Worry and metacognitions as predictors of the development of anxiety and paranoia

Recent studies have shown that worry and related negative metacognitions are characteristic in generalized anxiety and paranoia respectively. However, most of these studies did not take into account common co-occurrence of anxiety and paranoia, and longitudinal modelling of the role of worry and metacognitions on the development of anxiety and paranoia is rare. The current study aimed at examining the bidirectional longitudinal relationship between anxiety and paranoia, as well as the importance of worry and metacognitions in the development of these symptoms. Our validated sample consisted of 2291 participants recruited from universities, among whom 1746 participants (76.21%) completed online questionnaires at baseline and at one year, reporting levels of anxiety, paranoia, worry, and negative metacognitions. Structural equation modeling analyses, followed by path comparisons, revealed that anxiety and paranoia mutually reinforced each other over time. Negative metacognitions, rather than worry itself, were contributive to the development of both symptoms over time. Negative metacognitions showed bi-directional relationships with anxiety over the time period assessed and showed uni-directional relationships with paranoia. Clinical implications of our findings are discussed.

paranoia 24 . Wells and Matthews 25,26 also proposed that a negative biased style of thinking is a common factor that underlies psychopathology and is caused by dysfunctional metacognitions. This approach was a grounding for Wells' metacognitive model of generalized anxiety disorder (GAD), in which negative beliefs about the uncontrollability, danger and meaning of worry initiate worry about worry (i.e. meta-worry), and in turn exacerbate worry and intensify anxiety symptoms 27 .
Among both clinical 13,28,29 and non-clinical populations 18,30 , a general tendency to worry has been shown to have an impact on paranoia. However, these studies relied on a crude measure of a general tendency to worry only, and did not control for level of anxiety, which makes interpretation of results problematic. It has been shown that increased meta-worry [31][32][33] and negative beliefs about worry [34][35][36][37] , rather than worry itself, distinguished patients with anxiety disorders from non-clinical worriers, and that negative beliefs about worry predicted subsequent emergence of anxiety 38,39 . However, whether negative metacognitions contribute to both paranoia and anxiety over time remains unclear.
A comprehensive assessment of worry and negative metacognitions would allow us to discern whether the impact on paranoia was attributable to processes that are equally common among non-clinical worriers (i.e. worry itself) or processes that are more central to the psychopathology of anxiety (i.e. negative metacognitions). As opposed to cross-sectional studies, a better design would be to assess levels of paranoia, anxiety, worry and negative metacognitions repeatedly, and then to model their inter-relationships simultaneously. It has been increasingly agreed that psychopathology research will be advanced by elucidating the similarities and differences across psychiatric phenotypes and by identifying potential transdiagnostic features and underpinnings [40][41][42][43] . Therefore, the current study was aimed to examine the temporal relationship between paranoia and anxiety, and the transdiagnostic roles of worry and negative metacognitions to the development of these symptoms over time. Identification of specific processes that contribute to paranoia would also inform design of worry and metacognitive interventions for paranoia reduction [44][45][46] .
The present study adopted a structural equation modeling (SEM) approach of a large sample of non-clinical participants, assessed repeatedly. We hypothesized that, after controlling for auto-correlations, baseline level of anxiety will predict level of paranoia at one year, and baseline level of paranoia will predict level of anxiety at one year. As postulated by the metacognitive model of GAD 27 , we also hypothesized that after controlling for covariances, negative metacognitions, rather than worry itself, will predict changes in anxiety and paranoia over time.

Results
Demographics and correlation analyses. The mean age of the current sample was 19.77 (SD = 1.41) at baseline, and the majority of them were female (65.9%). Clinical characteristics of participants are summarized in Table 1. Correlations among anxiety, paranoia and each indicators of worry process at baseline and one year are presented in Table 2. All major variables were significantly correlated within and across time points (ps < 0.001).

Longitudinal
Comparisons between paths suggested that the regression coefficients of the path between negative metacognitions at baseline and anxiety at one year were not significantly different from that of the path leading from negative metacognitions at baseline to paranoia at one year (∆β = 0.051, p = 0.302).
The latent interaction between baseline anxiety and negative metacognitions significantly predicted paranoia at one year (B = 0.22, SE = 0.05, p < 0.001). Moreover, after adding the interaction term into the model, the path between anxiety at baseline and paranoia at one year became non-significant (p = 0.27). Post hoc analysis revealed that for individuals with high levels of negative metacognitions (i.e. 1 SD above the mean), baseline anxiety significantly predicted paranoia at one year (B = 0.16, SE = 0.04, p < 0.001), whereas for individuals with low levels of negative metacognitions (i.e. 1 SD below the mean), the link between baseline anxiety and paranoia at one year was not significant (p = 0.15). The prediction of anxiety at one year by the latent interaction between baseline paranoia and negative metacognitions was significant (B = −0.08, SE = 0.03, p = 0.01). For individuals with low levels of negative metacognitions, paranoia at baseline predicted anxiety at one year significantly (B = 0.08, SE = 0.03, p = 0.018), whereas for individuals with high levels of negative metacognitions, the prediction of anxiety at one year by paranoia at baseline was not significant (p = 0.76).

Discussion
The current study was the first attempt to examine longitudinal relationships between paranoia, anxiety, worry and negative metacognitions using a rigorous SEM approach with a large non-clinical sample. We conceptualized paranoia and anxiety as dimensional latent constructs, being indexed with observed symptoms or ideas, and considered cross-lagged relationships between variables using a model comparison approach.
Our results extended the understanding of the longitudinal relationship between paranoia and anxiety. Having controlled for auto-correlations, anxiety predicted change in paranoia, which was consistent with previous longitudinal cohort and experimental studies 8,[18][19][20] . The reverse was also true, so that paranoia and anxiety mutually reinforced each other over one year. However, when worry and negative metacognitions were added to the original SEM model, the paths from paranoia at baseline to anxiety at one year were no longer statistically significant. www.nature.com/scientificreports www.nature.com/scientificreports/ Together with a recently published cluster analysis study, which found that paranoid individuals tended to be anxious but anxious individuals might not be paranoid 14 , the current finding lends support to the hierarchical structure of paranoia in the general population, which suggests that ideas of persecution build upon more common anxious concerns 5 .
Our study provided evidence for the importance of negative metacognitions in the development of both paranoia and anxiety. After controlling for auto-correlations within variables and cross-lagged associations across variables, negative metacognitions, rather than worry itself, were predictive of both paranoia and anxiety a year after. The magnitudes of predictions of negative metacognitions were comparable for the two symptoms. This is in support of Wells's metacognitive model, which emphasizes that it is the appraisal of and response to worry that matters in mental ill-health 27 . While a reciprocal relationship between anxiety and negative metacognitions over www.nature.com/scientificreports www.nature.com/scientificreports/ time was found, the path leading from paranoia to negative metacognitions was not significant. In other words, negative metacognitions and anxiety maintain each other, but such downward spiral is absent in paranoia.
Not only exerting direct influences on both symptoms, our exploratory analyses suggested that negative metacognitions also moderate the interplay between paranoia and anxiety. Anxiety was associated with an increase in paranoia in those scoring high in negative metacognitions only. However, paranoia was associated with subsequent anxiety in those low in negative metacognitions at baseline. The finding that the nature of the temporal relationship between anxiety and paranoia was dependent on level of negative metacognitions is intriguing. It suggests an important role of negative metacognitions in driving the development of paranoid thinking from anxious concerns 5 , and hence more prudent consideration is warranted on theories that make direct links between anxiety and paranoia.
Worry did not emerge as an independent predictor of paranoia when controlling for covariances of negative metacognitions and anxiety, which suggests that the anxiety theory of paranoia should take into account the role www.nature.com/scientificreports www.nature.com/scientificreports/ of negative metacognitions and further evaluate the relative importance of worry itself and the related metacognitions. In sum, our preliminary results suggest that negative beliefs about worry and meta-worry, rather than worry itself, may be transdiagnostically relevant to anxiety and paranoia. This claim warrants further tests using repeated measures in clinical populations.
Several limitations of current study are noticed. Firstly, we only collected data at two time points, which limited us from investigating the forms of changes over time (e.g. using latent growth curve analysis) 47 , and from examining indirect effects of worry and metacognitions on paranoia via change in anxiety. Secondly, we recruited participants from tertiary institutions. It remains unclear how our results would differ if the sample spans across a wider range of age and education level. Lastly, we appreciate the possibility that development of paranoia and anxiety may also involve other unmeasured mechanisms beyond worry and metacognitions.

conclusion
The current findings lend support to the predictive roles of negative metacognitions on changes in both anxiety and paranoia over time. Negative metacognitions showed bi-directional relationships with anxiety over the time period assessed and showed uni-directional relationships with paranoia. Moreover, it appears that negative metacognitions may be important modifiers of the nature of the relationship between the two symptoms, with anxiety being associated with an increase in paranoia among those scoring high in negative metacognitions. Transdiagnostic investigations of the role of metacognitions across psychopathologies will shed light on the use of process-based interventions targeting these processes on treatment for various phenotypes.

Method
Ethics approval for this study was granted by the Survey and Behavioural Research Ethics Committee (SBREC) of the Chinese University of Hong Kong. All methods were carried out in accordance with relevant guidelines and regulations. Informed consent were obtained from all participants upon recruitment.
participants. We included full-time undergraduate students in Hong Kong (age [18][19][20][21][22][23][24][25] and excluded individuals with a current or previous psychiatric diagnosis. Recruitment was carried out using various methods, such as university mass mailing, distribution of leaflets, and promotion on a social media site (i.e. Facebook). At baseline, a total of 2796 participants responded to our survey. We removed 115 participants due to repeated responses and 96 due to invalid responses (not local undergraduate students: n = 19; reporting a current or past psychiatric diagnosis: n = 72; invalid contact information: n = 5). Another 294 participants failed to meet validity criteria (completion time, long-string responses, and odd-even consistency), leaving a final sample of 2291 participants.
All respondents were approached again after one year via e-mails and phone calls. Out of the valid sample of 2291 participants at baseline, 1746 (76.21%) completed the follow-up online survey. 48 , which consists of 32 items that measure level of reference ideation and level of persecutory ideation over the past month. Good internal consistencies (α = 0.69 to 0.92) and test-retest reliability (r = 0.81 to 0.88) were established 48 . GPTS has been translated into Chinese for the current study (unpublished) and yielded excellent reliability (α = 0.96).

Measures. Paranoia was measured by the Green et al. Paranoid Thought Scales (GPTS)
Anxiety was measured by the Generalized Anxiety Disorder 7-Item Scale (GAD-7) 49 . Symptom severity of generalized anxiety was assessed on a 4-point scale over the last two weeks. The Chinese version of GAD-7 yielded good test-retest reliability (α = 0.86) and convergent validity (r = 0.66 to 0.84) 50 . The current sample also reported excellent reliability (α = 0.92).
Worry was measured by the Worry Domain Questionnaire (WDQ) 51 , which is a 25-item questionnaire assessing worries about daily events including relationships, lack of confidence, aimless future, work incompetence and finances. The original WDQ has an acceptable level of internal consistency and good construct validity 51 , and has been translated (with permission) into Chinese for the current study (unpublished). The current sample yielded excellent reliabilities for subscales of WDQ (α = 0.83 to 0.91).
Negative metacognitions were measured by the Anxious Thought Inventory (AnTI) 52 , and the Metacognitions Questionnaire-Short Form (MCQ-30) 53 . AnTI is a 22-item questionnaire comprising social worry, health worry and meta-worry subscales. It has been translated into Chinese for the current study (unpublished). MCQ-30 assesses meta-cognitive beliefs on five dimensions: positive beliefs about worry (PB; "Worrying helps me cope"), negative beliefs about worry concerning uncontrollability and danger (U/D; "Worrying is dangerous for me"), negative beliefs concerning the need to control worry (NC; "I should be in control of my worry all the time"), cognitive confidence (CC), and cognitive self-consciousness (CSC) 53 . For the current purpose, only the meta-worry subscale of AnTI, and the two subscales addressing negative beliefs about worry (i.e. U/D and NC) of MCQ-30 were included. The original versions revealed good internal validities and test-retest reliabilities 52,54 . All three subscales achieved excellent levels of reliability (α = 0.90, 0.83 and 0.78 respectively) in the present study.
Analysis. Structural equation modeling (SEM) with anxiety, paranoia and worry processes being treated as latent variables was conducted using Mplus version 7 55 . Negative metacognition was indexed with three theory-driven indicators, namely U/D, NC and meta-worry. Worry was indexed with the five subscales of WDQ measuring worry about five distinct domains of life. Paranoia was indexed with two indicators, namely reference ideation and persecutory ideation. Average scores of each subscale were used for all above-mentioned indicators. Anxiety was indexed with the seven items of GAD-7, each capturing a symptom of generalized anxiety.
Correlation analyses between all observed variables were firstly performed. Confirmatory factor analyses (CFA) were then conducted to define the pattern of observed variables for latent constructs and evaluate their factorial invariance between baseline and one year 56 . Full SEM models with uni-directional paths leading from each latent variable at baseline to all latent variables at one year, were carried out subsequently to examine the Scientific RepoRtS | (2019) 9:14723 | https://doi.org/10.1038/s41598-019-51280-z www.nature.com/scientificreports www.nature.com/scientificreports/ cross-lagged relationships between latent variables. To test hypothesis 1, a CFA model concerning paranoia and anxiety across time points was tested. This was followed by a SEM model with bidirectional cross-lagged paths between paranoia and anxiety. To test hypothesis 2, models concerning worry, negative metacognitions, paranoia and anxiety were evaluated. Non-significant paths were eliminated to yield a more parsimonious model 57 .
Parameters were estimated by using the robust maximum likelihood (MLR) method 55,58 . Missing values were dealt with full information ML 59 . Model fit was evaluated by chi-square statistic and three goodness-of-fit indices, namely Comparative Fit Index (CFI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR) 60,61 . Competing models were compared using chi-square difference test with Satorra-Bentler correction 62 , and Akaike's Information Criterion (AIC) with a lower score indicating better model fit 63 .
As an exploratory analysis, the standardized regression weight of the cross-lagged path leading from anxiety at baseline to paranoia at one year was compared against the path leading from paranoia at baseline to anxiety at one year. Standardized regression weights of paths leading from worry and metacognitions at baseline to paranoia at one year were also compared against corresponding paths leading from that variable at baseline to anxiety at one year. Latent interaction effects between negative metacognitions and each symptom at baseline on the other symptom at one year were explored by using the latent moderated structural equations (LMS) method 64 . Significant results were followed by post hoc analysis using simple slope method 65 .