Study design and study population

This is a single-centre, longitudinal COVID-19 ancillary study performed with the well-established Kardiovize population-based sample representing 1% of the inhabitants of the city of Brno, Czech Republic.^{Reference Novotný, Gonzalez-Rivas, Kunzová, Skladaná, Pospíšilová and Polcrová12,Reference Movsisyan, Vinciguerra, Lopez-Jimenez, Kunzová, Homolka and Jaresova14} Participants of the Kardiovize study were selected randomly from databases of health insurance companies, to represent a probabilistic sample of the general population. Detailed descriptions of the recruitment processes for the Kardiovize study has been published elsewhere.^{Reference Movsisyan, Vinciguerra, Lopez-Jimenez, Kunzová, Homolka and Jaresova14} The entire Kardiovize study cohort was invited by email to participate in the COVID-19 add-on study.^{Reference Novotný, Gonzalez-Rivas, Kunzová, Skladaná, Pospíšilová and Polcrová12} The inclusion criterion for the analysis of the COVID-19 add-on study results was the availability of data on stress and depressive symptoms at baseline pre-COVID-19 and at least one of the COVID-19 waves. This inclusion criterion was used to ensure that all of the analysed data were longitudinal, while avoiding intercept and growth slope bias caused by a larger pre-COVID-19 sample. Although all citizens and permanent residents in the Czech Republic are registered with health insurance companies, the selection bias related to the most disadvantaged of society (for example, not responding to the invitation, not having or regularly using an email address, etc.) cannot be excluded, but has been estimated to be minimal. The COVID-19 add-on study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.^{Reference von Elm, Altman, Egger, Pocock, Gøtzsche and Vandenbroucke15}

Procedure

In the COVID-19 add-on study, we examined mental health in the same population-based sample before the COVID-19 pandemic and during the three consecutive COVID-19 waves (Fig. 1). Mental health data were collected (a) before the COVID-19 pandemic, in 2014 and 2015; (b) during the first COVID-19 wave, between 23 April and 27 May 2020; (c) during the second COVID-19 wave, between 31 August and 23 October 2020; and (d) during the third COVID-19 wave, between 4 March and 7 April 2021. All participants completed the Kardiovize COVID-19 e-questionnaire (in Supplementary Appendix 1 available at https://doi.org/10.1192/bjo.2023.620), which includes a battery of psychological scales, through an online survey administered with validated RedCap software (version 10.3.8, REDCap Consortium, Vanderbilt University, TN, USA; see https://www.project-redcap.org/).

Fig. 1

Timeline of the COVID-19 pandemic and restrictive measures in the Czech Republic. The lines show trends in daily numbers of new SARS-CoV-19 cases, SARS-CoV-19-related hospital admission and deaths over the examined time period of the pandemic (data are projected on a square-root-transformed scale). The upper horizontal bars show the time periods when key epidemiological measures were imposed. The grey sections indicate individual waves of the COVID-19 pandemic. The indicators above the plot show the time periods (start and end) of data collection in each COVID-19 wave. Source of the data: Ministry of Health of the Czech Republic (https://onemocneni-aktualne.mzcr.cz/covid-19). WHO, World Health Organization.

The primary outcomes were the stress levels, measured with the Perceived Stress Scale (PSS),^{Reference Cohen, Williamson, Spacapan and Oskamp16} and the severity of depressive symptoms, measured with the Patient Health Questionnaire (PHQ).^{Reference Löwe, Wahl, Rose, Spitzer, Glaesmer and Wingenfeld17} The PSS is a ten-item scale scored with a five-point Likert scale (0 ‘never’ to 5 ‘very often’), with a possible range of 0–40. Stress levels were categorised as low (score of 0–13), medium (score of 14–26) or high (score of 27–40). The reliability of the Czech version of the PSS in a single-factor approach was ɑ = 0.91.^{Reference Figalová and Charvát18} The cross-sectional McDonald's omega was ω = 0.90, 0.90 and 0.91 in the first, second and third waves of the study, respectively, and the longitudinal stability (intraclass correlation coefficient (ICC)) was 0.88. The presence and severity of depressive symptoms were assessed with the two identical items from the PHQ-9 (before the COVID-19 pandemic) and PHQ-4 (during the COVID-19 pandemic). Responses were measured with a four-point Likert scale (0 ‘not at all’ to 3 ‘almost every day’), with a possible range of 0–6 points. Despite its brevity, the PHQ-4 has proven to be a valid tool for rapid assessment of the general presence of depressive symptoms.^{Reference Löwe, Wahl, Rose, Spitzer, Glaesmer and Wingenfeld17,Reference Mendoza, Frondozo, Dizon and Buenconsejo19–Reference Christodoulaki, Baralou, Konstantakopoulos and Touloumi21} Reliability in the original study was ɑ = 0.78; the cross-sectional Spearman–Brown composite reliability was r* = 0.82, 0.83 and 0.82 in the first, second and third waves of the study, respectively; and the longitudinal stability (ICC) was 0.74. Depressive symptoms were considered present if the sum of the score of the two PHQ items was ≥3. Since stress, depression, anxiety and other related terms represent different negatively connoted aspects of mental health and the general terminology is not fully established, we used the term mental health difficulties to collectively refer to moderate-to-high stress and increased severity of depressive symptoms, as an expression of the negative effects of these terms on mental health.^{Reference Sultana, Khan, Hossain and Hasan22–Reference Hendrikx, Williamson, Baumann and Murphy24}

To identify factors affecting mental health, we collected information on several primary and secondary stressors, using well-established measurement tools. Primary stressors are factors that are inherent and arise directly from a negative event (such as fear of contracting COVID-19 and its possible health consequences such as death, etc.). In contrast, secondary stressors are non-inherent to the event. They arise partially from the consequences of this negative event (such as lockdowns, separation from loved ones and disruption of finances) and partially from longer-term socioeconomic disadvantage and lack of resilience to adversities, etc.^{25–Reference Lock, Rubin, Murray, Rogers, Amlôt and Williams27} Potential risk factors were selected by the Kardiovize researchers based on literature review. A brief review of the literature supporting the inclusion of each selected factor is presented in the Supplementary Table 1.

Primary stressors included cognitive and emotional perception of COVID-19, assessed with the Brief Illness Perception Questionnaire (B-IPQ).^{Reference Basu and Poole28} This eight-item scale uses a ten-point Likert scale (1–10 with higher scores indicating a more threatening perception of the illness), with a possible range of 10–70, and measures cognitive and emotional perception of the COVID-19 disease as threatening across different domains. The cross-sectional reliability in the first, second and third waves of the study, respectively, was ω = 0.51, 0.57 and 0.54 for cognitive perception and ω = 0.60, 0.58 and 0.61 for emotional perception of COVID-19 disease. These values are acceptable given that the B-IPQ measures different aspects of COVID-19 perception at the item level, which increases the heterogeneity of higher-order factors.^{Reference Cronbach29,Reference Osburn30} The longitudinal stability (ICC) was 0.71 for cognitive perception and 0.87 for emotional perception of COVID-19 disease. In agreement with the meta-analytic study about the B-IPQ,^{Reference Broadbent, Wilkes, Koschwanez, Weinman, Norton and Petrie31} we consider the validity and reliability of the instrument to be sufficiently established. A higher B-IPQ score implies that COVID-19 illness is perceived as more threatening.

Secondary stressors included feelings of loneliness, lack of resilience and the effects of the COVID-19 pandemic and associated measures on finances and work, social contacts, sleep and physical activity. The feeling of loneliness was assessed with the UCLA 3-Item Loneliness Scale.^{Reference Hughes, Waite, Hawkley and Cacioppo32} Responses are scored on a three-point Likert scale (1 ‘seldom’ to 3 ‘often’), providing a total score from 3 to 9 points. The UCLA 3-Item Loneliness Scale showed a reliability of ɑ = 0.89–0.94 in the original study^{Reference Russell33} and ω = 0.75, 0.79 and 0.77 in the first, second and third waves of this study, respectively, with a longitudinal stability (ICC) of 0.81. Higher scores indicate greater feelings of loneliness. Presence of the feeling of loneliness was defined as score ≥6.

Resilience was assessed with the Connor–Davidson Resilience Scale.^{Reference Vaishnavi, Connor and Davidson34} This short two-item instrument with a five-point Likert scale (0 ‘not at all’ to 4 ‘almost always’) and a score range of 0–8, demonstrated very good validity and reliability, with ɑ = 0.79.^{Reference Vaishnavi, Connor and Davidson34} The reliability was r* = 0.70, 0.72 and 0.70 in the first, second and third waves of the study, respectively, and the longitudinal stability (ICC) was 0.84. Presence of resilience was defined as low (score of 0–5), medium (score of 6–7) or high (score of 8).

The impact of the COVID-19 pandemic on finances, work, social contacts, sleep and physical activity was measured with the Kardiovize COVID-19 e-questionnaire (Supplementary Appendix 1). The items in this section of the e-questionnaire were designed by the Kardiovize researchers based on a literature review. This section has not been validated. All participants also provided general demographic and health data.

Statistical analysis

Missing values correspond only to respondents who dropped out in each wave. No imputation of missing values was applied. The nature of the missing data was examined with a correlation analysis of the relationship between the presence of missing values (using a binary matrix representing missingness) and the observed values. The results showed that the data are missing completely at random, as there was no relationship between missingness and the observed values (mean absolute correlation r = 0.046, maximum absolute correlation r = 0.2; all P > 0.05). No sensitivity analysis was performed.

Cross-sectional reliability of the instruments used in each wave of the study was verified with McDonald's omega (ω) and the Spearman-Brown composite reliability (r*) for the two-item scales. Long-term stability was verified with the ICC. The Wilcoxon rank-sum test and the chi-squared test were used to verify representativeness of the COVID-19 study population. Differences in the prevalence of mental health difficulties were verified with the G-test, with Benjamini–Hochberg correction for multiple comparisons.

Longitudinal changes in the stress levels, severity of depressive symptoms and the effects of different secondary stressors were analysed with latent growth curve models (LGCMs). LGCM simulates change as a function of time, which is represented by the latent variables, referred to as growth factors.^{Reference Ferrer and McArdle35–Reference Muthén and Curran37} A latent intercept and a latent slope (i.e. the growth factors) are estimated based on the individual trajectories. Growth factors provide an estimate of the average trajectory and the individual variation around that trajectory over time. These parameters provide an insight into the average change and the individual difference surrounding that change. The inclusion of additional time-variant and -invariant covariates then allow us to assess the effects of these stressors on the primary variable and the possible changes in its growth curve. Compared with analysis-of-variance methods, the LGCM also allows for the inclusion of respondents with missing values at some time points in the study (if there is more observed information than estimated information, i.e. a sufficient number of cases have at least three repeated measures), thereby increasing the overall N and the power of the results.^{Reference Curran, Obeidat and Losardo38}

As part of the analysis, we first modelled four types of growth (linear, quadratic, cubic and relative change) for the two primary outcomes, and compared their overall fit. Linear growth was specified by constraining the loadings of the latent growth parameters on the observed outcomes to assume incremental change per increase in unit time. Quadratic and cubic growths were modelled by fixing factor loadings of the quadratic and cubic terms to assume exponential change. Growth based on relative change was modelled by fixing the first loading to 0 and the last loading to 1, with the other loadings being freely estimated by the growth function itself. By convention, the intercept and slope parameters were allowed to co-vary.^{Reference Bollen and Curran39–Reference Tu and Gilthorpe41} Model fits were evaluated primarily based on the Akaike information criterion/Bayesian information criterion, with further checking of Comparative Fit Index, Tucker–Lewis index and root-mean-square error of approximation values. We then analysed the selected growth model in the context of the primary outcomes, stress levels and severity of depressive symptoms. Finally, we tested the effect of secondary stressors on the observed growth curve of both primary outcomes by adding the stressors as covariates to the LGCM. To keep the model parsimonious, only stressors with significant influence were retained in the final model. We also verified that the fit of the final model is better than the base full model with all predictors included (Supplementary Table 2). Given the non-normal distribution of some stressor scores, we have used and report a robust estimator (maximum likelihood estimation with Huber–White s.e.). All two-sided P < 0.05 were considered to be significant. Data analysis was performed with RStudio (version 2022.02.3, R version 4.2.0 for Windows, Posit, PBC, Boston, MA, USA; see https://posit.co/download/rstudio-desktop/).

Ethics statement

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human patients were approved by the Internal Review Board and the St. Anne's University Hospital ethics committee (reference number 28 V/2019, permission granted 28 April 2020). All participants provided written informed consent to participate in this study.