Untangling Potential Links between Childhood Trauma and the Psychological Response to the COVID ‐ 19 Pandemic: A Prospective Study Testing Multiple Mediation

Childhood maltreatment (CM)has been associated with adverse psychosocial outcomes during the pandemic, but the underlying mechanisms are unclear. In a prospective online study using baseline and 10-week follow-up data of 391 German participants, we applied multiple mediation analyses to test to what extent COVID-19 perceived stressors mediate the association between CM and later adverse psychosocial outcomescompared to established mediators of rumination and insecure attachment. We also explored the relative importance of different COVID-19 related stressors in predicting adverse psychological trajectories using elastic net regression.Results showed that CM was longitudinally associated with all adverse psychosocial outcome. COVID-19 perceived stressors, rumination, and insecure attachment mediated this relationship and full mediation was observed for the outcomes anxiety, stress and psychological well-being. COVID-19-related concerns about the future wasmost strongly and consistently associated with adverse psychosocial functioning.These ndings provide preliminary evidence that COVID-19 perceived stressors, in particular concerns about the future, may be a key mechanismunderlying the development of adverse psychosocial outcomes in individuals with a CM history. Thus, COVID-19 perceived stressors may require a higher priority for prevention and treatment efforts in vulnerable groups.Our results warrant replication in more representative cross-cultural samples. Dependent Variable; CI Condence Interval (bootstrapped); Standardised. are total, total indirect, specic indirect, and direct effects of the different multiple mediation models.


Introduction
It is obvious that the COVID-19 pandemic and its associated social restrictions places an exceptional strain on individuals leading to a deterioration of mental health and well-being world-wide (1). Speci cally, the pandemic has resulted in unprecedented major stressors that can pose enormous psychological challenges including a virtual standstill of our public and private lives, anxieties about getting infected, the course of disease, and receiving appropriate medical care as well as but not limited to job uncertainties and nancial di culties. In a recent study, we could show that a greater impact of such COVID-19-speci c stressors during the pandemic was associated with increased psychological di culties in a German general population sample (2). Moreover, representative cohort studies comparing changes in individuals before versus in the rst few weeks of the initial lockdowns have suggested signi cant increases in mental health symptomatology (3,4), also summarised in a recent meta-analysis (5). In addition, longitudinal studies have identi ed heterogeneous trajectories of mental health symptomatology during the pandemic. Here, younger age, female sex, lower income levels, economic inactivity, and pre-existing mental health conditions have been associated with worse longitudinal psychological trajectories in terms of depression, anxiety, and loneliness (e.g., [6][7][8][9][10][11]. This demonstrates the importance of inter-individual differences in mental health trajectories and emphasises that identi cation of important risk factors and accompanying underlying mechanisms is key, which could allow for targeted care or prevention approaches. One group of individuals that may be particularly vulnerable to the effects of the COVID-19 pandemic on mental health are those who experienced childhood maltreatment (12). Childhood maltreatment (CM), which includes traumatic experiences of abuse and neglect, is arguably the most consistent transdiagnostic risk factor across psychiatric disorders and lower psychological well-being as shown in in multiple retrospective case-control and longitudinal studies (e.g., [13][14][15][16][17][18][19]. In the context of the COVID-19 pandemic, CM exposed individuals showed greater mental health di culties in terms of symptoms of anxiety, depression, and posttraumatic stress disorder (PTSD), compared to non-exposed individuals in initial cross-sectional (20)(21)(22)(23) and longitudinal (24) studies. However, not all CM exposed individuals develop mental health di culties in adulthood and, so far, little is known about the exact pathways through which CM leads to an increased mental health risk (14). In order to improve treatment or even prevent an adverse mental health cascade during the current pandemic, it is therefore crucial not only to identify vulnerable groups by environmental strati cation based on CM criteria, but also to deepen our understanding of potential core mechanisms linking psychopathology to CM.
Stress sensitisation by early exposure to CM has been proposed as a key transdiagnostic mechanism leading to the evolvement of later psychopathology (25)(26)(27)(28). In the current pandemic, for instance, CM exposed individuals may be sensitised and particularly reactive to stress, which could lead to the perception that COVID-19-related stressors are particularly stressful. In turn, this could increase levels of adverse psychosocial outcomes. We are aware of only one longitudinal study amidst the current COVID-19 pandemic that showed that perceived stress mediated the association of early life adversity and depressive symptom severity in adolescents (29). Yet, this study did not differentiate stressors speci c versus unspeci c to the COVID-19 pandemic. Further, it is unclear whether this mediation generalises to mental health conditions other than depression, and if it also occurs in adults. Finally, it is unclear to what extent perceived stress still plays a relevant mechanistic role when compared to established transdiagnostic mediators between CM and mental health such as rumination (e.g., 14,30-32) and insecure attachment (33,34). This can be tested using multiple mediation analyses adjusting for important confounding factors such as age, sex, income, educational attainment, and pre-existing mental health conditions. This prospective study in individuals from the general population aims at investigating the relationship between CM and subsequent psychopathology and psychological well-being as well as the relative mediation via COVID-19 perceived stressors, rumination, and insecure attachment. Based on previous research and the theoretical considerations described above, we hypothesised i) that CM is associated with more adverse psychosocial outcomes in terms of depression, anxiety, stress, loneliness, paranoia, and psychological well-being, and ii) that these associations are mediated by COVID-19 perceived stressors, rumination, and insecure attachment. Understanding the factors linking higher rates of mental health di culties during the current pandemic to CM can inform the development of targeted prevention and psychosocial treatment efforts. Since not all COVID-19 stressors may be equally important in predicting adverse psychosocial outcomes, we further explored their relative importance.

Page 4/25
A longitudinal survey was conducted in adults (18 + years old) with varying levels of CM who were recruited online via social media platforms and university mailing lists. The secure online survey software (LimeSurvey) was used for assessments, which was set up using a forced response format to prevent missing data and questionnaire block randomisation to circumvent potential carry over effects. As a reimbursement participant could be included in a prize draw by entering their email address at the end of the survey. Recruitment and initial assessment took place between April 2020 -May 2021 with follow-up assessment 10 weeks after the initial assessment.
All participants provided informed consent prior to participation, the study was conducted in accordance with the Declaration of Helsinki (35)  Measures Exposure (at baseline) Childhood Trauma Questionnaire (CTQ) CM was assessed by the CTQ self-report questionnaire that comprises ve subscales: emotional, physical, and sexual abuse as well as emotional and physical neglect (36; German version: 37). Each subscale consists of ve items, which are rated on a 5-point Likert scale ranging from 1 (not at all) to 5 (very much). Reversed items were recoded. The subscale scores were calculated to represent a total score. Scores on the total scale range from 25 to 125 with higher scores indicating more severe CM. In previous research good psychometric properties of the questionnaire have been reported (36) and internal consistency of the total score at baseline was excellent in the present study (Cronbach's α = 0.92).
Proposed Mediators (at baseline)

COVID-19-speci c stressor impact index
The COVID-19-speci c stressor impact index of the COVID-19 Pandemic Mental Health Questionnaire (CoPaQ) (38) was used to assess COVID-19 perceived stressors over the past two weeks. The subscale includes different COVID-19 stressors (e.g., quarantine/curfew, small accommodation/home-o ce, nancial di culties, childcare responsibilities, and physical health concerns), which are rated on a 5-point Likert scale ranging from 0 (not at all) to 4 (very much). The "not applicable" answer option was recoded as zero. Scores are summed to form a total score, which can range from 0 to 56 and higher scores indicate greater COVID-19 perceived stressors. Preliminary psychometric evaluation of this subscale was reported to be sound (39).

Perseverative Thinking Questionnaire (PTQ)
Rumination was assessed with the PTQ (40), which consists of 15 items rated on a 5-point Likert scale ranging from 0 (Never) to 4 (Almost always). Items on content-independent negative ruminative thinking can be summed to a total score, which can range from 0 to 60. Higher scores indicate higher levels of ruminative thinking. Ehring et al. (40) reported good psychometric properties of the scale and we observed excellent internal consistency (α = 0.96).

Relationship Styles Questionnaire (RSQ)
We used the RSQ (41; German version: 42) to measure insecure attachment styles. The scale is comprised of 30 items, which are rate on a 5-point Likert scale ranging from 1 (not at all) to 5 (very much). Reversed items were recoded. Attachment avoidance and attachment anxiety were de ned as We used the DASS-21 to measure levels of depression, anxiety, and stress during the preceding week (41; German version: 42). Items are rated on a 4-point Likert scale of 0 (did not apply to me at all) to 3 (applied to me very much or most of the time). Subscale scores can each range from 0 to 28, respectively, and higher scores indicate greater levels of psychopathology. In clinical and non-clinical samples good psychometric properties of the scales have been reported (46). In our study, internal consistency estimates ranged from acceptable to excellent for each subscale (α Depression = 0.93, α Anxiety = 0.79, and α Stress = 0.89).

Revised-Green et al Paranoid Thoughts Scale (R-GPTS)
Paranoia over the past fortnight was assessed with the total score of the 18-item R-GPTS (44; German version: 45). 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 scale have been reported (47). The German version was translated from the original English version following common guidelines for forward and backward translation (49). The nal version was approved by one of the authors of the original version (D.F.). At T2, internal consistency of the total score was excellent (α = 0.91).

UCLA Loneliness Scale (UCLA-LS)
Loneliness was assessed using the UCLA-LS (47; German version: 48), which includes 20 items rated on a 5-point Likert scale ranging from 1 (not at all) to 5 (totally). Items that were reversed were recoded. The average score was built to represent the mean, with higher scores indicating greater loneliness. Good psychometric properties of the scale have been reported in previous research (51). In our sample, internal consistency was excellent at the second assessment timepoint (α = 0.94).

WHO (Five) Well-Being Index (WHO-5)
Psychological well-being was assessed with the WHO-5 (49; German version: 50). Five items are rated on a 6-point Likert scale ranging from 0 (at no time) to 5 (all of the time). Items were summed to represent the total score, which can range from 0 to 25, higher scores indicate greater well-being. Sound psychometric properties of the scale have been reported in previous research (54). In our sample, internal consistency was excellent at follow-up (α = 0.90).
First, bivariate Pearson's correlation coe cients and Chi-square tests (χ 2 ) were conducted to test associations between variables of interest.
Second, multiple mediation analyses using the maximum likelihood estimator were performed. CM was used as predictor variable and depression, anxiety, stress, loneliness, paranoia, and psychological wellbeing as outcome variables in separate mediation analyses. Mediator variables of COVID-19 perceived stressors, rumination, and attachment (anxious-and avoidant) were included simultaneously in the analyses and were allowed to correlate with each other. Standard errors were calculated using 10,000 bootstrap samples because some variables did not fully adhere to a normal distribution. We report biascorrected 95% bootstrapped con dence intervals (CI) for the total (c), direct (c'), and indirect effects.
Subdivision of the total indirect effect to speci c indirect effects allowed comparisons of the standardised effect sizes of each mediator. Standardised ordinary least squares regression coe cients are reported for all paths. To account for potential in uences of age, sex, income, educational attainment, and pre-existing mental health conditions, variables were included as covariates in each multiple mediation model. Of note, since multiple comparisons were performed we controlled for false positive rate by using the Benjamini-Hochberg procedure (59).
Finally, to explore the relative importance of each of the different perceived COVID-19 stressors contributing to the total COVID-19 stressor index, we performed elastic net regression analyses (60) to explore the relative predictive value of individual stressors for adverse psychosocial functioning. This technique is an extension of ordinary least squares regression that better accounts for collinearity between baseline predictors and simpli es the statistical model using regularisation. It includes the two hyperparameters α (tuning parameter of 0 to 1, which controls the type of shrinkage and, thus, the estimation method) and λ (penalty parameter of 0 to 1, which controls the amount of shrinkage with higher values leading to greater penalisation). The optimal hyperparameter combination was determined using grid search by selecting the model with smallest root mean squared error (RMSE) within 10-fold cross-validation. Of note, regularisation with α = 1 equals Least Absolute Shrinkage and Selection Operator (LASSO) regression (61) and α = 0 equals Ridge (62) regression; α values between 0 to 1 re ect the relative balance between the two regression models. Elastic net regression models were controlled for age and sex.

Sample characteristics
Six hundred sixty-eight participants completed the survey at baseline (T1) and 429 at the 10-week followup assessment (T2). To ensure high data quality, we excluded participants at baseline who answered more than 1 bogus item incorrectly (e.g., not checking "very much" for the item "Please, indicate 'very much'") (n = 58). In addition, participants with response times less than 25 minutes at baseline (n = 8) and less than 15 minutes at follow-up (n = 8) were excluded, which were deemed unlikely response times on the basis of personal experiences and response time descriptive statistics (baseline response time: median = 48 min, 1st quartile = 38 min, 3rd quartile = 61 min; follow-up: median = 23 min, 1st quartile = 29 min, 3rd quartile = 40 min). We also removed 30 participants who did not have a matching id variable between baseline and follow-up assessment, which resulted from a rare failure of the software. Taken together, this led to a nal sample of 391 individuals (77.43% females) on which analyses are based (age: mean = 30.99, standard deviation[sd] = 11.52). The nal sample consisted of 91.82% participants who indicated German nationality, 47.83% were single (see Table 1 for more demographic and clinical characteristics). Of note, participants only completing the baseline assessment did not differ signi cantly from the follow-up sample in terms of age, sex, nationality, employment status, marital status, and preexisting mental health conditions (p > 0.05; see Supplementary Table 1).

Multiple mediation models
Descriptive statistics of proposed exposure, mediator, and outcome variables can be found in Table 2. As a prerequisite for mediation analyses, we observed bivariate correlations between all proposed exposure and mediator variables (path a), mediator and outcome variables (path b), and exposure and outcome variables (path c) as shown in Supplementary  Note. SD = standard deviation. IQR = inter quartile range.
In separate multiple mediation models and as indicated by the total indirect effect, we observed evidence supporting overall mediation of effects of CM on the different outcome variables (i.e., depression, anxiety, stress, paranoia, loneliness, and psychological well-being) via COVID-19 perceived stressors, rumination, and attachment (anxious and avoidant) (see Table 3 and Fig. 1). Speci c indirect effects of the proposed mediators were signi cant for COVID-19 perceived stressors, rumination, and attachment avoidance across outcome variables. Attachment anxiety did not contribute additionally to the indirect effect in ve out of six separate multiple mediation analyses; except paranoia. Full mediation between CM and anxiety as well as stress was observed as the direct effect (c') was no longer signi cant after inclusion of the proposed mediators. Partial mediation was observed for the relationship between CM and depression, paranoia, loneliness, and psychological well-being since the direct effect remained signi cant. This suggests that our proposed set of mediators did not fully explain this relationship. Of note, all results from multiple mediation analyses were adjusted for the potential confounders age, sex, income, educational attainment, and pre-existing mental health conditions and the mediators were allowed to correlate with each other. When applying alpha correction for multiple testing, results remained largely unchanged but the speci c indirect effect of attachment avoidance was no longer signi cant for the outcomes stress and paranoia. Moreover, full mediation was observed for psychological well-being (see Table 3). Note. DV = Dependent Variable; CI = Con dence Interval (bootstrapped); Std. = Standardised. Depicted are total, total indirect, speci c indirect, and direct effects of the different multiple mediation models. P corrected = false discovery rate corrected p value.

Elastic Net Regression
We explored the relative variable importance of different COVID-19 stressors in predicting psychosocial outcomes using elastic net regression and results from the selected models are depicted in Fig. 2. Of note, the selected elastic net regression model for outcome loneliness was ridge regression (α = 0), so none of the predictor variables were set exactly to zero, and LASSO (α = 1) for outcomes stress, paranoia, and wellbeing. The hyperparameter α was in between 0 to 1 for outcomes depression and anxiety re ecting the optimal balance between ridge and LASSO regression selected for these outcomes during cross-validation. Of note, analyses were adjusted for age and sex. See Supplementary Table 3 for hyperparameters of selected models.
A few COVID-19 stressors were rather consistently associated with adverse outcomes. Among them were "the current pandemic" (stressor 1), "worries of not being able to get medical care" (stressor 10), "increased con icts with people close to me" (stressor 11), and "fears of what the future will bring, or that I won't be able to cope with everything" (stressor 14). In contrast, stressors such as "living in a small accommodation" (stressor 2), "being in home o ce" (stressor 7), and "uncertainties regarding my job, training place, studies or school" (stressor 13) were consistently unpredictive of adverse psychosocial outcomes. For some outcomes, the impact of COVID-19 perceived stressors was predominately associated with a few stressors that explained most of the variance. For example, levels of psychological well-being were only predicted by three COVID-19 stressors, namely "the current pandemic" (stressor 1), "increased con icts with people close to me" (stressor 11), and "fears of what the future will bring, or that I won't be able to cope with everything" (stressor 14). For other psychosocial outcomes such as anxiety and loneliness, variable importance was rather equally distributed across stressors.

Discussion
This prospective study scrutinised the pathway between CM and multiple key psychosocial outcomes (depression, anxiety, stress, paranoia, loneliness, and psychological well-being) during the COVID-19 pandemic regarding potential mediators: 1) COVID-19 perceived stressors, 2) rumination and 3) insecure attachment while adjusting for important confounders of age, sex, income, educational attainment, and pre-existing mental health conditions. Our ndings showed that COVID-19 perceived stressors may be a robust mechanism mediating the effect between CM and adverse psychosocial outcomes during the current pandemic. We also observed that rumination and insecure attachment (particularly attachment avoidance) acted as important additional mediators of the relationship between CM and psychosocial outcomes. The relative contribution of each potential mediator suggested that rumination was most strongly associated with adverse psychosocial outcomes. Of note, full mediation was observed for anxiety, stress and psychological well-being but not depression, paranoia, and loneliness. Taken together, our study adds to the accumulating research that highlights the importance of considering early adverse experiences when evaluating the psychological response to the current pandemic (21-23, 28, 62). In addition, our ndings suggest two important modi able therapeutic targets, i.e. rumination and COVID-19 perceived stressors, the latter being speci c to the pandemic for prevention and treatment efforts.
Our ndings correspond well to the stress sensitisation hypothesis (27), which argues that early adverse experiences lead to stress sensitisation making individuals more reactive to later stressors and, thus, increasing the susceptibility for adult psychopathology. Indisputable, the current pandemic and associated countermeasures pose a heavy strain on many individuals with anxieties about the disease, temporary closures of educational institutes, enforced social isolation, job losses, nancial di culties to name only a few of the many stressors associated with the pandemic. Our ndings show that the subjective perception of such COVID-19 related stressors matters in the relationship between CM and later adverse psychosocial outcomes during the pandemic. Moreover, we show that the impact of some stressors may be more relevant than others. Here, our explorative analyses highlight that perceived stress related to the pandemic itself, interpersonal con icts, concerns about medical care resources, and worries about the future were among the stressors that mattered most for almost all adverse psychosocial outcomes. In sum, these results advance previous research on the relationship between CM and later adverse psychosocial outcomes by quantifying the relevance of perceived stressors speci c to the pandemic (29).
For the outcomes of depression, loneliness, and paranoia, we observed only partial mediation suggesting that other mechanisms may be relevant to explain the relationship between early CM experiences and later adverse psychosocial outcomes. In our study, we focused on relatively established subjective psychological mediators of perceived stressors (speci c to the COVID-19 pandemic), rumination, and insecure attachment. Yet, there may be other potentially relevant psychological but also biological mediators of the relationship that we did not assess and which may help to identify additional factors that are amenable to treatment. Psychologically, for instance, recent research has highlighted the importance of perceived lack of social support during the current pandemic (63) and its transdiagnostic importance has also been discussed before the pandemic (14). Biologically, proin ammatory processes have been implicated in depressive symptomology speci cally (64, 65) and are proposed to explain the link between CM and later depression (66-68). As such, our study may have bene tted from inclusion of these factors and future longitudinal research should test their relative relevance in mediating the relationship between CM and later psychopathology in order to identify speci c factors that can be therapeutically targeted.

Strengths and limitations
Strengths of the present study include i) the prospective study design, ii) assessment of perceived stressors speci cally related to the COVID-19 pandemic, iii) simultaneous integration of established mediators of insecure attachment and rumination, iv) sophisticated analytical techniques, and v) inclusion of a broad array of key psychosocial outcomes. Yet, the study also has some important limitations. First, assessments are based exclusively on self-report questionnaires, so future research would bene t from inclusion of observer-based ratings, for example, by conducting structured interviews to assess attachment using the gold standard Adult Attachment Interview (69) or the Adult Attachment Projective (70). Second, although we report prospective data, only two timepoints of a 10-week time window were included and CM was assessed retrospectively. As several studies have shown a discrepancy between prospective and retrospective measures of CM (71,72), one has to be aware that CM scores always represent a subjective recall of adverse events during childhood which does not mean that this recall is invalid or less relevant. However, to minimise a potential recall bias, for example, due to current mood, we adjusted for self-reported life time mental health diagnoses. Yet, there may be other important confounders, for which we did not control. We also did not assess the speci c forms, duration, or frequency of CM, which may be important moderators but beyond the scope of the present study.
Finally, we report data of an online survey unrepresentative of the German population, in which female sex and relatively young participants were overrepresented. Therefore, replication in more representative or cross-national samples would allow for greater generalisability of ndings.

Conclusion
The current prospective study investigated the impact of CM on later adverse psychosocial outcomes during the COVID-19 pandemic. After adjusting for key confounding variable, we showed that subjective perception of COVID-19 stressors matters in the psychological response to the pandemic in addition to more established mediators of rumination and insecure attachment. Our ndings underscore the importance of stress sensitisation in childhood trauma-exposed individuals that is also present in the current pandemic. Importantly, identi ed mediators of COVID-19 perceived stressors, rumination, and insecure attachment fully explain the relationship between CM and anxiety, stress or psychological wellbeing and are amenable to psychological interventions. Thus, researchers and clinicians who encounter patients with a history of CM during the pandemic are advised to assess these key psychological mediators for more targeted prevention and treatment efforts in order to prevent deterioration of mental health.  Relative importance of COVID-19 related stressors in predicting psychosocial outcomes Note. Relative importance was extracted from elastic net regression models with optimal hyperparameters selected during cross-validation (see Supplementary Table 1 of not being able to get medical care, Stressor 11 = increased con icts with people close to me, Stressor 12 = nancial worries, Stressor 13 = uncertainties regarding my job, training place, studies or school, Stressor14 = fears of what the future will bring, or that I won't be able to cope with everything.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.