.27, −0.12, p < 0.01) was associated with greater initial likelihood of cannabis use.

Slopes:

We did not find significant associations between changes in trouble sleeping (r = 0.02, CI: −0.12, 0.16, p = 0.77), or social jetlag (r = −0.003, CI: −0.10, 0.09, p = 0.99) and increased likelihood of cannabis use over time. Weekday TIB was positively associated with cannabis use (r = 0.11, CI: 0.01, 0.22, p = 0.04) such that increases in TIB were associated with increased likelihood of cannabis use. After controlling for mental health symptoms, the association remained (r = 0.19, CI: 0.06, 0.31, p < 0.01). When stratified by gender, the association was found for females (r = 0.28, CI: 0.12, 0.43, p < 0.01) but not males (r = 0.08, CI: −0.12, 0.28, p = 0.43). Weekend TIB was negatively associated with cannabis use (r = −0.13, CI: −0.24, −0.01, p = 0.03) such that a steeper decline in weekend TIB was associated with increased likelihood of cannabis use over time. After controlling for mental health symptoms across waves, the association remained statistically significant (r = −0.14, CI: −0.27, −0.004, p = 0.04). Gender stratified models found a marginally significant association for males (r = −0.20, CI: −0.40, 0.001, p = 0.05) but not for females (r = −0.07, CI: −0.23, 0.01, p = 0.44). Weekday bedtime was significantly associated with cannabis use (r = 0.18, CI: 0.09, 0.27, p < 0.01) such that going to bed later was associated with increases in the likelihood of cannabis use. After controlling for mental health at each timepoint, the association remained (r = 0.134, CI: 0.03, 0.24, p = 0.02). Gender-stratified models found this association for males (r = 0.21, CI: 0.05, 0.38, p = 0.01) but not females (r = 0.07, CI: −0.08, 0.21, p = 0.36). Similarly, weekend bedtime was significantly associated with cannabis use (r = 0.24, CI: 0.15, 0.33, p < 0.01) such that going to bed later on weekends was associated with increases in the likelihood of cannabis use. After controlling for mental health at each timepoint, the association remained (r = 0.19, CI: 0.09, 0.29, p < 0.01). This association was found for both males (r = 0.25, CI: 0.09, 0.41, p < 0.01) and females (r = 0.14, CI: 0.01, 0.27, p = 0.03).

Discussion

Sleep problems among adolescents and emerging adults are highly prevalent and are associated with an increased risk of a number of adverse health conditions, including alcohol and cannabis use, which also increase during these developmental periods. In turn, alcohol and cannabis use among adolescents and emerging adults can set the stage for subsequent individual and societal consequences, including increased risk for alcohol and other drug use disorders, unemployment, mental and physical health morbidity, and increased healthcare utilization. To date; however, few studies address longitudinal associations between sleep problems and alcohol and cannabis use, despite the high propensity for problems in each of these behaviors during the adolescent and emerging adulthood years. To our knowledge, the current study is the first to examine the prospective relationship between sleep health and alcohol and cannabis use across 6 annual waves of data collection which span the crucial developmental transition from adolescence into emerging adulthood. Understanding these patterns during this developmental transition is critical to identify potential novel opportunities for intervention, and to mitigate the lasting consequences of alcohol and cannabis use as these youth become adults.

Parallel process models provided an opportunity to examine the unique association between changes in sleep and substance use over time, after controlling for initial levels of each construct and their effects on the slopes of each construct over time. Given that both sleep and substance use exhibit developmental changes during adolescence and emerging adulthood, understanding this dynamic association over time is crucial for identifying potential targets for intervention.

First, in terms of associations between initial levels of one construct (e.g. sleep) predicting changes in the other construct (e.g. alcohol use), although we found only a few significant cross-process associations, the evidence was consistent with prior work suggesting bidirectional associations between sleep and substance use [38, 73]. Specifically, increased likelihood of alcohol and cannabis use at baseline predicted a greater increase in weekday time in bed (TIB) over time. These findings may reflect that as teens become young adults and have more flexibility in schedules upon exiting high school, those who use alcohol or marijuana in younger years may opt for later schedules allowing for more time in bed during the week. Importantly, since TIB does not account for sleep latency or wakefulness after sleep onset, it is possible that adolescents who were using alcohol or cannabis had more disrupted sleep at baseline but spent more time in bed over time. In terms of the reverse direction (e.g. sleep predicting substance use), and consistent with prior work [45, 74], teens with greater social jetlag and later bedtimes at the initial assessment showed greater increases in the probability of cannabis use over time.

Over and above these cross-process intercept and slope associations as well as the within-process intercept and slope associations, we found that specific dimensions of sleep, including increasingly later bedtimes, increases in trouble sleeping (as an indicator of sleep quality), and smaller reductions in social jetlag were associated with increased likelihood of alcohol use over time. In addition, although mental health symptoms also increase during this time frame [75], and are associated with both sleep [29] and substance use [30], most associations between sleep (i.e. later bedtimes and trouble sleeping) and likelihood of alcohol use remained significant even after controlling for mental health symptoms. Given that some youth may “self-medicate” with alcohol or cannabis to manage mental health symptoms (e.g. depression or anxiety [76]), and given known bidirectional associations between sleep problems and mental health [77, 78], findings from the present study can help disentangle causal associations [38]. In contrast to these findings, the association between changes in social jetlag and increases in alcohol use probability was reduced to non-significance after controlling for mental health symptoms, which is consistent with findings by Tavernier et al. [48], and may reflect that circadian disruptions are important symptoms, if not precursors of adolescent mental health problems [79]. From an intervention perspective, findings suggest complex and dynamic associations between sleep, mental health symptoms and substance across adolescence, and that programs should target these mutually reinforcing behaviors and symptoms.

For cannabis use, we found that decreases in weekend TIB, as well as later weekend and weekday bedtimes over time, were associated with an increased likelihood of cannabis use. In addition, increases in weekday sleep TIB were associated with increases in cannabis use likelihood over time. As mentioned previously, there was also a significant association between the initial probability of cannabis use and increases in weekday TIB, suggesting that cannabis use in adolescence may be associated with changes in adolescent sleep-wake behaviors, particularly when teens become young adults and ostensibly have more flexibility in their weekday schedules (as opposed to high schools). All significant associations remained even after controlling for the time-varying influence of mental health symptoms. Unlike the alcohol use models, neither social jetlag nor trouble sleeping was associated with cannabis use over time.

Collectively, findings emphasize the importance of measuring multiple dimensions of sleep health, as different sleep dimensions were associated with alcohol versus cannabis use, which can help inform targeted intervention efforts. For example, given associations between increases in trouble sleeping and likelihood of alcohol use trajectories, behavioral interventions that target sleep problems may be important to reduce the likelihood of drinking. On the other hand, given that increases in cannabis use over time were associated with initial weekday TIB and increases in weekday TIB, as well as decreases in weekend TIB, findings suggest that early cannabis use may predict changes in sleep-wake patterns over time. Despite the limited and equivocal data on cannabis and sleep, cannabis is widely perceived to be beneficial for sleep [80]. Therefore, it is important to identify reasons for cannabis use in teens and to continue to conduct and disseminate research on the effects of cannabis on sleep. Finally, associations between later bedtimes with both alcohol and cannabis use and social jetlag with alcohol use suggest that other strategies that focus on issues of sleep timing or circadian misalignment may be important.

Our findings also add to the limited extant literature on gender differences in the association between sleep and substance use outcomes. Importantly, most prior literature on this topic has come from cross-sectional studies [81, 82], and evidence is mixed. For example, Johnson and Breslau [83] found a stronger cross-sectional association between sleep problems and illicit substance use in females than in males; however, this gender difference was no longer evident after adjustment for internalizing and externalizing symptoms. In contrast, Wong et al. [84] found that sleep problems in childhood (ages 3–8) predicted increased risk of early onset of alcohol, cigarette, and marijuana use in boys during adolescence (ages 12–17 years). In girls; however, childhood sleep problems only predicted onset of alcohol use in adolescence. For the most part, our findings are consistent with Wong et al., in that with only one exception, we found stronger associations between changes in sleep and changes in likelihood of alcohol or cannabis use for males compared to females. Although exploratory in nature, findings provide an important look at how patterns may differ for males and females.

Study findings must be considered in light of certain limitations. First, we did not control for timing of surveys as some participants completed them during the school year and others during summer, and sleep patterns may vary across these periods for high school students. However, controlling for timing of assessments would have added considerable complexity as a time-varying covariate, and may be less relevant at post-high school assessments. Another limitation is that both sleep and substance use were measured via self-report, which may have introduced bias or common method variance. Furthermore, differences in timeframes for substance use (i.e. past month) versus some of the sleep measures (e.g. “usual” bedtimes/ wake-up times) may have introduced heterogeneity in the results. Given low rates of substance use in earlier waves of assessment, we included binary measures of substance use, which may have limited our ability to detect differences in problem drinking or marijuana use. Future research focused on the frequency of use may yield different outcomes. It is important to note, however, that substance use rates in our sample match national rates [85], and other epidemiologic sleep studies have used similar measures to ours [63, 86] and have shown modest correlations with actigraphy-assessed sleep time spent asleep [87]. Finally, sleep measures were added when youth were approximately 16 years of age and we, therefore, did not have measures of pre-existing sleep disorders or problems.

Despite limitations, this study makes several important contributions to the literature. First, our longitudinal data covers the critical transition between two key developmental periods, late adolescence and emerging adulthood, both of which are associated with increased risk for sleep problems and substance use and disorders [4, 88, 89]. Most studies are cross-sectional or address sleep and substance use during one of these time periods. In contrast, we include 6 annual waves of data, and utilize a sophisticated modeling approach that provides a more granular analysis of how sleep health and alcohol or cannabis use associate over time as youth transition from adolescence into emerging adulthood. Second, we focus on alcohol and cannabis use, as these are the two most commonly used substances during adolescence and emerging adulthood, and we include multiple dimensions of sleep health, all of which are modifiable and salient to these developmental stages. Finally, our inclusion of mental health as a time-varying covariate allows us to isolate the effects of sleep, mental health, and alcohol or cannabis use over time.

Future longitudinal research should consider how developmental shifts throughout adolescence and early adulthood influence specific sleep and substance use patterns, including frequency of use or substance use problems and disorders. In addition to exploring gender differences, it is also important for research to consider the degree to which other factors such as race/ ethnicity may moderate longitudinal sleep and substance use associations across developmental transitions.

Our findings provide robust evidence of the dynamic and mutually reinforcing association between sleep and substance during the transition from adolescence to early adulthood. Therefore, it is critical to develop and utilize developmentally appropriate intervention efforts that can address sleep problems and substance use as well as their reciprocal influence on each other, during this unique developmental transition. For example, given the increased autonomy and greater reliance on peer social networks as sources of support and information that is characteristic of this developmental transition, interventions involving digital technologies (e.g. reminders that it is time for bed) [90], motivational interviewing [91], or social network interventions [92] focused on sleep health or substance use, may be particularly relevant for this age group. In addition, transdiagnostic approaches that position sleep and circadian problems as common, transdiagnostic mechanisms across mental and behavioral health problems, including substance use, may be particularly useful given evidence for bidirectional sleep and substance use associations [93, 94].

Disclosure Statements

Financial Disclosure: The authors have no financial conflicts of interests to disclose.

Non-financial Disclosure: The authors have no non-financial conflicts of interest to disclose.