Psychological mediators of the associations between pandemic-related stressors and suicidal ideation across three periods of the COVID-19 pandemic in Canada

Abstract

Background

The COVID-19 pandemic's mental health impact is well-established. While early evidence suggested suicide deaths remained stable or declined, suicidal ideation (SI) became more prevalent than before the pandemic. Our study: (1) examined the prevalence and distribution of SI among Canadian adults, (2) compared SI among those with and without pre-existing mental illnesses, and (3) evaluated associations between pandemic-related stressors (i.e., unemployment, insecure employment, loss of income, medical vulnerability, COVID-19 exposure) with SI, and whether such associations were mediated by depression, thwarted belongingness, perceived burdensomeness, or perceived discrimination.

Methods

The sample was comprised of data gathered at three timepoints (Wave 1 08/18/2020–10/01/2020, n = 6629; Wave 2 12/21/2020–03/31/2021, n = 5920; Wave 3 09/07/2021–12/07/2021, n = 7354). Quota-based responses from survey research panels which matched the geographic, age, and sex distribution of the Canadian population were supplemented with convenience-sampled responses.

Results

The prevalence of SI was 4.1 % (Wave 1), 5.3 % (Wave 2), and 5.8 % (Wave 3). Odds of SI were higher for respondents under the age of 35 years and with pre-existing mental illnesses. SI was associated with quarantining due to suspected or confirmed COVID-19 exposure, potential COVID-19 exposure at work, medical vulnerability toward COVID-19, insecure employment or unemployment, and income loss. These associations were mediated by psychological experiences, particularly depression and thwarted belongingness.

Limitations

This cross-sectional, observational study cannot establish temporality or causality.

Conclusion

Results highlight groups who may benefit from enhanced screening for depression and suicide risk. Reducing depression and increasing sense of belonging should be prioritized.

1. Introduction

The global healthcare landscape demonstrates that the COVID-19 pandemic negatively impacted the mental health of people with and without pre-existing mental illnesses (Pfefferbaum and North, 2020). Relative to pre-pandemic estimates, rates of anxiety, depression, alcohol use, loneliness, and sleep problems significantly increased during the pandemic (Czeisler, 2020; Kwong et al., 2021; O'Connor et al., 2021; Veldhuis et al., 2021). Many of these mental health problems are also correlated with suicidal thoughts and behaviors (Beautrais, 2000; Blakely et al., 2003; Pigeon et al., 2012; Stickley and Koyanagi, 2016), raising concern that rates of suicide may swell during and following the pandemic (Reger et al., 2020). Likewise, the pandemic brought about sudden changes to the economy and social landscape that resulted in increased financial insecurity and social isolation, both of which are robust correlates of suicidal thoughts and behaviors at the individual and population levels (Calati et al., 2019; Mathieu et al., 2022), furthering concerns that suicide rates may spike during the pandemic.

Despite these early fears, suicide rates reported globally during the pandemic varied widely. While an analysis of real-time suicide mortality data showed a decline in suicide deaths in the early months of the pandemic in 12 of 21 areas or countries (Pirkis et al., 2021), suicide deaths remained stable in Canada (McIntyre et al., 2021) but increased in Nepal (Acharya et al., 2022) and Japan (Sakamoto et al., 2021). Suicidal ideation (SI), defined as thoughts about wanting to be dead or taking action to end one's life (Posner et al., 2007), is an important predictor of suicide as approximately one third of people with suicidal thoughts act on these thoughts (Nock et al., 2018), sometimes with fatal consequences. Moreover, even when thoughts are not acted upon, SI elevates risk for various negative outcomes, including mortality from natural causes, poor physical health, and disability (Batterham et al., 2013; van Spijker et al., 2011). It is therefore important to know not only whether more people died by suicide than would be expected during this unprecedented time, but also whether more people are thinking about suicide, who is most likely to experience such thoughts, and what aspects of their pandemic experience drive these thoughts. In the U.S. and U.K., the prevalence of SI rose two- to three-fold relative to pre-pandemic surveys (Czeisler, 2020; O'Connor et al., 2021). Similarly in Canada, the rate of SI was estimated to be 6 % in the spring and early summer of 2020 (Daly et al., 2021; Jenkins et al., 2021) and 8 % in Winter of 2021 (Mental Health Research Canada, 2021), which is 2 to 3 times higher than the pre-pandemic rate of 3 % (Government of Canada, 2017). Epidemiologically, these findings are vital to better understand psychological distress and vulnerabilities to suicide. This could also emphasize the need for surveillance, targeted screening and prevention that will be needed to monitor and mitigate the corresponding risks.

A handful of studies have explored demographic and pandemic-related risk factors of SI during the pandemic. Characteristics that have been consistently associated with higher rates of SI during the pandemic include younger age, lower educational attainment, being required to quarantine or self-isolate, loss of employment or a significant portion of income due to the pandemic, and being a frontline or essential worker (Czeisler, 2020; Daly et al., 2021; Gratz et al., 2020; Liu et al., 2021). Other characteristics that have been inconsistently associated with SI during the pandemic include the female gender, family caregiving status, being non-White in majority White countries, and being unemployed, with some studies finding positive associations with a risk of SI, and others finding no association (Czeisler, 2020; Gadermann et al., 2021; Liu et al., 2021; O'Connor et al., 2021). To our knowledge, only one study has examined suicide in rural versus urban residents, documenting nearly identical rates of SI in American respondents living in these areas (Czeisler, 2020).

Very few studies have examined risk factors associated with SI during the pandemic in Canada, and those that have are limited to the earliest period of the pandemic, namely in the Summer and Autumn of 2020 (Daly et al., 2021; Jenkins et al., 2021; Liu et al., 2021). Identifying risk factors of SI in Canadian populations at later periods in the pandemic would be useful to further develop preventative measures and public health interventions to prepare for future pandemics. Additionally, the availability of minimal evidence associated with pre-existing mental health conditions within current published research is a significant limitation to assess true-impact (Czeisler, 2020; Daly et al., 2021; Jenkins et al., 2021; Liu et al., 2021). Hence, it is unclear if people with mental health conditions that preceded the pandemic were able to access mental health care services during the pandemic and, for those who self-managed their conditions, what the scope of their needs were. Finally, there is a need to examine psychological experiences that may account for the observed associations between demographic characteristics and SI, to illuminate why certain groups are vulnerable. So far, evidence has suggested that thwarted belongingness mediated the association between stay-at-home orders and suicide risk (Gratz et al., 2020), insomnia mediated the association between COVID-19 worries and SI (Killgore et al., 2020), nightmares fully mediated the association between trauma exposure and SI in healthcare workers (Que et al., 2022), and depression mediated the association between stressful events and SI in adolescents (Cheng et al., 2021). Further examination of the roles of depression, perceived burdensomeness, thwarted belongingness – three important risk factors for SI (Joiner, 2005; Klonsky et al., 2021) – as well as the potential role of perceived discrimination given apparent concentration of SI in racialized groups, is warranted.

This influenced us to conduct an exploratory observation study to identify and report the prevalence and distribution of suicidal thoughts in a Canadian population recruited at three time points of Autumn 2020 [08/18/2020–10/01/2020], Winter 2020/21, [12/21/2020–03/31/2021] and Autumn 2021 [09/07/2021–12/07/2021]. We examined several pandemic related stressors, including: (1) medical vulnerability toward COVID-19 infection, (2) potential exposure to COVID-19 at work, (3) requirement to self-isolate (i.e., quarantine) due to confirmed or suspected COVID-19 exposure, (4) unemployment, (5) insecure employment, and (6) loss of personal or household income. We categorised our sample to: Group 1; mental illness naïve and without current SI; Group 2; mental illness naïve but with current SI; Group 3; with a mental illness without current SI; and Group 4; with a mental illness with a current SI. We hypothesized respondents who endorsed SI without a pre-existing mental illness would report a higher burden of pandemic-related stressors. We further examined whether psychological symptoms such as depression, perceived burdensomeness, thwarted belongingness, and perceived discrimination, mediated the associations between pandemic-related stressors and SI.

2. Methods

2.1. Participants & procedures

Canadian adults, aged 17 and over, were recruited via an established social research survey firm (MARU/Blue) using quota sampling to match the age, sex, and geographic of the Canadian population as per the 2019 Census, with over-sampling from the least populous provinces (i.e., Saskatchewan, Manitoba, and the Atlantic provinces), and under-sampling from the most populous provinces (i.e., Ontario and Quebec) to permit province-level analyses. We used quota-based sampling rather than true probability sampling due to the time-sensitive nature of this survey. Quota-based responses were supplemented with recruitment via advertisements posted on social media, community research organizations, and non-profit listservs aimed to increase responses from the youngest (i.e., 17–24-year-olds) and oldest (i.e., 65+ year-olds) demographics. Surveys were available in both official languages, and 9.8–15.5 % of respondents completed surveys in French. Respondents from MARU/Blue were offered a small monetary incentive (<$5) for their participation, while all other respondents were entered into a $50 gift card raffle. All participants were provided with a list of national and provincial mental health resources during consent and at the end of the survey. Data collection for Wave 1 (W1; n = 6629, 92.1 % Maru/Blue and 7.9 % convenience sampled) occurred between 08/18/2020 and 10/01/2020, Wave 2 (W2; n = 5920, 88.3 % Maru/Blue and 11.7 % convenience sampled) occurred between 12/21/2020 and 03/31/2021, and Wave 3 (W3; n = 7354, 81.4 % Maru/Blue and 18.6 % convenience sampled) occurred between 09/07/2021 and 12/07/2021. Our total sample size was comprised of 16,972 unique respondents.

2.2. Measures

Survey content was developed in collaboration with an international research consortium led by co-authors PP and SR. The surveys contained a set of common items present in all national surveys such as standardized measures of depression, anxiety, loneliness, and alcohol use. Other items were adapted to the national context such as those pertaining to regional public health restrictions and government financial supports. In addition, we included items that were specific to the Canadian surveys, including validated measures of thwarted belongingness, perceived burdensomeness and perceived discrimination (see below), as well as questions assessing receipt of the Canadian Emergency Response Benefit or Employment Insurance.

2.2.1. Suicidal ideation

SI was assessed with a single item, “Have you ever experienced suicidal thoughts?”. The primary outcome for the present analyses was the portion of respondents who endorsed current suicidal ideation (“Yes, currently”, assigned a score of 1), versus past ideation only (“Yes, in the past”), no ideation (“No”), or “not sure” (all assigned a score of 0). While single item assessment of SI is common in epidemiological surveys (e.g., Liu et al., 2021), because this item had not been previously validated, we compared endorsement against a validated and commonly used measure - the ninth item of the Patient Health Questionnaire (“Over the last 2 weeks, have you been bothered by.... Thoughts of being better off dead or of hurting yourself in some way”, with response options of Not at all [0], Several days [1], More than half the days [2], and Nearly every day [3]; Kroenke et al., 2001). Additionally, in W3 we added three items assessing passive SI (i.e., “wishing you were dead or wishing you could go to sleep and not wake up”), active SI (i.e., “actually had any thoughts of killing yourself”), and suicide planning (i.e., “thinking about how you might kill yourself”) within the past three months, adapted from the Columbia Suicide Severity Rating Scale (Posner et al., 2011), with each rated as Yes (1) or No (0), to further assess the sensitivity of primary outcome item.

2.2.2. Demographics

Demographic information including respondents' age, biological sex, geographical area, educational background, mental illness status, race and ethnicity were explored. Further information is provided in Appendix 1.

2.2.3. Pandemic experiences

An evidence synthesis indicated that people's experiences due to the pandemic differed (Phiri et al., 2021). These findings informed our exploration of pandemic-related stressors in the Canadian population. Participants reported on various stressors relevant to the COVID-19 pandemic, including whether they: were medically vulnerable toward COVID-19 (e.g., by virtue of a pre-existing medical condition, immune suppressing treatment, or advanced age); were required to have direct contact with people displaying COVID-19 symptoms at work; had been required to self-isolate due to confirmed or suspected exposure to COVID-19; had ever tested positive for COVID-19; or, had lost a significant portion of their personal or household income due to the pandemic. Employed respondents reported whether they were working remotely, worked in an Essential Service sector, had been asked to take an unpaid leave, or were worried about losing their current job. Unemployed respondents reported whether they had received unemployment benefits (e.g., Employment Insurance or the Canadian Emergency Response Benefit) since the pandemic began. All items were scored Yes (1) or No (0).

2.2.4. Psychological mediators

In each Wave, participants completed validated measures of depression (the Patient Health Questionnaire [PHQ-9], with only the first eight items scored to avoid conflating scores with the SI outcome; Kroenke et al., 2001), anxiety (the Generalized Anxiety Disorder seven item screener [GAD-7]; Spitzer et al., 2006) and alcohol use (the Alcohol Use Disorder Identification Test – Concise version [AUDIT-C]; Bush et al., 1998). In W2 and W3 only, participants completed validated measures of perceived burdensomeness and thwarted belongingness (10-item Interpersonal Needs Questionnaire; Bryan, 2011) and perceived discrimination (short [5 item] version of the Everyday Discrimination Scale; Sternthal et al., 2011). Scale internal consistencies appear in Table 1 .

Table 1.

Sample composition by wave and recruitment source, and scale internal consistencies.

	Wave 1	Wave 2	Wave 3	ϕ	MARU/Blue	Social media	ϕ
	% (n)				% (n)
Male	38.9a (2343)	44.2b (2112)	42.4b (2509)	0.044	44.6a (6388)	24.1b (576)	0.146
Under 35	27.7a (1672)	37.3b (1807)	41.5c (2505)	0.125	31.0a (4476)	60.7b (1508)	0.220
Racialized persons	17.5a (1033)	20.2b (948)	19.8b (1157)	0.030	18.7a (2632)	21.1b (506)	0.022
Less than BA	52.8a (3096)	51.7a (2428)	56.4b (3291)	0.041	55.0a (7711)	46.4b (1104)	−0.061
Rural resident	17.0a (1028)	16.1a (777)	15.4a (927)	0.018	17.4a (2517)	8.7b (215)	−0.084
Live alone	21.9a (1331)	20.7a (1001)	30.0b (1814)	0.097	24.8a (3581)	22.7a (565)	−0.017
Live with children	31.9a (1836)	33.3a (1535)	15.3b (921)	0.190	26.7a (3747)	54.5b (22.9)	−0.030
Prior MH concern	33.3a (2029)	21.0b (1019)	24.7c (1492)	0.116	24.3a (3527)	40.7b (1013)	0.131
Unemployed	11.8a (448)	17.2b (606)	16.2b (709)	0.064	13.8a (1334)	21.0b (429)	0.076
Medically vulnerable	17.1a (1040)	30.2b (1374)	32.8c (1963)	0.161	26.9a (3818)	22.9b (559)	−0.032
Exposure @ work	8.8a (497)	8.7a (397)	10.3b (591)	0.025	8.8a (1209)	12.1b (276)	0.040
Self-isolating	6.1a (372)	5.7a (274)	11.9b (716)	0.105	6.5a (947)	16.7b (415)	0.132
Tested positive	0.4a (27)	1.6b (79)	3.4c (203)	0.093	1.4a (202)	4.3b (107)	0.077

	α
	W1	W2	W3
PHQ-8	0.93	0.93	0.93
GAD-7	0.94	0.93	0.93
AUDIT-C	0.75	0.76	0.79
INQ-TB	–	0.95	0.95
INQ-PB	–	0.83	0.85
EDS	–	0.75	0.80

Notes. Column superscripts indicate significant pairwise differences, p < .05. PHQ-8 scores are used throughout this study to avoid conflating the ninth item, which assesses recent suicidal thoughts, with the outcome in our models.

2.3. Data analysis

2.3.1. Data cleaning and preliminary analyses

The study analyses were conducted using SPSS 27 and MPlus 8.5. Prior to analyses, we used a quality control step to ensure all participants met the protocol specified eligibility criteria, and then implemented data cleaning procedures recommended for survey studies (DeSimone and Harms, 2018; Goldammer et al., 2020), including verifying correct responses to two simple attention-check questions and removing responses with improbably fast completion time or completion of fewer than 10 % of the items in the survey. Total exclusions, including participants who were automatically directed out of the survey once enrolment quotas were met, resulted in removal of 2012 responses from Wave 1, 5887 from Wave 2, and 4481 from Wave 3. Further details regarding data cleaning are available at https://osf.io/4d6tq/?view_only=975765c631bd4d76a00db38dd21d08e0. We compared the sample compositions across waves and recruitment sources, and subsequently examined agreement between SI items (see Preliminary analyses, below). Data were weighted prior to analyses to match the sex, age, and provincial distribution of the Canadian population per the 2019 Census.

2.3.2. Primary analyses

To explore associations between population characteristics and pandemic-related stressors linked to SI, we computed simple odds ratios to reflect the unadjusted, bivariate associations between each predictor and outcome. Given the large number of tests, we used a threshold of p < .001 for these tests.

To examine SI among respondents with (vs. without) prior mental health diagnoses, we used cross-tabulations to compare the compositions of the four groups: no mental health diagnosis without current SI; no mental health diagnosis with current SI; prior mental health diagnosis without current SI; prior mental health diagnosis with current SI. Because we expected low cell numbers for the latter two groups, we collapsed data across waves and used unweighted data given weights were sample specific. We compare these groups on the proportion of respondents who were: females, 35 years old, racialized persons, had less than a Bachelor's degree, rural residents, medically vulnerable toward COVID-19, potentially exposed to COVID-19 at work, suspected or confirmed to have COVID-19 exposure, unemployed, and insecurely employed. Further, we used a one-way ANCOVA to compare psychological symptoms in these groups, controlling for all aforementioned demographic and pandemic-related experiences. Post-hoc pairwise comparisons used a Bonferonni corrected p < .016 to indicate significant differences.

Finally, we used Structural Equation Modeling (SEM) to examine the direct and indirect associations between pandemic-related stressors and SI. We constructed and refined a factor model comprising of latent factors for each potential psychological mediator of depression, perceived burdensomeness, thwarted belongingness, and perceived discrimination. We consulted modification indices to adjust the factor model until an adequate fit was indicated by a combination of parsimony-adjusted and incremental indices, including: (1) the root mean square error of approximation (RMSEA) with values below 0.05 indicating good fit, (2) the standardized root mean squared residual (SRMR) with values below 0.08 indicating good fit, (3) the Comparative Fit Index (CFI) with values above 0.90 indicating adequate fit and values near or above 0.95 suggesting good fit, and (4) the Tucker-Lewis Index (TLI), with values above 0.90 suggesting adequate fit (Hu and Bentler, 1999). Once an adequate measurement model was identified, we fit the mediation models, including direct associations between each exogenous pandemic-related stressor and SI. Indirect associations between each exogenous variable and SI via the latent psychological variables were also evaluated. Bias-corrected confidence intervals for the direct and indirect effects were reported based on 1000 bootstrapped samples for an unadjusted model as well as a model adjusting for the effects of sex, age, race, and education on current SI and each of the latent mediators. Fig. 1 indicates the conceptual map developed for the purpose of the study.

Fig. 1.

Conceptual structural model, testing the direct and indirect paths from COVID-related stressors and demographic characteristics to current SI through depression, perceived burdensomeness, thwarted belongingness, and perceived discrimination.

Notes. PB = perceived burdensomeness. TB = thwarted belongingness. PD = perceived discrimination. Exposure to COVID-19 includes suspected or confirmed exposure. The covariates gender, age, education, and ethnicity were entered as separate predictors.

3. Results

3.1. Preliminary analyses

3.1.1. Cross-wave and recruitment source comparisons

Removal of potential duplicate respondents across waves based on unique panel IDs and/or IP addresses resulted in a total unweighted sample of 16,972 for cross-wave comparisons (W1 n = 6091; W2 n = 4883, W3 n = 6038). As shown in Table 2 , there were small but significant differences in sample compositions by wave and recruitment source (ϕ = −0.088 to 0.220), with most falling below ϕ = 0.10 (i.e., a small effect). Given that (a) differences were small, (b) the inclusion of the social media/community recruited sample increased representation of demographic groups that were of substantive interest (i.e., younger adults, racialized persons, and people living with children, with pre-existing MH concerns, and who were self-isolating), and (c) we were primarily interested in understanding the magnitude of associations between various experiences and SI, we report results from the total sample, except where reporting prevalence estimates which are likely to be sensitive to the underlying sample composition.

Table 2.

Bivariate, unadjusted associations between demographics and pandemic experiences with current suicidal ideation.

	Wave 1			Wave 2			Wave 3
	Current suicidal ideation (4.1 %)			Current suicidal ideation (5.3 %)			Current suicidal ideation (5.8 %)
	Non-index% with SI	Index% with SI	OR 95 % CI	Non-index% with SI	Index% with SI	OR 95 % CI	Non-index% with SI	Index% with SI	OR 95 % CI
Demographic
Female sex	4.5	3.6	0.78 0.61–1.01	5.3	4.8	0.94 0.71–1.15	6.0	5.0	0.82 0.67–1.01
Age under 35	3.0	7.0	2.48 1.94–3.18	3.8	9.0	2.48 1.96–3.14	3.9	10.6	2.93 2.39–3.58
Racialized persons	3.9	5.0	1.28 0.94–1.73	4.9	7.0	1.47 1.11–1.95	5.5	7.3	1.37 1.07–1.75
Less than bachelors	3.1	5.1	1.67 1.29–2.17	4.8	5.8	1.22 0.97–1.55	4.9	6.5	1.35 1.10–1.65
Rural resident	4.1	4.0	0.97 0.69–1.36	5.3	5.0	0.94 0.67–1.32	5.9	5.4	0.92 0.69–1.23
Living alone	3.8	5.0	1.32 1.00–1.74	5.1	6.1	1.22 0.94–1.59	–	–	–
Living with children	3.8	4.9	1.30 0.99–1.68	5.3	5.7	1.08 0.84–1.39	–	–	–
Pre-existing MH dx	1.3	10.1	8.26 6.15–11.09	3.1	14.8	5.35 4.22–6.79	3.6	13.9	4.28 3.50–5.25
Family caregiving	–	–	–	4.8	6.0	1.27 0.94–1.71	5.4	7.1	1.35 1.07–1.70
Pandemic experiences
Unemployed	4.2	9.8	2.48 1.76–3.51	5.6	9.8	1.83 1.28–2.62	6.7	13.1	2.10 1.50–2.94
Insecurely employed	3.1	7.0	2.33 1.72–3.15	4.1	10.1	2.65 2.00–3.53	5.0	13.5	3.00 2.38–3.78
Medically vulnerable	3.8	5.4	1.42 1.06–1.92	4.7	6.2	1.33 1.04–1.70	5.4	6.3	1.17 0.95–1.44
Potential for exposure @ work	3.8	7.5	2.03 1.40–2.94	4.8	10.0	2.21 1.54–3.16	5.4	11.9	2.39 1.80–3.16
Suspected or confirmed exposure	3.6	9.5	2.83 2.08–3.87	4.9	10.1	2.16 1.49–3.14	5.0	11.9	2.59 2.02–3.31
… required to self-isolate	3.6	9.3	2.75 2.01–3.76	5.0	8.8	1.81 1.18–2.78	5.2	10.7	2.18 1.67–2.86
… tested positive	4.1	9.7	2.52 0.76–8.34	4.9	16.3	3.75 2.12–6.63	5.4	16.6	3.47 2.39–5.03
Income loss	–	–	–	4.1	8.9	2.27 1.77–2.92	4.5	11.7	2.82 2.27–3.49
Of those working…
Essential worker	–	–	–	4.5	5.7	1.27 0.98–1.64	6.0	7.8	1.34 1.07–1.68
Remote working	3.3	4.4	1.34 0.83–2.15	5.7	5.5	0.97 0.74–1.29	6.8	6.6	0.98 0.78–1.23
Of those not working…
Received any benefits (EI or CERB)	8.4	10.6	1.30 0.67–2.53	8.3	8.9	1.07 0.60–1.92	14.0	10.7	0.74 0.44–1.24

Notes. Bolding indicates bivariate associations p < .001. SI = suicidal ideation. EI = employment insurance. CERB = Canada Emergency Response Benefit.

3.1.2. Evaluation of SI items

The single SI item was endorsed by 256 respondents (4.0 % of the unweighted sample) in W1, 333 respondents (5.6 % of the unweighted sample) in W2, and 500 respondents (6.8 % of the unweighted sample) in W3. The mean on item nine of the PHQ-9 was 0.24 (SD = 0.63, with 15.3 % [n = 1011] endorsing a rating >0) in W1, 0.30 (SD = 0.70, 18.8 % [n = 1120] reporting a score >0) in W2, and 0.32 (SD = 0.72, 19.2 % [n = 1420] reporting a score >0) in W3. Agreement between the single SI item and ninth item of the PHQ-9 was good (W1: 88 %, W2: 86.8 %, W3: 86.5 %), with most discrepancies attributable to endorsement of the ninth PHQ item but not the single SI item (W1: 11.8 %; W2: 12.8 %; W3: 12.9 %). On the additional items administered in W3, 19.7 % of respondents (n = 1450; 88.5 % agreement with single SI item) endorsed passive SI in the past three months; 10.7 % (n = 784; 77.5 % agreement with single SI item) endorsed active SI; and 9.8 % (n = 724; 68.3 % agreement with single SI item) endorsed suicidal planning. These results suggest that the item used to define our outcome had adequate sensitivity. Given the ninth item of the PHQ-9 has been criticized as being overly inclusive as a screening tool for SI (Na et al., 2018), we decided to continue with the single-item to operationalize the outcome of interest.

3.2. Primary analyses

3.2.1. Distribution of suicidal ideation during the pandemic

After sample weighting, the prevalence of SI was 4.1 %, 5.3 %, and 5.8 % in W1, W2, and W3, respectively. The prevalence of SI was 3.9 %, 4.7 %, and 4.7 % among survey panel respondents, and 7.0 %, 9.8 %, and 13.1 % among convenience-sampled respondents in each respective Wave. Distribution of SI by age, level of education, and urban density for each Wave and subsample is presented in Supplementary Table 1. The associations between demographic characteristics and pandemic-related experiences and SI are summarized in Table 3 . In all three Waves, younger age (i.e., under 35 years) and pre-existing mental health concerns were associated with significantly elevated odds of SI. Racialized participants reported higher odds of SI relative to White participants in W2 and W3. Lower educational attainment was associated with SI in W1 and W3. Living alone was associated with higher odds of SI in W1 but not W2. Family caregiving responsibilities were associated with higher odds of SI in W3 but not W2. Across all waves, rates of SI did not differ by sex or rural versus urban residency.

Table 3.

Demographic characteristics, pandemic experiences, and adjusted estimates of psychological symptoms among respondents with and without prior mental health concerns, with and without current suicidal ideation.

	A. No prior MH or SI (n = 11,093)	B. Prior MH, no current SI (n = 3717)	C. Prior MH with SI (n = 575)	D. No prior MH with SI (n = 353)	χ2(3)	p	ϕ
	Proportion	Proportion	Proportion	Proportion
Demographic characteristics
Female sex	54.1a	71.3b	63.2c	48.1a	361.09	<0.001	0.15
Age under 35	30.0a	43.5b	61.0c	52.8c	470.51	<0.001	0.171
Racialized persons	19.8a	14.9b	19.7a	29.2c	70.70	<0.001	0.067
Less than bachelors	51.7a	57.2b	64.1c	60.9b,c	67.78	<0.001	0.066
Rural resident	16.6a	15.7a	15.4a	12.4a	6.47	0.091	0.020
Pandemic-related stressors
Medically vulnerable	23.4a	30.1b	39.1c	26.7a,b	125.97	<0.001	0.089
Potential exposure @ work	12.3a	19.2b	30.1c	20.8b,c	137.20	<0.001	0.120
Suspected or confirmed exposure	7.3a	15.0b	23.4c	13.5b	319.97	<0.001	0.141
Unemployed	6.4a	13.2b	18.2c	9.9a,b	189.77	<0.001	0.125
Insecurely employed	23.1a	26.9b	40.7c	40.5c	112.87	<0.001	0.101

Psychological symptoms	M* (SE)	M* (SE)	M* (SE)	M* (SE)	F (df)	p	η2
Depression	5.15 (0.07)a	8.84 (0.12)b	14.26 (0.31)c	13.21 (0.37)c	549.32(3, 8352)	<0.001	0.165
Generalized anxiety	3.92 (0.06)a	6.79 (0.11)b	10.27 (0.29)c	9.53 (0.34)c	340.45(3, 8352)	<0.001	0.109
Alcohol use	2.73 (0.03)a	2.88 (0.06)a	3.43 (0.14)b	3.22 (0.17)b	10.78(3, 8352)	<0.001	0.004
PB**	1.63 (0.02)a	1.98 (0.04)b	3.37 (0.09)c	3.42 (0.09)c	190.31(3, 4722)	<0.001	0.108
TB**	3.14 (0.02)a	3.76 (0.05)b	4.57 (0.10)c	4.78 (0.10)c	137.93(3, 4722)	<0.001	0.081
PD**	0.65 (0.02)a	0.84 (0.04)b	1.57 (0.10)c	1.50 (0.10)c	42.83(3, 4722)	<0.001	0.026

Notes. MH = mental health concern or diagnosis. PB = perceived burdensomeness. PD = perceived discrimination. TB = thwarted belongingness. SI = suicidal ideation. *Table reports estimated marginal means adjusted for sex, age, racialized persons, education, medical vulnerability, work exposure, suspected/confirmed exposure, unemployment and insecure employment. **PB, TB, and PD are conducted on W2 and W3 combined samples only (n = 3594 no MH or SI, n = 853 MH no SI, n = 145 MH and SI, n = 142 no MH with SI). Column superscripts indicate significant pairwise differences, with Bonferonni correction, at p < .05.

Regarding pandemic-related experiences, being unemployed, insecurely employed (i.e., required to take unpaid leave or fear of losing one's job), working in an environment that required direct contact with people displaying symptoms of COVID-19, and having suspected or confirmed exposure to COVID-19 (i.e., required by public health to self-isolate due to exposure, or a positive test result) were associated with increased odds of SI at all three Waves. Medical vulnerability to negative outcomes of COVID-19 infection was associated with higher odds of SI in W1 and W2, but not W3. Loss of income due to the pandemic, only assessed in W2 and W3, was associated with higher odds of SI in both those Waves. Among respondents who were working, being an essential worker was associated with higher odds of SI in W3 but not W2. Neither the ability to work remotely (among those working) nor receiving unemployment benefits (among those not working) were associated with SI in any Wave.

3.2.2. Correlates of SI in people with and without pre-existing MH concerns

As shown in Table 3, relative to respondents without pre-existing MH concerns and without SI, respondents without pre-existing MH concerns who reported current SI were more likely to be under 35 years old, identify as racialized, have less than a Bachelor's degree, report potential for COVID-19 exposure at work, have a suspected or confirmed COVID-19 exposure, and be unemployed or insecurely employed. Relative to respondents with pre-existing MH concerns and current SI, these respondents were more likely to be male and to be a racialized person but were less likely to be medically vulnerable toward COVID-19 infection, have a suspected or confirmed COVID-19 exposure, or be unemployed. With respect to psychological symptoms, respondents with SI scored similarly regardless of whether they had pre-existing MH concerns or not, and consistently reported worse psychological and behavioral functioning compared to respondents with pre-existing MH concerns who did not report SI, and those with neither MH concerns nor SI. Additional pairwise comparisons are shown in Table 3.

3.2.3. Psychological mediators of the associations between pandemic stressors and SI

Steps and results of the initial latent factor construction are presented in Supplementary Table 2. Consulting the modification indices, we performed between two and four sequential adjustments until the measurement models demonstrated adequate fit to the data. Next, we added the direct and indirect paths from pandemic experiences to SI through depression, perceived burdensomeness, thwarted belongingness, and perceived discrimination (see Fig. 1). The final structural models fit the data well (see Table 4 ), and the adjusted models explained 55.2 % (SE of R² = 0.030) and 47.8 % (SE of R² = 0.022) of the variance in SI in W2 and W3, respectively.

Table 4.

Unadjusted and adjusted standardized coefficients and 95 % confidence intervals of the total, total indirect, specific indirect, and direct effects of the mediation models in Waves 2 and 3.

Model fit	W2		W3
	Unadjusted model	Adjusted model	Unadjusted model	Adjusted model
	RMSEA = 0.029 (0.027, 0.030)	RMSEA = 0.029 (0.028, 0.031)	RMSEA = 0.029 (0.027, 0.030)	RMSEA = 0.029 (0.028, 0.030)
	CFI = 0.954, TLI = 0.944	CFI = 0.948, TLI = 0.937	CFI = 0.958, TLI = 0.950	CFI = 0.955, TLI = 0.945
	SRMR = 0.027	SRMR = 0.029	SRMR = 0.026	SRMR = 0.027
	Est (95 % CI)	Est (95 % CI)	Est (95 % CI)	Est (95 % CI)
Unemployment
Total effect	0.131 (0.085, 0.176)	0.078 (0.026, 0.120)	0.154 (0.119, 0.187)	0.110 (0.075, 0.150)
Total direct	−0.046 (−0.089, −0.002)	−0.068 (−0.122, −0.027)	−0.023 (−0.056, 0.009)	−0.026 (−0.059, 0.009)
Total indirect	0.177 (0.154, 0.200)	0.146 (0.119, 0.171)	0.177 (0.160, 0.195)	0.136 (0.120, 0.154)
Via depression	0.049 (0.037, 0.066)	0.040 (0.028, 0.053)	0.074 (0.061, 0.088)	0.051 (0.040, 0.063)
Via PB	0.007 (0.001, 0.013)	0.003 (−0.001, 0.007)	0.012 (0.006, 0.018)	0.007 (0.003, 0.012)
Via TB	0.095 (0.070, 0.120)	0.088 (0.066, 0.114)	0.080 (0.063, 0.101)	0.072 (0.058, 0.091)
Via PD	0.025 (0.016, 0.039)	0.015 (0.007, 0.026)	0.010 (0.005, 0.017)	0.006 (0.002, 0.011)
Insecure employment
Total effect	0.159 (0.114, 0.203)	0.142 (0.091, 0.192)	0.152 (0.109, 0.187)	0.122 (0.078, 0.160)
Total direct	−0.008 (−0.053, 0.036)	−0.002 (−0.052, 0.049)	−0.003 (−0.042, 0.035)	−0.008 (−0.053, 0.029)
Total indirect	0.166 (0.146, 0.189)	0.144 (0.122, 0.168)	0.156 (0.138, 0.173)	0.130 (0.114, 0.148)
Via depression	0.044 (0.032, 0.057)	0.041 (0.029, 0.054)	0.065 (0.053, 0.078)	0.052 (0.041, 0.064)
Via PB	0.011 (0.002, 0.021)	0.006 (−0.002, 0.013)	0.016 (0.008, 0.025)	0.011 (0.003, 0.019)
Via TB	0.073 (0.056, 0.095)	0.067 (0.049, 0.089)	0.055 (0.042, 0.070)	0.051 (0.039, 0.067)
Via PD	0.039 (0.023, 0.055)	0.030 (0.017, 0.048)	0.019 (0.009, 0.031)	0.016 (0.007, 0.027)
Medical vulnerability
Total effect	0.086 (0.037, 0.126)	0.116 (0.068, 0.164)	0.043 (0.001, 0.079)	0.060 (0.021, 0.098)
Total direct	0.065 (0.017, 0.104)	0.080 (0.037, 0.128)	0.026 (−0.010, 0.062)	0.026 (−0.012, 0.063)
Total indirect	0.021 (0.001, 0.039)	0.036 (0.016, 0.055)	0.017 (0.003, 0.031)	0.034 (0.020, 0.053)
Via depression	0.011 (0.005, 0.017)	0.019 (0.012, 0.028)	0.014 (0.009, 0.022)	0.024 (0.018, 0.032)
Via PB	0.001 (0.000, 0.004)	0.002 (0.000, 0.005)	0.003 (0.001, 0.005)	0.003 (0.001, 0.007)
Via TB	0.002 (−0.009, 0.014)	0.006 (−0.006, 0.019)	−0.004 (−0.012, 0.003)	0.001 (−0.007, 0.011)
Via PD	0.007 (0.002, 017)	0.009 (0.002, 0.019)	0.004 (0.001, 0.008)	0.005 (0.002, 0.011)
Suspected or confirmed exposure
Total effect	0.065 (0.025, 0.103)	0.051 (0.002, 0.091)	0.123 (0.089, 0.155)	0.080 (0.047, 0.113)
Total direct	0.006 (−0.033, 0.043)	0.009 (−0.034, 0.047)	0.056 (0.027, 0.086)	0.042 (0.010, 0.074)
Total indirect	0.059 (0.040, 0.081)	0.042 (0.021, 0.062)	0.067 (0.052, 0.081)	0.038 (0.023, 0.053)
Via depression	0.019 (0.013, 0.028)	0.015 (0.010, 0.024)	0.033 (0.025, 0.041)	0.018 (0.012, 0.025)
Via PB	0.005 (0.001, 0.010)	0.003 (−0.001, 0.007)	0.006 (0.003, 0.009)	0.002 (0.001, 0.006)
Via TB	0.015 (0.005, 0.028)	0.009 (−0.005, 0.021)	0.018 (0.011, 0.028)	0.010 (0.002, 0.018)
Via PD	0.020 (0.012, 0.032)	0.015 (0.008, 0.026)	0.010 (0.005, 0.017)	0.008 (0.003, 0.014)
Loss of income
Total effect	0.083 (0.032, 0.126)	0.067 (0.016, 0.117)	0.095 (0.060, 0.133)	0.106 (0.067, 0.142)
Total direct	0.009 (−0.035, 0.052)	0.000 (−0.051, 0.049)	0.013 (−0.024, 0.047)	0.022 (−0.014, 0.057)
Total indirect	0.074 (0.050, 0.095)	0.067 (0.044, 0.087)	0.082 (0.065, 0.097)	0.084 (0.067, 0.100)
Via depression	0.021 (0.014, 0.031)	0.021 (0.013, 0.030)	0.035 (0.026, 0.043)	0.035 (0.026, 0.045)
Via PB	0.002 (0.000, 0.005)	0.001 (0.000, 0.003)	0.006 (0.003, 0.010)	0.004 (0.001, 0.008)
Via TB	0.027 (0.014, 0.043)	0.025 (0.010, 0.040)	0.028 (0.020, 0.041)	0.031 (0.021, 0.043)
Via PD	0.024 (0.014, 0.036)	0.020 (0.011, 0.032)	0.014 (0.007, 0.023)	0.013 (0.006, 0.022)

Notes. PB = perceived burdensomeness. PD = perceived discrimination. TB = thwarted belongingness.

Table 4 summarizes the total direct and indirect effects from each pandemic-related stressor and SI, as well as the specific indirect associations via depression, perceived burdensomeness, thwarted belongingness and perceived discrimination for both an unadjusted model and a model adjusting for sex, age, race, and education. The associations between unemployment and insecure employment with current SI were fully mediated by the set of psychological experiences. The largest indirect path between unemployment and SI occurred through thwarted belongingness in both Waves, while the indirect paths from insecure employment and SI for thwarted belongingness and depression were of similar magnitude in both Waves. Medical vulnerability toward COVID-19 had significant direct and indirect associations with SI, the largest of which was indirectly through depression. Having a suspected or confirmed exposure to COVID-19 had only an indirect association with SI in W2, but both direct and indirect associations with SI in W3. The largest indirect effect occurred via depression, with additional indirect paths through perceived discrimination in W2 and W3, and thwarted belongingness in W3. Loss of personal income was only indirectly associated with SI in both Waves, with the significant indirect associations via depression, thwarted belongingness, and perceived discrimination in all three waves, and a small but significant indirect path through perceived burdensomeness in W3.

4. Discussion

Given that mental health symptoms have increased, on average, during the COVID-19 pandemic (Wu et al., 2021), continued study of drivers of suicidal thoughts during this public health crisis is warranted to identify interventions or policies that could both prevent tertiary, pandemic-related mortality and improve psychological wellbeing. This study expands what was previously known about the impacts of the pandemic on SI in three ways. First, it extends previous single-point Canadian studies conducted early in the pandemic (i.e., mostly May through September 2020), showing that rates of SI remained high in the second year of the pandemic. The rise in SI in the Waves 2 and 3 surveys, relative to Wave 1, corresponded to much higher COVID-19 infection rates in Canada (Wave 1 average weekly cases = 852, range = 358 to 1978; Wave 2 average weekly cases = 4736, range = 2831 to 7957; Wave 3 average weekly cases = 3230, range = 2288 to 4387; Government of Canada, 2022) and approval and distribution of COVID-19 vaccines throughout the later part of Wave 2 and all of Wave 3 (approximately 2 % of Canadians had received a single vaccine dose by the end of Wave 2, while 76 % had received one dose and 6 % had received at least one booster by the end of Wave 3; Little, 2020). Although the timing and strength of public health orders (e.g., restriction of in-person gatherings, business operations, and school delivery modes) varied provincially, Wave 3 also corresponded with eased restrictions in most provinces, relative to the previous two Waves (Canadian Institute for Health Information, 2022). Finer-grained analyses may inform the relative contributions of these factors to fluctuations in the rates of SI observed throughout the pandemic. Second, it characterizes an important group that has not previously been examined – people who reported no pre-existing MH concerns but endorsed SI during the pandemic. Third, we show that the associations between unemployment, insecure employment, suspected or confirmed COVID-19 exposure, and loss of personal income and SI are fully mediated by psychological experiences, namely depression and thwarted belongingness. Together, these findings help to further characterize who is at risk of SI, and why these risks are elevated, in the context of the COVID-19 pandemic.

Comparing what we are learning about SI during the pandemic to what was known prior, a few trends stand out. First, several studies, this one included, have not found sex or gender differences in SI during the pandemic (Czeisler, 2020; Liu et al., 2021; O'Connor et al., 2021). This contrasts well-established gender differences that were robustly documented before the pandemic, with higher rates of SI found among women (Beautrais, 2002; Schrijvers et al., 2012). Given that SI appears to have become more prevalent in the pandemic, this likely reflects escalating rates of SI in males during this time, and suggests there is value in screening and prevention efforts targeted toward men, particularly because the transition between suicidal thoughts and actions is more rapid in males than females (Schrijvers et al., 2012).

Second, while pre-pandemic studies suggested that unemployment, job and financial precarity, and financial hardship are associated with SI at both the individual and population level (Blakely et al., 2003; Mathieu et al., 2022; McIntyre and Lee, 2020), surveys during the pandemic have generally shown significant but small, or non-significant, differences in rates of SI according to employment status (Liu et al., 2021; McIntyre et al., 2021). Our findings contrast these latter results, showing roughly two-fold higher odds of SI among working-age Canadians who were unemployed relative to those who were working or out of the workforce (e.g., students, homemakers). Moreover, we found that receiving government financial support in Canada, either Employment Insurance or Canadian Emergency Response Benefits (CERB), did not reduce these associations, although this may reflect underlying demographic differences in CERB recipients (Morissette et al., 2021). These findings are consistent with a recent literature review suggesting that while financial policies such as enhanced unemployment benefits, worker protections, and increased minimum wages are associated with reduced suicidal thoughts and behaviors at the population level, these polices have unclear effects on individual risk of SI as very little evidence has directly assessed their impact (Mathieu et al., 2022). In our study, unemployment was associated with SI at both waves primarily through its association with thwarted belongingness––that is, via a diminished sense of meaningful connection with other people. Elevated risk of SI of a similar magnitude was also identified in our surveys among Canadians who were employed but were either worried about losing their job or had been asked to take an unpaid leave by their employer (i.e., insecurely employed); these survey respondents had two- to three-fold higher odds of reporting SI relative to securely employed respondents. Insecure employment was associated with SI through thwarted belongingness, depression, and perceived discrimination. Given that loss of personal income was likewise (albeit, more weakly) indirectly associated with current SI via thwarted belongingness, depression and perceived discrimination, the pattern of results suggests that overall financial insecurity (e.g., household debt, employment precarity, housing affordability) may provide a more holistic picture of SI risk (Choi et al., 2021; Raifman et al., 2020). Although further replication is warranted, our results suggest that supporting a sense of meaningful connection for those experiencing financial hardship may be beneficial.

Third and finally, these results are in accord with a robust literature associating mental health concerns and suicide (Bertolote and Fleischmann, 2002). In our surveys, the odds of endorsing current SI were four to eight-fold higher among respondents who reported having a prior mental health diagnosis or treatment. Put another way, by the Winter of 2021, nearly one in seven survey respondents with prior mental health concerns reported they were experiencing suicidal thoughts. This finding emphasizes the potential benefit of proactively screening of people who are receiving mental health services (e.g., psychotropic medication, counselling) so that appropriate and potentially life-saving support can be offered (Hofstra et al., 2020). At the same time, people with pre-existing mental health concerns accounted for only two thirds of the total SI cases in our sample; the remaining one third of cases were reported by people who reported no history of formal mental health diagnosis or concerns. This accords with recent perspectives emphasizing that viewing suicide as a consequence of mental illness, or mental illness as a necessary precursor to SI, is overly reductionistic and likely to miss an important set of people experiencing suicidal thoughts because of other sources of seemingly intractable pain (Klonsky et al., 2021). Relative to people with known MH concerns, respondents who endorsed SI that was not accompanied by diagnosed mental health concern(s) were more likely to be male and non-White, and less likely to be medically vulnerable toward COVID-19 infection or unemployed. These results align with recent meta-analytic findings that people who died by suicide without receiving mental health services were more likely to be male, younger or older (vs. middle) aged, ethnic minorities, and living in rural communities (Tang et al., 2021; Walby et al., 2018). Such findings underscore the importance of developing varied, culturally sensitive supports to effectively reach and engage people who may be at risk.

4.1. Limitations

Several limitations merit acknowledgment. First, although we were able to recruit a well-stratified national sample using demographic and regional quotas, possible response biases remain that could limit generalizability to the full Canadian population, particularly given we used exclusively online recruitment and described the study as focusing on mental health. While prevalence estimates should be interpreted cautiously, it is worth noting that our results closely match the distribution of mental health concerns reported in pre-pandemic representative samples (Government of Canada, 2017) and rates of SI reported in nationally representative surveys conducted during the pandemic (Liu et al., 2021; McAuliffe et al., 2021). Second, we used a single item to assess current suicidal thoughts. Although many prior studies have used single items, for instance the ninth item of the PHQ-9 (Canadian Mental Health Association, 2020) or purpose-built items (Liu et al., 2021; McAuliffe et al., 2021), to assess SI, this practice may introduce individual biases in respondents' interpretation of the term “suicidal thoughts” as well as the time frame implied by the response option “currently”. While our analysis suggests that the item selected in this study provides a conservative estimate of the prevalence of SI, future studies that include more nuanced questions regarding the nature of these thoughts (e.g., passive versus active ideation, with or without planning) would help further characterize at-risk populations. Moreover, self-report measures of SI may underestimate rates of suicidal thoughts in respondents who may be motivated to under-report mental health problems, for instance due to stigmatizing beliefs about SI or mental illness, or distrust that their responses are truly anonymous. Third and finally, although our three-wave design allowed some comparisons regarding rates of SI at three points in the pandemic, the cross-sectional nature of each sample precludes conclusions regarding changes in SI. Longitudinal studies examining within-person changes in mental health, and SI specifically, would help illuminate how changing exposures over the course of the pandemic contribute to risk (and, we hope, recovery).

5. Conclusions

These limitations notwithstanding, this study contributes incrementally to an emerging body of literature answering two important questions: who is experiencing suicidal thoughts during the COVID-19 pandemic, and why are these thoughts more common in certain groups? Our findings highlight financial insecurity and heightened vulnerability or exposure to COVID-19 as experiences that predict increased rates of suicidal thinking, and depression and thwarted belongingness as drivers of these thoughts. Although the dynamics of this pandemic are ever evolving, continued attention to its mental health consequences is warranted.

Role of the funding source

This work was funded by a COVID-19 Mental Health and Substance Use Service Needs and Delivery Operating Grant co-funded by the Canadian Institutes of Health Research and Health Research BC (202007MS6-450104; COV-2020-2258), and a Health Research seed grant from the BC Ministry of Health, and was supported by a Health Research BC Scholar Award (#18240) to the first author.

CRediT authorship contribution statement

All authors have reviewed and take responsibility for this work. The survey was designed by PP, SR, BT and TP. Writing and analyses were performed by BT, with assistance from AS, BW, NL and MG. PP, SR and TP critically reviewed and revised the paper.

Conflict of interest

The authors have no conflicts of interest to declare.

Appendix A. Supplementary data

Supplementary material