COVID-19 cumulative exposure
We used a COVID-19 exposure index developed, tested and validated in previous research [
17]. In brief, we adopted a methodology from the frailty research field. In a frailty index, individual deficits indicative of frailty are combined; where a score of 0 means no deficits and a score of 1 means all deficits are present. The index score is calculated by dividing the sum of items by the number of items [
20‐
22]. Likewise, to operationalize exposure to pandemic-related adversity, a 35-item COVID-19 exposure index was developed based on direct and indirect exposures to the COVID-19 pandemic. These included COVID-19 infection of participants and their relatives, consequences of COVID-19 infection, of social restrictions and pandemic-related pressure on the health care system. For a full overview of items, see supplementary file 1. For each item, participants scored 0 (no) or 1 (yes). We calculated a cumulative exposure index based on the two COVID-19 measurements. If a participant tested positive for COVID-19, was admitted to hospital or intensive care unit due to COVID-19, their score on the respective item would be 2, regardless of whether they reported this once or twice as we considered COVID-19 disease as having more impact than the other items and it was less likely to be reported twice compared to other items. For the remaining items, we calculated the sum score of the items (0, 1 or 2). Next, we summed all scores, resulting in a combined COVID-exposure index with a range of 0–70 [
17]. Although this index was developed to use among Dutch older adults, the approach could be of use in general and non-Western populations.
Mental health outcomes
Depressive symptoms were assessed with the Centre for Epidemiologic Studies Depression scale (CES-D, short version, 10-item scale). This self-report questionnaire measuring depressive symptoms in the general population has good psychometric properties and validity in older populations. The sum score ranges from zero to 30, with higher scores indicating more severe depressive symptoms. For the 10-item variant a cut-off score of ≥ 10 is used to determine a probable depression [
23‐
25]
Anxiety symptoms were measured with the Hospital Anxiety Depression Scale—Anxiety subscale (HADS-A) [
26]. The HADS-A subscale consists of seven items for measuring anxiety. The sum score ranges from zero to 21, with higher scores indicating more severe anxiety. The HADS-A has a cut-off of eight or higher to indicate clinically relevant anxiety.
Loneliness was assessed using the 11-item De Jong Gierveld scale [
27]. Loneliness scores range from zero to 11, higher scores indicate more severe loneliness. A score of three or higher indicates loneliness.
In the statistical models the three mental health outcomes were used as continuous variables.
Potential protective factors
Following literature on resilience in diverse contexts, we included a range of potential protective factors that cover somatic, lifestyle, social, psychological, and socioeconomic domains [
28,
29].
Functional limitations were assessed by seven activities: going up and down the stairs 15 steps without stopping, using public transportation on their own, cutting own toenails, getting (un)dressed, sitting down, and standing up from a chair, walking outside for five minutes without stopping, taking shower or bath on their own. Participants could answer on a 5-point Likert scale (without difficulty, with some difficulty, with much difficulty, only with help, cannot). Participants that reported at least some difficulties were considered limited for the activity. The number of activities with limitations was counted (number of items with at least some difficulty or worse, range zero to seven).
COVID-19 vaccination status was assessed with the question if the participant was vaccinated (no/yes) in the 2021 survey.
Frequency of praying/meditating was measured with a seven-point Likert scale (never, less than once a month/once a month, a few times a month, once a week, a few times a week, once a day, multiple times a day). In the analyses praying was categorized into never, up to once a day and more than once a day.
The personality trait of neuroticism, or emotional stability, was measured with an abbreviated (15 items) version of the NPV, the Nederlandse Persoonlijkheids Vragenlijst (DPQ, the Dutch Personality Questionnaire). Higher scores indicate less emotional stability [
30].
Mastery was measured with the seven-item Pearlin Mastery Scale [
31]. Sum scores could range from seven to 35. Higher scores indicate a stronger internal locus of control reflecting the perception that events in one's life relate to one's own actions rather than to external sources like powerful other persons, institutions, or circumstances.
Partner status was measured with the question if the participant had a partner (no/yes).
Network size was defined as the number of network members (≥ 18 years) with whom the participant had important/frequent contact and measured with a network delineation methodology described in more detail by Cochran et al. [
32]. Network size ranged from zero to 79.
Internet use for social contact was measured with the question if the participant used internet to keep contact with other people (no/yes). Participants who did not have internet access or a device were categorized non-users.
Highest completed educational level was asked in nine categories, which were recoded to the nominal number of years it takes to complete that level (range 5–18 years).
Monthly household income was measured in 25 categories, ranging from 0–453 euros (lowest income category) to 5446 or more euros (highest income category). We recoded the categories to their median values, divided the median values by 500 and used this as a continuous variable.
All potential protective factors, except COVID-19 vaccination status, were available in the last regular LASA measurement. In the two COVID-19 surveys protective factors were not assessed.
Statistical analyses
Baseline characteristics are presented as means with their standard deviation for continuous variables and as percentages for categorical variables. Analyses were conducted separately for depressive symptoms, anxiety symptoms and loneliness.
To answer the first research question, we used paired t-tests to describe and test average changes in depressive symptomatology, anxious symptomatology, and loneliness between the three included time points (i.e., the pre-pandemic time point (2018/19), and the first and second COVID-19 time points).
To answer the second research question, we estimated the effect of cumulative COVID-19 exposure on change scores of depressive symptoms, anxiety symptoms and loneliness using linear regression (model 1). Because our study focused on cumulative COVID-19 exposure and longer-term changes in mental health, in this analysis we used change scores between the pre-pandemic time point and the second COVID-19 time point. Positive change scores indicated worsening of mental health and negative scores indicated improvement. In model 2, we adjusted for pre-pandemic mental health.
To answer the third research question, we estimated main effects of protective factors adjusted for pre-pandemic mental health and all other protective factors (model 3). Unstandardized B’ s are shown for clinical interpretation and standardised B’s are shown to enable comparing outcomes (significance p < 0.05).
To answer our fourth research question, we examined buffering effects of protective factors. For this, we included a COVID-19 exposure index-by-protective factor interaction term, in separate models for each protective factor. If the interaction term indicated that the expected association between COVID-19 exposure and change in mental health was weaker in persons in whom the protective factor was present, this was interpreted as evidence for a buffering effect (significance p < 0.10) [
33].
For the second, third and fourth research question we performed (nonhierarchical) multiple linear regression analysis.There were missing data on at least one of the potential protective factors or the exposure indices in 13.5% of the participants: functional limitations (0.2%), COVID-19 vaccination status (1.0%), internet use (1.1%), mastery (1.3%), network size (1.4%), COVID-19 exposure index at the second COVID-19 questionnaire (1.7%) and first COVID questionnaire (3.1%), neuroticism (3.3%) and income (4.6%). Missing data were handled using multiple imputation (predictive mean matching, 100 iterations, 14 imputations). Because imputing all items for COVID-19 exposure indices directly was not feasible, we used passive imputation; in our imputation model, the items of each exposure index were predicted by the other items from that exposure index, the total score of the other exposure index and all covariates, potential protective factors, and outcomes. We did not impute missing data on depressive symptoms, anxiety symptoms or loneliness as these were the outcomes of our study but included these as auxiliary variables in the imputation model. Analyses were conducted with IBM SPSS version 26.0 and the mice package in R version 4.0.3 [
34].