A Meta-Analysis of Bi-Directional Associations between Sleep Health and Substance Use among U.S. Youth: Racial/Ethnic Differences

Abstract

Purpose of review:

This paper aimed to estimate pooled bi-directional associations between multidimensional sleep health and substance use among youth and investigate whether these associations differed as a function of race/ethnicity.

Recent findings:

Empirical observational studies providing quantitative data on the association of sleep health (duration [sleep obtained per 24 hours], satisfaction [subjective assessment of sleep], alertness [ability to maintain attentive wakefulness], and timing [placement of sleep]) and substance use (i.e., alcohol and cannabis), and racial/ethnic demographic information among U.S. youth (10–25) were identified through a systematic literature search. Random effects meta-analyses were conducted using 95 effect sizes extracted from 38 studies.

Summary:

We found evidence for a bi-directional relationship between total sleep duration and substance use. Sleep satisfaction predicted substance use, but findings were inconclusive whether substance use predicted sleep satisfaction. Sleep alertness predicted alcohol (but not cannabis) use, whereas sleep timing predicted cannabis (but not alcohol) use. Nuanced racial/ethnic differences were also found in these sleep-substance use relationships, which differed across sleep domains and types of substance.

INTRODUCTION

Substance use among youth (defined as individuals aged 10 to 25, overarching developmental age spans of adolescence and young adulthood [1]) is a major public health concern in the United States (U.S.). Alcohol and cannabis are two of the most used substances by youth [2], with the median age of initiation for both substances around age 15 [3]. Substance use quickly escalates throughout youth as the rate of past-year alcohol use surges from 17% among youth aged 12 to 17 to nearly 68% among youth aged 18 to 25 and rates of past-year cannabis use similarly rising from 11% to a substantial 38% [2]. The legal drinking age (i.e., 21; [4] and increasing legalization of recreational cannabis use [5]) in the U.S. likely coincide with the growing incidence of alcohol and cannabis use across youth. Early initiation of substance use coupled with rapid escalation, confers risk for numerous biopsychosocial adverse outcomes for youth [6-7], including poor sleep health [8-9].

Poor sleep health is also pervasive among youth [10]. Sleep health is posited to be multifaceted, comprising dimensions of regularity, satisfaction, alertness, timing, efficiency, and duration [11]. Sleep health can be influenced by individual factors as well as broader socioecological contexts, which may be salient for youth [12-14]. For example, normative neurobiological changes in circadian rhythm and sleep homeostasis, accompanied by competing social pressures (e.g., early school start time; academic and work demands; increased responsibility with independence from parents) in youth, may result in shorter sleep duration and poor sleep satisfaction, alertness, timing, and efficiency [15-17]. Diverse sleep domains are shown to differentially predict adverse health outcomes [11] and high-risk health behaviors such as youth substance use [18-19]. Given the drastic initiation and escalation of substance use along with poor sleep health occurring in youth make it a critical developmental period to better understand the sleep-substance use associations.

Substance use and poor sleep health often co-occur in youth [20-21] and are theorized to be interconnected, with both potentially preceding and serving as markers of risk for the other prospectively [22]. Limited and mixed findings reported by available systematic reviews and meta-analyses as well as individual studies, however, preclude an understanding of the theorized bi-directional sleep-substance use association. For example, a recent meta-analysis and systematic review focusing on young adults (ages 18–30) demonstrated different uni-directional sleep-substance associations as a function of specific substance types and sleep domains [23]. Although meta-analyses could not be conducted on the bi-directional sleep-substance use associations due to insufficient effect size estimates, the authors reported pooled odds ratios for the uni-directional association of sleep disturbances (OR=1.24) and sleep duration (OR=0.95) with alcohol use frequency and quantity [23]. Notably, while the authors found significant heterogeneity in the sleep-alcohol associations, assessed moderators (i.e., population [college student], cut-off scores for sleep satisfaction, alcohol use indicator) did not account for the study variability [23]. Thus, despite the rapidly growing synthesized findings on complex bi-directional sleep-substance use associations, significant gaps remain in the direction and magnitude of the bi-directional sleep-substance use relationships across varying sleep domains and substance use indicators during the critical developmental stage of youth (ages 10 to 25) as well as potential moderating factors associated with between-study variability.

Race/ethnicity is one underexplored factor that may be associated with the variability in aggregated effect size estimates in sleep-substance use associations. Racial/ethnic disparities among youth have been well-documented in separate sleep and substance use literature. Findings indicate that racially/ethnically minoritized (i.e., non-White) youth may be at increased risk for poor sleep health [24-27] as well as escalating rates of substance use and related consequences [28-29] compared to their White counterparts. Research on sleep and substance use disparities has primarily focused on Black and White individuals aged 18 and older. However, burgeoning research has expanded to include Asian, Multiracial, American Indian/Alaska Native, Native Hawaiian/Other Pacific Islander and Hispanic/Latino racial/ethnic youth [30-33], whose populations are soaring in the U.S. [34]. Yet, few studies have examined racial/ethnic disparities in sleep-substance use associations, with available findings limited in scope regarding various sleep health domains and substance use indicators, as well as in racial/ethnic representation. Given the demonstrated racial/ethnic disparities in sleep and substance use in youth, it is exigent for researchers to investigate the role of race and ethnicity in the bi-directional sleep-substance use association among youth and incorporate a more representative range of racial/ethnic groups, which are reflective of the rapidly evolving U.S. demographic landscape.

The current meta-analysis represents a quantitative synthesis of available findings on the sleep-substance use relationship among youth. Specifically, this meta-analysis aimed to estimate pooled bi-directional associations between sleep and substance use (with a focus on alcohol and cannabis use) among youth. While a recent well-conducted meta-analysis in 2022 indicated insufficient effect sizes to estimate bi-directional relationships among young adults (18 to 30), the rapid growth of this literature in the last two years warrants further investigation incorporating new findings. Further, this meta-analysis aimed to capture developmental age spans of both adolescence and young adulthood, and thus additional investigation may help to clarify the sleep-substance association among youth (ages 10 to 25). Further, this meta-analysis aimed to investigate whether the direction and magnitude of the sleep-substance associations differ as a function of race/ethnicity. In addition, differences in other sociodemographic (i.e., sex; developmental period [adolescent; young adult]) and methodological (i.e., study design; assessment of substance use) factors were examined given their well-established associations with youth sleep [35-37] and substance use [38].

METHOD

Literature Search

A systematic literature search for peer-reviewed empirical articles reporting associations between sleep and substance use among youth was performed using PsycInfo, PubMed/Medline, and Scopus until January 2024. The following search terms were used: (alcohol* OR cannabis* OR marijuana*) AND (sleep* OR insomnia* OR circadian*) AND (adolescen* OR teen* OR youth* OR college* OR university* OR "young adult"). Secondary searches were also conducted by manually reviewing reference lists from eligible articles and related systematic reviews and meta-analyses. This manuscript was prepared in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria [39]. This study was pre-registered using Prospero (ID: CRD42023463084) [40].

Inclusion Criteria and Study Selection

Eligible studies were (1) empirical observational studies providing original and quantitative data on substance use (i.e., alcohol or cannabis) and sleep, (2) reported the sample age between 10 to 25 years, (2) published in a peer-reviewed English language journal (as opposed to unpublished case studies, thesis/dissertation data, conference abstracts), (3) included sample race/ethnicity demographic information, and (4) were conducted in the U.S. as race/ethnicity are social constructs and their health implications may differ across countries [41]. Determination of observational, non-interventional studies was made to capture naturally occurring associations of sleep and substance use in real life without experimental manipulation or intervention.

Elimination of articles was performed through title/abstract and full-text review. Studies were excluded if they (1) were non-original (i.e., systematic reviews and meta-analyses), animal studies, intervention (as opposed to observational), or qualitative studies, (2) reported the average sample ages outside of 10 to 25, (3) did not report sleep and substance use associations, (4) did not report sample racial/ethnic breakdowns, (5) were not conducted in the U.S., or (6) reported duplicate sample data in a prior study. Restrictions were not placed on year of publication.

Data Extraction

Sleep and substance use effect size estimates and demographics (i.e., proportion of racial/ethnic groups, proportion male, developmental stage, and study design) were extracted from each study. Restrictions were not placed on specific sleep nor substance use indicators. As the literature on the relationship between sleep and substance use is burgeoning, synthesizing pooled estimates across multiple indicators of sleep and substance use, rather than focusing on a single assessment or indicator may offer a more comprehensive understanding of this relationship.

Effect size estimates were obtained for six sleep health domains [11]: total duration (i.e., total amount of sleep obtained per 24 hours), weekday duration, weekend duration, satisfaction (i.e., subjective assessment of sleep), alertness (i.e., ability to maintain attentive wakefulness) and timing (i.e., placement of sleep within the 24-hour day). Effect size estimates were obtained for two substance types (i.e., alcohol, cannabis) and substance use indicators: alcohol use (i.e., past-day and -week quantity; past-day, -week, -month, and -year, and lifetime frequency) and cannabis use (i.e., past-day, -week, -month, and -year, and lifetime frequency).

Pearson’s r was used as the effect size estimate for sleep-substance associations. For studies that reported estimates other than r, their effect sizes were converted to r using the formulae for converting odds ratios (for dichotomous data) and eta-squared [42], and Peterson & Brown’s formula for converting standardized β [43]. When multiple sleep or substance use domains were reported in the same study, a separate effect size for each domain was extracted so long as each study contributed only one effect size estimate to each meta-analysis.

Data were also extracted for exploratory moderation analyses. Regarding racial/ethnic differences, because only one study reported racial/ethnic group differences in sleep-substance associations [87], the proportion of participants in the sample identifying with each racial and ethnic group were obtained to determine whether proportion of a specific racial or ethnic sample participant group was associated with the strength and direction of the sleep and substance use associations. Study participants in each racial/ethnic group were coded as White, Black, Asian, Multiracial, American Indian/Alaska Native, Native Hawaiian/Other Pacific Islander or Hispanic/Latino/Spanish origin (hereafter, Hispanic), consistent with the U.S. Census and guidelines for reporting race/ethnicity [34]. Studies that reported on an “other” racial group were further extracted.

Proportion of male (coded to assume binary classifications [male/female] if not specified), developmental stage (i.e., adolescent; young adult), and study design (i.e., cross sectional; longitudinal) were also obtained for moderation analyses. Further, due to the limited number of effect sizes available, only alcohol (but not cannabis) use indicators (i.e., frequency, quantity) were extracted for moderation analyses.

To evaluate study quality, the domain-based Risk of Bias Assessment Tool for Nonrandomized Studies (RoBANS; [44]) was used. Assessed domains included selection of participants, confounding variables, measurement of exposure, incomplete outcome data, and selective outcome reporting. The domain of blinding of outcome assessments was excluded as blinding was not applicable to study design of included studies (i.e., observational/self-report studies, as opposed to intervention studies).

Study inclusion and exclusion, sleep, substance use, and demographic information were extracted by two independent coders (F.D. and A.V.) using piloted forms. Interrater agreement was excellent for continuous variables (median intraclass correlation coefficient = 0.96), with discrepancies resolved with a third independent coder (E.S.).

Data Synthesis

Meta-analytical synthesis was conducted using Comprehensive Meta-Analysis (CMA), version 3 [45]. Separate meta-analyses were conducted to examine the bi-directional association of varying indicators of sleep and substance use. Pearson’s r was calculated as a measure of effect size to assess the strength and direction of the sleep-substance association. Random effects models were estimated under the assumption that variation in the observed effect sizes was due to variability in study outcomes and designs (e.g., cross-sectional; longitudinal [including prospective and intensive longitudinal findings]; [44]). Between-study heterogeneity was estimated using Cochran’s Q (i.e., significance testing of ratio of heterogeneity across studies to heterogeneity within studies; [45]) and I² (i.e., inconsistency across effect sizes after controlling for number of included studies with 25% representing low, 50% moderate, and 75% high heterogeneity; [44]). ‘One-study-removed’ analyses were conducted as an estimate of sensitivity to evaluate each study’s influence on the effect size estimate and to clarify the role of outliers on pooled associations [45].

Exploratory Moderation Analyses

When significant between-study heterogeneity was found (Cohran’s Q statistic), exploratory moderation analyses were conducted to estimate whether a moderator was associated with between-study heterogeneity [45]. Continuous and categorical moderators (i.e., proportion race/ethnicity; proportion male, developmental stage [adolescent reference group]; study design [cross-sectional reference group]; and alcohol use indicators [frequency reference group]) were tested using mixed-effects meta-regression for effect size estimates of sleep and substance use associations with at least four effect sizes [45].

Publication Bias

Publication bias for meta-analysis was assessed using Egger’s regression test for meta-analyses and Duval and Tweedie’s Trim and fill test for meta-analyses with at least ten effect size estimates [44-47].

RESULTS

Study Selection and Characteristics

Study identification, screening, eligibility, and selection identified 38 qualifying studies, as shown in Figure 1. Characteristics of included studies are shown in Table 1. Publication years ranged from 2001 to 2022, with 42% (k=16) of studies published since 2019. White, Black, and Hispanic participants comprised the largest percentages of the study sample compared to other reported racial groups (e.g., Asian, Multiracial, American Indian/Alaska Native, and Native Hawaiian/Other Pacific Islander). The percentage of male participants ranged from 0%–56%. Articles varied in developmental stage (i.e., adolescent [k=24]; young adult [k=14]) as well as study design (i.e., cross-sectional [k=20]; longitudinal [k=18]). Most longitudinal studies reported uni-directional associations [k=31], whereas only limited findings reported prospective bi-directional relationships between sleep and substance use [k=7].

Figure 1.

PRISMA 2020 flow diagram for new systematic reviews which included searches of databases, registers and other sources

From: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. For more information, visit: http://www.prisma-statement.org/

Table 1.

Summary of studies included in meta-analysis (k=38)

Study	Year	Sample Size and Dataset	Sample Demographics	Study Design	Predictor(s)	Outcome(s)
Cusack et al. [55]	2022	932 college students enrolled in a Mid-Atlantic university	Mage=18.5 (SD=0.42); 26% Male; 50% White, 20% Black, 16% Asian, 13% Other	Longitudinal: three annual assessments	Sleep satisfaction (PSQI)	Alcohol use (past-month frequency and quantity; AUDIT)
Ehlers et al. [56]	2018	800 Mexican American and American Indian young adults recruited from eight geographically contiguous reservations	Mage=23 (SD=0.01; range=18 to 30); 44% Male; 27% American Indian/Alaska Native; 73% Hispanic	Cross-sectional	Alcohol use (quantity during heaviest drinking period during adolescence; TLFB)	Sleep satisfaction (PSQI)
Fucito, et al. [57]	2017	42 heavy-drinking college students at-risk for alcohol use disorder recruited from local colleges in New England	Mage=20.52 (SD=1.31); 52% Male; 69% White, 5% Black, 19% Asian, 5% Multiracial, 19% Hispanic, 2% Other	Longitudinal: seven-day protocol	Sleep total duration (PSD) Sleep satisfaction (PSD) Alcohol use (past-day quantity; DDQ)	Alcohol use (past-day quantity; DDQ) Sleep total duration (PSD) Sleep satisfaction (PSD)
Goodhines, Desalu et al. [58]	2020	414 high school students from an urban, socioeconomically disadvantaged high school in the northeastern U.S. (Project Teen)	Mage=16 (SD=1.08; range=9th to 11th grade); 42% Male; 22% White, 41% Black, 18% Asian, 17% Multiracial, 12% Hispanic, 2% Other	Cross-sectional	Weekday sleep duration (one-item) Weekend sleep duration (one-item) Sleep satisfaction (ISI) Alcohol use (past-year frequency; NIAAA)	Alcohol use (past-year frequency; NIAAA) Sleep satisfaction (ISI)
Goodhines, Gellis et al. [59]	2019	171 college students enrolled in a four-year university in the northeastern U.S.	Mage=19 (SD=1.35); 32% Male; 74% White, 6% Black, 12% Asian, 6% Multiracial, 1% American Indian/Alaska Native	Longitudinal: two waves, two months apart	Sleep total duration (four-items) Sleep satisfaction (ISI) Sleep timing (MEQ) Alcohol use (past-2-month frequency; NIAAA)	Alcohol use (past-2-month frequency; NIAAA) Sleep total duration (four-items) Sleep satisfaction (ISI)
Goodhines, Gellis, Ansell et al. [60]	2019	83 college students endorsing past-month cannabis and alcohol sleep aid	Mage=19 (SD=1.11); 30% Male; 73% White, 8% Black, 11% Asian, 5% Multiracial, 3% Other	Longitudinal: 14-day protocol	Cannabis use (past-day frequency; one-item)	Sleep total duration (one-item) Sleep satisfaction (PSQI)
Goodhines, Wedel et al. [61]	2022	407 high school students from an urban, socioeconomically disadvantaged high school in the northeastern U.S. (Project Teen)	Mage=16 (SD=1.08; range=13 to 19); 42% Male; 22% White, 41% Black, 18% Asian, 17% Multiracial, 1% American Indian/Alaska Native, 2% Native Hawaiian/Other Pacific Islander 12% Hispanic	Longitudinal: baseline; follow-up one year later	Sleep satisfaction (ISI) Cannabis use (past-year frequency; one-item)	Cannabis use (past-year frequency; one-item) Sleep satisfaction (ISI)
Graupensperger, et al. [62]	2022	409 young adults reporting one past-month occasion of simultaneous alcohol and cannabis use, and reported drinking alcohol 3+ times in the past month were recruited from the community in Seattle, Washington	Mage=21.61 (SD=2.17; range=18 to 25); 49% Male; 48% White, 4% Black, 16% Asian, 11% Multiracial, 16% Hispanic, 4% Other	Longitudinal: five 14-day bursts separated by four months	Sleep total duration (two-items) Alcohol use (past-day dichotomized; one-item) Cannabis use (past-day dichotomized; one-item)	Alcohol use (past-day use dichotomized; one-item) Cannabis use (past-day use dichotomized; one-item) Sleep total duration (two-items)
Graupensperger, Fairlee, Vitiello et al. [63]	2021	409 young adults reporting one past-month occasion of simultaneous alcohol and cannabis use, and reported drinking alcohol 3+ times in the past month were recruited from the community in Seattle, Washington	Mage=21.61 (SD=2.17); 49% Male; 48% White, 4% Black, 16% Asian, 11% Multiracial, 16% Hispanic, 4% Other	Longitudinal: five 14-day bursts separated by four months	Sleep satisfaction (PSQI)	Alcohol use (past-day use dichotomized; one-item) Cannabis use (past-day use dichotomized; one-item)
Hasler, Bruce et al. [64]	2019	36 alcohol drinkers reporting ≥1, standard drink per week over the past 30 days were recruited from the community in Pittsburgh	Mage=21.3 (range=18 to 22); 38% Male, 69% White, 11% Black, 11% Asian, 0.08% Multiracial, 0.05% Hispanic	Longitudinal: 14-day protocol	Sleep timing (one-item)	Alcohol use (past-day quantity; TLFB)
Hasler, Franzen et al. [65]	2017	729 adolescents in the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) study	Mage=15.9 (SD=2.4; range=12 to 21); 49% Male; 75% White, 12% Black, 7% Asian American; 12% Hispanic, 6% Other	Cross-sectional	Sleep weekday duration (STQ) Sleep weekend duration (STQ) Sleep satisfaction (PSQI) Sleep alertness (CASQ-5)	Alcohol use (past-year frequency; CDDR) Cannabis use (past-year dichotomized; CDDR)
Hasler, Graves et al [66]	2022	831 adolescents in the National Consortium on Alcohol and Neurodevelopment in Adolescence (NCANDA) study	Mage=16.2 (SD=2.5); 49% Male; 71% White, 11% Black, 7% Asian, 11.6% Hispanic, 9% Other	Longitudinal: six annual assessments	Sleep satisfaction (PSQI) Sleep timing (CSM-4)	Alcohol use frequency (past-year dichotomized; CDDR) Cannabis use frequency (past-year dichotomized; CDDR)
Haynie et al. [67]	2018	2,785 adolescents across 81 schools stratified by nine U.S. census divisions, as part of NEXT Generation Health Study	Age range=10th to 12th graders; 45% Male; 62% White, 14% Black, 19% Hispanic, 5% Other	Longitudinal: three annual assessments	Sleep weekday duration (two-items) Sleep weekend duration (two-items) Sleep timing (one-item)	Alcohol use (past-month frequency; one-item)
Jackson et al. [68]	2020	7,765 participants from the 2017 cohort of the Monitoring the Future (MTF) study	Age range=8th and 10th graders; 49% Male; 61% White, 13% Black, 26% Hispanic	Cross-sectional	Cannabis use (past-month dichotomized; one-item)	Sleep total duration (one-item)
Johnson & Breslau [69]	2001	13,381 adolescents from a nationally representative sample from the 1996-8 U.S. National Household Survey on Drug Abuse (NHSDA)	Mage=14 (range=12 to 17); 51% Male; 68% White, 14% Black, 4% Asian, 13% Hispanic, 0.7% Other	Cross-sectional	Alcohol use (past-year frequency; one-item)	Sleep satisfaction (one-item)
Kwon et al. [70]	2021	14,638 high school students from the 2017 Youth Risk Behavior Surveillance System (YRBSS)	Mage=16 (range=14 to 18); 49% Male; 58% White, 11% Black, 9% Hispanic, 22% Other	Cross-sectional	Alcohol use (past-month frequency; one-item) Cannabis use (past-month frequency; one-item)	Sleep weekday duration (one-item)
Lund et al. [71]	2010	1,125 college students enrolled in large private midwestern U.S. university	Mage=20 (SD=1.3; range=17 to 24); 37% Male; 86% White, 2% Black, 2% Multiracial, 1% American Indian/Alaska Native, 0.4% Native Hawaiian/Other Pacific Islander, 2% Asian American/Pacific Islander, 2% Other	Cross-sectional	Alcohol use (past-week quantity; one-item)	Sleep satisfaction (PSQI)
Marmorstein [72]	2017	127 youth from Camden Youth Development Study	Mage=13.2 (SD=0.8; range=6th or 7th grade); 50% Male; 5% White, 32% Black, 6% American Indian/Alaska Native, 71% Hispanic	Cross-sectional	Sleep satisfaction (PSQI) Sleep alertness (two-items)	Alcohol use (past-4-month frequency; one-item)
Maultsby et al. [73]	2021	2,770 adolescents from NEXT Generation Health Study	Age range=10th to 12th graders; 45% Male; 56% White, 20% Black, 19% Hispanic, 5% Other	Longitudinal: three annual assessments	Sleep weekday duration (two-items) Sleep weekend duration (two-items) Sleep satisfaction (two-items) Sleep timing (two-items)	Cannabis use (past-month frequency; one-item)
McGlinchey & Harvey [74]	2015	3,843 adolescents from waves 2 (1996) and 3 (2001-2002) of the National Longitudinal Study of Adolescent Health (Add Health)	Mage=16; 48% Male; 58% White, 24% Black, 4% Asian, 3% American Indian/Alaska Native, 7% Hispanic	Longitudinal: two waves, six years apart	Sleep timing (one-item)	Alcohol (past-year drunkenness dichotomized; one-item)
McKnight-Eily et al. [75]	2011	12,154 high school students from the 2017 YRBSS	Mage=16; 50% Male; 63% White, 15% Black, 16% Hispanic, 7% Other	Cross-sectional	Sleep weekday duration (one-item)	Alcohol (past-month; one-item) Cannabis (past-month; one-item)
Miller et al. [76]	2017	385 first-year, heavy-drinking college students at a private university in the Northeastern U.S.	Mage=18.6 (SD=0.4); 48% Male; 65% White, 4% Black, 12% Asian, 10% Multiracial, 0.5% American Indian/Alaska Native	Longitudinal: nine-week protocol	Sleep satisfaction (PSQI)	Alcohol (past-day quantity; one-item)
Negriff et al. [77]	2011	262 adolescent girls enrolled in an ongoing longitudinal study (Dorn et al., 2008) and recruited from the community	Mage=14.93 (SD=2.17; range=11 to 18); 0% Male, 63% White, 32% Black, 5% Other	Cross-sectional	Sleep timing (M/E)	Alcohol (past-month frequency; DISC) Cannabis (past-month frequency; DISC)
Nguyen-Louie et al. [78]	2017	95 adolescents from an ongoing longitudinal neuroimaging project in San Diego, California	Mage=13.4 (SD=0.07); 53% Male; 68% White, 3% Black, 6% Asian, 17% Multiracial, 20% Hispanic, 5% Other	Longitudinal: three annual assessments	Sleep timing (M/E) Sleep alertness (SWPBS)	Alcohol use (lifetime frequency; CDDR) Cannabis use (lifetime frequency; CDDR)
O’Brien & Mindell [79]	2005	388 adolescents from the Sleep Habits Survey and Youth Risk Behavior Survey on high school students in Philadelphia, Pennsylvania	Mage=16.62 (range=9th to 12th grade); 56% Male; 78% White, 7% Black, 9% Asian, 2% Multiracial, 3% Other	Cross-sectional	Sleep alertness (DSS)	Alcohol use (past-month frequency; one-item)
Pasch, Laska et al. [80]	2010	242 youth recruited from within 7-county metropolitan area of Minneapolis-St. Paul, Minnesota	Mage=16.4 (range=10 to 16); 49% Male; 93% White, 1% Black, 0.4% Asian, 5% Multiracial	Cross-sectional	Sleep weekday duration (NEDS) Sleep weekend duration (NEDS)	Alcohol use (past-month frequency; one-item) Cannabis use (past-month frequency; one-item)
Pasch, Latimer et al. [81]	2012	704 adolescents from the Identifying the Determinants of Eating and Activity (IDEA) or Etiology of Childhood Obesity (ECHO) studies	Mage=14.7 (SD=1.83); 49% Male; 86% White, 5% Black, 1% Asian, 0.4% American Indian/Alaska Native, 0.1% Native Hawaiian/Other Pacific Islander, 5% Hispanic, 6% Other	Longitudinal: two waves, one year apart	Sleep total duration (NEDS) Alcohol use (past-month frequency dichotomized; one-item) Cannabis use (past-month frequency dichotomized; one-item)	Alcohol use (past-month frequency dichotomized; one-item) Cannabis use (past-month frequency dichotomized; one-item) Sleep total duration (NEDS)
Roane & Taylor [82]	2008	4,494 adolescents from Add Health	Mage=15.83 (SD=1.83, range=12 to 18); 48% Male; 59% White, 21% Black, 3% Asian, 0.6% American Indian/Alaska Native, 4% Hispanic, 12% Other	Cross-sectional	Sleep satisfaction (three-items)	Alcohol use (Quantity; one-item) Cannabis use (Lifetime frequency dichotomized; one-item)
Roberts et al. [83]	2008	3,134 adolescents from Teen Health 2000 in Houston, Texas	Mage=14.93 (SD=2.17, range=11 to 17); 51% Male, 35% White, 35% Black, 25% Hispanic, 5% Other	Longitudinal: two waves, one year apart	Sleep satisfaction (three-items)	Alcohol use (past-year frequency; DISC)
Taylor, Bramoweth et al. [84]	2013	1,039 undergraduate students from the University of North Texas, Texas	Mage=20.39 (SD=3.93); 28% Male; 66% White, 12% Black, 5% Asian/Pacific Islander, 10% Hispanic, 4% Other	Longitudinal: one-week protocol	Sleep satisfaction (PSQI) Sleep alertness (ESS)	Alcohol use (AUDIT)
Taylor, Clay et al. [85]	2011	838 undergraduate students from the University of North Texas, Texas	Mage=19.78 (SD=1.89, range=17 to 26); 25% Male; 70% White, 13% Black, 10% Hispanic, 5% Asian/Pacific Islander	Cross-sectional	Sleep timing (MEQ)	Alcohol use (AUDIT) Cannabis use (past-week frequency; MPS)
Terry-McElrath et al. [86]	2016	154,611 8th graders, 139,316 10th graders, and 85,960 12th graders from the 1991-2014 MTF study	Age range=8th grade to 12th graders; 48-49% Male; 58-65% White, 12-14% Black, 11-13% Hispanic, 10-13% Other	Cross-sectional	Sleep total duration (MTF)	Alcohol use (past-year frequency; one-item) Cannabis use (past-year frequency; one-item)
Troxel, Ewing et al. [87]	2015	2,539 youth from 16 middle schools across three school districts in Southern California	Mage=15.5 (SD=0.68; range=14 to 19); 46% Male; 21% White, 2% Black, 21% Asian, 10% Multiracial, 0.83% American Indian/Alaska Native, 0.79% Native Hawaiian/Other Pacific Islander, 44% Hispanic, 14% Other	Cross-sectional	Sleep weekday duration (two-items) Sleep weekend duration (two-items)	Alcohol use (past-month frequency; CHKS) Cannabis use (past-month frequency; CHKS)
Troxel, Rodriguez et al. [88]	2021	3,265 adolescents and emerging adults in South California	Mage=16.2 (SD=0.7); 47% Male; 20% White, 2% Black, 20% Asian, 12% Multiracial, 47% Hispanic	Longitudinal: six annual assessments	Sleep satisfaction (one-item) Sleep timing (one-item)	Alcohol use (past-month frequency; one-item) Cannabis use (past-month frequency; one-item)
Wahlstrom et al. [89]	2017	8,261 high school students in five school districts in Wyoming, Colorado, and Minnesota for Teen Sleep Habits Survey	Age range=9th to 12th graders; 49% Male; 73% White, 4% Black, 6% Asian, 5% Multiracial, 9% Hispanic, 3% Other	Cross-sectional	Sleep total duration (TSHS)	Alcohol use (past-2-week frequency dichotomized; one-item)
Winiger et al. [90]	2022	4,637 high school students from the 2019 administration of the Healthy Kids Colorado Survey	Age range=9th to 12th graders; 51% Male; 53% White, 5% Black, 35% Hispanic, 8% Other	Cross-sectional	Sleep total duration (two-items) Cannabis use (past-month frequency dichotomized; one-item)	Cannabis use (past-month frequency dichotomized; one-item) Sleep total duration (two-items)
Wong et al. [91]	2019	7,307 college students across ten universities from ten different U.S. states	Mage=20.21 (SD=3.14); 30% Male; 75% White, 17% Hispanic	Cross-sectional	Sleep satisfaction (ISI)	Cannabis use (past-month; CUDIT)
Yurasek et al. [92]	2019	267 undergraduate students from large southeastern university who reported using cannabis at least three times in the past month	Mage=19.9 (SD=1.4); 38% Male; 67% White, 6% Black, 17% Asian, 25% Hispanic, 4% Other	Cross-sectional	Sleep satisfaction (ISI)	Cannabis use (past-month frequency; one-item)

Note. PSQI=Pittsburgh Sleep Quality Index; AUDIT=Alcohol Use Disorders Identification Test; TLFB=Timeline Follow Back; PSD=Pittsburgh Sleep Diary; DDQ=Daily Drinking Questionnaire; NIAAA=National Institute of Alcohol Abuse and Alcoholism; ISI=Insomnia Severity Index; MEQ=Morning Eveningness Questionnaire; STQ=Sleep Timing Questionnaire; CDDR=Customary Drinking and Drug Use Record; CASQ-5=Cleveland Adolescent Sleepiness Questionnaire; CSM-4=Composite Scale of Morningness; M/E=Morningness/Eveningness; DISC=Diagnostic Interview Schedule for Children; SWPBS=Sleep/Wake Problems Behavior Scale; DSS=Daytime Sleepiness Scale; NEDS=Night Eating Diagnostic Scale; ESS=Epworth Sleepiness Scale; MEQ=Morningness-Eveningness Questionnaire; Marijuana Problem Scale; CHKS=California Healthy Kids Survey; TSHS=Teen Sleep Habits Survey; CUDIT=Cannabis Use Disorders Identification Test.

Regarding sleep health domains, predictors (total duration [k=7]; weekday duration [k=7]; weekend duration [k=6]; satisfaction [k=17]; timing [k=10]; alertness [k=5]) and outcomes (total duration [k=7]; weekday duration [k=1]; satisfaction [k=9]) varied. Similarly, regarding substance use indicators, predictors (alcohol [frequency k=5; quantity k=3] and cannabis [frequency k=7] use) and outcomes (alcohol frequency [k=20]; alcohol quantity [k=7]; cannabis frequency [k=18]) varied.

Study Quality

No studies included in the meta-analysis were rated as high on any risk of bias domain assessed. The most common sources of methodological concern, where most studies were rated as unclear (as opposed to being high for bias risk), arose from incomplete outcome data (i.e., unclear handling of missing data; k=16) and selective outcome reporting (i.e., no pre-registered hypotheses; k=37). See Supplementary Table 1 for complete study quality ratings.

Sleep Duration and Substance Use

As shown in Table 2, we found evidence for a bi-directional association between total sleep duration and substance use. Shorter total sleep duration predicted alcohol (k=7, r=.08 [.03, .12], p<.001) and cannabis (k=6, r=.11 [.08, .14], p<.001) use. In the other direction, alcohol (k=5, r=.06 [.05, .08], p<.001) and cannabis (k=5, r=.23 [.09, .35], p=.001) use predicted shorter total sleep duration. While shorter weekday sleep duration predicted both alcohol (k=7, r=.07 [.02, .11], p=.001) and cannabis (k=5, r=.06 [.03, .09], p<.001) use, shorter weekend duration only predicted alcohol use (k=5, r=.04 [.01, .08], p=.007); results were inconclusive regarding cannabis use.

Table 2.

Pooled Pearson’s correlations and heterogeneity estimates from random effects models of the sleep health and substance use

	Alcohol Use predicted by Sleep Health						Sleep Health predicted by Alcohol Use
	Random Effects Model			Heterogeneity			Random Effects Model			Heterogeneity
	k	r [95% CI]	p-value	Q	p-value	I2	k	r [95% CI]	p-value	Q	p-value	I2
Total Duration	7	.08 [ .03, .12]	<.001	677.18	<.001	99%	5	.06 [ .05, .08]	<.001	0.95	.91	0%
Weekday duration	7	.07 [ .02, .11]	.001	55.78	<.001	89%
Weekend duration	5	.04 [ .01, .08]	.007	6.37	.17	37%
Satisfaction	11	.09 [ .04, .13]	<.001	48.36	<.001	79%	7	.06 [−.01, .14]	.10	53.57	<.001	88%
Alertness	5	.18 [ .03, .32]	.01	43.74	<.001	90%
Timing	9	.05 [−.04, .15]	.25	179.76	<.001	95%
	Cannabis Use predicted by Sleep Health						Sleep Health predicted by Cannabis Use
	Random Effects Model			Heterogeneity			Random Effects Model			Heterogeneity
	k	r [95% CI]	p-value	Q	p-value	I2	k	r [95% CI]	p-value	Q	p-value	I2
Total Duration	6	.11 [ .08, .14]	<.001	268.48	<.001	98%	5	.23 [ .09, .36]	.001	426.97	<.001	99%
Weekday duration	5	.06 [ .03, .09]	<.001	10.52	.03	61%
Weekend duration	4	.06 [−.05, .17]	.29	58.44	<.001	94%
Satisfaction	7	.07 [ .01, .13]	.01	86.21	<.001	93%	3	−.02 [−.10, .06]	.62	3.13	.20	36%
Alertness	3	.04 [−.08, .16]	.52	7.91	.01	74%
Timing	6	.09 [ .01, .17]	.03	52.57	<.001	90%

Note. k = number of studies; r = pooled Pearson’s correlation coefficient (values range from 0-1, with greater values indicating greater association); CI = confidence interval; Q = null hypothesis significance test for presence of heterogeneity within studies; I² = percentage of between-study variability

Given significant and large between-study heterogeneity in the sleep duration-substance use associations as indicated in Table 2, exploratory moderation analyses were conducted. After accounting for proportion of White and Hispanic youth in the same model, as proportion of Black (b=1.66, SE=0.43, p<.001) and “other” (b=3.40, SE=1.12, p<.001) youth increased, the association of alcohol use predicted by total sleep duration also increased. Insufficient available data disallowed for moderation analyses with proportion of Asian, Multiracial, American Indian/Alaska Native, or Native Hawaiian/Other Pacific Islander youth. The association of alcohol use predicted by total sleep duration was weaker as the proportion of young adult participants (vs. adolescent participants; b=−0.12, SE=0.06, p=.03) and as the number of effect sizes employing longitudinal study designs (vs. cross-sectional; b=−0.11, SE=0.05, p=.01) increased. Moderation analyses were inconclusive for proportion male and alcohol indicators in the total duration-alcohol use association. The moderating role of race/ethnicity, male, developmental stage, study design, and alcohol indicators were also inconclusive for the total sleep duration predicted by alcohol use association.

After accounting for proportion of White youth in the same model, as proportion of Black youth increased, the association of cannabis use predicted by total sleep duration decreased (b=−2.62, SE=0.85, p=.002). Similarly, compared to cross-sectional studies, as the number of longitudinal studies increased, a stronger association in cannabis use predicted by total sleep duration was found (b=0.62, SE=0.06, p<.001). Moderation analyses with proportion male and developmental stage were inconclusive.

Regarding weekday (as opposed to total) sleep duration, after accounting for each other proportion in the same model, as proportion of Black (b=0.32 SE=0.15, p=.03) and “other” (b=1.45, SE=0.71, p=.04) youth increased, the association of alcohol use predicted by weekday sleep duration decreased. The moderating role of male, developmental stage, study design, and alcohol use indicators were inconclusive for alcohol use predicted by weekend sleep duration. While the moderating roles of race/ethnicity and developmental stage were inconclusive, as proportion of male (b=1.10, SE=0.37, p=.003) and cross-sectional study designs (vs. longitudinal; b=−0.04, SE=0.01, p=.01) increased, the relationship of cannabis use predicted by weekday sleep duration increased.

Sleep Satisfaction and Substance Use

Sleep satisfaction positively predicted alcohol (k=11, r=.09 [.04, .13], p<.001) and cannabis use (k=7, r=.07 [.01, .13], p=.01), as shown in Table 2. Conversely, results were inconclusive regarding sleep satisfaction predicted by either alcohol or cannabis use. There was significant and large between-study heterogeneity in alcohol and cannabis use predicted by sleep satisfaction and thus exploratory moderation analyses were conducted (Table 2). Moderation analyses could not be conducted with proportion of “other”, American Indian/Alaska Native, or Native Hawaiian/Other Pacific Islander youth due to insufficient available data.

As proportion of Multiracial (b=2.48, SE=0.92, p=.007) and Hispanic (b=0.81, SE=0.27, p=.003) youth increased, and the proportion of Black (b=−0.55, SE=0.22, p=.01) and Asian (b=−1.8, SE=0.86, p=.03) youth decreased the association of alcohol use predicted by sleep satisfaction increased, after accounting for proportion of White youth in the same model. The moderating associations of male, developmental stage, study design, and alcohol indicators were inconclusive in the association of alcohol use predicted by sleep satisfaction. Results were also inconclusive regarding the respective moderation of race/ethnicity, male, developmental stage, or study design in the relationship of cannabis use predicted by sleep satisfaction.

Sleep Alertness with Substance Use

Sleep alertness positively predicted alcohol use (k=5, r=.18 [.03, .32], p=.01), but results were inconclusive regarding cannabis use (Table 2). Exploratory moderation analyses with race/ethnicity, male, developmental stage, study design, and alcohol indicators were conducted due to significant and large between-study heterogeneity (Table 2). Moderation analyses could not be conducted with proportion of Multiracial, American Indian/Alaska Native, other, or Native Hawaiian/Other Pacific Islander study participants due to insufficient available data.

As proportion of White (b=−2.09, SE=0.46, p<.001) and Black (b=−5.84, SE=1.32, p<.001) youth increased, the association of alcohol use predicted by sleep alertness decreased, when accounting for the other proportion in the same model. Moderation results were inconclusive for male, developmental stage, study design, or alcohol indicators in the association of alcohol use predicted by sleep alertness.

Sleep Timing with Substance Use

Sleep timing positively predicted cannabis use (k=6, r=.09 [.008, .17], p=.03) but results were inconclusive on alcohol use, as shown in Table 2. Given significant and large between-study heterogeneity in the timing-cannabis association, exploratory moderation analyses were conducted. As proportion of Black decreased (b=−0.82, SE=0.29, p=.004) and Hispanic youth increased (b=0.31, SE=0.13, p=.02), the association of cannabis use predicted by sleep timing increased, while controlling for White and “other” youth proportions in the same model. Moderation with proportion of Asian, Multiracial, American Indian/Alaska Native, or Native Hawaiian/Other Pacific Islander youth could not be conducted due to insufficient available data. Results were inconclusive whether male, developmental stage, study design, or alcohol assessment moderated the relationship of cannabis use predicted by sleep timing association.

Sensitivity Analyses

‘One-study-removed’ analyses yielded the same pattern of significance and inconclusive findings for all meta-analyses conducted, indicating overall meta-analyses results were not driven by a single effect size and that there were no extreme outliers.

Publication Bias

The only association that provided at least ten effect size estimates was the association of alcohol use predicted by sleep satisfaction. Results from the Egger’s regression test did not demonstrate presence of bias and, while the Trim and Fill method suggested imputation of two effect size estimates, the adjusted pooled association remained similar (r=.10 [.06, .15]).

DISCUSSION

This exploratory meta-analysis examined the bi-directional associations of multidimensional sleep health (i.e., duration, satisfaction, alertness, and timing) and substance use (i.e., alcohol and cannabis) among youth in the U.S., utilizing 98 effect sizes extracted from 38 total studies. Pooled correlations (r = .09 to .23) demonstrated distinct bi-directional patterns of sleep health dimensions and substance use for youth. However, the sleep-substance associations varied across different sleep domains and substance types/indicators, and significant between-study heterogeneity was accounted for by race/ethnicity, sex, developmental stage, and study design. Overall, consistent with previous meta-analyses on uni-directional sleep-substance use associations among young adults [23], this synthesis underscores the need for multidimensional assessment of sleep health, substance types and substance use indicators to understand the bi-directional relationship of youth sleep and substance use.

Meta-analytic findings indicated a bi-directional association of sleep duration and substance use, suggesting that alcohol and cannabis use may predict shorter sleep duration, which may in turn predict greater alcohol and cannabis use in youth. On the other hand, while youth sleep satisfaction predicted alcohol and cannabis use, results were inconclusive whether sleep satisfaction were predicted by alcohol or cannabis use. Although bi-directional associations could not be tested with other sleep health dimensions due to insufficient available data, we found positive uni-directional predictions of shorter weekday and weekend sleep duration and alertness with alcohol use (but not cannabis use); findings also indicated positive prediction of weekday sleep duration and timing with cannabis use (but not with alcohol use). Findings from this exploratory synthesis support burgeoning literature demonstrating that sleep health dimensions differentially predict substance use outcomes for youth [19, 23, 48-51]. Notably, a majority of articles that were extracted for this meta-analysis focused on sleep satisfaction as the sole domain of sleep health, alcohol use frequency as the only substance and substance use indicator, and uni-directional, rather than bi-directional relationships between sleep health and substance use. Thus, to enhance the sleep-substance use literature, a multidimensional focus on sleep health and broader range of substance use is needed to disentangle the complex temporal sleep-substance relationship. Such research is exigent for informing targeted interventions aimed at mitigating adverse consequences during the salient developmental epoch of youth.

Moderation analyses revealed significant variations in sleep health-substance use associations as a function of race/ethnicity, both between- and within- racial/ethnic groups. As the proportion of Black youth increased in a study sample, the association of alcohol use predicted by total and weekday sleep duration increased, but alcohol use predicted by sleep satisfaction and alertness and total sleep duration predicted by cannabis use decreased. Similarly, as the proportion of Hispanic youth increased, the association of alcohol use predicted by sleep satisfaction and cannabis use predicted by sleep timing increased, but alcohol use predicted by total sleep duration decreased. As the proportion of White youth increased, the association of alcohol use predicted by alertness decreased but cannabis use predicted by timing increased. Findings from this meta-analysis also demonstrated understudied racial/ethnic groups to moderate the sleep-substance relationship. As the proportion of Multiracial youth increased and proportion of Asian youth decreased, the relationship of alcohol use predicted by sleep satisfaction increased. Further, as the proportion of the “other” racial group increased, the associations of alcohol use predicted by total and weekday sleep duration also increased. Overall, these novel, albeit preliminary meta-analytic findings suggest that diverse racial/ethnic youth may differentially report associations of sleep health and substance use. Efforts to characterize racial/ethnic differences in bi-directional sleep health and substance use association among youth is an important avenue for future research. This line of research could critically enhance our nuanced understanding of the sleep health-substance use feedback loop over time, which may be driving health-related disparities for minoritized racial and ethnic youth.

Moderation analyses also demonstrated that proportion of male study participants positively moderated the relationship of cannabis use predicted by weekday sleep duration. However, results were inconclusive whether proportion male moderated additional sleep health-substance use associations. Given that over the past decade, prevalence of substance use and disorder have increased at a larger rate for females than males and that females report greater rates of sleep dissatisfaction than males, future research with more contemporary samples is needed to investigate inequities in sleep and substance use as a function of sex [93]. Further, investigating intersectionality of race/ethnicity with sex or other social identities [53] may provide greater insight into observed differences in the sleep and substance use associations.

Limitations and Future Directions

Findings of this meta-analysis should be interpreted in light of several limitations, which in turn highlight avenues for future research. First, this meta-analysis excluded articles that did not report race/ethnicity data, potentially contributing to differential associations across sleep and substance use indicators. Second, many studies included in analyses relied on uni-directional, cross-sectional data, which may have limited our ability to accurately assess the bi-directional relationship between sleep and substance use. Indeed, cross-sectional findings positively moderated the magnitude of the alcohol use predicted by total sleep duration relationship. Third, this synthesis amalgamated youth. While assessing youth may capture important temporal stability of the sleep and substance use relationship, developmental risk and protective factors in adolescence and young adulthood may be important to disentangle. For example, moderation analyses demonstrated that adolescents (as opposed to young adults) increased the magnitude of the relationship of alcohol use predicted by total sleep duration. Lastly, the limited literature disallowed for investigation of additional sleep health indicators (e.g., efficiency; regularity) and substance use domains (e.g., nicotine). Further, while substance use is typically assessed in terms of frequency, quantity, and impact [54], substance use indicators were amalgamated in analyses, potentially obfuscating important patterns in substance use behavior, although moderation analyses were inconclusive.

Conclusion

This exploratory meta-analysis represents a critical step toward characterizing nuanced bi-directional sleep health and substance use associations among youth. Greater efforts are needed to characterize the prospective relationship between multidimensional sleep health and substance use indicators and include diverse and understudied racial and ethnic groups. Such research can inform targeted interventions that may address the downward sleep-substance feedback loop and reduce health disparities in the U.S. youth.