The clinical sample (n = 108) was recruited as part of the LMU Biobank study and was composed of psychiatric inpatients from the Department of Psychiatry and Psychotherapy of the LMU University Hospital Munich. Participants indicated demographic information and filled out self-report questionnaires (order: CoPaQ, DASS-21, R-GPTS, WHO-5, UCLA, SNI, & BRS) using paper–pencil. Psychiatric inpatients with insufficient comprehension of German, an acute psychotic or manic episode, or acute suicidality were excluded from participation.
The non-clinical control sample was recruited online from the general German population using advertisements on social media (Facebook) and via university mailing lists. Assessments were made via a secure online survey software (LimeSurvey). This study is part of an ongoing longitudinal survey into the mental health consequences of the pandemic. The non-clinical sample completed the same questionnaire batterie, which was presented in a block randomised order to reduce carry-over effects and using a forced response format. At the end, participants were asked to enter their email address to be included in a prize draw. The sample consisted of adults (18+ years). In total, 387 (77.87%) identified as women, 108 (21.73%) as men, and 2 (0.40%) as diverse with an age range from 18 to 75 years (mean = 30, standard deviation (SD) = 11).
To obtain a more comparable case–control sample in terms of key sociodemographic factors, the clinical and non-clinical samples were matched on age, sex, and employment status using R software and the MatchIt (v4.1.0) package . Matching is preferable over sole adjustment of potential confounders in regression analyses since it increases sample comparability and efficiency of analyses as similar numbers of cases and controls are present across confounder strata . After matching, clinical and non-clinical samples were comparable in age and sex (age: t(212.56) = − 1.47, p = 0.142; sex: χ2(1) = 0.07, p = 0.785), but differences remained for employment status (χ2(6) = 27.22, p < 0.001).
Ethical approval and informed consent
The study was subject to ethics committee approval (clinical sample [Project Number: 18-716]; non-clinical sample [Project Number: 20-118]) and conducted in accordance with the Declaration of Helsinki . All participants provided informed consent. Recruitment in both study groups took place between April-December 2020.
Data integrity and quality control
Integrity of participants’ responses and data was ascertained in multiple pre-processing steps (see Supplementary Methods and Supplementary Fig. 1 for an overview).
COVID-19 Pandemic Mental Health Questionnaire (CoPaQ)
The CoPaQ (https://osf.io/3evn9/)  is a newly developed and highly comprehensive self-report measure assessing the psychosocial impact of the COVID-19 pandemic. For the purpose of this study, we included data of an index assessing the impact of COVID-19-specific stressors over the past 2 weeks from the CoPaQ. Individual stressors included among others quarantine/curfew, small accommodation/home-office, financial difficulties, childcare responsibilities, and physical health concerns; we provide a full list of items in Table 1 and Supplementary Fig. 2 depicts COVID-19-specific stressors inter-item correlations. Each stressor was rated using a 5-point Likert scale ranging from 0 (Not at all) to 4 (Very much) and participants’ responses of “Not applicable” were recoded as 0. A sum score of all items was calculated as an index of COVID-19-specific stressors with higher scores indicating a greater stressor impact. We observed an acceptable internal consistency of the COVID-19-specific stressors scores with McDonald’s Omega (ω) = 0.79 (95% confidence interval [CI]: 0.75–0.84). It is important to note, however, that stressors are likely to occur relatively independently, so a high internal consistency was not necessarily presumed.
Psychosocial outcome measures
We selected a diverse range of psychosocial outcome measures that have been reported to be of relevance during the current pandemic [12, 28,29,30,31]. This includes mental health symptomatology measures of stress, anxiety, depression, and paranoia; transdiagnostic mental health factor measures of loneliness and rumination; and positive psychological functioning measures of psychological well-being and resilience.
Mental health symptomatology
Depression, Anxiety and Stress Scales-21 (DASS-21)
The German version of DASS-21 [32, 33] was used to measure anxiety, depression, and stress during the preceding week. Items are rated on a Likert scale of 0 (did not apply to me at all) to 3 (applied to me very much or most of the time). Higher scores indicate greater levels on each of the respective subscales. In clinical and non-clinical samples good psychometric properties of the scales have been reported . In our study, DASS-21 subscale scores’ internal consistency ranged from good to excellent: ωAnxiety = 0.84 (95% CI: 0.79, 0.88), ωDepression = 0.93 (95% CI: 0.92, 0.95), and ωStress = 0.89 (95% CI: 0.86, 0.91).
Revised-Green et al. Paranoid Thoughts Scale (R-GPTS)
The total score of the German version of the 18-item R-GPTS [35, 36] that includes two subscales of ideas of reference (e.g., “People definitely laughed at me behind my back”) and ideas of persecution (e.g., “I was certain people did things in order to annoy me”) assessed over the past fortnight were used to measures paranoia. Items are rated on a 5-point Likert scale ranging from 0 (not at all) to 4 (totally). Scores can range from 0 to 72; higher scores indicate higher levels of paranoia. Excellent psychometric properties of the scales have been reported for the English version . In our study, the R-GPTS subscale scores ranged from good to excellent with ωPart A = 0.88 (95% CI: 0.85, 0.91) and ωPart B = 0.91 (95% CI: 0.88, 0.94).
Transdiagnostic mental health factors
Perseverative Thinking Questionnaire (PTQ)
The PTQ  consists of 15 items and is a self-report scale, which measures content-independent negative ruminative thinking. Items are rated on a 5-point Likert scale ranging from 0 (Never) to 4 (Almost always). Higher scores indicate higher levels of ruminative thinking and scores can range from 0 to 60. Good psychometric properties have been reported in previous research . In our study, the internal consistency of the PTQ was excellent ω = 0.97 (95% CI: 0.97, 0.98).
UCLA Loneliness Scale (UCLA)
The German version of the UCLA [38, 39] was used to assess loneliness. The intensity and frequency of feelings of loneliness are assessed with 20 items using a 5-point Likert scale ranging from 1 (not at all) to 5 (totally). Reversed items were recorded and then averaged to form a mean score, with higher scores indicating greater loneliness. The German version of the UCLA has been reported to show high internal consistency and discriminant validity . We observed an excellent internal consistency with ω = 0.93 (95% CI: 0.91, 0.94).
Positive psychological functioning
Brief Resilience Scale (BRS)
The German version of the six items BRS [40, 41] was used to assess resilience. Items are rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Reversed items were recoded to calculate mean scores with higher scores indicating greater resilience. Sound psychometric properties of the self-report questionnaire were reported in previous research . In our study, internal consistency was good with ω = 0.88 (95% CI: 0.85, 0.91).
WHO (Five) Well-Being Index (WHO-5)
Participants were asked to complete the German version of the WHO-5 [42, 43] which assesses well-being over the past 2 weeks. Items are rated on a 6-point Likert scale ranging from 0 (not present) to 5 (constantly present). Scores are summed, with higher scores indicating greater well-being. Good psychometric properties have been reported in previous research . We observed an excellent internal consistency with ω = 0.91 (95% CI: 0.89, 0.93).
All analyses were conducted in R (v4.0.3; R Foundation for Statistical Computing) with packages psych (v1.8.12) , lavaan (v0.6-3.1295) , careless (v1.1.3) , apaTables (v2.0.5) , MBESS (v4.8.0) , and missForest .
After conducting the different steps to ensure data integrity and quality (see Supplementary Fig. 1), we imputed missing values. Since we had continuous and categorical mixed-type data, missing data were handled by applying the non-parametric, iterative MissForest imputation, which is based on a random forest algorithm . Out-of-bag (OOB) estimates per sample for the imputation error were OOBPFC < 0.001 for the non-clinical and OOBPFC = 0.153 for the clinical sample.
First, internal consistency was calculated for the COVID-19-specific stressors index and all outcomes variables using McDonald's Omega [ω; 51] instead of Cronbach's α since assumptions are rarely met in practice [52; see “Measures”]. Descriptive statistics and the strength of statistical association between variables were tested using bivariate Pearson’s correlation coefficients, Chi-square tests (χ2), and unpaired two-sample t tests (Welch t test) when appropriate. We report magnitudes of effect sizes according to Cohen : correlation coefficients of 0.10 are considered “small”, those of 0.30 are “medium”, and those of 0.50 are “large” with 95% CI using 5000 bootstrapped samples with replacement.
Multiple linear regression analyses
We ran multiple linear regression analyses to evaluate associations of case–control status, COVID-19-specific stressors and their interaction with mental health outcomes in the matched sample. These regression analyses were conducted unadjusted and adjusted for age, sex, and employment status. All independent variables were standardised to facilitate interpretation of regression coefficients (βs) and main effects. In an additional step, we repeated regression analyses using psychosocial outcome variables on their original scale and standardising these variables; results for outcome variables on original scales are presented in Tables and Figures and results for standardised outcome variables are presented in the Results section to facilitate comparison to other manuscripts and between scales, respectively. To assess the robustness of results, also against violations of homoscedasticity, we provide 95% bootstrapped CI using 5000 bootstrapped samples with replacement. All hypothesis testing was two-tailed according to α = 0.05. R2 is reported when appropriate.
To explore the respective impact of COVID-19-specific stressors on the different psychosocial outcome variables and in clinical and non-clinical samples separately, we performed additional group-stratified multiple regression analyses, again adjusted for age, sex, and employment status. For these analyses, both dependent and independent variables were standardised to allow effect size comparisons of the COVID-19-specific stressors predictor between samples and outcome variables.
To analyse the robustness and consistency of results, we applied four sets of sensitivity analyses. First, the same multiple linear regression analyses were repeated in the larger sample (n = 605) that was not matched on age, sex, and employment status, but also adjusted for these variables. Second, we repeated our primary analyses in the matched sample by excluding COVID-19-specific stressor items related to ‘living in a small accommodation’, ‘office work’, ‘customer service’, ‘childcare’, ‘running school lessons’, and ‘employment uncertainties’. This was done to explore consistency of results for those COVID-19-specific stressors that applied equally well to community-dwelling individuals and psychiatric inpatients and, thus, are of relevance across contexts. Third, multiple linear regression analyses were repeated in the matched sample while additionally adjusting for essential work activity for the maintenance of critical infrastructure (i.e., participants were grouped into the following categories (a) health care worker, (b) essential worker but non-healthcare worker, and (c) non-essential worker) and, finally, in separate analyses we controlled for the date of assessment using a linear and quadratic effect of time in addition to the matching variables.