This study was conducted in accordance with the principles of the Declaration of Helsinki and received ethics approval by the Health Research Ethics Board at the University of Alberta (Pro00101048), University of Calgary Conjoint Health Research Ethics Board (REB20-1544), Human Research Ethics Board at the University of British Columbia (H20-01732) and University of Waterloo Human Research Ethics Committee (ORE#42494). Operational approvals from participating AL homes were obtained.
It is reported per STROBE guidelines (Table S1, Additional File 1) [26].
Study design and setting
This cross-sectional study is part of a prospective cohort investigation, COVCARES-AB/BC (COVID-19 and Caregivers of Assisted living Residents: their Experiences and Support needs), underway in Alberta and British Columbia, Canada. Both provinces initiated “essential visitors” policies in AL in mid-March 2020, allowing a single caregiver to enter for end-of-life visits or to provide assistance with feeding or mobility [27]. However, the decision to impose more stringent or complete bans on visitors during the first wave was at the discretion of individual homes.
Study sample
We invited all eligible AL homes in Alberta and British Columbia to participate in COVCARES-AB/BC. Eligibility criteria included homes that were licensed and publicly subsidized (in Alberta these are known as designated supportive living), in operation for at least 6 months, not primarily serving psychiatric clients and with a minimum number of residents aged 65 + years (4 for small, 10 for large homes, based on region’s median bed-size). Based on available provincial registries (listing all publicly subsidized AL homes in the two provinces), we identified 163 eligible homes in Alberta and 137 in British Columbia at study onset. We contacted each eligible AL home by email and invited a key contact (e.g., facility administrator or director of care) to participate in our study. We sent up to three reminder emails in two-week intervals and contacted remaining non-respondents by telephone. Further description about AL in the two provinces is provided in Table S2, Additional File 1.
Consenting homes distributed our study materials and open web-based survey link to all identified caregivers via their internal email listing and/or in-person for caregivers visiting the home. This survey link was also distributed via social media, websites, email lists and newsletters affiliated with our study team and government/caregiver stakeholder partners in both provinces. Those who identified as the primary adult caregiver (i.e., aged 18 + years and most informed or involved in care) of a resident aged 65 + years who had lived in the AL home for three or more months prior to March 1, 2020, were eligible to participate.
The baseline caregiver survey was administered online, between October 28, 2020, and March 31, 2021, by the Survey Research Centre (SRC) at the University of Waterloo. The SRC has been in operation for over 20 years and employs robust and standardized methods, training, and protocols for survey research. These robust measures ensured that fraudulent survey completions were minimized (and tracked for removal) and that any duplicate and ineligible survey responses were identified and removed from the data. Caregivers completing the survey received a $25 coffee gift card.
Measures
Our caregiver survey included standardized and validated measures used in an earlier study of AL and nursing home residents and their caregivers in Alberta [18, 28] and by the Canadian Longitudinal Study on Aging [29, 30]. Included were items regarding caregivers’ sociodemographic characteristics, physical and mental health status, and social support. Items capturing changes in the frequency and nature of their visits with residents and involvement in resident care (during the 3-months prior to and post March 1/20) were also included. We selected this timeframe as it encompasses the first wave of the pandemic, during which visitor restrictions were the most stringent. Survey measures of pandemic-related stressors included loss of employment and income (and concern level), absence of opportunities to stay well-informed/engaged in care of the resident, and increased levels of concern about resident’s physical and mental well-being.
Anxiety and depressive symptoms
Caregiver anxiety was assessed with the 7-item Generalized Anxiety Disorder scale (GAD-7) [31, 32]. The items capture how often (over past 2 weeks) respondents have been bothered by feelings associated with anxiety. We used a cut-point of ≥ 8 to define a clinically significant anxiety disorder (sensitivity 92%, specificity 70%) [33].
Depressive symptoms were assessed with the Center for Epidemiologic Studies Depression Scale, Short Form (CES-D10) [34] that includes 10 items regarding how often (in past week) respondents have experienced signs or symptoms of depression. We used a cut-point of ≥ 10 to identify clinically significant depressive symptoms (sensitivity 89%, specificity 47% for major depressive disorder) [35].
Both instruments have been extensively used in previous caregiver studies [29, 30, 36, 37]. Consistent with scale development [31, 32, 34], respondents with missing values for only one item were assigned the mean of their other item responses before totalling their scale score. The terms ‘anxiety’ and ‘depressive symptoms’ are used in the remaining text to reflect significant symptomatology.
Covariates
Based on the Stress Process Model [24], background factors examined included caregiver age, gender, marital status, ethnicity, highest education, pre-pandemic household income, and province of AL home. Exacerbating and ameliorating factors [25] examined included caregivers’ number of chronic conditions, and perceived emotional/informational social support. The latter was assessed using the relevant subscale of the RAND Medical Outcomes Study – Social Support Survey [MOS-SSS]) [38].
We also examined pandemic-specific stressors supported by the Appraisal Model [25], including change in employment status and income reduction (combined with level of concern) during the 3 months post-March 1, 2020. Covariates reflecting caregiving stressors included their belief as to whether the AL home/staff created opportunities for them to be well-informed and involved in care of the resident, the change (comparing the 3 months post- vs pre-March 1/20) in their level of concern about the resident’s depression status, and whether they considered moving the resident out of the AL home during the 3 months post-March 1/20 because of COVID-19 [39].
AL home characteristics, including bed size, urban/rural location and ownership status (private for-profit vs non-profit) were also examined.
Statistical analysis
Descriptive analyses examined the distribution of caregivers’ characteristics, overall and by the presence/absence of anxiety or depressive symptoms.
We employed unadjusted and adjusted Poisson regression models, modified for binary outcomes [40], to examine associations between caregivers’ characteristics and our two mental health outcomes.
Guided by our theoretical models [24, 25], we developed our final models in a staged approach by first including caregivers’ background and health characteristics, followed by general and caregiving-specific pandemic related stressors and lastly emotional/informational social support. Each covariate was added one at a time while examining for potential collinearity issues. To provide final parsimonious multivariable models, we removed statistically non-significant factors, but retained those significant for only one outcome (but not the other) for comparative purposes. For both outcomes, we retained household income (though non-significant in full models) to allow for an interpretation of pandemic-related income loss in the context of baseline socioeconomic status. We did not include self-rated health in our models, given its conceptual overlap with our two mental health outcome measures. Though our survey asked whether caregivers tested positive for COVID-19 during the first pandemic wave, it was not feasible to explore this measure in our models as only three caregivers responded positively to this question.
Missing data analysis
For most covariates, the proportion of respondents with missing values was small (e.g., < 1–2%). For household income, the proportion was larger (13.7%), so we allowed missing values to represent one level of this measure. Missing values were more common for anxiety and depressive symptoms (e.g., 75 and 54 respondents with 2 + missing scale items for the GAD-7 and CES-D10, respectively), so we compared the distribution of caregiver characteristics by the presence/absence of missing data for these two outcomes. As missing values were also relatively more common for emotional/informational social support (n = 41), we showed final models with and without this measure.
Sensitivity analyses
We assessed whether accounting for clustering of caregivers within AL homes affected model estimates by including a robust sandwich variance estimator in our models [41]. Intra-cluster correlations were small (0.009 for anxiety, 0.000 for depression) and p values of the home-level covariance component were not statistically significant, suggesting no meaningful home-level clustering. We also conducted multiple imputation analyses using the fully conditional method and specifying a generalized logit distribution [42, 43].
All analyses were two-tailed with statistical significance defined as p ≤ 0.05. SAS version 9.4 (SAS Institute Inc., Cary, NC) was used to conduct all analyses.