Bidirectional associations between depressive symptoms and cigarette, e-cigarette, cannabis, and alcohol use: Cross-lagged panel analyses among young adults before and during COVID-19

Abstract

Background:

The literature regarding bidirectional relationships of depressive symptoms to cigarette and alcohol use is mixed, and limited regarding e-cigarette and cannabis use. Moreover, COVID-19 has significantly impacted mental health and substance use, especially among young adults. Thus, this is a critical period for focused research on these relationships among young adults.

Methods:

We analyzed longitudinal data (assessments in Fall 2018, 2019, and 2020) from 3,006 young adults (M_age = 24.56 [SD = 4.72], 54.8% female, 31.6% sexual minority, 71.6% White, 5.3% Black, 12.2% Asian, 11.4% Hispanic) from 6 US metropolitan statistical areas. Cross-lagged panel models were conducted to examine bidirectional associations between depressive symptoms and past 30-day use of cigarettes, e-cigarettes, cannabis, and alcohol (respectively), controlling for sociodemographics.

Results:

During the study period, depressive symptoms decreased before the pandemic but increased during, cigarette and e-cigarette use decreased in both periods, alcohol use showed no change before but increases during the pandemic, and cannabis use increased in both periods. Additionally, each outcome demonstrated greater stability before versus during COVID-19. Finally, greater antecedent depressive symptoms correlated with more days of subsequent cigarette (β = 0.03, SE = 0.01, p =.011) and e-cigarette use (β = 0.03, SE = 0.01, p =.021), but fewer days of alcohol use (β = −0.02, SE = 0.01, p =.035). W2 cannabis use and alcohol use, respectively, were related to W3 depressive symptoms (cannabis: β = 0.09, SE = 0.02, p <.001; alcohol: β = 0.06, SE = 0.02, p =.002). No other cross-lagged associations were significant.

Conclusions:

Intervention efforts targeting depression and substance use should explicitly address the potential for onset and escalation of substance use and depressive symptoms, respectively, especially during societal stressors.

1. Introduction

The prevalence of depressive disorders has increased in the US (Weinberger et al., 2018; Kessler et al., 2003; Hasin et al., 2018; Substance Abuse and Mental Health Services Administration, 2019), particularly during COVID-19 (Czeisler et al., 2020). One study using national data documented that the proportion of US adults experiencing mild, moderate, and severe depressive symptoms before versus during COVID-19 increased from 16.2% to 24.6%, 5.7% to 14.8%, and 0.7% to 5.1%, respectively (Ettman et al., 2020). Another study documented that prevalence of depressive disorder quadrupled from 2019 to 2020 (Czeisler et al., 2020). These pandemic-related increases may have resulted from several factors (e.g., loss of employment or income, social isolation, increased childcare responsibilities) (Romm et al., 2021; Brave et al., 2022; Saladino et al., 2020). Depressive disorders are particularly prevalent among young adults (National Institute of Mental Health, 2022); among young adults (ages 18–34), prevalence increased from 2018 (7.8%) to 2020 (15.2%) (Daly et al., 2021). The prevalence of depressive disorders also differs by other sociodemographic characteristics, with higher rates among women, White and multiracial individuals (National Institute of Mental Health, 2022), some foreign-born subgroups (Sun et al., 2020; Budhwani et al., 2015), and those of higher socioeconomic status (per education and income) and who are married (National Institute of Mental Health, 2022). Additionally, during COVID-19, parents reported greater increases in depressive symptoms and stress, perhaps due to increased childcare demands (Romm et al., 2021).

Depression has been shown to be prospectively associated with later substance use behaviors, including cigarette (Duan et al., 2021; Carroll et al., 2020), e-cigarette (Moustafa et al., 2021; Bandiera et al., 2017), cannabis (Substance Abuse and Mental Health Services Administration, 2019; Bandiera et al., 2017; Wilkinson et al., 2016; Lev-Ran et al., 2014), and alcohol use (Wilkinson et al., 2016; Peirce et al., 2000; McCarty et al., 2009). Several sociodemographic subgroups are particularly impacted by substance use, including younger adults (Substance Abuse and Mental Health Services Administration, 2019; National Institute on Drug Abuse, 2020; Cornelius et al., 2020; Osibogun et al., 2018; Petersen et al., 2020; Schulenberg et al., 2021), men, and sexual minorities (Hughes et al., 2016; Grant et al., 2017; Jeffers et al., 2021; National Cancer Institute, 2017), with sociodemographic subgroups disproportionately impacted across substances (e.g., Whites for alcohol and tobacco (National Cancer Institute, 2017; Szaflarski et al., 2011), Blacks for cannabis (Jeffers et al., 2021), some foreign-born immigrants for alcohol and tobacco (Szaflarski et al., 2011; Bosdriesz et al., 2013), higher socioeconomic status for alcohol (Szaflarski et al., 2011), and less education for cannabis and cigarettes (Jeffers et al., 2021; National Cancer Institute, 2017).

Use prevalence of these substances has changed in recent years (National Institute on Drug Abuse, 2020; Cornelius et al., 2020; Grucza et al., 2018; Harrell et al., 2017). COVID-19 has marked a pivotal time for substance use; estimates of the proportion of US adults initiating or increasing their substance use ranges from 10% (Czeisler et al., 2020) to 18.2% (McKnight-Eily et al., 2021). Outside of COVID-19, other societal changes may have impacted use, for example, the dramatic evolution of the US tobacco market, including e-cigarettes (Cornelius et al., 2020; Harrell et al., 2017; Kasza et al., 2017). The past decade also marked rapidly changing cannabis regulation, with 33 states currently allowing medical cannabis use (Global, 2020) and 11 states and DC allowing recreational use (Macnamara, 2020). Within this context, changes in young adult use differed across substances: past-month use prevalence from 2018 to 2020 decreased for cigarettes (from 7.8% in 18–24 year-olds and 16.5% in 25–44 year-olds (Creamer et al., 2019) to 7.4% and 14.1%, respectively (Cornelius et al., 2022)) and alcohol (55.1% in 18–25 year-olds (Substance Abuse and Mental Health Services Administration, 2018) to 51.5% (Cornelius et al., 2022)), but increased for e-cigarettes (from 7.6% in 18–24 year-olds and 4.3% in 25–44 year-olds (Creamer et al., 2019) to 9.4% and 5.2% (Cornelius et al., 2022)) and cannabis (from 21.5% in 18–25 year-olds (Substance Abuse and Mental Health Services Administration., 2018) to 23.0% (Substance Abuse and Mental Health Services Administration, 2020)).

Many studies have shown bidirectional relationships between depressive symptoms and substance use (Fluharty et al., 2017; Boden and Fergusson, 2011). As posited by several theoretical explanations, these relationships may result from various underlying pathways (Borsboom, 2017; Baskin-Sommers, Foti, 2015). One macro-level theory is network theory, which suggests that mental health symptoms are causally connected through myriads of biological, psychological, and societal mechanisms (Borsboom, 2017). Among the most well-studied neurobiological mechanisms is reward processing (Baskin-Sommers and Foti, 2015); individuals with depressive symptoms may engage in substance use to self-medicate (i.e., to reduce negative affect) (Gehricke et al., 2007; Boden et al., 2010) and/or achieve a high or enjoyment (Baskin-Sommers and Foti, 2015). Additionally, substance use may lead to depressive symptoms due to related neurophysiological and metabolic changes, which ultimately evolves to bidirectional associations (Boden and Fergusson, 2011). Connections between depression and substance use can be amplified by various psychosocial factors (e.g., parent/peer use) or buffered (e.g., resilience, social support) (Borsboom, 2017).

Bidirectional associations between depressive symptoms and substance use have been most extensively examined with regard to cigarette smoking and alcohol use. For example, a 2017 systematic review of 148 studies on cigarette smoking found evidence of positive associations in both directions (Fluharty et al., 2017): nearly half indicated baseline depression/anxiety predicted some type of later smoking behavior, and over a third showed smoking exposure predicted later depression/anxiety (Fluharty et al., 2017). A 2011 review covering 30 years of published research indicated that alcohol use disorder or major depressive disorder doubled the risks of the other disorder, asserting that alcohol use is more likely to increase the risk of depression (vs. the alternate explanation) (Boden and Fergusson, 2011).

Limited research has examined bidirectional relationships between depression and alternative tobacco products or cannabis, thus undermining related intervention efforts addressing potential co-occurring behavioral concerns and aiming to reduce the risk of one leading to the other (e.g., e-cigarettes leading to depression). A 2021 review of e-cigarette use and depression in adolescents and young adults showed mixed findings, with some indicating that e-cigarette use predicted later depressive symptoms and some no associations (Becker et al., 2021). Another longitudinal study of young adults found that antecedent depressive symptoms predicted future e-cigarette use, but not the reverse (Bandiera et al., 2017). In contrast, an adolescent study revealed a bidirectional relationship between depressive symptoms and e-cigarette use. (Lechner et al., 2017) Regarding cannabis, one study of adolescents and young adults found that cannabis use predicted subsequent depressive symptoms, but not vice versa (Hoffmann, 2018). Another young adult study found that depressive symptoms predicted cannabis use, but not vice versa (Wilkinson et al., 2016). However, a national study documented bidirectional associations between cannabis use and depressive symptoms, but indicated that cannabis use more strongly predicted depression relative to the reverse association (Pacek et al., 2013).

There are several key opportunities to advance the literature. First, research regarding e-cigarette and cannabis use in relation to depressive symptoms is limited, thus warranting further examination. Second, limited research has examined bidirectional relationships across cigarette, e-cigarette, cannabis, and alcohol use within the same study, which would mitigate the implications of differential sample characteristics on results. Third, COVID-19 has had a major impact on mental health and substance use in the US and globally over the past 2 years (Ettman et al., 2020; Czeisler et al., 2020; McKnight-Eily et al., 2021), thus calling for research to inform related interventions during this pivotal period.

This study analyzed a longitudinal dataset from a diverse sample of young adults across 6 US metropolitan statistical areas (MSAs) surveyed annually with 2 surveys before the COVID-19 pandemic (Fall 2018, Fall 2019) and one survey during the pandemic (Fall 2020). We assessed bidirectional associations between depressive symptoms and use of cigarettes, e-cigarettes, cannabis, and alcohol. We hypothesized: 1) greater increases in depressive symptomatology and substance use during versus before the pandemic; 2) antecedent depressive symptoms correlate with greater subsequent use of each substance; and 3) use of each substance correlates with greater subsequent depressive symptoms.

2. Materials & methods

2.1. Study design

This study analyzed survey data among young adults (aged 18–34) in a 2-year longitudinal cohort study, the Vape shop Advertising, Place characteristics and Effects Surveillance (VAPES) study, which examines vape retail and consumer impact. This study draws participants from 6 MSAs (Atlanta, Boston, Minneapolis, Oklahoma City, San Diego, Seattle) with varied tobacco and cannabis legislative contexts (Public Health Law Center, 2020). This study, described elsewhere (Berg et al., 2021), was approved by the Emory University Institutional Review Board.

2.2. Participants

Potential participants were recruited via social media in Fall 2018 and were assessed annually (with mid-year check-ins). Ads posted on Facebook and Reddit targeted individuals: 1) using indicators reflecting those eligible (i.e., 18–34 years old, residing in zip codes of the 6 aforementioned MSAs, English speaking); 2) by identifying work groups or activities of interest that appeal to young adults (e.g., sports/athletics, entertainment, arts, lifestyle, technology), as well as tobacco-related interests (e.g., Marlboro, Juul); and 3) by posting advertisements including images of young adults of diverse racial/ethnic backgrounds socializing in bars and/or outdoor spaces, young adult professionals in professional work settings, etc. Purposive, quota-based sampling was used to ensure sufficient proportions of the sample represented e-cigarette and cigarette users and to obtain roughly equal numbers of men and women and 40% racial/ethnic minority; subgroup enrollment was capped by MSA.

After clicking an ad, individuals were directed to a webpage with a consent form, and then completed an online eligibility screener, which assessed age and home address zip code, as well as sex, race/ethnicity, and cigarette and e-cigarette use in order to facilitate achieving subgroup enrollment quotas. If eligible and allowed to advance, they then completed the online Wave 1 (W1) survey. Then, participants were notified that 7 days later they would receive an email to confirm their participation. Upon confirming, they were officially enrolled and emailed their first incentive ($10 e-gift card).

Of 10,433 who clicked on ads, 9,847 consented, of which 2,751 (27.9%) were not allowed to advance due to: 1) ineligibility (n = 1,472); and/or 2) their subgroup target being met (n = 1,279). Of the 7,096 allowed to advance, 3,636 (51.2%) began the survey but did not provide complete data (the majority of whom discontinued the survey in the initial section of the survey). Of the 3,460 (48.8%) who provided complete data, 3,006 (86.9%) confirmed participation at the 7-day follow-up (Berg et al., 2021). This study uses data from W1 (2018; n = 3,006), W2 (2019; n = 2,375, 79.0%), and W3 (2020; n = 2,476, 82.4%); overall, 797 (26.5%) were lost to attrition at either W2 or W3.

2.3. Measures

2.3.1. Outcomes: Depressive symptoms and substance use.

At each wave, we administered the Patient Health Questionnaire – 2 item (PHQ-2), which asks about depressive symptoms (“little interest or pleasure in doing things”, “feeling down, depressed or hopeless”) in the past 2 weeks (0 = not at all to 3 = nearly every day); higher sum scores indicate greater symptomatology (range = 0–6; Cronbach’s alpha = 0.87) (Kroenke et al., 2003). At each wave, substance use was assessed using measures from national surveillance systems (Substance Abuse and Mental Health Services Administration. 2020; Harlow et al., 2019): “During the past 30 days, how many days did you: smoke cigarettes? use electronic cigarettes (or e-cigarettes)? use marijuana/cannabis? drink alcohol?” (continuous variables: ranges = 0–30).

3. Covariates

Sociodemographic covariates (assessed at W1) included: age, sex, sexual orientation, race, ethnicity, US versus foreign born, education level, employment status, relationship status, children in the home, and MSA.

3.1. Data analysis

We conducted: 1) descriptive analyses to characterize the sample; and 2) bivariate analyses to assess sociodemographics in relation to depressive symptoms and use of each substance, separately. The distributions of depressive symptoms were approximately normal (skewness 0.88–1.0). However, the distributions of substance use were slightly skewed (skewness varying 0.23–3.25), so we conducted log transformation for each substance use variable (skewness varying 0.11–1.61).

To assess changes in depressive symptoms and substance use across the 3 waves, we conducted multilevel linear regression modeling, for depressive symptoms and log transformed substance use variables with a random intercept to account for the clustering of repeated measures within each participant. Time was included as a categorical variable; changes over time (W1-W2, W2-W3, W1-W3) were estimated.

Cross-lagged panel models (CLPM, i.e., a type of longitudinal structural equation modeling) were used to assess bidirectional associations between depressive symptoms and substance use. CLPM assesses cross-lagged relationships between 2 variables (i.e., an antecedent variable predicting a second variable at a later time point), as well as autoregressive relationships (or stability) of a single variable over time, accounting for contemporary correlations between the residuals of 2 variables measured concurrently. Other analytic approaches were considered, specifically random intercept CLPM (Hamaker et al., 2015); however, this approach was rejected because it is recommended for data including at least 4 waves of data (Bolger and Laurenceau, 2013), only captures temporal fluctuations around individual person means – ignoring potential effects of between-person causes (Asendorpf and Rauthmann, 2021; Orth et al. 2021; Andersen, 2021), and has limited ability to control for unobserved confounders in estimating cross-lagged effects (Lüdtke and Robitzsch, 2021).

For each substance, the initial CLPM estimated cross-lagged and autoregressive paths freely without constraints. We then used Wald tests to assess equality of cross-lagged relationships between depressive symptoms and substance use. If Wald tests yielded non-significant results (suggesting associations did not differ over time), paths were constrained to be equal over time. Robust maximum likelihood was used to account for possible skewness. Non-significant Chi-square tests (p >.05), Root Mean Square Error of Approximation (RMSEA) ≤ 0.08, Tucker–Lewis index (TLI) > 0.95, and comparative fit index (CFI) > 0.95 were used to indicate good model fit (Kline, 2015). Models included MSA of residence and sociodemographic covariates, selected based on the literature (Little and Card, 2013). (Note: Preliminary analyses indicated no differences in results if covariate selection was restricted, e. g., omitting nativity, children.) Mplus 8.1 was used for CLPM, missing values were accounted for with full information maximum likelihood, and significance level was set at p <.05.

4. Results

4.1. Participant characteristics

Participants were on average 24.56 years old (SD = 4.71), 56.5% female, 31.6% sexual minority, 71.6% White, 5.3% Black, 12.2% Asian, 10.9% other race, and 11.4% Hispanic (Table 1). Bivariate analyses (Table 1) indicated that greater depressive symptoms correlated with more days of use of each substance and weak-moderate correlations in substance use behaviors (range = 0.06–0.25; p’s < 0.001). Significant (p <.05) correlates of depressive symptoms included being younger, sexual minority, White or other race, Hispanic, less educated, and unemployed. More days of use of each substance correlated with being a sexual minority (except for alcohol), US-born, and married/living with a partner/ other marital status (vs. single). Additional details are provided in Table 1.

Table 1.

Sociodemographics, depressive symptoms & substance use among young adults in 6 US metropolitan statistical areas, Fall 2018 (Wave 1 [W1]), N = 3,006.

Variables	Total	W1 PHQ-2		W1 Cigarette Use		W1 E-cigarette Use		W1 Cannabis Use		W1 Alcohol Use
Sociodemographics	N (%) or *M (SD)	M (SD) or *r	^p	M (SD) or *r	^p	M (SD) or *r	^p	M (SD) or *r	^p	M (SD) or *r	^p
Age *	24.56 (4.71)	−0.10	<0.001	0.16	<0.001	−0.05	0.007	0.00	0.875	0.13	<0.001
Male #	1,271 (43.5)	1.66 (1.69)	0.567	4.00 (8.69)	0.119	7.77 (11.83)	<0.001	5.48 (9.94)	0.138	6.10 (6.96)	0.030
Female	1,648 (56.5)	1.62 (1.65)		3.50 (8.65)		5.75 (10.70)		4.94 (9.45)		5.57 (6.17)
Sexual minority	950 (31.6)	2.19 (1.80)	<0.001	4.65 (9.64)	<0.001	7.51 (11.73)	0.004	7.02 (10.78)	<0.001	5.76 (6.81)	0.938
No	2,056 (68.4)	1.43 (1.57)		3.35 (8.23)		6.25 (11.02)		4.45 (9.13)		5.78 (6.40)
Race
White	2,151 (71.6)	1.69 (1.68)	<0.001	3.84 (8.84)	<0.001	7.30 (11.73)	<0.001	5.57 (9.99)	<0.001	6.28 (6.67)	<0.001
Black	159 (5.3)	1.38 (1.57)		5.09 (9.99)		3.35 (7.52)		5.04 (9.94)		4.54 (6.19)
Asian	367 (12.2)	1.44 (1.57)		1.68 (5.72)		3.58 (8.62)		2.35 (6.31)		3.45 (4.74)
Other	329 (10.9)	1.95 (1.81)		4.96 (9.60)		7.47 (11.48)		6.60 (10.66)		5.68 (6.85)
Hispanic	343 (11.4)	1.89 (1.77)	0.011	4.88 (9.12)	0.012	7.06 (11.06)	0.472	5.64 (9.93)	0.442	5.63 (6.38)	0.656
No	2,663 (88.6)	1.64 (1.67)		3.62 (8.66)		6.60 (11.29)		5.21 (9.73)		5.79 (6.55)
Foreign born	286 (9.5)	1.59 (1.66)	0.408	2.62 (7.39)	0.020	4.49 (9.65)	0.001	2.59 (6.74)	<0.001	4.84 (6.46)	0.011
No	2,720 (90.5)	1.68 (1.68)		3.88 (8.84)		6.88 (11.40)		5.55 (9.97)		5.87 (6.53)
Education ≥ BA degree	2,203 (73.3)	1.53 (1.59)	<0.001	2.91 (7.71)	<0.001	5.43 (10.40)	<0.001	4.33 (8.75)	<0.001	6.31 (6.55)	<0.001
No	803 (26.7)	2.07 (1.85)		6.08 (10.69)		10 (12.78)		7.83 (11.72)		4.31 (6.26)
Employment
Student	825 (27.5)	1.63 (1.64)	<0.001	1.86 (6.04)	<0.001	5.34 (10.14)	<0.001	3.76 (8.16)	<0.001	4.48 (5.39)	<0.001
Unemployed	244 (8.1)	2.26 (1.96)		7.60 (12.13)		7.11 (11.84)		7.00 (11.49)		3.67 (5.42)
Full-time	1,202 (40.0)	1.46 (1.61)		4.22 (9.18)		6.68 (11.48)		4.96 (9.59)		7.08 (7.22)
Part-time	735 (24.5)	1.87 (1.69)		3.87 (8.63)		7.94 (11.76)		6.88 (10.68)		5.79 (6.39)
Relationship status
Single	1,876 (62.4)	1.73 (1.71)	0.052	3.12 (7.92)	<0.001	5.99 (10.70)	<0.001	4.77 (9.12)	<0.001	5.41 (6.22)	<0.001
Married/living with partner	1,089 (36.2)	1.57 (1.63)		4.75 (9.78)		7.66 (12.00)		5.96 (10.58)		6.39 (7.00)
Other	41 (1.4)	1.63 (1.64)		6.88 (10.93)		10.12 (13.57)		9.61 (12.49)		5.95 (6.54)
Children in the home	611 (20.3)	1.66 (1.75)	0.851	5.88 (10.65)	<0.001	8.24 (12.17)	<0.001	6.16 (10.74)	0.011	4.35 (6.21)	<0.001
No	2,395 (79.7)	1.68 (1.66)		3.22 (8.07)		6.25 (10.99)		5.03 (9.47)		6.14 (6.56)
W1 depressive symptoms * §	1.67 (1.68)	–	–	0.17	<0.001	0.11	<0.001	0.16	<0.001	0.03	0.080
W1 depressive symptoms ≥ 3	2,290 (76.2)	4.15 (1.13)		6.11 (10.29)		8.26 (11.62)		6.95 (11.03)		5.76 (6.68)
No	716 (23.8)	0.89 (0.89)	<0.001	3.03 (8.03)	<0.001	6.15 (11.10)	<0.001	4.74 (9.26)	<0.001	5.78 (6.49)	0.954
W1 number of days used past 30 *
Cigarettes	3.76 (8.72)	0.17	<0.001	–	–	–	–	–	–	–	–
E-cigarettes	6.65 (11.26)	0.11	<0.001	0.15	<0.001	–	–	–	–	–	–
Cannabis	5.26 (9.75)	0.16	<0.001	0.25	<0.001	0.24	<0.001	–	–	–	–
Alcohol	5.77 (6.53)	0.03	0.080	0.14	<0.001	0.06	0.002	0.12	<0.001	–	–
W1 past 30-day use status
Cigarettes	808 (26.88)	2.23 (1.83)	<0.001	–	–	571 (70.67)	<0.001	497 (61.59)	<0.001	675 (83.54)	<0.001
No	2,198 (73.12)	1.47 (1.58)		–	–	562 (25.57)		681 (31.03)		1628 (74.07)
E-cigarettes	1,133 (37.69)	2.01 (1.73)	<0.001	571 (50.40)	<0.001	–	–	678 (60.05)	<0.001	925 (81.64)	<0.001
No	1,873 (62.31)	1.47 (1.62)		237 (12.65)		–	–	500 (26.70)		1378 (73.57)
Cannabis	1,178 (39.24)	1.98 (175)	<0.001	497 (42.19)	<0.001	678 (57.56)	<0.001	–	–	1044 (88.62)	<0.001
No	1,824 (60.76)	1.47 (1.60)		310 (17.00)		451 (24.73)		–	–	1257 (68.91)
Alcohol	2,303 (76.61)	1.67 (1.67)	0.921	675 (29.31)	<0.001	925 (40.17)	<0.001	1044 (45.37)	<0.001	–	–
No	703 (23.39)	1.67 (1.72)		133 (18.92)		208 (29.59)		134 (19.12)		–	–

^*Continuous variables reported as M (SD) in total column and r for each W1 outcome. Otherwise, categorical variables: N (%) reported in total column and M (SD) for each outcome.

^{^}p-values reflect tests (t-tests or ANOVA for categorical variables; Pearson correlations for continuous variables) between the respective outcome (W1 depressive symptoms or substance use) and the indicated covariate.

^#Other gender reported by 87 participants.

^§Depressive symptoms scores not log transformed.

Past 30-day use prevalence was highest at W1 for cigarettes (26.9%) and e-cigarettes (37.7%), and roughly equal across waves for cannabis (39.2–40.0%) and alcohol (71.2–76.9%; Table 2). As presented in Table 3, multilevel modeling indicated significant W1-W2 decreases in depressive symptoms (p =.024), but significant W2-W3 increases (p <.001); decreases in past 30-day cigarette and e-cigarette use across both periods (p’s < 0.001); increases in cannabis use in each period (W1-W2: p =.028; W2-W3: p =.004); and W2-W3 decreases in alcohol use (p <.001), but no W1-W2 change. After adjusting for covariates, findings were similar, except change in depressive symptoms at W1-W2 (non-significant).

Table 2.

Average depressive symptoms and days of cigarette, e-cigarette, cannabis & alcohol use (in the past 30-days) among young adults in 6 US metropolitan statistical areas assessed in Fall 2018 (Wave 1 [W1]), 2019 (W2) and 2020 (W3), respectively.

	W1 N = 3,006		W2 N = 2,375		W3 N = 2,476
Variable	M (SD)	N (%) with score ≥ 3	M (SD)	N (%) with score ≥ 3	M (SD)	N (%) with score ≥ 3
Depressive symptoms	1.67 (1.68)	716 (23.8)	1.58 (1.67)	514 (21.6)	1.80 (1.72)	612 (24.7)
Past 30-day substance use	Days of use among all participants	Reported any past 30-day use	Days of use among all participants	Reported any past 30-day use	Days of use among all participants	Reported any past 30-day use
	M (SD)	N (%)	M (SD)	N (%)	M (SD)	N (%)
Cigarettes	3.76 (8.72)	808 (26.9)	2.79 (7.42)	572 (24.1)	2.48 (7.10)	483 (19.5)
E-cigarettes	6.65 (11.26)	1,133 (37.7)	4.94 (9.94)	776 (32.7)	4.41 (9.75)	623 (25.2)
Cannabis	5.26 (9.75)	1,178 (39.2)	4.94 (9.37)	945 (40.0)	5.72 (10.20)	990 (40.0)
Alcohol	5.77 (6.53)	2303 (76.6)	5.47 (6.16)	1,827 (76.9)	5.57 (6.79)	1,763 (71.2)

Table 3.

Change of depressive symptoms, days of cigarette, e-cigarette, cannabis & alcohol use (in the past 30-days) over time, in Fall 2018 (Wave 1 [W1]), 2019 (W2) and 2020 (W3).

	Cigarette				E-cigarette				Depressive symptoms
	Estimated change over time				Estimated change over time				Est. change over time
Time	b (SE)	p	Adjusted* b (SE)	p	b (SE)	p	Adjusted* b(SE)	p	b (SE)	p	Adjusted* b (SE)	p
W1	Ref.		Ref.		Ref.		Ref.		Ref.		Ref.
Change W1 to W2	−0.07 (0.02)	<0.001	−0.06 (0.02)	<0.001	−0.10 (0.02)	<0.001	−0.09 (0.02)	<0.001	−0.08 (0.03)	0.024	−0.05 (0.03)	0.14
Change W2 to W3	−0.08 (0.02)	<0.001	−0.07 (0.02)	<0.001	−0.13 (0.02)	<0.001	−0.13 (0.02)	<0.001	0.22 (0.04)	<0.001	0.21 (0.04)	<0.001
Change W1 to W3	−0.14 (0.02)	<0.001	−0.13 (0.02)	<0.001	−0.23 (0.02)	<0.001	−0.22 (0.02)	<0.001	0.15 (0.03)	<0.001	0.16 (0.03)	<0.001
	Alcohol				Cannabis
	Estimated change over time				Estimated change over time
Time	b (SE)	p	Adjusted* b (SE)	p	b (SE)	p	Adjusted* b (SE)	p
W1	Ref.		Ref.		Ref.		Ref.
Change W1 to W2	−0.02 (0.02)	0.185	−0.03 (0.02)	0.095	0.04 (0.02)	0.028	0.04 (0.02)	0.024
Change W2 to W3	−0.07 (0.02)	<0.001	−0.07 (0.02)	<0.001	0.05 (0.02)	0.004	0.05 (0.02)	0.005
Change W1 to W3	−0.09 (0.02)	<0.001	−0.10 (0.02)	<0.001	0.10 (0.02)	<0.001	0.10 (0.02)	<0.001

^*Adjusted models adjust for the covariates including age, sex, sexual orientation, race, ethnicity, US versus foreign born, education level, employment status, relationship status, children residing in the home, and MSA of residence. Substance use variables (i.e., days of use) are log transformed.

4.2. Cross-lagged panel model (CLPM) on depressive symptoms and substance use

The initial CLPM models (using log transformed use variables) did not fit the data. Thus, based on modification indices, second-order autoregressive paths of W3 on W1 for both depressive symptoms and each substance use variable were included for each model. In each case, the model fit was significantly improved. Then, we assessed equality of cross-lagged paths for antecedent depressive symptoms correlating with later substance use over time and cross-lagged paths for antecedent substance use correlating with depressive symptoms over time. For cigarettes and e-cigarettes, Wald tests suggested non-significant differences in each path over time, so we constrained the cross-lagged paths to be equal over time. For cannabis and alcohol, Wald tests showed significant differences over time for antecedent use in relation to later depressive symptoms, but not the reverse associations; thus, the latter were constrained to be equal over time.

Model fit was excellent for cigarettes (Chi-square = 2.570[df = 4], p =.632, RMSEA = 0.000, CFI = 1.0, TLI = 1.0, SRMR = 0.002), e-cigarettes (Chi-square = 6.698[df = 4], p =.153, RMSEA = 0.015, CFI = 0.999, TLI = 0.987, SRMR = 0.003), cannabis (Chi-square = 6.515[df = 3], p =.089, RMSEA = 0.020, CFI = 0.999, TLI = 0.973, SRMR = 0.003), and alcohol (Chi-square = 4.148[df = 3], p =.246, RMSEA = 0.011, CFI = 1.0, TLI = 0.989, SRMR = 0.002).

Greater antecedent depressive symptoms correlated with more days of subsequent cigarette (standardized path coefficient, β = 0.03, SE = 0.01, p =.011) and e-cigarette use (β = 0.03, SE = 0.01, p =.021), but fewer days of alcohol use (β = −0.02, SE = 0.01, p =.035; see Fig. 1). W2 cannabis use and alcohol use, respectively, were related to W3 depressive symptoms (cannabis: β = 0.09, SE = 0.02, p <.001; alcohol: β = 0.06, SE = 0.02, p =.002). No other cross-lagged associations were significant.

Fig. 1. Cross-lagged panel models (CLPM) on the associations between depressive symptoms and substance use among young adults assessed in Fall 2018 (Wave 1), 2019 (Wave 2) and 2020 (Wave 3), respectively.

Notes: Standardized path coefficients and SEs. ** p <.01. * p <.05. Bolded lines indicate significant estimates and dashed lines insignificant. Substance use variables are log-transformed. CIG, ECIG, CAN, ALC, and DEP indicate use of cigarettes, e-cigarettes, cannabis, and alcohol, and depressive symptoms, respectively. All models adjusted for the following covariates (not illustrated in figure): age, sex, sexual orientation, race, ethnicity, US versus foreign born, education level, employment status, relationship status, children residing in the home, and MSA of residence.

Based on standardized path coefficients (β), depressive symptom stability was modest-large for the first-order autoregressive path (β varying between 0.33 and 0.51) and small-modest for the second-order path (0.26–0.28), and stability of cigarette, e-cigarette, cannabis, and alcohol use was large for the first-order autoregressive path (0.50–0.76) and weak for the second-order path (0.19–0.25).

5. Discussion

In this diverse sample of US young adults, findings indicated that depressive symptomatology and substance use behaviors were more stable during the year prior to (2018–2019) versus during the pandemic (2019–2020). These findings might suggest the potential role of pandemic-related societal instability and stressors in impacting depressive symptoms and substance use (Romm et al., 2021; Brave et al., 2022; Saladino et al., 2020). Furthermore, depressive symptoms and cannabis use increased in both periods (particularly during the pandemic), while cigarette and e-cigarette use decreased in both periods, and alcohol use decreased during the pandemic. These findings somewhat align with national data, indicating increases in depressive symptoms and cannabis use and decreases in cigarette and alcohol use from 2018 to 2020 (Daly et al., 2021; Creamer et al., 2019; Cornelius et al., 2022; Substance Abuse and Mental Health Services Administration, 2018; Substance Abuse and Mental Health Services Administration, 2019); however, our findings differ from national data that indicate increases in e-cigarette use (Substance Abuse and Mental Health Services Administration, 2019; Creamer et al., 2019; Cornelius et al., 2022; Substance Abuse and Mental Health Services Administration, 2018).

Notably, no hypothesized bidirectional associations between depressive symptoms and substance use were found. Instead, earlier depressive symptoms were associated with more days of subsequent cigarette and e-cigarette use before and during the pandemic, while earlier cannabis and alcohol use were associated with later depressive symptoms, but only from Fall 2019 to Fall 2020. Interestingly, depressive symptoms were associated with fewer days of alcohol use during both periods. Current findings must be contextualized with the existing literature regarding each substance and reflecting on the various potential underlying pathways, as posited by several theoretical explanations (Borsboom, 2017; Baskin-Sommers and Foti, 2015).

Regarding cigarettes and e-cigarettes, current findings are consistent with some literature, for example the 2017 systematic review that showed that depressive symptoms consistently predict subsequent cigarette smoking (Fluharty et al., 2017). However, our findings contradict the systematic review’s findings regarding bidirectionality, which were undermined by a very small number of studies reporting bidirectionality, particularly for the relationship between depression and smoking level (n = 2) (Fluharty et al., 2017). Current findings align with earlier study results indicating that antecedent depressive symptoms in young adults predicted future e-cigarette use, but not reverse associations (Bandiera et al., 2017), but contradict earlier findings indicating a bidirectional relationship between depressive symptoms and e-cigarette use in adolescents (Lechner et al., 2017). It is worth noting that the first-order stability (autoregressive) coefficients were larger for cigarette and e-cigarette use than depressive symptoms, making it difficult to observe cross-lagged effects for these use variables. Nevertheless, current findings and the available literature suggest that, at least in young adults, depressive symptoms may precede cigarette and e-cigarette use and progression, and their potential use for affect regulation (Gehricke et al., 2007; Mineur and Picciotto, 2009), which contributes to the literature, particularly the limited and mixed literature regarding e-cigarettes (Bandiera et al., 2017; Becker et al., 2021; Lechner et al., 2017).

Current results regarding cannabis and alcohol use align with some research indicating that later depressive symptoms are associated with earlier cannabis (Hoffmann, 2018) and alcohol use (Boden and Fergusson, 2011), respectively. However, these relationships were only documented during the pandemic period; that is, Fall 2019 cannabis and alcohol use, respectively, were associated with greater increases in depressive symptoms by Fall 2020. Moreover, we found no association between earlier depressive symptoms and increased cannabis or alcohol use, contradicting existing findings that depressive symptoms predicted cannabis use (Wilkinson et al., 2016) and showing bidirectional correlations between depression and cannabis (Wilkinson et al., 2016; Lev-Ran et al., 2014; Pacek et al., 2013; Rasic et al., 2013) and alcohol use (Carroll et al., 2020; Wilkinson et al., 2016; Peirce et al., 2000; McCarty et al., 2009; Marsden et al., 2019; Ranjit et al., 2019). In fact, current findings indicated that fewer prior depressive symptoms were associated with more subsequent alcohol use. Collectively, these findings could be interpreted by the literature suggesting that cannabis and alcohol may be used for social or other reasons but may also increase depressive symptoms (Bresin and Mekawi, 2019; Votaw and Witkiewitz, 2021; Cooper et al., 2016), which may be particularly true during societal stressors or periods of isolation.

These findings have implications for research and practice. First, findings suggest that use of one substance does not equate to the others in terms of depressive symptoms as an underlying mechanism, and vice versa. In young adults – who represent the age group at highest risk for using cigarettes, e-cigarettes, cannabis, and alcohol (Substance Abuse and Mental Health Services Administration 2019; National Institute on Drug Abuse, 2020; Cornelius et al., 2020; Osibogun et al., 2018; Petersen et al., 2020; Schulenberg et al., 2021) – the trajectories and reasons for use may be particularly distinct (e.g., social reasons for cannabis and alcohol) (Berg et al., 2020; Berg et al., 2018; Berg et al., 2021; Haardorfer et al., 2021; Patterson et al., 2020; Bierhoff et al., 2019; Gorfinkel et al., 2020). Moreover, how substance use behaviors have been impacted by COVID-19 have also differed. While some studies indicated increases in use of cigarettes (Vanderbruggen et al., 2020), e-cigarettes (Yach, 2020), cannabis (Dumas et al., 2020; Romm et al., 2021; Sharma et al., 2020; van Laar et al., 2020), and alcohol (Vanderbruggen et al., 2020; Romm et al., 2021; Sharma et al., 2020), some indicate decreases in cigarette (Romm et al., 2021; Sharma et al., 2020) and e-cigarette use (Dumas et al., 2020; Stokes, 2020). However, the broader literature (over time) is even more mixed in terms of the associations between these substance use behaviors and depressive symptoms, potentially due to differences across study samples (e.g., age of cohort, sample size), sociodemographic compositions of samples, the measures of depression and substance use employed, the length and number of follow-up periods, and the specific analytic approaches used, among other reasons. Thus, we must advance the literature, with a particular eye toward ensuring rigorous methods. For example, future research assessing substance use and psychological factors across multiple time-points (e.g., intensive, short-term diary studies or experience sampling methods) could use random intercept CLPM to advance our understanding of within- and between-person dynamics and of the strengths and limitations of these different analytic approaches (Hamaker et al., 2015; Asendorpf and Rauthmann, 2021; Lüdtke and Robitzsch, 2021). In practice, intervening on depression and substance use simultaneously among individuals with co-occurring depression and substance use disorders is the optimal approach (Hides et al., 2010; Minkoff, 2001), and current findings underscore the need to monitor substance use over time among patients reporting only depression, and vice versa.

Study limitations include limited generalizability to other US young adults. Substance use rates should not be interpreted as prevalence rates, as our sampling design aimed to achieve a sample with roughly a third being current e-cigarette and cigarette users – and thus, they likely have higher other substance use rates. This may also explain the particularly high rates of depressive symptoms in this sample (~21%-24%) across periods relative to the national estimates (~8–15%) (Daly et al., 2021). In addition, these analyses did not account for polyuse, as we aimed to assess bidirectional associations between use of each specific substance in relation to depressive symptoms and documented differential associations. Notably, no strong collinearity was found among the substance use variables and depressive symptoms. Nonetheless, future research might use longitudinal latent class analysis to address polyuse. Finally, our measures were limited in scope (i.e., not exhaustive of all potentially important factors) and by their self-reported nature.

6. Conclusions

No bidirectional associations were documented in this study. Instead, earlier depressive symptoms were associated with greater cigarette and e-cigarette use before and during the pandemic, while earlier cannabis and alcohol use were related to later depressive symptoms, but only from Fall 2019 to Fall 2020. Moreover, depressive symptomatology and substance use behaviors were more stable prior to versus during COVID-19, underscoring the impact of this major societal stressor. These findings should inform intervention efforts to optimize the physical and mental health outcomes among young adults, especially as the population continues to experience the impact of COVID-19 and may face other societal stressors in the future. Furthermore, research should further elucidate explanatory mechanisms of associations between depressive symptoms and use of specific substances, across specific populations, in order to advance this area of research.