A Longitudinal Cohort Study of Youth Mental Health and Substance use Before and During the COVID-19 Pandemic in Ontario, Canada: An Exploratory Analysis.

BACKGROUND: Youth mental health appears to have been negatively impacted by the COVID-19 pandemic. The impact on substance use is less clear, as is the impact on subgroups of youth, including those with pre-existing mental health or substance use challenges.
OBJECTIVE: This hypothesis-generating study examines the longitudinal evolution of youth mental health and substance use from before the COVID-19 pandemic to over one year into the pandemic among youth with pre-existing mental health or substance use challenges.
METHOD: A total of 168 youth aged 14-24 participated. Participants provided sociodemographic data, as well as internalizing disorder, externalizing disorder, and substance use data prior to the pandemic's onset, then every two months between April 2020-2021. Linear mixed models and Generalized Estimating Equations were used to analyze the effect of time on mental health and substance use. Exploratory analyses were conducted to examine interactions with sociodemographic and clinical characteristics.
RESULTS: There was no change in internalizing or externalizing disorder scores from prior to the pandemic to any point throughout the first year of the pandemic. Substance use scores during the pandemic declined compared to pre-pandemic scores. Exploratory analyses suggest that students appear to have experienced more mental health repercussions than non-students; other sociodemographic and clinical characteristics did not appear to be associated with mental health or substance use trajectories.
CONCLUSIONS: While mental health remained stable and substance use declined from before the COVID-19 pandemic to during the pandemic among youth with pre-existing mental health challenges, some youth experienced greater challenges than others. Longitudinal monitoring among various population subgroups is crucial to identifying higher risk populations. This information is needed to provide empirical evidence to inform future research directions.

Introduction

Since the COVID-19 pandemic began in early 2020, important community-based public health measures to mitigate transmission have led to widespread unintended consequences in other domains of health.[1] This includes direct and indirect losses of key social, educational, and physical opportunities especially important during the formative years of development.[2-5] As such, there is growing concern about the impact of the COVID-19 pandemic on youth mental health.[6,7]

In Ontario, Canada, measures to decrease interpersonal contacts were put in place in March 2020, including stay-at-home orders, recurring school closures, and limitations on recreational activities and social interactions; many of these measures remained in place in certain Ontario jurisdictions for over a year.[8] Moreover, there was a slower vaccine rollout for younger youth, with Health Canada’s approval of a vaccine for children 12–15 not occurring until in May 2021—five months after the approval for the general public—and vaccines for children below 12 approved only in fall 2021.[9]

Globally, the negative mental health impacts of the COVID-19 pandemic and related public health measures on youth appear to be well established.[10] The impacts include worsened anxiety symptoms,[2] depressive symptoms,[11] and suicidal ideation.[12] Vulnerable youth are particularly affected, including LGBTQ + youth and others already impacted by intersecting social determinants of health.[10,13] A cross-sectional study in Canada found that youth with a pre-existing psychiatric diagnosis experienced higher rates of mental health deterioration than youth without a pre-existing diagnosis during the first wave of COVID-19.[3] Another Canadian study showed stronger declines in youth from the general population compared to those with pre-existing mental health challenges.[14] Despite a sizeable body of research on youth mental health during the pandemic, data directly linking youth mental health prior to the pandemic and during the pandemic remains scant. One such study suggested a very different trajectory, i.e., a decrease in anxiety and no change in depression during the pandemic compared to before the pandemic.[15] Given the importance of understanding youth mental health during this time and the conflicting pattern of findings, there has been a call for more prospective longitudinal research.[16]

The impact of the pandemic on youth substance use has been unclear, but should also be considered. While some youth reported using substances to cope with the pandemic,[17] other research has shown varied results, including increases, decreases, or stable rates of use.[18,19] One study found no major effects on the overall frequency of substance use, but mixed effects for different substances, as well as changes in reasons and contexts of use.[20]

Aside from the direct mental health impacts of public health measures, indirect impacts include delayed diagnosis and treatment, increased exposure to adverse childhood experiences, and decreased opportunities for social development.[21-23] These impacts have been detected across health systems worldwide; for instance, a United States analysis comparing 2019 rates of mental health-related emergency department visits to 2020 rates showed a 24% and 31% increase for children 5–11 and youth 12–17, respectively.[24] A similar trend of increasing youth mental health-related emergency department visits has been observed in Canada.[25]

As adolescence and young adulthood constitute a period of substantial risk for the emergence of mental health problems, further observational studies are needed to monitor the impacts of the pandemic on youth mental health, including longitudinal studies that account for evolving public health meaures.[26] Notably, there is a need to understand pre-pandemic factors that may place youth at risk for worsened mental health problems during a pandemic, such as clinical and sociodemographic characteristics.

In this hypothesis-generating study, our primary aim was to explore the longitudinal evolution of youth mental health and substance use from before the COVID-19 pandemic through to over one year into the pandemic in a clinical sample of youth who had mental health and/or substance use challenges prior to the pandemic. As an exploratory aim, we explore sociodemographic factors that may have influenced mental health and substance use status over the course of the pandemic. Understanding changes to youth mental health and substance use at discrete time points during the evolving pandemic can inform future research directions.

Methods

Design and Participants

The study cohort included a subgroup of participants who took part in a parent study conducted at the Centre for Addiction and Mental Health (CAMH) in Toronto, Canada. In the parent study, participants were invited from three pre-existing clinical cohorts and one non-clinical cohort at CAMH to participate in a rapid-response COVID-19 study launched in April 2020.[14] The present study included the following pre-existing clinical cohorts. The first was the Early Identification of Psychosis Spectrum Symptoms (EIPSS) study, a longitudinal cohort study of youth aged 14–24 that aimed to identify early evidence of psychosis spectrum symptoms. The second was the YouthCan IMPACT project, a pragmatic randomized-controlled trial of an integrated youth service hub model of service delivery; participants included youth aged 14–18 at intake.[27] The third cohort included participants in a cross-sectional study of youth aged 14–24 entering CAMH services in Youth Addictions and Concurrent Disorder Service (YACDS), an out-patient treatment program for youth with substance use challenges, with or without concurrent mental health problems.[28] The three source cohorts were combined into one sample, as they were all youth aged 24 or under seeking services in a single tertiary care hospital during 2018–2021.

Participants from the three source cohorts in the parent study were included in the current study if they met two criteria: 1) They participated in the longitudinal COVID-19 study, providing at least one time point of data on the Global Appraisal of Individual Needs Short Screener (GAIN-SS; see below for details), and 2) they had a pre-COVID-19 GAIN-SS score available in the study from which they originated, which they completed between April 2018 and October 2019. The clinical cohort was compised of N = 276 participants; 108 participants without pre-COVID GAIN-SS scores were excluded (Figure 1). The analytic sample consisted of N = 168 (participant rate = 61%) youth aged 14 to 24 at their initial assessment. Of the 168 participants, 56, 39, and 73 were from YouthCan IMPACT, EIPSS, and YACDS, respectively. Participants provided informed consent for both the source study and the COVID-19 study. All studies were approved by the CAMH Research Ethics Board.

Figure 1.

Sample flow diagram to display study selection for the pre-pandemic sample. Diagram includes source cohorts, screening process, and the final study sample. YCI represents YouthCan IMPACT cohort, EIPSS represents the Early Identification of Psychosis Spectrum Symptoms cohort, and YACDS represents the Youth Addictions and Concurrent Disorder Service cohort.

Data Collection Waves

This study included eight time points. The first time point (T1)—pre-COVID-19—was collected between April 2018 and October 2019 as part of the source study. The remaining time points were collected during the pandemic; web-based data collection occurred in seven COVID-19 time points, every two months beginning in April 2020. COVID-19 data was then linked to pre-pandemic data.

The COVID-19 data collection waves occurred during notable events in Ontario’s COVID-19 experience. T2 (April 2020) occurred at the peak of Ontario’s first wave of COVID-19. T3 (June 2020) took place during an optimistic period, when some regions had reduced restrictions.[29] During T4 (August 2020), Ontario reached a low for daily new cases and further reduced restrictions.[30] During T5 (October 2020), Ontario broke the previous record for daily COVID-19 infections and the second wave was announced. T6 (December 2020) occurred when the first vaccine was approved and administered in Canada, but also when Ontario was reporting another all-time daily high for case counts.[29,30] At T7 (February 2021), Ontario entered a third lockdown amidst rising cases during the third wave. T8 (April 2021) corresponded to the opening of Ontario’s vaccination program beyond high-risk populations to mass vaccine delivery.[31]

Measures

Sociodemographic. A demographic information form was used to collect time-variant sociodemographic variables, including age (14 to 28), government financial support, student status, housing (stable vs. precarious), and NEET status (not in employment, education, or training)[32]; these were collected at each data collection wave. Time-invariant sociodemographic variables were also collected at intake into the source study: gender (trichotomized: women, men, transgender/non-binary), ethnicity (dichotomized: racialized/non-racialized), and country of birth (dichotomized: Canada, other).

Global Appraisal of Individual Needs Short Screener (GAIN-SS). The GAIN-SS is a screener that supports clinical decision-making in four dimensions: a) internalizing disorders; b) externalizing disorders; c) substance use disorders; and d) crime/violence.[33] The GAIN-SS (version 3) includes 29-items, with sub-screeners of 5 to 7 items for each dimension, and several single-item screeners.[33,34] When used among service-seeking youth (aged 17 to 24 years), the GAIN-SS has shown good internal consistency, with an alpha of 0.91.[35] For each GAIN-SS item, respondents rate how recently they have experienced significant difficulties with the described, ranging from never to past month.[36] While past-year cutoffs are available,[35] we used continuous scores for past month symptoms as multiple time points were collected within a year. Scores ranged from 0 to 5 (substance use, crime/violence), 0 to 6 (internalizing), and 0 to 7 (externalizing).[33] GAIN-SS Crime/Violence was excluded from the analyses due to low endorsement rates; the supplementary single-item screening questions were not examined. Diagnostic plots were produced to assess normality. As the GAIN-SS substance use screener scores were positively skewed, scores in this screener were then dichotomized; scores = 0 were classified as having a low likelihood of clinical diagnosis, while scores ≥ 1 were classified as having a moderate to high likelihood of clinical diagnosis.[33] Moderate to high scores suggest the need for treatment.[33]

Pre-existing mental health diagnoses. Mental health diagnoses were available for the YouthCan IMPACT and EIPSS study participants, derived at T1 by research staff using the Diagnostic Interview for Affective Symptoms for Children and the Structured Clinical Interview for DSM.[37,38] Due small cell sizes, only major depressive disorder and anxiety disorders (dichotomous) were included as predictors in the models to avoid overestimating associations.

Statistical Analysis

Descriptive statistics were reported for demographics at T1. All analyses were performed in SPSS Version 25 (IBM Corp., NY, USA). Two-sided p-values < 0.05 were considered statistically significant; no adjustment for multiple testing was performed given the hypothesis-generating nature of this study.

Primary analyses. GAIN-SS internalizing and externalizing scores were analyzed as continuous measures, as in prior studies.[39,40] Linear mixed-effects models were used to analyze the main effect of time on GAIN-SS scores (internalizing and externalizing) across 8 time points, adjusting for time-varying age and source cohort.[41] To account for repeated measures within individuals, participants were treated as random subject effects. Type III tests of fixed effects were used to analyze the main effects for the dependent variables (GAIN-SS scores) across time.[42] Estimated marginal means and their corresponding 95% confidence intervals (CI) were reported. Estimates were calculated with Restricted Maximum Likelihood, which is robust against missing responses.

GAIN-SS substance use scores were dichotomized and analyzed as a binary outcome due to its skewed distribution. Generalized Estimated Equation (GEE) models were used to analyze the main effect of time on GAIN-SS substance use scores, adjusting for age and source cohort.[43] The results of the models are presented using estimated marginal means and 95% CI.

Exploratory analyses. Employing the same overall approach, linear mixed effects models and GEE models were used to analyze interactions between sociodemographic/clinical characteristics and time on the GAIN-SS sub-screeners.[41] These hypothesis-generating models included the main effect of time, sociodemographic/clinical characteristics, and time-by-sociodemographic/clinical characteristics interaction as fixed effects, also adjusting for age at T1 and source cohort. Pairwise contrasts between each pandemic timepoint versus baseline were generated. Based on significant effects, time trends of GAIN-SS scores by student status (time variant) and country of birth (time invariant) are reported as exploratory post-hoc models. Additional exploratory models are reported in the Supplementary Materials.

Results

Demographics

Of the 168 participants in the study, data were available for 168 (100%) at T1, 160 (95%) at T2, 121 (72%) at T3, 103 (61%) at T4, 101 (60%) at T5, 94 (56%) at T6, 92 (55%) at T7, and 109 (65%) at T8. The mean age at T1 was 17.9 years (range = 14–24). Participants identified as a woman/girl (64.3%), man/boy (30.4%), or transgender or non-binary (5.4%). Slightly more than half of participants identified as Caucasian (54.2%). Additional baseline participant characteristics are shown in Table 1.

Table 1.

Demographic and Clinical Characteristics at T1 (n = 168).

		Frequency (%)
Gender	Man/boy	51 (30.4%)
	Woman/girl	108 (64.3%)
	Transgender or non-binary	9 (5.4%)
Ethnic origin	Caucasian	91 (54.2%)
	Asian (East)	10 (6.0%)
	Asian (South and South East)	12 (7.2%)
	Black (African and Caribbean)	15 (8.9%)
	Mixed heritage	22 (13.1%)
	Latin American	5 (3.0%)
	Indigenous	3 (1.8%)
	Another background	10 (6.0%)
Born in Canada		143 (85.1%)
First language English		159 (95.2%)
Student		118 (70.2%)
Not in employment, education, or training		22 (13.3%)
Living situation	Own apartment/home	12 (7.2%)
	With parents or other family members	132 (79.0%)
	With friends/peers	10 (6.0%)
	Precarious housing	13 (7.8%)
Legal system involvement		38 (22.6%)
Receiving government support		29 (17.4%)
GAIN Short Screener (past month diagnosis)a	Internalizing disorder	108 (64.3%)
	Externalizing disorder	59 (35.1%)
	Substance use disorder	45 (26.8%)
	Crime/violence	2 (1.2%)
	Total screener	147 (87.5%)

aGAIN Short Screener scores are used here to determine the proportion of participants who met the criteria for a high likelihood of a diagnosis in the past month.

Primary Analyses: Mental Health and Substance use Over Time

Overall, the internalizing (p = 0.97) and externalizing (p = 0.21) disorder scores of participants did not significantly differ across the 8 timepoints (1 pre-pandemic and seven pandemic assessment times) (Table 2). However, there were significant differences in substance use across the 8 timepoints (p = 0.02). Using the pre-pandemic time point as the reference group (T1), pairwise contrasts showed significant decreases in substance use at T3 (p = 0.034), T5 (p = 0.034), T6 (p = 0.047), T7 (p = 0.003) and T8 (p = 0.013).

Table 2.

Mental Health and Substance use in Youth Across 8 Timepoints. Estimated Marginal Means (and 95% Confidence Intervals) and Type III Test p-Values of the Main Effects of Time are Reported.

Outcomes	T1 (n = 168)	T2 (n = 160)	T3 (n = 121)	T4 (n = 103)	T5 (n = 101)	T6 (n = 94)	T7 (n = 92)	T8 (n = 109)
Estimated Marginal Means (95% CI) from Linear Mixed Models									P-value
Internalizing disorders	3.28	3.25	3.24	3.29	3.09	3.18	3.18	3.15	0.97
	(2.95–3.61)	(2.96–3.53)	(2.93–3.55)	(2.96–3.62)	(2.76–3.42)	(2.84–3.52)	(2.83–3.53)	(2.82–3.48)
Externalizing disorders	1.83	2.06	2.03	1.85	2.03	2.16	2.18	2.24	0.21
	(1.57–2.09)	(1.82–2.29)	(1.77–2.29)	(1.57–2.12)	(1.75–2.31)	(1.87–2.44)	(1.89–2.48)	(1.96–2.52)
Estimated Proportion of Participants with Moderate to High Likelihood of Clinical Diagnosis for Substance Use (95% CI) from Generalized Estimating Equations (GEE) models									P-value
Substance use disorders	0.48	0.42	0.37	0.50	0.34	0.35	0.27	0.31	0.02
	(0.40–0.57)	(0.33–0.51)	(0.28–0.46)a	(0.39–0.61)	(0.25–0.45)a	(0.25–0.46)a	(0.19–0.38)a	(0.22–0.41)a

Legend: Symbol a represents statistical significance for the pairwise contrast between each pandemic timepoint vs. T1.

Notes: Each model used T1 as the reference group and adjusted for age and source cohort. Data was collected from April 2018 to October 2019 (T1), April 2020 (T2), June 2020 (T3), August 2020 (T4), October 2020 (T5), December 2020 (T6), February 2021 (T7), and April 2021 (T8).

Exploratory Analyses: Interactions Between Sociodemographic Characteristics and Time on Mental Health and Substance use

The linear mixed models and GEE models by student status and born-in-Canada status, across the internalizing, externalizing and substance use sub-screeners, are shown in Table 3. Graphs of the estimated marginal means with T1 as the reference group, are shown in Figure 2. For the remaining exploratory models, see the supplementary materials.

Table 3.

Interactions Between Sociodemographic Characteristics and Time on Mental Health and Substance use. Estimated Marginal Means (95% Confidence Intervals) and Estimated Proportion of Participants with Moderate to High Likelihood of Clinical Diagnosis (95% Confidence Intervals) are Reported.

Outcomes	T1 (n = 168)	T2 (n = 160)	T3 (n = 121)	T4 (n = 103)	T5 (n = 101)	T6 (n = 94)	T7 (n = 92)	T8 (n = 109)
	Estimated Marginal Means (95% CI) from Linear Mixed Models								P value
Internalizing disorders
Time-by-student status									0.01
Student	3.12	3.29	3.21	3.20	3.14	3.36	3.38	3.34
	(2.75–3.48)	(2.97–3.62)a	(2.85–3.56)	(2.81–3.59)	(2.77–3.51)a	(2.98–3.74)a	(2.98–3.77)a	(2.96–3.72)a
Non-student	3.65	3.15	3.31	3.47	2.93	2.77	2.74	2.76
	(3.19–4.11)	(2.72–3.59)a	(2.82–3.8)	(2.98–3.96)	(2.38–3.48)a	(2.21–3.32)a	(2.19–3.29)a	(2.27–3.25)a
Time-by-birthplace									0.03
Born in Canada	3.29	3.16	3.13	3.27	3.01	3.16	3.12	3.29
	(2.95–3.63)	(2.85–3.47)	(2.8–3.47)	(2.92–3.62)	(2.66–3.37)	(2.79–3.53)	(2.73–3.5)	(2.93–3.65)
Born Elsewhere	3.31	3.76	3.88	3.42	3.50	3.33	3.50	2.51
	(2.56–4.06)	(3.02–4.5)	(3.07–4.7)	(2.51–4.33)	(2.65–4.35)	(2.48–4.18)	(2.67–4.33)	(1.72–3.3)
Time-by-NEETb									0.23
NEET	3.85	3.13	3.4	3.55	3.02	3.06	2.97	2.57
	(3.21–4.49)	(2.63–3.63)	(2.79–4)	(2.9–4.19)	(2.22–3.81)	(2.36–3.76)	(2.29–3.64)	(1.86–3.29)
In EETc	3.18	3.28	3.20	3.23	3.10	3.21	3.22	3.22
	(2.84–3.52)	(2.97–3.59)	(2.87–3.53)	(2.88–3.58)	(2.76–3.45)	(2.84–3.57)	(2.85–3.6)	(2.88–3.57)
Externalizing disorders
Time-by-student status									0.01
Student	1.76	2.15	2.01	1.67	2.19	2.34	2.31	2.30
	(1.46–2.07)	(1.88–2.42)	(1.71–2.32)	(1.34–2.01)	(1.87–2.51)a	(2.01–2.68)a	(1.97–2.65)	(1.98–2.63)
Non-student	2.00	1.85	2.07	2.17	1.61	1.70	1.89	2.06
	(1.60–2.39)	(1.47–2.23)	(1.64–2.50)	(1.74–2.61)	(1.12–2.1)a	(1.21–2.2)a	(1.4–2.38)	(1.62–2.49)
Time-by-birthplace									0.42
Born in Canada	1.74	1.96	1.97	1.80	2.00	2.07	2.04	2.26
	(1.46–2.01)	(1.71–2.21)	(1.69–2.25)	(1.51–2.1)	(1.69–2.3)	(1.76–2.39)	(1.71–2.36)	(1.96–2.56)
Born elsewhere	2.45	2.58	2.36	2.04	2.2	2.59	2.86	2.16
	(1.84–3.06)	(1.98–3.18)	(1.68–3.04)	(1.26–2.82)	(1.49–2.92)	(1.87–3.3)	(2.17–3.56)	(1.51–2.81)
Time-by-NEET									0.04
NEET	2.06	1.66	1.70	2.29	1.21	1.75	1.91	2.08
	(1.49–2.63)	(1.22–2.11)	(1.16–2.25)	(1.71–2.87)	(0.48–1.93)a	(1.12–2.38)	(1.3–2.52)	(1.43–2.73)
In EET	1.82	2.16	2.12	1.75	2.14	2.25	2.25	2.25
	(1.54–2.09)	(1.91–2.42)	(1.84–2.4)	(1.44–2.05)	(1.85–2.44)a	(1.94–2.55)	(1.93–2.57)	(1.96–2.54)
Estimated Proportion of Participants with Moderate to High Likelihood of Clinical Diagnosis for Substance Use (95% CI) from Generalized Estimating Equations (GEE) models									P-value
Substance use
Time-by-student status									0.01
Student	0.41	0.42	0.36	0.50	0.34	0.39	0.36	0.38
	(0.31–0.52)	(0.32–0.53)a	(0.26–0.47)	(0.38–0.62)	(0.24–0.46)	(0.28–0.51)a	(0.25–0.49)a	(0.28–0.50)a
Non-student	0.69	0.44	0.41	0.53	0.39	0.30	0.16	0.21
	(0.51–0.82)	(0.31–0.58)a	(0.27–0.56)	(0.34–0.72)	(0.22–0.60)	(0.16–0.49)a	(0.08–0.29)a	(0.12–0.34)a
Time-by-birthplace									0.33
Born in Canada	0.47	0.40	0.36	0.48	0.37	0.37	0.27	0.33
	(0.37–0.56)	(0.30–0.49)	(0.27–0.46)	(0.37–0.60)	(0.27–0.49)	(0.26–0.50)	(0.18–0.39)	(0.23–0.45)
Born elsewhere	0.59	0.55	0.39	0.59	0.18	0.24	0.28	0.22
	(0.41–0.75)	(0.34–0.74)	(0.18–0.65)	(0.31–0.82)	(0.05–0.48)	(0.08–0.53)	(0.09–0.61)	(0.07–0.50)
Time-by-NEET									0.46
NEET	0.68	0.42	0.34	0.45	0.49	0.23	0.16	0.29
	(0.37–0.89)	(0.26–0.6)	(0.18–0.56)	(0.23–0.7)	(0.23–0.75)	(0.08–0.52)	(0.06–0.35)	(0.13–0.54)
In EET	0.45	0.42	0.37	0.51	0.34	0.38	0.31	0.32
	(0.36–0.55)	(0.33–0.52)	(0.27–0.47)	(0.40–0.63)	(0.25–0.44)	(0.27–0.49)	(0.21–0.43)	(0.23–0.43)

Legend: Symbol a represents statistical significance for the pairwise contrasts between pandemic timepoints versus T1, symbol b represents Not in Education, Employment, or Training, and symbol c represents Education, Employment, or Training.

Notes: Each model used T1 as the reference group and adjusted for age and source cohort. P-values are reported for the time interaction. Data was collected from April 2018 to October 2019 (T1), April 2020 (T2), June 2020 (T3), August 2020 (T4), October 2020 (T5), December 2020 (T6), February 2021 (T7), and April 2021 (T8). The following formulas were used: (Non-Student EMM TX - Student EMM TX) - (Non-Student EMM T1 - Student EMM T1) for student status, (Born Elsewhere EMM TX - Born in Canada EMM TX) - (Born elsewhere EMM T1 - Born in Canada EMM T1) for birthplace, and (In EET EMM TX – Not in EET EMM TX) - (In EET EMM T1 – Not in EET EMM T1).

Figure 2.

Line plot of internalizing, externalizing, and substance use scores for sociodemographic characteristics and time. Sociodemographic characteristics include student status, birthplace, and employment, education, or training status. Each model was adjusted for age and source cohort. Estimated Marginal Means are reported for internalizing and externalizing scores, and estimated proportion of participants with moderate to high likelihood of clinical diagnosis are reported for substance use scores. Error bars indicate the standard error.

Internalizing Disorders

The change in internalizing disorders over time differed significantly between students and non-students (time-by-student status interaction p = 0.011). Compared to T1 scores, non-students experienced lower internalizing scores in each pandemic time point, except for T3 and T4, while students’ scores remained relatively consistent (Figure 2a). There was also a significant interaction for time by birthplace (p = 0.028). Participants born in Canada appeared to have marginally lower internalizing scores compared to participants born elsewhere based on the significant interaction, despite non-significant pairwise comparisons. The interaction between time and gender, pre-pandemic diagnoses, ethnicity, NEET status, and government support were not significant.

Externalizing Disorders

The change in externalizing disorders over time also differed significantly between students and non-students (time-by-student status interaction p = 0.014), with significant effects at T5 and T6 (Figure 2b). During October (T5) and December (T6) 2020, students reported significantly higher externalizing symptoms than non-students. The marginal means for students were lower than for non-students at T1, T3 (June 2020), and T4 (August 2020). There was also a marginally significant interaction for time by NEET status (time-by-NEET interaction p = 0.043). Participants who were NEET appeared to have lower externalizing scores compared to participants who were engaged in education, employment, or training at T5 (October 2020). Time interactions for the remaining characteristics were not significant.

Substance Use Disorders

The change in substance use over time showed significant differences between students and non-students (time-by-student status interaction p = 0.01). Larger decreases in the substance use disorder scores were observed among non-students compared to students in each pandemic time point (Figure 2c). Time interactions for the other sociodemographic and clinical characteristics were not significant.

Discussion

This study longitudinally examined the mental health and substance use of youth with pre-existing mental health and substance use challenges in Ontario, beginning before the COVID-19 pandemic. The findings indicate that internalizing and externalizing disorder scores did not change during the pandemic, but substance use disorder scores declined. Exploratory analyses showed potential differences in trajectories based on sociodemographic characteristics. Notably, internalizing disorders, externalizing disorders, and substance use varied over time differently for students than for non-students.

Research on mental health during the pandemic has generally pointed to negative impacts. A recent meta-analysis[44] found that the global prevalence of depressive and anxiety symptoms among children and youth in the general population had doubled since the onset of the pandemic. However, there is a paucity of longitudinal research utilizing pre-pandemic data in clinical samples of youth. Those that are available have reported mixed findings.[45] For instance, one study found a decrease in anxiety and no change in depression in a school-based sample of youth in South West England compared to a pre-pandemic time point.[15] The current study, which focused not on the general population of youth, but on a sample of youth with pre-existing mental health or substance use challenges, found that mental health difficulties did not worsen over the first year of the pandemic. These findings, along with findings from the present study, could suggest the influence of local context on a region’s mental health outcomes through the pandemic, and to the importance of examining impacts prospectively, using pre-pandemic data.

Alternatively, our findings may point to differential impacts of the sample based on pre-existing mental health or substance use challenges.[17] Some youth reported positive impacts of the pandemic, related to an increase in free time that they used for positive activities and family interaction.[17] It may be that for youth with pre-existing mental health challenges, these positive impacts offset the negative effects of public health restrictions to some degree, obviating observable change. It could also be that a ceiling effect limited the ability to detect meaningful change, or that the measure was not sensitive to the type of change participants experienced.

Reflecting previous research, this study showed a decline in substance use in this sample during the pandemic. A longitudinal population-based study among Icelandic youth aged 13 to 18 years reported a decrease in substance during the pandemic,[11] while a cross-sectional study among Canadian high school students found a decrease in the percent of youth that use substances, but an increase in frequency of use.[46] Social factors, such as popularity and peer pressure are predictors of substance use among youth.[47,48] These experiences and exposures may have been limited during the pandemic by reduced social contact. It is possible that public health measures meant to limit in-person contacts may have had a protective effect against substance-use disorders, known to be impacted by social influences in youth, possibly alongside changes in living situations. What remains unclear, however, is how these social factors evolved among youth over the pandemic and the role they played in substance use behaviors.

Pandemic-related disruptions, such as disruptions to in-person schooling, have been argued to negatively impact mental health.[5,49,50] In Ontario, there was a staggered reopening of schools in September 2020, with schools closing again in several public health regions in January 2021 and reopening again in February 2021.[5] School can serve as a protective factor for youth mental health via social connectedness and attendance[51]; by disrupting these protective factors, school closures may have had negative repercussions on mental health. Reduced exposure to positive factors at school (e.g., lack of routine), [52-54] in tandem with and increased exposure to negative factors at home (e.g., child abuse and neglect)[55] may have further exacerbated mental health problems. It remains important for schools to implement action plans that prioritize student mental health.[50,56] For those youth who were not in school, future research should seek to confirm the role of pandemic-related employment and social disruption on mental health and substance use.

Many studies have shown the disproportionate impact of COVID-19 on racialized, precariously employed, and other marginalized communities.[57-59] In the current study, gender, ethnicity, and government support were not significantly associated with worse internalizing, externalizing, and substance use scores during the pandemic. Local contextual factors may have reduced the impact for some marginalized groups. For example, the Canada Emergency Response Benefit (CERB) funding that was available across sociodemographic groups during the pandemic may have eased stress due to pandemic-related unemployment.[60] Other studies have highlighted the benefit of social safety nets such as CERB on youth mental health during the pandemic.[61] However, CERB is no longer available in Ontario; longer-term longitudinal studies should examine the impact of the disruption of CERB. Future research should also test whether youth born outside of Canada and whether youth who are NEET had differential mental health impacts from the pandemic, given the marginal findings observed in the current study.

Strengths and Limitations

This study has several key strengths. The longitudinal design, with multiple COVID-19 time points and pre-pandemic data is a strength. As pandemic-related disruptions varied among youth with different characteristics (e.g., students compared to non-students), this study had the strength of examining mental health impacts among subgroups of youth. However, there are also several limitations. Notably, the T1 data collection was not designed explicitly for this purpose as the pandemic was not predicted; T1 time points vary over the course of a year and participants were at different stages in research projects and treatment. Some may have been receiving clinical services at the onset of the pandemic, while others were not; this may have biased the findings and limited generalizability. Using diagnostic data is a strength; however, the diagnoses were conducted by research staff, may have introduced interviewer bias, and were only available for a portion of the sample. Moreover, dichotomized variables may have been subject to a ceiling effect for some youth, and limited power may have eliminated important nuances. Bias arising from attrition may be a concern due to attrition from the cohort over time, although 78% of participants responded to 4 or more of the 8 timepoints. Lastly, selection bias may also be a concern; the study sample might not extend beyond the context of urban Canadian youth seeking mental health services before the pandemic.

Conclusion

This study shows that the mental health of youth with pre-existing mental health challenges was stable over the course of the pandemic, but that substance use declined. However, the exploratory analyses suggests that this potentially not be the case for all: students and youth who were not born in Canada may have experienced greater challenges than their counterparts. A comprehensive understanding of the mental health impacts of COVID-19 among various population subgroups is critical in identifying higher risk populations and to inform future research directions.

Click here for additional data file.

Supplemental material, sj-pdf-1-cpa-10.1177_07067437221097906 for A Longitudinal Cohort Study of Youth Mental Health and Substance use Before and During the COVID-19 Pandemic in Ontario, Canada: An Exploratory Analysis by Natasha Y. Sheikhan, Lisa D. Hawke, Clement Ma, Darren Courtney, Peter Szatmari, Kristin Cleverley, Aristotle Voineskos, Amy Cheung and Joanna Henderson in The Canadian Journal of Psychiatry