Trends in the co-occurrence of substance use and mental health symptomatology in a national sample of US post-secondary students from 2009 to 2019

Abstract

Objective:

This study examined joint trends over time in associations between substance use (heavy drinking, cannabis, and cigarette smoking) and mental health concerns (depression, anxiety, and suicidal ideation) among US post-secondary students.

Participants:

Data came from 323,896 students participating in the Healthy Minds Study from 2009 to 2019, a national cross-sectional survey of US post-secondary students. Weighted two-level logistic regression models with a time by substance interaction term were used to predict mental health status.

Results:

Use of each substance was associated with a greater odds of students endorsing depression, anxiety, and suicidal ideation. Over time, the association with mental health concerns strengthened substantially for cannabis, modestly for heavy drinking, and remained stable for smoking.

Conclusion:

Given co-occurrence is common and increasing among post-secondary students, college and university health systems should prioritize early identification, psychoeducation, harm-reduction, and brief interventions to support students at risk.

Introduction

Depression, anxiety, and suicidality have been rapidly increasing among college and university students over the past decade, with suicide being the second leading cause of death among this population.^1,2 A majority of college and university (hereby referred to as post-secondary) students are within the developmental stage of emerging adulthood, a period typically including those who are 18–25 years of age, characterized by being “in-between” adolescence and adulthood.³ Relative to adolescents and adults, emerging adults experience the highest incidence and prevalence of mental illness, substance use, and substance use disorders (SUDs).^4,5 Substance use has been associated with a greater odds of experiencing depression and anxiety among post-secondary students^6–8 with the most commonly used substances being alcohol, cannabis, and tobacco products.^9,10 More generally, use of these substances during adolescence and emerging adulthood is significantly associated with the onset or worsening of suicidal behaviors and mental health problems.^3–6 However, trends in substance use over time do not follow the same pattern as mental health problems.

Over the past two decades, there have been slight decreases in alcohol use, large decreases in cigarette use, and slight increases in cannabis use among emerging adults in the US and Canada.^11–14 The decline in cigarette use appears universal across age groups¹¹ while trends in cannabis and alcohol vary across developmental age groups. Specific to US post-secondary students, data from the Monitoring the Future national survey have demonstrated similar decreases in heavy episodic drinking, large decreases in cigarette smoking, and slight increases in cannabis use over time.¹⁵ Notably, the prevalence and trends in substance use differ between post-secondary and non-post-secondary age-matched peers. Specifically, alcohol use is higher among post-secondary students with slower declines over time compared to non-post-secondary attending peers, while the opposite is true for cigarette smoking prevalence and trends.^15,16 There are also differences in the prevalence of types of SUDs, whereby post-secondary students are more likely to experience alcohol use disorder and non-post-secondary attending peers are more likely to experience nicotine and other SUDs.¹⁷ Overall, the patterns in substance use over time are distinct from the consistent and dramatic increases in mental health problems observed among adolescents and emerging adults, and the post-secondary context seems to play a role in these trends.^1,2,11,18,19

To our knowledge, trends in the co-occurrence of substance use and mental health problems have not yet been explored among post-secondary students. There are a few studies using nationally representative repeated cross-sectional surveys of adolescents and adults that have found inconsistent and divergent temporal patterns depending on developmental age, specific substance, and mental health concern being examined.^20–27 Specifically, among US adolescents, there appears to have been a decoupling of the association between past 2-week heavy episodic drinking and depression between 1991 and 2018²³ but a strengthening in the association between past month cannabis use and depression between 2004 and 2016.²¹ Among Canadian and US adults, three studies have found increases in the co-occurrence of cannabis use and depression and suicidality over the past 2 decades (15–60; 2002–2012²⁴; 12+, 2005–2017²⁵; 20–59, 2005–2016²⁷), one found increases in the co-occurrence of high anxiety and cannabis (18+, 2008–2017²⁶), and one found no temporal changes in the co-occurrence of cannabis use and distress (18+, 2008–2016²⁰). Related to smoking, increased connections between smoking (before age 25) and mental health problems have also been observed among more recent US birth cohorts.²² Thus, it is important to explore the role of substance use in the evolving post-secondary mental health landscape, namely their joint trends over time, to inform future prevention and intervention efforts within post-secondary institutions.

This study seeks to examine the trends in associations between common substance use (e.g., alcohol, cannabis, and smoking) and mental health concerns (e.g., depression, anxiety, and suicidal ideation) using a large national sample of US post-secondary students between 2009 and 2019. Of note, prior research has consistently found sex and gender differences in prevalence and patterns of mental health concerns and substance use; males are typically more likely to use any substance and have a SUD, though gender gaps in substance use are closing among adults,²⁸ whereas females experience higher rates of depression and anxiety4 and more co-occurring substance use and mental health concerns.^{25,26,29–31} However only two of the existing temporal studies directly examined sex or gender differences in changes over time with inconsistent results.^23,24 Further, as previously discussed, there are also notable differences in temporal changes in mental health and substance use across developmental age. Accordingly, secondary objectives of this study include exploration of gender and developmental age differences in these substance use and mental health trends among post-secondary students. Using large-scale data, this study is able to fill gaps in knowledge at a population-level and variations therein, with implications both at a system-level and also for addressing behavioral health inequalities.

Materials and methods

Data

Data for the current study came from the Healthy Minds Study (HMS) from 2009 to 2019. HMS is a national cross-sectional Web-based survey of US post-secondary students conducted annually since 2009. HMS was approved by the institutional review boards on all campuses. A random sample of 4,000 students was selected from each participating school (or an entire population when a school had <4,000 students). Students had to be at least 18 years old to participate; there were no other exclusion criteria. Students were recruited via email with three reminders (with the exception of 2009 which included mail and email) and students were entered into a cash sweepstakes drawing regardless of participation. Students completed informed consent online prior to entering the survey. Response rates were as follows: 42% in 2009 (15 schools), 25% in 2010 (26 schools), 26% in 2011 (11 schools), 23% in 2012 (31 schools), 16% in 2013 (17 schools), 26% in 2014–2015 (37 schools), 27% in 2015–2016 (23 schools), 23% in 2016–2017 (54 schools), 23% in 2017–2018 (60 schools), 16% in 2018–2019 (79 schools), and 16% in 2019–2020 (70 schools).

To account for potential differences between respondents and the full random samples at each school, sampling probability weights were computed based on school administrative data related to gender, academic level, race/ethnicity, and grade point average. Logistic regression was used to estimate the probability of being a responder based on these administrative sociodemographic characteristics, and weights were subsequently generated by taking 1 divided by the predicted probability of responding. This method follows the nonresponse adjustment weight used in the National Comorbidity Survey Replication.³² These sampling weights give equal aggregate weight to each school, so the estimates are not dominated by larger schools.¹

This paper includes data from all students within US schools that received mental health questionnaires and at least one substance use item. Only complete data were used for analyses with original sampling weights. Detailed item-level missing data analyses are available in Supplementary Materials. Thus, the final samples for analysis included 323,896 students in 269 schools between 2009 and 2019 for depression and suicidal ideation analyses and 263,292 students in 226 schools between 2013 and 2019 for anxiety analyses (when the anxiety screen was added to HMS). Sample sizes across years in the final weighted sample are as follows: 4.2% (n = 13,929) 2009, 7.5% (n = 24,706) 2010, 3.1% (n = 10,337) 2011, 8.2% (n = 26,955) 2012, 4.5% (n = 14,792) 2013, 8.9% (n = 29,457) 2014, 5.4% (n = 17,802) 2015, 12.1% (n = 39,725) 2016, 10.6% (n = 35,029) 2017, 16.3% (n = 53,570) 2018, 19.1% (n = 62,879) 2019. Please see supplementary materials for more details regarding institutional characteristics across years in the final analysis sample.

Measures

Mental health outcomes

Depression was measured with the Patient Health Questionnaire-9 (PHQ9)³³ scale, which assesses the presence and frequency of 9 depression symptoms over the past 2 weeks, scored on a scale from not at all (0) to nearly every day (3). Scores were summed (min 0, max 27) where higher scores reflect a greater severity of depressive symptomatology. Depression was analyzed categorically, using a clinical cut-point of ≥10 as a proxy for meeting criteria for major depression. This cut-point has been consistently found to have good sensitivity and specificity for diagnosing major depression in previous studies and meta-analyses and performs better than other cut-points.³⁴

Anxiety was measured with the Generalized Anxiety Disorder-7 (GAD7)³⁵ scale, which assesses the presence and frequency of 7 symptoms of GAD over the past 2 weeks, scored on a scale from not at all (0) to nearly every day (3). Scores were summed (min 0, max 21) where higher scores reflect a greater severity of anxiety symptomatology. Anxiety was analyzed categorically, using a clinical cut-point of > =10 as a proxy for meeting criteria for GAD. Cutoff scores between 7 and 10 have demonstrated good sensitivity and specificity, with 10 being most commonly used.³⁶ Of note, the GAD7 was not included until the 2013 survey year.

Suicidal Ideation (SI) was measured with a single item “In the past year, did you ever seriously think about attempting suicide?” Response options included yes (1) and no (0).

Substance use

Heavy Episodic Drinking (HED) was measured by asking students “Over the past 2 weeks, on how many occasions have you had 4 [female]/5 [male]/4–5 [transgender, other] drinks in a row?” Response options included none, once, twice, 3–5 times, 6–9 times, and 10 or more times. Responses were collapsed into any HED in the past 2 weeks (1) compared to none (0).

Cannabis use was measured by asking students about any use over the past 30 days. Wording slightly differed across years. From 2009 to 2011, students were asked “In the past 30 days, have you used any of the following drugs?” with a list of drugs including “Marijuana (also known as grass, weed, pot, hash, or hash oil).” From 2012 to 2014, students were asked “How often, if ever, have you used any of the substances listed below? Do not include anything you used prescribed by your doctor.” with a list of drugs including “Marijuana (or hashish, blunts, Spice, K2)” and response options from never (0) to used in the past 30 days (4). From 2015 to 2019, students were asked “Over the past 30 days, have you used any of the following drugs?” with a list of drugs including “Marijuana.” A binary variable was created whereby responses were coded as any cannabis use in the past 30 days (1) compared to none (0).

Cigarette Smoking was measured by asking students, “Over the past 30 days, about how many cigarettes did you smoke per day?” with response options including 0, less than 1 cigarette, 1–5 cigarettes, about one-half pack, and 1 or more packs. Some years had response options that extended to 2 or more packs a day. Of note, 2013 and 2014 only asked about any use in the past 30 days. Response options were collapsed into any cigarette use in the past 30 days (1) versus none (0).

Student sociodemographic covariates

Age group was included categorically as 18–20 (underage emerging adults), 21–25 (legal drinking age emerging adults), and 26 and over (mature student age). Moderating effects of age groups were explored.

Gender was included as female, male, or transgender or gender non-conforming (TGNC). Gender moderation effects were explored.

Undergraduate student status was included as a covariate defined as: enrollment in an Associate’s or Bachelor’s degree, enrollment in a graduate or professional degrees, or enrollment in an ‘other’ degree type, online degree, or non-degree.

Race was included as Asian, Black, Hispanic, White, other race (including American Indian or Alaskan Native, Native Hawaiian or Pacific Islander, Middle Eastern, Arab, or Arab American, or other self-identified race), or Multiracial in the analyses. Across all years of the data, students were asked to identify their race and ethnicity based on a set of responses with instructions to “select all that apply,” which was recoded for analysis.

First generation students were captured by recoding the parental education question which asked students, “What is the highest level of education completed by one of your parents or stepparents” for two parents or caregivers. Response options included: 8th grade or lower, between 9th and 12th grade but no high school degree, high school degree, some college but no degree, Associate’s degree, Bachelor’s degree, or Graduate degree. The highest parental education response was used for the analyses and recoded into first generation student (1) if indicating the highest level of parental education was an Associate’s degree or less.

Institutional covariates

Several institutional characteristics were included as covariates to account for changes in school participation over time, given prior research has found institutional characteristics are associated with variation in mental health outcomes in this sample.³⁷ First, schools were coded as public or private.³⁸ Second, institution type was coded based on Carnegie Classification³⁹ of: Associate’s Colleges, Baccalaureate Colleges, Special Focus Institutions, Master’s Colleges and Universities, or Doctorate-granting Universities. Third, school size was included based on student body size as per institution websites and coded in five categories: under 1,000, 1,000–4,999, 5,000–9,999, 10,000–19,999, and 20,000+. Fourth, school selectivity was included based on proportion of applicants admitted.^40,41 Lastly, the census geographic region in which the school was located was adjusted for (e.g., West, Midwest, Northeast, South).

Analysis

All analyses were performed using SAS Enterprise guide 7.1. Descriptive statistics were calculated within the larger sample used for depression and SI. To explore temporal differences in demographics, a series of multilevel regressions predicting demographics were used to explore any significant effects of continuous year. Further, differences in participating school were explored through school-level correlations and Chi-square tests between institutional characteristics and year. Next, the weighted prevalence of HED, cannabis use, and cigarette use were plotted over time. Weighted 2-level (students within school) logistic regressions were used to predict change in prevalence of substance use over time, adjusted for student sociodemographic and school institutional characteristics. Subsequently, the weighted prevalence of categorized mental health concerns (e.g., PHQ9 > =10, GAD7 > =10, any SI) were plotted over time, stratified by dichotomous substance use variables. Lastly, a series of 2-level logistic regressions were performed separately for associations between substance use and mental health outcomes. All models were estimated using generalized linear mixed models with maximum odds estimation by adaptive Gaussian quadrature, which is an integral approximation approach to allow for the inclusion of sampling weights². Model 1A included all student and school covariates, year, and substance use. Model 1B additionally included an interaction term between substance and year. To note, curvilinear relationships between year and mental health outcomes were explored by comparing model fit with and without the inclusion of polynomials (e.g., year² and year³). Model fit was negligibly different with and without polynomials; thus, year was estimated as a linear effect. 10-year temporal trends were generated by multiplying the log odds (i.e., β) of the time by substance interaction by 10 and: (1) taking the exponent for a 10-year OR indicating change relative to 2009, and (2) calculating the ORs and 95% confidence intervals (CIs) for students using substances in 2009 and 2019 by combining the substance use main effect with the substance by year interaction. The same steps were repeated for 6-year anxiety trends (2013–2019). An OR of 1.5 was considered clinically meaningful corresponding to a cohen’s d of ~0.2,⁴² which has also been used in previous studies.²⁴ In the second set of models, 3-way interaction terms between substance, time, and gender (model 2A) and age group (model 2B) were added. A significance level of p < 0.05 was used.

Results

The overall sample was predominantly composed of students in Associate’s or Undergraduate degrees (80.1%) with 42.8% 18–20 years old, 37.0% 21–25 years old, and 20.2% 26 years and older. The sample was 56.1% female, 42.2% male, and 1.7% TGNC. The majority of students identified as White (65.2%); 10.1% were Asian, 7.0% Hispanic, 6.8% Black, 3.2% other race, and 8.1% Multiracial. Further, 36.5% were first-generation post-secondary students. Over time, there were significantly more females and less White students, with no other student demographics significantly varying across years after accounting for school clustering. These findings reflect previously documented improvements in access to post-secondary education and diversity among student populations over the past decade.⁴³ At the school level, the proportion of participation across regions and school types varied significantly across years and there was a positive correlation between admission selectivity and time (i.e., less selective schools were more likely to participate in recent years). The proportion of public versus private schools and school size were not correlated with year.

General substance use trends

Comparing the weighted prevalence in 2009 and 2019, there was about a 20% decrease in HED, 40% increase in cannabis, and 50% decrease in smoking. See Figure 1 for the weighted prevalence of substance use over time. Multilevel regressions adjusting for student sociodemographic and institutional characteristics revealed significant time trends, whereby: HED decreased over time (OR = 0.96, 95% CI 0.94 to 0.97; p < 0.0001), cannabis use increased over time (OR = 1.04, 95% CI 1.03 to 1.06; p < 0.0001), and smoking decreased over time (OR = 0.92, 95% CI 0.91 to 0.93; p < 0.0001). This corresponds to a 10-year 34% decrease in the odds of HED, 54% increase in the odds of cannabis, and 57% decrease in the odds of smoking.

Figure 1.

Weighted unadjusted prevalence of substance use over time.

Depression

The proportion of students surpassing clinical cutoffs for depression significantly increased over time (See Table 1 and Figure 2 for results). Being an emerging adult (18–20 and 21–25), female, TGNC, Undergraduate/Associate’s or other degree student, Hispanic, Asian, Multiracial, of other race, and first generation student were consistently associated with greater odds of surpassing cutoffs for depression.

Table 1.

Multilevel regressions for examining substance use associations with depressive symptoms surpassing clinical cut-points from 2009 to 2019.

	Depression on HED		Depression on cannabis		Depression on smoking
	(N = 323,896)		(N = 323,896)		(N = 323,896)
	OR (95% CI); p value		OR (95% CI); p value		OR (95% CI); p value
Year	1.07 (1.06 to 1.09); <.0001	1.07 (1.05 to 1.09); <.0001	1.07 (1.05 to 1.08); <.0001	1.06 (1.05 to 1.07); <.0001	1.08 (1.07 to 1.1); <.0001	1.08 (1.07 to 1.1); <.0001
Substance	1.29 (1.25 to 1.34); <.0001	1.2 (1.12 to 1.29); <.0001	1.84 (1.78 to 1.91); <.0001	1.51 (1.38 to 1.65); <.0001	2.05 (1.97 to 2.14); <.0001	2.09 (1.93 to 2.26); <.0001
Year*Substance		1.01 (1 to 1.02); 0.0166		1.03 (1.02 to 1.04); <.0001		1 (0.99 to 1.01); 0.6804
18–20 (ref 26+)	1.33 (1.25 to 1.41); <.0001	1.33 (1.25 to 1.41); <.0001	1.28 (1.2 to 1.35); <.0001	1.28 (1.21 to 1.36); <.0001	1.4 (1.32 to 1.49); <.0001	1.4 (1.32 to 1.49); <.0001
21–25	1.25 (1.19 to 1.32); <.0001	1.25 (1.19 to 1.32); <.0001	1.24 (1.18 to 1.31); <.0001	1.24 (1.18 to 1.31); <.0001	1.33 (1.26 to 1.4); <.0001	1.33 (1.26 to 1.4); <.0001
Female (ref Male)	1.34 (1.3 to 1.38); <.0001	1.34 (1.3 to 1.38); <.0001	1.38 (1.34 to 1.42); <.0001	1.38 (1.34 to 1.42); <.0001	1.38 (1.34 to 1.43); <.0001	1.39 (1.35 to 1.43); <.0001
TGNC	4.06 (3.65 to 4.51); <.0001	4.05 (3.64 to 4.5); <.0001	3.95 (3.54 to 4.4); <.0001	3.94 (3.54 to 4.39); <.0001	3.96 (3.55 to 4.42); <.0001	3.95 (3.54 to 4.41); <.0001
Black (ref White)	0.99 (0.93 to 1.06); 0.7455	0.99 (0.92 to 1.05); 0.6946	0.97 (0.91 to 1.04); 0.3953	0.97 (0.91 to 1.04); 0.4071	1 (0.94 to 1.07); 0.9373	1 (0.94 to 1.07); 0.8889
Hispanic	1.1 (1.05 to 1.15); <.0001	1.1 (1.05 to 1.15); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001
Asian	1.09 (1.04 to 1.14); 0.0006	1.09 (1.04 to 1.14); 0.0006	1.13 (1.08 to 1.19); <.0001	1.13 (1.08 to 1.19); <.0001	1.05 (1 to 1.1); 0.0435	1.05 (1 to 1.1); 0.0394
Other Race	1.36 (1.28 to 1.45); <.0001	1.36 (1.28 to 1.44); <.0001	1.33 (1.25 to 1.42); <.0001	1.33 (1.25 to 1.42); <.0001	1.28 (1.2 to 1.36); <.0001	1.28 (1.21 to 1.37); <.0001
Multiracial	1.33 (1.27 to 1.39); <.0001	1.32 (1.27 to 1.38); <.0001	1.3 (1.24 to 1.35); <.0001	1.3 (1.24 to 1.36); <.0001	1.31 (1.26 to 1.37); <.0001	1.32 (1.26 to 1.37); <.0001
Undergrad or associate’s (ref graduate and professional)	1.41 (1.34 to 1.48); <.0001	1.41 (1.34 to 1.48); <.0001	1.38 (1.31 to 1.45); <.0001	1.38 (1.31 to 1.45); <.0001	1.38 (1.31 to 1.45); <.0001	1.38 (1.31 to 1.45); <.0001
Other Degree Type	1.17 (1.06 to 1.28); 0.001	1.17 (1.07 to 1.28); 0.0007	1.15 (1.06 to 1.26); 0.0018	1.16 (1.06 to 1.26); 0.0017	1.16 (1.06 to 1.27); 0.0018	1.16 (1.06 to 1.27); 0.0015
First Generation Student	1.22 (1.19 to 1.26); <.0001	1.22 (1.19 to 1.26); <.0001	1.22 (1.19 to 1.26); <.0001	1.22 (1.19 to 1.26); <.0001	1.22 (1.19 to 1.26); <.0001	1.22 (1.19 to 1.26); <.0001
10-year change	1.12 (d = 0.06)		1.34 (d = 0.16)		0.98 (d=−0.01)

Sampling weights were applied, and all institutional variables were adjusted across models. The table includes models 1A and 1B related to PHQ9≥10 for each substance.

Figure 2.

Weighted unadjusted prevalence of depression (PHQ9>=10) stratified by substance use.

In Model 1A, all substances were associated with a greater odds of surpassing clinical cut-points for depression (OR_HED= 1.29; OR_CAN= 1.84 OR_CIG= 2.05). Model 1B revealed the associations between substance use and depression significantly increased over time for HED and cannabis but not smoking. Temporal effect sizes for HED (OR_1year=1.01; OR_10year=1.12) were negligible and for smoking were non-significant indicating these associations have remained relatively stable over time (i.e., not strengthening nor decoupling). Cannabis yielded the strongest temporal trend (OR_1year=1.03; OR_10year=1.34). Put differently, in 2009, students who endorsed past month cannabis use had 1.51 times (95% CI 1.38 to 1.64) higher odds of surpassing cutoffs for depression compared to those who did not endorse use, while in 2019 the odds were 2.01 times higher (95% CI 1.93 to 2.12). Thus, there was evidence of strengthening in the association between cannabis and depression surpassing a priori thresholds for meaningful effects (i.e., OR > =1.5).

Gender and age group did not moderate the magnitude of the change in associations between substance use and depression over time. Of note, when examining 2-way interactions between substance use and gender and age without including the time interaction, gender and age did impact the magnitude of the main effect of substance use. Compared to males, females had larger effects of cannabis use (OR = 1.09 [95% CI 1.02 to 1.16]) while TGNC students had lower effects of cannabis use (OR = 0.78 [95% CI 0.65 to 0.94]). There were smaller associations between HED and depression among legal age emerging adults (OR_21–25 0.85 [95% CI 0.79 to 0.92]) compared to mature students (26+) while there were larger ssociations between smoking and depression for emerging adults (OR_18–20 1.35 [95% CI 1.21 to 1.50]; OR_21–25 1.19 [95% CI 1.08 to 1.32]). There were no differences in associations between depression and HED or cannabis and developmental age. Thus, gender and developmental age do seem to play a role in the relationship between substance use and depression, but these demographics did not impact how trends changed over time.

Suicidal ideation

The proportion of students reporting SI significantly increased over time (See Table 2 and Figure 3 for results). Being an emerging adult (18–20 and 21–25), Undergraduate/Associate’s or other degree student, female, TGNC, Multiracial, and first generation student were consistently associated with greater odds of indicating SI. Being Hispanic was related to a lower odds of SI.

Table 2.

Multilevel regressions for examining substance use associations with suicidal ideation from 2009–2019.

	SI on HED		SI on Cannabis		SI on Smoking
	N = 323,896		N = 323,896		N = 323,896
	OR (95% CI); p value		OR (95% CI); p value		OR (95% CI); p value
Year	1.09 (1.07 to 1.1); <.0001	1.08 (1.06 to 1.09); <.0001	1.08 (1.07 to 1.1); <.0001	1.07 (1.06 to 1.09); <.0001	1.1 (1.08 to 1.11); <.0001	1.1 (1.08 to 1.11); <.0001
Substance	1.33 (1.27 to 1.39); <.0001	1.1 (1 to 1.21); 0.0474	2.04 (1.95 to 2.13); <.0001	1.75 (1.56 to 1.96); <.0001	2.08 (1.97 to 2.2); <.0001	2.15 (1.91 to 2.41); <.0001
Year*substance		1.03 (1.01 to 1.04); 0.0001		1.02 (1.01 to 1.04); 0.0056		1 (0.98 to 1.01); 0.6202
18–20 (ref 26+)	1.67 (1.54 to 1.82); <.0001	1.68 (1.55 to 1.82); <.0001	1.6 (1.47 to 1.73); <.0001	1.6 (1.47 to 1.74); <.0001	1.78 (1.64 to 1.94); <.0001	1.78 (1.64 to 1.94); <.0001
21–25	1.41 (1.31 to 1.51); <.0001	1.41 (1.32 to 1.52); <.0001	1.39 (1.29 to 1.49); <.0001	1.39 (1.29 to 1.49); <.0001	1.51 (1.41 to 1.62); <.0001	1.51 (1.41 to 1.62); <.0001
Female (ref male)	1.02 (0.98 to 1.07); 0.244	1.02 (0.98 to 1.06); 0.2771	1.06 (1.02 to 1.1); 0.0062	1.06 (1.02 to 1.1); 0.0062	1.06 (1.02 to 1.1); 0.0061	1.06 (1.02 to 1.11); 0.0057
TGNC	3.87 (3.49 to 4.29); <.0001	3.89 (3.51 to 4.32); <.0001	3.77 (3.4 to 4.17); <.0001	3.76 (3.4 to 4.17); <.0001	3.78 (3.39 to 4.2); <.0001	3.76 (3.38 to 4.18); <.0001
Black (ref white)	1.07 (0.99 to 1.16); 0.0752	1.07 (0.99 to 1.16); 0.0774	1.06 (0.98 to 1.15); 0.1412	1.06 (0.98 to 1.15); 0.1424	1.1 (1.01 to 1.19); 0.0203	1.1 (1.01 to 1.19); 0.0215
Hispanic	0.91 (0.84 to 0.99); 0.0223	0.91 (0.84 to 0.99); 0.0237	0.92 (0.85 to 1); 0.0416	0.92 (0.85 to 1); 0.0424	0.92 (0.85 to 1); 0.0394	0.92 (0.85 to 0.99); 0.0368
Asian	0.99 (0.93 to 1.05); 0.6752	0.99 (0.93 to 1.05); 0.6833	1.05 (0.99 to 1.12); 0.079	1.05 (0.99 to 1.12); 0.0821	0.95 (0.9 to 1.01); 0.0895	0.95 (0.9 to 1.01); 0.0917
Other race	1.05 (0.96 to 1.16); 0.287	1.05 (0.96 to 1.16); 0.2806	1.03 (0.93 to 1.13); 0.5794	1.03 (0.93 to 1.14); 0.5619	0.99 (0.89 to 1.09); 0.7801	0.98 (0.89 to 1.09); 0.7463
Multiracial	1.36 (1.28 to 1.45); <.0001	1.36 (1.28 to 1.44); <.0001	1.33 (1.25 to 1.41); <.0001	1.33 (1.25 to 1.41); <.0001	1.35 (1.27 to 1.43); <.0001	1.35 (1.27 to 1.44); <.0001
Undergrad or associate’s (ref graduate and professional)	1.39 (1.29 to 1.49); <.0001	1.39 (1.3 to 1.49); <.0001	1.35 (1.25 to 1.45); <.0001	1.35 (1.25 to 1.45); <.0001	1.35 (1.26 to 1.45); <.0001	1.35 (1.26 to 1.45); <.0001
Other degrees type	1.4 (1.19 to 1.65); <.0001	1.41 (1.2 to 1.66); <.0001	1.39 (1.18 to 1.63); <.0001	1.39 (1.18 to 1.63); 0.0001	1.4 (1.19 to 1.65); <.0001	1.39 (1.18 to 1.64); 0.0001
First generation	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001	1.11 (1.06 to 1.16); <.0001
10-year change	1.31 (d = 0.15)		1.24 (d = 0.12)		0.96 (d=−0.02)

Sampling weights were applied, and all institutional variables were adjusted across models. The table includes models 1A and 1B related to endorsing SI for each substance.

Figure 3.

Weighted unadjusted prevalence of suicidal ideation stratified by substance use.

In Model 1A, all substances were associated with a greater odds of endorsing SI (OR_HED= 1.33; OR_CAN= 2.04; OR_CIG= 2.08). Model 1B revealed the associations between substance use and SI significantly increased over time for HED and cannabis but not smoking. Temporal effect sizes for HED (OR_1year=1.03; 10-year OR_10year=1.31) and cannabis (OR_1year=1.02; OR_10year 1.24) were small in magnitude. In 2009, students engaging in HED were negligibly more likely to endorse SI compared to those who did not (OR = 1.10 [95% CI 1.002 to 1.21]), while in 2019 the odds were 1.44 times higher (95% CI 1.34 to 1.54). For cannabis use, students using cannabis in 2009 had 1.75 times (95% CI 1.56 to 1.96) higher odds of SI compared to those who did not, while in 2019 the odds were 2.16 times higher (95% CI 2.03 to 2.29). Thus, for SI there is some evidence of strengthening in the association with HED (although below a priori thresholds) and cannabis (surpassing a priori thresholds) but stable associations with smoking.

Female gender was associated with slightly smaller temporal changes in the association between HED and SI (OR = 0.97 [95% CI 0.95 to 0.99); p = 0.03). However, 2-way interaction models found, in general, females had larger associations between cannabis use and SI (OR = 1.13 [95% CI 1.05 to 1.22]; p = 0.001) and smoking and SI (OR = 1.05 [1.0002 to 1.09]; p = 0.049) and no other gender differences emerged for HED. Emerging adults had larger associations between cigarette smoking and SI (OR_18–20 1.28 [95% CI 1.11 to 1.46]; OR_21–25 1.16 [95% CI 1.02 to 1.33]) but no age differences emerged for HED or cannabis.

Anxiety

The proportion of students surpassing clinical cut-points for GAD significantly increased over time (See Table 3 and Figure 4 for results). Being an emerging adult (18–20 and 21–25), Undergraduate/Associate’s student, female, TGNC, Multiracial, other race, and first generation student were consistently associated with greater odds of surpassing cutoffs for GAD. Black and Asian students had a lower odds of surpassing GAD clinical cutoffs.

Table 3.

Multilevel regressions for examining substance use associations with anxiety symptoms surpassing clinical cut-points from 2013 to 2019.

	Anxiety on HED		Anxiety on cannabis		Anxiety on smoking
	N = 263,292		N = 263,292		N = 263,292
	OR (95% CI); p value		OR (95% CI); p value		OR (95% CI); p value
Year	1.14 (1.12 to 1.15); <.0001	1.13 (1.11 to 1.14); <.0001	1.13 (1.11 to 1.14); <.0001	1.12 (1.1 to 1.13); <.0001	1.14 (1.13 to 1.16); <.0001	1.14 (1.13 to 1.16); <.0001
Substance	1.23 (1.18 to 1.27); <.0001	1.13 (1.05 to 1.22); 0.0014	1.66 (1.6 to 1.72); <.0001	1.4 (1.3 to 1.5); <.0001	1.82 (1.73 to 1.91); <.0001	1.91 (1.73 to 2.11); <.0001
Year*substance		1.02 (1 to 1.04); 0.0215		1.04 (1.02 to 1.06); <.0001		0.99 (0.96 to 1.01); 0.2758
18–20 (ref 26+)	1.36 (1.28 to 1.45); <.0001	1.36 (1.28 to 1.46); <.0001	1.32 (1.24 to 1.41); <.0001	1.32 (1.24 to 1.41); <.0001	1.43 (1.34 to 1.52); <.0001	1.43 (1.34 to 1.52); <.0001
21–25	1.35 (1.27 to 1.42); <.0001	1.35 (1.27 to 1.42); <.0001	1.33 (1.26 to 1.41); <.0001	1.33 (1.26 to 1.41); <.0001	1.42 (1.34 to 1.5); <.0001	1.42 (1.34 to 1.5); <.0001
Female (ref male)	1.76 (1.7 to 1.83); <.0001	1.76 (1.7 to 1.83); <.0001	1.81 (1.75 to 1.88); <.0001	1.81 (1.74 to 1.87); <.0001	1.81 (1.75 to 1.88); <.0001	1.81 (1.75 to 1.88); <.0001
TGNC	3.76 (3.41 to 4.14); <.0001	3.75 (3.41 to 4.13); <.0001	3.67 (3.33 to 4.05); <.0001	3.66 (3.32 to 4.03); <.0001	3.67 (3.32 to 4.06); <.0001	3.67 (3.32 to 4.06); <.0001
Black (ref White)	0.76 (0.71 to 0.82); <.0001	0.76 (0.71 to 0.82); <.0001	0.75 (0.7 to 0.81); <.0001	0.75 (0.7 to 0.81); <.0001	0.77 (0.72 to 0.83); <.0001	0.77 (0.72 to 0.83); <.0001
Hispanic	0.97 (0.91 to 1.03); 0.3463	0.97 (0.91 to 1.03); 0.3365	0.98 (0.91 to 1.04); 0.4521	0.98 (0.92 to 1.04); 0.4862	0.98 (0.91 to 1.04); 0.4758	0.98 (0.91 to 1.04); 0.4713
Asian	0.8 (0.76 to 0.85); <.0001	0.8 (0.76 to 0.84); <.0001	0.83 (0.79 to 0.88); <.0001	0.83 (0.79 to 0.88); <.0001	0.78 (0.73 to 0.82); <.0001	0.78 (0.73 to 0.82); <.0001
Other race	1.2 (1.11 to 1.3); <.0001	1.2 (1.11 to 1.29); <.0001	1.18 (1.09 to 1.27); <.0001	1.18 (1.1 to 1.27); <.0001	1.14 (1.06 to 1.23); 0.0009	1.14 (1.06 to 1.23); 0.0008
Multiracial	1.2 (1.14 to 1.26); <.0001	1.2 (1.14 to 1.26); <.0001	1.18 (1.12 to 1.24); <.0001	1.17 (1.12 to 1.24); <.0001	1.19 (1.13 to 1.25); <.0001	1.19 (1.13 to 1.25); <.0001
Undergrad or associate’s (ref graduate and professional)	1.22 (1.16 to 1.29); <.0001	1.23 (1.16 to 1.29); <.0001	1.2 (1.14 to 1.27); <.0001	1.2 (1.14 to 1.27); <.0001	1.2 (1.14 to 1.27); <.0001	1.2 (1.14 to 1.27); <.0001
Other degrees type	0.99 (0.9 to 1.09); 0.7884	0.99 (0.9 to 1.09); 0.7798	0.98 (0.89 to 1.08); 0.6744	0.98 (0.89 to 1.08); 0.676	0.98 (0.89 to 1.08); 0.6647	0.98 (0.89 to 1.08); 0.6651
First generation student	1.19 (1.15 to 1.22); <.0001	1.19 (1.15 to 1.22); <.0001	1.18 (1.15 to 1.22); <.0001	1.18 (1.15 to 1.22); <.0001	1.18 (1.14 to 1.22); <.0001	1.18 (1.14 to 1.22); <.0001
6-year change	1.13 (d = 0.07)		1.29 (d = 0.14)		0.92 (d=−0.05)

Sampling weights were applied, and all institutional variables were adjusted across models. The table includes models 1A and 1B related to GAD7> = 10 for each substance.

Figure 4.

Weighted unadjusted prevalence of GAD7>=10 stratified by substance use.

In Model 1A, all substances were associated with a greater odds of surpassing clinical cutoffs for GAD (OR_HED= 1.23; OR_CAN=1.66; OR_CIG= 1.82). Model 1B revealed the associations between substance use and anxiety significantly increased over time for HED and cannabis but not smoking. Temporal effect sizes for HED were small (OR_1year=1.02; OR_6year=1.13) and cannabis were small-moderate (OR_1year=1.04; OR_6year 1.29). This translates to students endorsing HED having small differences in rates of GAD in 2013 (OR = 1.13 [95% 1.05 to 1.22]) but those engaging in HED in 2019 had 1.28 times higher odds (95% 1.21 to 1.35) of surpassing thresholds for GAD. Similarly, students who endorsed cannabis use in 2013 had small differences in rates of GAD (OR = 1.4 [95% CI 1.30 to 1.50]) compared to those who did not use, while in 2019 the odds were 1.8 times higher (95% CI 1.7 to 1.9) for those using cannabis. Thus, there is some evidence of strengthening in the association between HED (below a priori thresholds) and cannabis (surpassing a priori thresholds) and anxiety but stable associations between smoking and anxiety.

Gender and age group did not moderate the magnitude of the change in associations between substance use and anxiety over time. In 2-way interaction models, there were smaller associations between smoking and anxiety among TGNC students (OR = 0.71 [95% CI 0.58 to 0.88]) but no other gender differences emerged. Regarding age, there were smaller effects among emerging adults related to cannabis (OR_18–20 0.82 [95% CI 0.74 to 0.92]; OR_21–25 0.82 [95% CI 0.74 to 0.90]) and HED (OR_18–20 0.89 [95% CI 0.81 to 0.98]; OR_21–25 0.85 [95% CI 0.77 to 0.94]) with larger effect sizes for smoking (OR_18–20 1.14 [95% CI 1.01 to 1.29]).

Discussion

This study presents novel findings related to trends in the associations between substance use and mental health among US post-secondary students between 2009 and 2019. Following other documented trends, heavy drinking and smoking have decreased, while cannabis has increased among post-secondary students.^15,44 In our study, use of all substances was related to a greater odds of students endorsing depression, anxiety, and suicidal ideation. Smoking had the strongest association with mental health outcomes, and this association did not change over time for any of the mental health outcomes examined. Heavy episodic drinking had the smallest association with mental health outcomes, with negligible increases between drinking and depression and small increases between drinking and suicidal ideation and anxiety over time (i.e., slight strengthening). Cannabis had the largest consistent increases in associations with mental health outcomes over time (i.e., strengthening), where by 2019 the magnitude of the associations between cannabis and mental health problems were similar to associations with smoking. Lastly, we found that gender and developmental age did appear to play a role in the relationship between substance use and mental health problems, though they did not impact how trends changed over time (i.e., no notable strengthening or decoupling within subgroups).

In our study, the strongest temporal trends were found between past month cannabis use and depression, suicidality, and anxiety. To our knowledge, ours is the first study to directly compare temporal effects across substances within the same study, but observed strengthening of associations with cannabis are consistent with results from nationally representative population studies of adolescents and adults in the US and Canada.^21,24–27 Though still clinically meaningful and increasing over time, associations between cannabis and anxiety were smaller than the associations with depression and suicidality in our sample of post-secondary students. Relatedly, a recent meta-analysis found cannabis use during adolescence predicted development of depression and suicidality in young adulthood but was not significantly related to developing anxiety.⁴⁵ Thus, though there appear to be differential cross-sectional and temporal associations with cannabis and mental health depending on the type of symptomatology people are experiencing, co-occurrence has been increasing among US post-secondary students.

There are two core hypotheses for observed increases in associations between cannabis use and mental health problems over time. First, given dramatic increases in the potency of cannabis^46,47, higher levels of tetrahydrocannabinol (THC) may be contributing to poorer mental health outcomes especially when paired with low cannabidiol (CBD).⁴⁸ On the other hand, people may be using cannabis to cope with mental health problems more-so today than in the past. Despite limited scientific evidence of benefits for cannabinoid-based psychotherapeutics⁴⁹, there have been increases in public perceptions of health benefits related to cannabis use⁵⁰ and declines in the perceived risks of using cannabis, particularly among those with depression.²⁵ Further, female students had stronger associations between cannabis and depression and suicidality than males, in line with prior studies; among several hypotheses, this may be partially driven by sex-differences in the effects of THC or gender differences in reasons and contexts for use.^25,29,30 Overall, these findings speak to the increasing need for post-secondary navigators and health providers to provide cannabis psychoeducation, such as by use of cannabis lower risk guidelines⁴⁸, including debunking myths and encouraging harm reduction strategies especially among female students and those with mental health concerns.

Our study found consistent co-occurrences of HED and depression, anxiety, and suicidality among post-secondary students with differential changes over time depending on type of mental health concern being explored. Of note, previous cross-sectional studies among post-secondary students are inconsistent regarding magnitude and directions of associations between HED and mental health concerns. For example, one study found negative associations between frequent HED and depression and positive associations with anxiety⁷, a different study found HED not to be associated with depression or anxiety⁵¹, and another study found students with depression reported more frequent HED.⁸ For depression, our study revealed slight, but negligible increases in the association with HED over time. This is different than findings among adolescents which found decreases in co-occurrence.²³ Further, within our cohort, there was pronounced strengthening between HED with SI and anxiety over time. Overall, differences in cross-sectional study findings in previous studies may be due to the years of data collection whereby more recent years may find more pronounced co-occurrences of HED and mental health concerns than earlier studies. These trends may suggest that declines in alcohol use at the population level may be driven more-so by students without mental health comorbidities, resulting in a greater proportion of students with mental health problems remaining engaged in heavy use. Focusing on mitigating substance-related harms rather than focusing on abstinence (i.e., harm reduction, use of protective behavioral strategies) may be particularly important for post-secondary students with poorer mental health.⁵² Discussing low risk drinking guidelines with students, highlighting potential implications related to mental health, is thus increasingly important.^53,54

With regards to smoking, there were strong associations between smoking and depression, suicidal ideation, and anxiety – particularly among emerging adults - but these associations did not change over time. Similarly, a prior US study found no changes over time in associations between smoking (before age 25) and depression.²² Overall, the association between smoking and internalizing mental health problems was strong and consistent over time indicating that although rates of smoking are decreasing among post-secondary students at a population level, the risk of comorbidity and the need for concurrent interventions is not changing. Given the myriad of negative physical and mental health effects associated with smoking,^55–57 post-secondary health and mental health providers should be equipped to provide smoking cessation counseling and Nicotine Replacement Therapy when needed.

The 2019 National Survey of Drug Use and Health found the prevalence of co-occurring SUDs and psychiatric disorders was stable among adolescents between 2015 and 2019 but increased among emerging adults (18–25) and older adults.¹¹ Our findings confirm these trends among post-secondary students, in particular, related to common substance use and depression, anxiety, and suicidal ideation. Importantly, although around 43% of emerging adults in the general US population with co-occurring problems received treatment in 2019, 37% received mental health treatment only.¹¹ Post-secondary students in particular are more likely to seek help informally from peers, with stigma remaining a strong barrier for many with mental health concerns.^58,59 Thus, substance use screening and assessment among post-secondary students that do access student mental health services is critical, and this assessment in and of itself may be a helpful intervention.⁶⁰ Brief interventions - defined as 1 to 2 sessions - have shown promise in reducing cannabis use and related problems among emerging adults⁶¹ and alcohol use among post-secondary students and adults.^62,63 To enhance existing brief interventions for post-secondary students, it is important to consider incorporating discussions of co-occurring mental health problems^61,64 and discussions related to alternative substance-free activities.^65,66 Further, technology-based interventions for co-occurring problems have shown promise and may be cost-effective and low-barrier ways to engage post-secondary students in concurrent treatment.^67,68

Limitations

The cross-sectional design inhibits our ability to infer causality, and thus these results should be interpreted as associations or co-occurrences of substance use and mental health symptomatology without respect for directionality. However, we utilized large national samples with annual data collection spanning 11 years allowing for an appropriately detailed exploration of temporality with respect to changes in co-occurrence. Cannabis, cigarette smoking, and suicidal ideation were captured by single item questions, which may impact the quality of reporting. However, single item questions have previously demonstrated acceptable properties in studies with respect to substance use⁶⁹ and suicide-related outcomes.⁷⁰ Further, we only looked at past month or past 2-week substance use and associations may be different for higher or lower frequencies of use. Regarding cannabis specifically, reporting may have been influenced by: (1) state specific legal status of medicinal and/or recreational cannabis use, (2) a lack of differentiation between medicinal and recreational cannabis use for all surveys with the exception of 2012–2014. We also only examined internalizing disorders and temporal changes between substance use and other psychiatric concerns may be different. For example, a previous study did not find strengthened connections between smoking and depression but did find increasing associations with SUDs, attention-deficit hyperactivity disorder, bipolar disorder, and antisocial personality disorder.²² Further, although we adjusted for commonly cited student and school level confounding variables, there may be other confounding variables that were not accounted for in this analysis (e.g., age of first use, physical health comorbidities, family history). Lastly, missing data and nonrandom sampling at the national level may result in different results than representative data sets. However, many of our results are consistent with existing studies in national representative samples of adults.

Conclusion

Among national samples of US post-secondary students from 2009 to 2019, heavy drinking, cannabis, and smoking have consistently been related to co-occurring depression, anxiety, and suicidal ideation. Over time, associations between mental health concerns and cannabis have been strengthening, and this is also true for mental health concerns and heavy drinking, but to a lesser extent. This suggests a greater prevalence of comorbidity today than in the past. Despite notable differences between population level increases in mental health concerns and substance use trends, substance use remains a concern among students with mental health challenges. Given co-occurrence of substance use and mental health concerns is common and increasing among post-secondary students, student mental health systems should prioritize early identification, psychoeducation, harm-reduction, and brief interventions.

Data access

Researchers can apply for access to the data used in this study on the following website: https://healthymindsnetwork.org/research/data-for-researchers/