Attachment and resilience as mediators or moderators in the relationship between trauma and psychotic-like experiences

Background: A large body of evidence has established a tight relation between traumatic experiences (TEs) and psychotic-like experiences (PLEs). Nevertheless, more comprehensive models involving multiple interactions of serial or parallel mediations and moderations still need to be elucidated. Among the many potential mediators or moderators, insecure attachment and resilience play a key role in the association of stress with PLEs. Hence, we aim to explore the complex pathways that lead from different types of TEs to PLEs, involving attachment and resilience modeled as mediators or moderators. Methods: One thousand ten high school students completed the International Trauma Exposure Measure (ITEM), the 11-item Resilience Scale for Adults (RSA-11), the 16-item Prodromal Questionnaire (iPQ-16), and the Relationship Questionnaire (RQ). A path analysis was conducted to assess mediation and moderation. Results: The final model showed that the impact of childhood TEs on PLEs was mediated by a pathway through anxious-insecure attachment styles (i.e., fearful and preoccupied, respectively, 8.75 % and 8.53 % of the total effect) and personal resilience resources. Conversely, the avoidant-insecure attachment was associated with lower interpersonal resilience (b = 0.14 [0.08, 0.20]), which in turn moderated the impact of recent TEs on PLEs (interaction term b = 0.34 [0.21, 0.47]). Conclusions: Our model examines a complex model that includes factors buffering the effect of traumatic experiences on PLEs. Our results highlight the importance of insecure-anxious attachment to personal resilience resources and of insecure-avoidant attachment to interpersonal resilience as potential targets for clinical practice.


Introduction
Psychotic-like experiences (PLEs) are defined as psychotic symptoms, that is, hallucinations and delusions, occurring in the absence of functional impairment or illness (Hinterbuchinger and Mossaheb, 2021). PLEs have an estimated lifetime prevalence of 5.8 % (McGrath et al., 2015). PLEs are not, by definition, associated with an impairment in global functioning; they index the risk of transitioning to a full-blown schizophrenia spectrum disorder (SSD) (Kelleher and Cannon, 2011;van Os et al., 2009).
The association between traumatic experiences (TEs) and PLEs is well established (McGrath et al., 2017b(McGrath et al., , 2017a. TEs, occurring with some kind of intention to harm, i.e., interpersonal violence, neglect, physical, psychological, or sexual abuse, as well as bullying or marginalization, have shown the highest impact on PLEs and/or subsequent SSD (Ciocca et al., 2021;Croft et al., 2019;Mongan et al., 2019;Moriyama et al., 2018), while non-intentional TE, such as injuries, loss, natural disasters, have shown a more negligible impact on PLEs (Arseneault et al., 2011;McGrath et al., 2017b). Additionally, the different age of exposure to TEs plays a role in the association with PLEs, with the most recent TEs showing larger effect sizes compared to the more remote ones (Croft et al., 2019).
Several potential intervening factors have been involved in explaining the trauma-PLEs association, including biological, cognitive, and psychological ones.
Psychological mediators include post-traumatic and dissociative symptoms, affective dysregulation, insecure attachment, social defeat, low self-esteem, loneliness, as well as a distorted appraisal of subsequent TEs (Boldrini et al., 2020;Quijada et al., 2015;Sheinbaum et al., 2020;Sideli et al., 2020;Williams et al., 2018). Although a wide body of research has focused on single mediators between trauma and PLEs, more comprehensive models involving multiple interactions of serial or parallel mediations and moderations need to be investigated in order to address the complex relationships among putative intervening factors.
To this end, insecure attachment and resilience are two broad constructs that could alternatively behave as mediators or moderators of the impact of different types of TEs occurring at different times.

Insecure attachment is associated with psychosis
In recent years, evidence has been built to explain the putative role of insecure attachment in the psychological pathways from TEs to PLEs and psychosis (Korver-Nieberg et al., 2014). The attachment system regulates infant-caregiver relationships and romantic relations in adulthood. Adult attachment is organized along two orthogonal dimensions, avoidance and anxiety, with four possible attachment styles: Secure Attachment, with low levels of anxiety and avoidance; Preoccupied Attachment, associated with low avoidance and high anxiety; Dismissing Attachment, combining high avoidance and low anxiety; Fearful Attachment associated with high avoidance and anxiety and high emotional dysregulation (Bartholomew and Horowitz, 1991;Shaver and Mikulincer, 2008). A large body of evidence shows that insecure attachment is a mediator of the effect of childhood trauma and subsequent psychotic symptoms in clinical and non-clinical samples (Chatziioannidis et al., 2019;Sheinbaum et al., 2020Sheinbaum et al., , 2014Sitko et al., 2014). However, evidence on which attachment dimension, i.e., anxiety, avoidance, or disorganized, conveys most of the effect of TE on psychosis is confused by methodological heterogeneity, including differences in conceptualizations of attachment and psychometric tools.
The mechanisms underpinning the association between insecure attachment and PLEs have been extensively reviewed, with putative mediators being grouped into affective (i.e., affective dysregulation and disturbances) and cognitive factors (i.e., negative self-schemata, low self-esteem, negative beliefs about self and others) (Partridge et al., 2021).
However, these mediators only marginally explain how insecure attachment makes individuals more fragile and prone to psychotic symptoms depending on subsequent traumatic experiences. In other words, what is left behind by these studies is how insecure attachment contributes to a general stress-vulnerability model, which, in the presence of a proximal traumatic event, could lead to the emergence of psychotic symptoms.

Resilience
Resilience is commonly conceptualized as one's ability to cope with difficult situations, bouncing back to a pre-stress state and eventually rising to a higher level of functioning. From a clinical perspective, resilience can be thought of as a set of protective factors that buffer the effect of risk exposures, such as stressful or traumatic events, and prevent or attenuate the onset of subsequent psychological or psychiatric symptoms (Hjemdal et al., 2006;Southwick et al., 2014).
Resilience has been proposed to be made up of at least two higherorder factors comprising individual assets and external resources (Fergus and Zimmerman, 2005), which are related, respectively, to personal and interpersonal resilience factors (Morote et al., 2017;Rossi et al., 2020a). Personal resilience factors, or assets, include individual dispositional attributes such as a positive outlook, one's ability to cope with stressful events, self-efficacy and self-esteem, the ability to uphold daily routines, organize one's environment, and social competence.
Resilience has been shown to exert a protective role across the whole spectrum of psychosis, ranging from PLEs (Rossi et al., 2020b) to fullblown schizophrenia (Rossi et al., 2017). A cluster analysis highlighted how psychotic patients with greater resilience showed positive features of recovery (Rossi et al., 2018) and, consistently, poorer resilience has been associated with greater severity of depression among individuals with schizophrenia.
We have recently demonstrated that, within the trauma-PLEs pathways, personal and interpersonal resilience exert a different role, with personal resilience being the only mediator of the effect of adverse childhood experiences on PLEs in adulthood and interpersonal resources being left out of this model (Rossi et al., 2020a).

Attachment and resilience
Attachment and resilience are two constructs that could be viewed as complementary, in the sense that secure attachment has been advocated as a potential core feature for positive adaptation and resilience (Atwool, 2006;Darling Rasmussen et al., 2019). A moderate degree of correlation between attachment and resilience has recently been confirmed by a meta-analysis and narrative review (Darling Rasmussen et al., 2019). Furthermore, secure attachment has been shown to moderate the impact of traumatic experiences on the development of subsequent psychopathology (Ensink et al., 2021).
The accumulated evidence so far does not allow further detail on the mechanisms by which attachment and resilience participate in PLEs.
Decomposing resilience in its personal and interpersonal components allows to better hypothesize the complex interactions with insecure attachment, with low personal resilience being putatively associated with negative self-schemata and negative self-beliefs. In contrast, avoidant attachment could affect the devolvement and the setup of interpersonal resilience that could ultimately predict thriving and successful personal adaptation (Feeney and Collins, 2015).

Hypothesis and aim
The present study aims to explore, in a non-clinical sample, the complex pathways that lead from different types of traumatic experiences to PLEs, involving attachment and resilience modeled as mediators or moderators, depending on different types of traumas on psychosis. To do so, we tested this hypothesis on a cohort of 1010 last year high school students.
Given the above premises, we hypothesize that personal and interpersonal resilience interacts with insecure attachment and proximal and distal traumatic experiences in a complex pathway involving mediation and moderation (Fig. 1).
In particular, we hypothesize that: 1) early TEs exert their effect on PLEs through insecure attachment and personal resilience assets. 2) recent TEs directly impact PLEs, moderated by interpersonal resilience resources. Interpersonal resilience is, in turn, negatively affected by early TEs and insecure attachment.

Methods
The present study is based on the Dual Trauma study cohort. The dual trauma study is a multipurpose epidemiological survey aimed at addressing the complex relations between different traumatic exposures and a number of relevant behavioral and psychopathological outcomes in a non-clinical sample. Non-clinical samples are helpful in psychosis research since psychotic phenomena are believed to lie on a continuum with normal experiences. Furthermore, the confounding effects of the treatment of psychosis or consequences of the diagnosis are absent in non-clinical samples.

Design, participants, and procedure
Dual trauma is a cross-sectional study conducted in 2019, 10 years after the 2009 L'Aquila earthquake, focused on a target population comprising all adolescents and young adults attending the last year of high school in the province of L'Aquila, central Italy. The estimated target population size is 2000. This population is of particular interest as virtually all people from L'Aquila and its surroundings, estimated to be half of the target population, were directly exposed to the 2009 earthquake when they were between 7 and 10 years old. The other half of the target population, coming from the surroundings of Avezzano and Sulmona, was not directly exposed to this event. Inclusion criteria were: attending the last year of high school and being >18 years old at the time of enrollment. Enrollment was limited to students >18 in order to avoid any selection bias derived from parental consent. Exclusion criteria included being on supported teaching as a proxy of mild to severe cognitive impairment.
A cluster sampling was designed according to geographic area (L'Aquila, Avezzano, and Sulmona), school, and class. In the area, 12 out of 14 high schools were identified and invited to participate; two schools did not provide their consent to participate. In the remaining ten schools, 50 % to 100 % of the last-year classes were allowed to participate. A sampling strategy was designed in order to collect representative data regarding school type, as this factor was associated with relevant differences in mental health outcomes in the Italian adolescent population (Lisi et al., 2020).
The local school council and headmasters provided the consent to participate, as well as the number of eligible classes at the school level. The participants provided their individual written consent. The questionnaires were collected on paper and pencil form by the University staff. The local ethics committee at the University of L'Aquila provided approval for the study. This study adheres to the Declaration of Helsinki.
Recruitment and data collection took place between November 2019 and January 2020.

Traumatic experiences
The Italian version of the International Trauma Exposure Measure (ITEM) is a newly developed checklist that aims to capture traumatic life events and their associated features, consistent with the ICD-11 description of a TE. The ITEM measures the exposure to different TEs during different developmental periods (childhood, adolescence, and adulthood). The ITEM is freely available at https://www.traumam easuresglobal.com to the research and clinical communities and may be used without permission. For this study, the following modifications were introduced: for each TE, the respondent was asked to rate whether the TE had occurred during childhood, adolescence, or in the last six months. Each TE was classified as intentional or unintentional by two independent researchers. The results were checked and discussed with the original authors of the ITEM. During the classification process, three TE were excluded because they were deemed irrelevant as intentional or unintentional TE. As a result, six binary variables were produced: the presence of any intentional TE during the three life periods. For this study, we focused only on intentional TEs that occurred during childhood and the last six months. This choice was made a priori in order to keep the model simpler. Intentional TEs were selected over unintentional TEs because they have shown the strongest association with psychosis in the literature (Moriyama et al., 2018).

Psychotic-like experiences
The Italian version of the Prodromal Questionnaire-16 (iPQ-16) (Azzali et al., 2018) was used to assess the presence of PLEs. iPQ-16 is a 16-item self-report instrument that explores the presence/absence of 16 PLEs, including perceptual aberrations/hallucinations, unusual thought content/delusions, and two negative symptoms and their associated psychological distress. iPQ-16 scores the number of actual PLE endorsed, ranging from 0 to 16, and a distress score on a 4-point Likert scale ranging from 0 to 48. Although the iPQ-16 was originally designed as a screening tool for UHR individuals in help-seeking populations, several studies have used this instrument in the general population as a measure of PLEs (  2019). Because different cut-off points have been proposed for different populations (Savill et al., 2018), we chose to use the iPQ-16 score as a continuous rather than binary variable to avoid sensibility/specificity issues. For clarity of exposure, we will present only the results of the iPQ-16 distress score.

Resilience
Resilience was assessed using the 11-item Resilience Scale for Adults (RSA-11) , which is based on the original 33-item version (Capanna et al., 2015;Friborg et al., 2003). The RSA is a 7point Likert scale of 11 items grouped into two factors, namely, personal and interpersonal resilience. Personal resilience includes items from the original factors: Perception of Self, concerning self-confidence and positive outlook; Planned Future, concerning a positive outlook on one's own future; Social Competence, concerning an individual's own perception of her/his ability to initiate verbal contact and flexibility in social interactions; and Structured Style, concerning goal-oriented planning ability. Interpersonal resilience includes items from the original factors: Family Cohesion, concerning shared values and cohesion within one's family; and Social Resources, concerning social support and feeling of cohesion outside the family. Psychometric details of the Italian version of the RSA-11 can be found elsewhere . In this sample, the alpha coefficient was 0.83.

Attachment style
Attachment style was measured using the Italian version of the Relationship Questionnaire (RQ) (Bartholomew and Horowitz, 1991). The RQ is made up of four short paragraphs, each describing a prototypical attachment style (secure, preoccupied, fearful, and dismissing). Participants are first asked to choose one of the four attachment styles, like the one that best describes them (categorical scoring), providing a single item variable with four possible values. Afterward, participants are asked to rate their degree of correspondence to each prototype on a 7-point scale, providing four continuous variables, one for each attachment style (dimensional scoring).
For our analysis, we used only the four-dimensional scores.

Statistical analysis
Given the non-normal distribution of PLEs in the sample, all of the following regression analyses were bootstrapped at 5000 repetitions.
As a preliminary analysis, we thought to address the impact of intentional childhood and recent TEs on PLEs. To do so, we conducted a multiple linear regression. As the impact of TEs occurring at different ages on PLEs might show some degree of overlap, we entered all TEs simultaneously as independent variables to adjust for the mutual influence of childhood and recent TEs.
Subsequently, we conducted a step-by-step fitting of a path model that included exposures, outcomes, resilience, and attachment. To do so, we independently tested a number of partial path models and finally put all models together. Path analysis was conducted using the sem (structural equation modeling) module in Stata 16®. In order to obtain standardized coefficients, all continuous variables were standardized. The fitted models included a number of complex analyses, i.e., parallel and serial mediation and moderation analyses.
Mediation analyses included the estimate of direct, indirect, and total effects and the percentage of the total effect mediated, estimated as indirect/total effect*100. Mediation was assessed by inspecting the bootstrapped bias-corrected confidence intervals of the indirect effects in the model, according to the syntax reported in (UCLA: Statistical Consulting Group, n.d.). For every indirect effect, bootstrapping was conducted with 5000 replications.
Moderation was assessed by testing the significance of the interaction terms, followed by marginal analysis.
The first step is largely based on the model we previously demonstrated (Rossi et al., 2020a): we selected childhood and recent TEs as exogenous variables and PLEs as the outcome. As suggested in previous literature, personal resilience was introduced as a mediator between childhood TEs and PLEs. At the same time, we tested the role of interpersonal resilience as a moderator of the effect of recent TEs, introducing an interaction term.
In the second step, we conducted a parallel mediation role with preoccupied, fearful, and dismissing attachment style, as measured by the continuous scores of the RQ in the pathway between childhood TEs and PLEs. Similarly to previous works (Sheinbaum et al., 2020), we conducted a parallel multiple mediation analysis in order to assess the concurrent mediation effect of the four attachment styles. Contrary to previous works using the RQ (Sheinbaum et al., 2020, we initially chose to include the secure attachment score because we thought it could play a key role in the association with resilience in subsequent sub-models. However, possibly due to its negative cross-correlation with the remaining attachment styles, we observed better fit indices in models excluding secure attachment. For the sake of simplicity, these analyses are not shown here.
In the third step, we explored the association between attachment style and resilience. We fitted a multivariate multiple regression model, including attachment styles as independent variables and personal and interpersonal resilience as outcomes.
Finally, we tested an overall model with childhood and recent TEs as exogenous variables, preoccupied and fearful attachment style, through personal resilience, as mediators of the effect of childhood TEs on PLEs, and secure and dismissing styles affect interpersonal resilience, which in turn moderates the impact of recent TEs. A graphical description of the hypothesized final model is presented in Fig. 1.

Cohort characteristics
One thousand ten individuals participated in the study, 501 (49.60 %) males, 498 (49.31 %) females, and 11 (1.09 %) preferred not to say. The mean age was 18.7 (SD = 0.65). The sample characteristics and descriptive statistics of key variables are reported in Table 1. A total of 933 individuals had complete data on the selected variables and were thus included in the analysis. Differences in gender, age, and school attended between participants with (n = 933) and without complete cases (n = 77) were tested by means χ 2 , finding no statistically significant differences between the subjects included in the analysis and those not included. The mean PQ-16 distress score was 5.92 (SD = 5.53), with 154 (15.25 %) participants scoring ≥11. Regarding traumatic exposure, 432 (42.77 %) and 180 (17.82 %) individuals reported at least one TE during childhood and the last six months.

Associations between TEs, resilience, and PLEs
The path model describing the interactions among intentional TEs, resilience, and PLEs is reported in Fig. 2

Attachment styles as mediators of the effect of childhood TEs on PLEs
Parallel mediation analysis of attachment styles is reported in Fig. 3 and

Association between attachment and resilience
The association between the four attachment styles and personal and interpersonal resilience is reported in Table 4

A global model of traumatic experiences, attachment, resilience, and PLEs
The final model built from the above-mentioned partial models is proposed in Fig. 1 and reported in Table 5

Discussion
In the present study, we tested a model of a complex interplay among attachment and resilience in the pathways from early and recent traumatic experiences to PLEs in a sample of late Italian adolescents. To the best of our knowledge, this is the first study to include both a direct measure of resilience and attachment styles in a path model of PLEs and TEs. In particular, moving from previous evidence of a mediating role of personal resilience in the relation between childhood adversities and PLEs (Rossi et al., 2020a), we included interpersonal resilience as a moderator of the impact of recent TEs on PLEs. This first finding is of particular interest as it allows the assignment of a clearer role to interpersonal resilience, which would otherwise be left out of a model of PLEs.
Furthermore, we proposed a model for the association between insecure attachment and low resilience, with anxious attachment styles, that is, preoccupied and fearful, mediating the impact of childhood    trauma on personal resilience, personal resilience mediating the impact of anxious attachment on PLEs, and dismissing attachment directly affecting interpersonal resilience.
In the context of research on PLEs, childhood TEs have already been shown to affect both insecure attachment (Sheinbaum et al., 2014) and resilience (Rossi et al., 2020a). The present model proposes personal resilience as a mediator of the effect of anxious attachment styles on PLEs.
Interpersonal resilience, that is, social and familiar connectedness, moderates the impact of recent TEs on PLEs, with recent TEs exerting a greater effect on PLEs in individuals with lower levels of social and familial support. In this model, dismissing attachment affects interpersonal resilience.
Because resilience, as measured by the Resilience Scale for Adults, includes different factors that are either related to one's selfrepresentation or interpersonal resources (Friborg et al., 2003), we explored the interplay of resilience with some adult attachment styles. In our model, preoccupied and fearful attachment styles mediated the effect of childhood TEs on PLEs. This indirect effect is, in turn, mediated by personal resilience. It is important to point out that these two attachment styles are associated with a negative perception of one's self (Bartholomew and Horowitz, 1991).
Other psychological factors related to one's personal representation have been suggested to mediate the early TEs -PLEs relation, including self-esteem (Fisher et al., 2013;Wickham et al., 2015). Thus, it is not surprising that attachment and resilience constructs related to one's own image are affected by early TEs, which ultimately convey a risk of PLEs. However, insecure attachment has been repeatedly proven to mediate the relationship between trauma and PLEs Lavin et al., 2019;Williams et al., 2018).
Although preoccupied, fearful attachment and personal resilience acted as serial mediators of childhood TEs on PLEs, this was not the case for dismissing attachment and interpersonal resilience. Moving from the evidence that avoidant attachment styles play a role in establishing and maintaining a supportive social network (Lee and Gillath, 2016;Zinn et al., 2017), we hypothesized that this attachment dimension, which is tightly related to the internal representation of the other as trustworthy and reliable in providing support, could influence interpersonal resilience. Intriguingly, personal resilience per se was not directly associated with PLEs; however, people with greater interpersonal resilience showed a smaller impact of a recent TE on PLEs. This effect is consistent with a buffering role of resilience against TEs on PLEs (Crush et al., 2020).
The model tested in the present study has a number of innovative insights together with other replicated findings. In particular, the association between dismissing attachment and interpersonal resilience was never tested.

Methodological considerations
This study has a number of relevant limitations. The first limitation is its reliance on retrospective self-report of TEs, which, inherently, could lead to a recall bias and a potentially high rate of false positives or negatives. Second, this study is based on cross-sectional data, and the fitted path model could not necessarily represent the actual connections between the variables involved. Although TEs and PLEs have a clear time organization, although retrospectively reported, the direction of the association between attachment and resilience we hypothesized was largely based on speculation.
Third, no information was collected on any currently diagnosed psychiatric condition or any psychological or psychiatric treatment, nor regarding socioeconomic status. Therefore, it was impossible to adjust the results for these variables.
Lastly, our sampling technique has been conducted in a convenient fashion. This means the status of young students, the limited geographical area and the lack of a number of relevant confounders could have posited a threat to the generalizability of the results. Furthermore, albeit the rigorous sampling procedure minimized the selection bias, we cannot exclude that subjects that did not participate in this study, (for example, youngsters that have dropped school before completing their education or students absent on the day of data collection) could have skewed our results.

Clinical considerations
Our results have some relevant clinical implications. As resilience represents an important buffer against life adversities that can promote the progression along the psychosis continuum (DeLuca et al., 2022), building effective resilience is of pivotal importance relative in the prevention of psychotic symptoms and psychotic disorders. Because insecure attachment is central to many psychological interventions for psychosis (Partridge et al., 2021), improving resilience could represent a secondary effect of psychological interventions that could promote stronger protection against the detrimental psychological effects of adverse life experiences and stress.

Conclusions
The current findings provide a more detailed psychological model of PLEs. They suggest that attachment and resilience-informed approaches may be useful in managing psychotic symptoms in people who have experienced traumatic events. In particular, our model highlights the importance of insecure-anxious attachment to personal resilience resources and the importance of insecure-avoidant attachment to interpersonal resilience as potential targets for clinical practice. Overall, examining a complex model that includes factors that buffer the effect of traumatic experiences is critical to advancing our understanding of psychosis and tailoring intervention strategies.

Declaration of competing interest
All the authors declare that they have no conflict of interest to disclose.