Psychotic experiences and dissociation in adolescents: Within-person analysis in the Tokyo teen cohort

Although many cross-sectional studies showed that psychotic experiences (PEs) and dissociation were closely related, the longitudinal association between them remains unknown. Therefore, the aim of the current study was to examine the longitudinal association of these two symptoms throughout adolescence, under the hypothesis that these two symptoms are bidirectionally associated. Data were obtained from a population-based cohort, the Tokyo Teen Cohort study (TTC; N = 3171). PEs and dissociation were assessed at 10, 12, 14, and 16 years of age. PEs were assessed using a total score from five-item self-report questionnaires derived from the Diagnostic Interview Schedule for Children (DISC-C). Dissociation was assessed using subscale scores of the Child Behavior Checklist (CBCL) completed by primary caregivers. We examined the longitudinal relationship between PEs and dissociation using the random intercept cross-lagged panel model (RI-CLPM). The within-person component of the RI-CLPM revealed no significant cross-lagged effect of dissociation on PEs at any time point. On the other hand, there was a significant ( p < 0.05) association between PEs at age 14 and dissociation at age 16 ( β = 0.106, 95 % CI 0.047 – 0.165). The between-person component revealed a significant time-invariant relationship between the two symptoms ( β = 0.324, 95 % CI 0.239 – 0.410). The longitudinal relationship between PEs and dissociation was limited at the within-person level, whereas the between-person correlation was significant. The only significant longitudinal pathway was from PEs to dissociation, suggesting that PEs may be a predictor of dissociation in mid-adolescence.


Introduction
Psychotic experiences (PEs) are weak expressions of psychotic symptoms observed in the general population and are also referred to as an extended and transdiagnostic psychosis phenotype (Sullivan et al., 2020;van Os and Reininghaus, 2016).The median lifetime prevalence of PEs has been reported to be about 6 % (Linscott andvan Os, 2013and McGrath et al., 2015), and they are more common in childhood and adolescence, with the median prevalence reported to be 17 % among children (aged 9 to 12) and 7.5 % among adolescents (aged 13 to 18) (Kelleher et al., 2012).PEs are known to be a risk factor for a variety of later mental illnesses, both psychotic (van Os et al., 2009;Zammit et al., 2013) and non-psychotic disorders; affective disorders, anxiety disorders, behavioral disorders and substance-use disorders, or suicidal thoughts and behaviors (Bourgin et al., 2020;Bromet et al., 2017;Healy et al., 2019;McGrath et al., 2016;Pignon et al., 2018).With regard to the prevention and treatment of subsequent mental illness, the determinants and associated factors of PEs should be investigated.
In the context of the relationship to PEs, dissociation has been focused as one of those that play an important role.Dissociation is a common symptom, with a reported prevalence of 6.3 % (Mulder et al., 1998).Several meta-analyses reported significant associations between dissociation and PEs, particularly auditory hallucinations (Bloomfield et al., 2021;Longden et al., 2020;Pilton et al., 2015).It has also been reported that dissociation mediates the relationship between traumatic experiences and PEs (Cole et al., 2016;Pearce et al., 2017;Perona-Garcelán et al., 2012;Schalinski et al., 2019;Sun et al., 2018;Varese et al., 2012a;Wearne et al., 2018;Williams et al., 2018;Yamasaki et al., 2016), but most of these previous studies were cross-sectional designs.A few longitudinal studies (Escher et al., 2002a(Escher et al., , 2002b;;Muenzenmaier et al., 2015) suggested that dissociation was associated with future PEs, but no study has examined the bidirectional longitudinal association in the large-sampled general population.
Furthermore, the effect of PEs on dissociation remains unknown.Since psychotic symptoms can be traumatic experiences (Berry et al., 2013;Buswell et al., 2021;Dunkley et al., 2015;Rodrigues and Anderson, 2017), PEs might lead to later dissociation.If so, the relationship between PEs and dissociation would be bidirectional, forming a vicious circle.Therefore, it would be important to investigate the bidirectional relationship between PEs and dissociation, but this has not been well explored.Despite the high prevalence of PEs and dissociation in the general population, the only study (Humpston et al., 2020) showed that these two symptoms were longitudinally associated with each other among clinical samples, with decreasing effect size over the long follow up.Thus, the bidirectional relationship between dissociation and PEs in the general population remains unknown.
Although cross-lagged panel modeling (CLPM) has been widely used in this field for the purpose of demonstrating causal inference, a limitation of CLPM is that CLPM cannot infer causality about within-person relationships because of the mixture of within-person and betweenperson relationships (Hamaker et al., 2015;Usami et al., 2019), resulting in inappropriately biased cross-lagged estimates (Lucas, 2023).The problems identified in the conventional CLPM have led to a growing number of studies in the field of psychosis and PEs examining longitudinal relationships at the within-person level (Stanyon et al., 2022;van Os et al., 2021).The within-person relationship was referred to as a relationship seen in the process of change within an individual, while the between-person relationship being a group level relationship.Since the main interest of this study was how dissociation affects later PEs (and vice versa) in an individual, we employed the random intercept cross-lagged panel model (RI-CLPM), which allows us to analyze these within-person effect, separated from time-invariant between-person effect.We primarily analyzed the data using the RI-CLPM, but we also employed the traditional CLPM for comparison and to examine the consistency of our results with previous studies that used traditional methods.The purpose of the current study was to examine the longitudinal relationship between PEs and dissociation using data from a large-scale population-based adolescent cohort study, under the hypothesis that PEs and dissociation have longitudinal bidirectional associations with each other in general adolescents.

Study design and participant
The data used in this study were from the Tokyo Teen Cohort (TTC) Study, a population-based multidisciplinary longitudinal cohort study (Ando et al., 2019).TTC has been conducted to investigate physiological and psychological development during adolescence, which has been known as a critical phase for future health and well-being.In the TTC, the first survey was conducted at the age of 10 years.This is a birth cohort study, and the target population was defined by the date of birth (children born between September 2002 andAugust 2004).In this cohort study, a regionally representative sample was randomly selected by Basic Resident Register.The subjects residing in the three municipalities in Tokyo, Setagaya-ku, Chofu-shi and Mitaka-shi.The second, third and fourth surveys of data collections were conducted at the age of 12 years, 14 years and 16 years, respectively.
A letter of request for survey cooperation was mailed to each household around the time of the children's 10th birthday, and trained interviewers visited the household and obtained written informed consent for participation.The interviewers also made efforts to obtain the understanding and assent of the children themselves.The recruitment and data collection for the second, third and fourth surveys were conducted in a similar manner.
The interviewers visited each target household twice to collect data through self-administered questionnaires and interviews.During the second visit, both the primary caregiver and the child were asked to complete the enclosed questionnaire at a distance from each other so that more reliable data on personal information could be collected in privacy.
A total of 3171 child-parent pairs were recruited at 10 years old.Of the 3171 participants, 3007, 2667 and 2616 cooperated at the age of 12, 14 and 16, respectively (follow-up rate: 94.8 %, 84.1 % and 82.5 %).Therefore, 555 participants were lost to follow-up at the age of 16.

Measurements
Information on PEs was obtained from the children's responses to the self-administered questionnaires.Self-administered questionnaires were also completed by the primary caregivers regarding children's age, sex and the Child Behavior Checklist (CBCL) (Achenbach, 1991;Itani et al., 2001).

Psychotic experiences (PEs)
PEs were assessed by five items, four of which were derived from the Diagnostic Interview Schedule for Children (DISC-C) (Costello et al., 1985), and one question was added about visual hallucinations.They were answered by the children themselves.The five items were as follows: "Have you ever heard voices that other people cannot hear?" "Have other people ever read your thoughts?""Have you ever had messages sent especially to you through the television or radio?" "Have you ever thought that people are following you or spying on you?" "Have you ever seen things that other people could not see?"Each question was scored on a 3-point Likert scale; no = 0, maybe = 1, yes = 2.The score for each item was summed up to a total PEs score, ranging from 0 to 10.A higher score means a stronger experience of PEs.

Dissociation
As used in the previous studies (Becker-Blease et al., 2004;Malinosky-Rummell and Hoier, 1991), dissociation was assessed by the six items from the CBCL.The six items were as follows: "acts too young for his/her age," "can't concentrate, can't pay attention for long," "confused or seems to be in a fog," "daydreams or gets lost in his/her thoughts," "stares blankly," and "sudden changes in mood or feeling."Each question was scored on a 3-point scale; not true = 0, somewhat or sometimes true = 1, very true or often true = 2.The score for each item was summed up to a total dissociation score, ranging from 0 to 12. Higher score means stronger dissociation.

Covariates
To estimate the longitudinal association between PEs and dissociation, we treated age in months and sex as potential cofounders in the analysis of the conventional cross-lagged panel model.Since correlation between the random intercepts captured the time-invariant effect of unmeasured between-person differences in RI-CLPM, the within-person component was controlled for time-invariant confounders, such as sociodemographic factors, environmental influences, and early-life experiences.

Statistical analyses
The longitudinal associations between PEs and dissociation were examined with both the random intercept cross-lagged panel model and the conventional CLPM for comparison (Figs. 1 and 2).We applied full information maximum likelihood estimation to complement missing data, under the assumption of missing at random.Model fit was assessed with the root mean square error of approximation (RMSEA), the comparative fit index (CFI), and the Tucker-Lewis index (TLI).CFI and TLI are reported to fit well with values above 0.95 and RMSEA with values below 0.06 (Hu and Bentler, 1999).The analyses were conducted using the lavaan package (Rosseel, 2012) in R. To examine longitudinal measurement invariance, we constructed a multiple indicator RI-CLPM model (Mulder and Hamaker, 2021).We then imposed constraints to compare model fit.We also applied STARTS, which has been reported to be suitable for studying complex psychological and developmental processes.However, the model is also known to have issues with convergence difficulty (Orth et al., 2021;Sorjonen et al., 2023).
Although it was post-hoc, due to concerns about deviations from normality in the current results, we examined the results using MLR (Maximum Likelihood Robust) estimation with listwise deletion in addition to the primary estimation method of ML (Maximum Likelihood).

Results
Table 1 shows the demographic information and descriptive statistics of the participants.There were significant differences in both the PEs and the dissociation scores across the four time points(both: Friedman's test p < 0.001).The mean PEs score at baseline (10 years old) did not differ significantly according to the drop out or missing values in the later follow-up (Mann-Whitney U test p = 0.92).Similarly, the later drop out or missing values of dissociation did not show the difference in the mean score of dissociation at baseline (p = 0.62).Table 2 shows the pairwise correlation matrix between the variables of PEs and dissociation.
Fig. 1 shows the conventional CLPM for the longitudinal association between PEs and dissociation.After adjusting for age in months and sex, the analysis revealed that the respective lagged estimates between the two symptoms were small but significant in both directions.
In contrast, the RI-CLPM revealed no significant lagged association in the time-variant within-person component except for the one from PEs at 14 y.o. to dissociation at 16 y.o.(Fig. 3).Significant autoregressions for lagged estimates of PEs and dissociation indicated substantial withinperson stability.In the time-invariant, between-person component of the model (Fig. 4), there was a significant positive covariance between the random intercepts of the two symptoms, which suggested those with stronger PEs tended to have stronger dissociation on average throughout the four timepoints.Fit indices of the model indicated a better fit than the conventional CLPM; CFI = 0.995, TLI = 0.985 and RMSEA = 0.031.Measurement invariance was examined using the multiple indicator RI-CLPM model and the chi-square difference test was significant, indicating that weak factorial invariance was not maintained, but the RMSEA values of the configural and the weak factorial invariance model were 0.025 and 0.033, respectively.The model did not converge when STARTS was applied.
In the estimation using MLR with listwise deletion, we confirmed that there were no major changes in the lag estimates and p-values of the RI-CLPM that would affect the interpretation of the results.Specifically, when using p < 0.05 as the cutoff, the significant parameters remained the same.On the other hand, in the conventional CLPM, the cross-lags from dissociation at age 10 to PEs at age 12, from PEs at age 12 to dissociation at age 14, and from dissociation at age 14 to PEs at age 16 were no longer significant.

Discussion
To our knowledge, this was the first study to examine the withinperson effect of the longitudinal associations between PEs and dissociation in the general adolescent population, using the RI-CLPM.The results of the analyses differed between the conventional CLPM and the RI-CLPM.The conventional CLPM showed that PEs and dissociation had small but significant longitudinal associations with each other.On the other hand, the RI-CLPM showed that the two symptoms had less association with each other at the within-person level and the only significant lagged estimate was from PEs at the age 14 to dissociation at age 16.In addition, there was a time-invariant between-person correlation between them.Since within-person processes and between-person differences were analyzed indistinguishably in the conventional CLPM, the significant cross-lagged association between the two symptoms found in the CLPM could be considered to reflect the between-person relationship found in the RI-CLPM.
The results of within-person component of the RI-CLPM appeared to be inconsistent with previous studies because previous studies did not distinguish between within-person and between-person relationships.Since the between-person component examined a similar relationship as cross-sectional methods, the results of the meta-analyses (Bloomfield et al., 2021;Longden et al., 2020;Pilton et al., 2015), which showed significant cross-sectional relationship between PEs and dissociation, were consistent with the present study.The previous studies that conducted longitudinal analyses (Escher et al., 2002a(Escher et al., , 2002b;;Humpston et al., 2020;Muenzenmaier et al., 2015) also did not use methods that allowed for within-person level examination, so their results were composites of within-person and between-person components, similar to The results of this study suggested that at the between-person level, there were common tendencies, environmental factors, genetic backgrounds, or past experiences that would make both dissociation and PEs more likely to occur.Traumatic experiences and adverse childhood experiences are well-known risk factors common to PEs and dissociation (Bailey et al., 2018;Bell et al., 2019;Bendall et al., 2008;Dalenberg et al., 2012;Morgan and Gayer-Anderson, 2016;Read et al., 2005;Varese et al., 2012b).Furthermore, the correlations between PEs and dissociations in the between-person level possibly resulted from an overlap in the constructive concept of them, or the difficult nature of separating them (Renard et al., 2017).Some researchers have even suggested that PEs could be conceptualized as dissociative (Longden et al., 2020).A challenge for the future would be to investigate the as yet unidentified factors that exist in the background.
The within-person component of the RI-CLPM revealed a significant association between PEs at age 14 and dissociation at age 16.We should consider the possibility of multiple comparisons, but similar trends were seen in the path from PEs at the age of 10 to dissociation at the age of 12, suggesting that psychotic experiences might lead to subsequent dissociation.This suggested that PEs might influence later dissociation when focused on a single individual, which would be explicable in the context of trauma.Since PEs are often perceived as strange and unusual by the general adolescents, PEs could be distressing (Kelleher et al., 2015) and become traumatic experiences.Thus, PEs as traumatic experiences may lead to dissociation in general adolescents.Even if the PEs were not experienced as traumatic, mental health problems such as anxiety, depression, and insomnia which were caused by the psychotic experiences could contribute to dissociation ( Černis et al., 2021).The fact that only the 14-to 16-year-old pathway was significant could be influenced by the tendency of individuals to perceive PEs as more bizarre as they  SD: Standard Deviation.PEs: Psychotic Experiences, defined as the sum of 5-item scores (possible range: 0-10).Dis: Dissociation, defined as the sum of 6-item scores (possible range: 0-12).

Table 2
Pairwise correlation matrix between the 8 variables.PEs: Psychotic Experiences.
mature, but this has not been investigated much.On the other hand, the lack of longitudinal measurement invariance suggested that the scales might not have measured the same constructs consistently over the study period.Consequently, the interpretation of the cross-lagged associations required caution.The observed significant cross-lagged effect might be influenced by changes in the measurement properties of the scales over time rather than representing true psychological processes.However, since the chi-square difference test might be significant due to the large sample size, we could not conclusively determine that invariance was not maintained based on such a small difference in RMSEA in this case.These results suggested that when dealing with adolescents who have PEs, we should pay attention to the subsequent dissociation and whether the PEs themselves have become traumatic experiences.Focusing on the within-person process, the presence of dissociation itself did not predict later PEs, but it should be noted that the two symptoms are at common risk and can coexist (Bloomfield et al., 2020).
This study has several strengths.It was the first large-scale study to examine the longitudinal relationship between PEs and dissociation in the general adolescent population.In addition, this study was also the first to use the RI-CLPM to examine the within-person and betweenperson associations of PEs and dissociation.The new finding was that most of the associations found in the previous studies were at the between-person level, while PEs were associated with later dissociation only at specific ages at the within-person level.This study also had several limitations.First, although the RI-CLPM can control for timeinvariant background factors, we could not control for time-varying covariates such as adversity experiences and the subsequent psychological distress between the ages of 10 and 16.Second, there may have been a relationship between the two symptoms that could have been revealed by observing at a shorter interval than every two years.Third, since the assessment of dissociation was based on parent-report questionnaires, children's dissociative symptoms may have been underestimated.Fourth, the assumption of RI-CLPM and CLPM included linear relationships between PEs dissociation.While non-linear methods exist, our focus was on incorporating latent variables to capture within-person relationships and control for individual differences.Future studies could explore non-linear models to further validate these findings.Additionally, this study's results raised concerns about type 1 errors due to the floor effect.The MLR estimation showed no significant changes in the results of the RI-CLPM, while the results of the CLPM differed.Although this was likely influenced by listwise deletion, it highlighted the need for caution in interpreting the CLPM results.
In conclusion, there was a significant between-person correlation in the relationship between PEs and dissociation, while there was less of a longitudinal within-person relationship.The only significant longitudinal cross-lagged relationship was from PEs at age 14 to dissociation at age 16, suggesting that PEs may be a predictor of dissociation in midadolescence.

Declaration of competing interest
None.

Fig. 1 .
Fig. 1.The conventional cross-lagged panel model of longitudinal associations between psychotic experiences (PEs) and dissociation (Dis).Each regression and covariance shows standardized coefficients with 95 % CIs, adjusted for sex and age in months.Solid lines represent estimates with 95 % CIs which do not include 0. Comparative Fit Index (CFI) = 0.911, Tucker-Lewis Index (TLI) = 0.836, Root Mean Square Error of Approximation (RMSEA) = 0.084.

Fig. 4 .
Fig. 4. The between-person component of the random intercept cross-lagged panel model.Covariance (standardized, with 95 % CIs) between the random intercepts of PEs and Dis indicates between-person differences.PEs: Psychotic Experiences, Dis: Dissociation, RI: random intercept.

Table 1
Demographic data and PEs/dissociation scores.