Introduction
Traumatic life experiences are highly common among people with psychosis, with dose–response relationships between traumatic life events and psychosis risk indicating a potential causal link (Pastore, de Girolamo, Tafuri, Tomasicchio, & Margari, Reference Pastore, de Girolamo, Tafuri, Tomasicchio and Margari2022). Accordingly, systematic reviews indicate that symptoms of post-traumatic stress disorder (PTSD) play a role in the pathway from trauma to psychosis (Alameda et al., Reference Alameda, Rodriguez, Carr, Aas, Trotta, Marino and Murray2020; Bloomfield et al., Reference Bloomfield, Chang, Woodl, Lyons, Cheng, Bauer-Staeb and Lewis2021; Sideli et al., Reference Sideli, Murray, Schimmenti, Corso, La Barbera, Trotta and Fisher2020). This may explain the phenomenological overlap between psychotic symptoms and post-traumatic sequelae, such as voices that overlap literally or thematically with traumatic memories; or a shared sense of current threat characterizing both paranoia and post-traumatic hypervigilance (Compean & Hamner, Reference Compean and Hamner2019; Van Den Berg et al., Reference Van Den Berg, Tolmeijer, Jongeneel, Staring, Palstra, Van Der Gaag and Hardy2023). Psychological interventions for psychosis are becoming increasingly trauma-focused to adequately address this overlap and improve outcomes (Hardy et al., Reference Hardy, Keen, van den Berg, Varese, Longden, Ward and Brand2023).
In the latest International Classification of Disease (11th Edition; ICD-11), so-called ‘core’ symptoms of PTSD include re-experiencing of traumatic events (e.g. memory intrusions, flashbacks), avoidance of reminders of the events (either internally or externally), and hyperarousal (e.g. hypervigilance) (World Health Organization [WHO], 2019). The trauma histories typical of people with psychosis (i.e. repeated interpersonal victimization often during childhood) pose an increased risk of complex PTSD (cPTSD), classified in the latest ICD (Karatzias et al., Reference Karatzias, Hyland, Bradley, Cloitre, Roberts, Bisson and Shevlin2019; Trauelsen et al., Reference Trauelsen, Bendall, Jansen, Nielsen, Pedersen, Trier and Simonsen2015). cPTSD includes the aforementioned ‘core’ symptoms of PTSD, alongside disturbances of self-organization (DSOs): emotional dysregulation, negative self-concept, and interpersonal difficulties (WHO, 2019). Emerging evidence suggests cPTSD may be more common among people with psychosis than PTSD, and that DSOs play a role in the trauma–psychosis pathway (Panayi et al., Reference Panayi, Berry, Sellwood, Campodonico, Bentall and Varese2022). Like core PTSD symptoms, there is phenomenological overlap between psychotic symptoms and DSOs (e.g. affective blunting in negative symptoms of psychosis and emotional dysregulation in DSOs) (Favrod et al., Reference Favrod, Nguyen, Tronche, Blanc, Dubreucq, Chereau-Boudet and Llorca2019).
To theoretically capture this overlap and guide intervention, psychological models of PTSD in psychosis have been developed. In a seminal model, Morrison and colleagues proposed that traumatic memory intrusions may be appraised as a threatening sensory experience occurring in the present, leading to distressing or unhelpful coping strategies that maintain the experience (Morrison, Frame, & Larkin, Reference Morrison, Frame and Larkin2003). This model has since been expanded following developments in the field that indicate multiple trauma-related mechanisms may maintain psychosis, including DSO-consistent factors (e.g. emotional dysregulation, negative self-concept) that may perpetuate intrusive trauma memories, their appraisals, and subsequent coping strategies (Hardy, Reference Hardy2017). Recent analytic innovations may be applied to explore complex interrelationships between symptoms of cPTSD and psychosis, and test whether they mirror relationships proposed by theoretical models to guide promising new directions for trauma-focused psychosis interventions.
Network theory, a relatively new perspective in mental health research, proposes that mental health difficulties arise from self-perpetuating relationships between individual symptoms as opposed to an underlying ‘disease’ entity (Borsboom, Reference Borsboom2017). These relationships between elements can be represented in a network of variables, where symptoms are represented by circles (i.e. ‘nodes’) connected by lines (i.e. ‘edges’) that demonstrate the strength and direction of their relationship. Derived from network theory, network analysis is a data-driven statistical technique that can be used to visualize such networks, identifying the importance of each node in maintaining the network (Fried & Cramer, Reference Fried and Cramer2017). Further metrics can be derived to gain further insight on the structure of the network, such as clustering (identification of subcommunities of closely related nodes based on relative edge strength; Golino & Epskamp, Reference Golino and Epskamp2017) and centrality (quantifying the relative importance of each node in the network; Bringmann et al., Reference Bringmann, Pe, Vissers, Ceulemans, Borsboom, Vanpaemel and Kuppens2016).
Network analysis has been increasingly used to model the complexity of psychological difficulties, with metrics such as clustering applied to identifying psychological symptom communities/domains, and centrality to the identification of potentially tractable targets for mental health interventions (Hardy, O’Driscoll, Steel, Van Der Gaag, & Van Den Berg, Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021; Levin et al., Reference Levin, Hyland, Karatzias, Shevlin, Bachem, Maercker and Ben-Ezra2021; McElroy et al., Reference McElroy, Shevlin, Murphy, Roberts, Makhashvili, Javakhishvili and Hyland2019). Several studies have already applied network analysis to examine the interrelationships between PTSD and psychosis in cross-sectional samples. Consistent with the models outlined above, psychotic and PTSD symptoms are often bridged by post-traumatic hyperarousal and trauma-related beliefs (Astill Wright et al., Reference Astill Wright, McElroy, Barawi, Roberts, Simon, Zammit and Bisson2023; Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021; Jin et al., Reference Jin, Xu, Wang, Li, Wang, Sun and Wang2022). Hyperarousal and negative trauma-related beliefs are arguably consistent with DSOs (i.e. emotional dysregulation and negative self-concept, respectively), tentatively modeling the overlap between cPTSD and psychosis. A recent network analysis explicitly modeled this overlap using the only validated diagnostic measure of cPTSD in a trauma-exposed sample of people with comorbid psychosis and PTSD (Frost, O’Driscoll, Peters, & Hardy, Reference Frost, O’Driscoll, Peters and Hardy2023). Affective dysregulation was related to voices, and negative self-concept linked to delusions via self-blame, affirming the importance of trauma-related mechanisms beyond episodic memory (i.e. DSOs and trauma-related beliefs) in the above models of post-traumatic stress in psychosis.
Like most network analyses, networks of (c)PTSD and psychosis to date have been cross-sectional (Contreras, Nieto, Valiente, Espinosa, & Vazquez, Reference Contreras, Nieto, Valiente, Espinosa and Vazquez2019; Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023; Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021). However, the lack of temporality in cross-sectional networks limits insight into the directionality of psychological mechanisms that could clarify processes of symptom formation and maintenance. A growing number of studies therefore use longitudinal data to capture the dynamic interplay between symptoms using temporal network models (Blanchard, Contreras, Kalkan, & Heeren, Reference Blanchard, Contreras, Kalkan and Heeren2022; Epskamp, van Borkulo, et al., Reference Epskamp, van Borkulo, van der Veen, Servaas, Isvoranu, Riese and Cramer2018). In contrast to traditional longitudinal designs capturing developmental change over extended periods, intense longitudinal designs (i.e. multiple measurements recorded over short periods) better suit the self-perpetuating stability of psychopathological networks (Epskamp, Reference Epskamp2020). Experience sampling methodology (ESM) is well-placed to capture data of this kind, involving several measurements per day over a given period (Myin-Germeys et al., Reference Myin-Germeys, Kasanova, Vaessen, Vachon, Kirtley, Viechtbauer and Reininghaus2018; Trull & Ebner-Priemer, Reference Trull and Ebner-Priemer2009). Several studies have applied temporal network analysis to ESM data using multilevel vector autoregressive (mlVAR) modeling to model the dynamic relationships between individual symptoms in the flow of daily life (Blanchard et al., Reference Blanchard, Contreras, Kalkan and Heeren2022).
The application of temporal network analysis to intensive longitudinal data generates three complementary networks: between-subject, contemporaneous, and temporal networks (Epskamp, Borsboom, & Fried, Reference Epskamp, Borsboom and Fried2018). Between-subjects networks depict partial correlations between symptoms on average throughout the testing period, and as such can be used to visualize relationships that may manifest over longer periods (Epskamp & Fried, Reference Epskamp and Fried2018; Epskamp, Waldorp, Mõttus, & Borsboom, Reference Epskamp, Waldorp, Mõttus and Borsboom2018). On the other hand, contemporaneous networks depict partial correlations between symptoms within a single moment, and temporal networks depict regression coefficients of relationships between symptoms from one moment to the next (i.e. ‘lagged’ relationships) (Epskamp & Fried, Reference Epskamp and Fried2018; Epskamp, Waldorp, et al., Reference Epskamp, Waldorp, Mõttus and Borsboom2018). Relationships between symptoms may operate over small timeframes, and as such emerge in contemporaneous, but not temporal, networks (Epskamp, van Borkulo, et al., Reference Epskamp, van Borkulo, van der Veen, Servaas, Isvoranu, Riese and Cramer2018). Conversely, relationships may also operate over longer periods and as such emerge in the between-subjects network, but not in contemporaneous or temporal networks.
The aforementioned cross-sectional network analysis of cPTSD and psychotic symptoms analyzed baseline clinical data from a randomized controlled trial of trauma-focused cognitive-behavioral therapy for psychosis (Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023; Peters et al., Reference Peters, Hardy, Dudley, Varese, Greenwood, Steel and Morrison2022). A subsample of this group completed an additional ESM study prior to randomization, which found that increases in both core PTSD and DSOs temporally predicted subsequent increases in paranoia, voices, and visions in the flow of daily life (Panayi et al., Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024). Yet, several queries remain about the dynamics of cPTSD and psychotic symptoms, including the interdependence of effects (given models were tested separately for each positive symptom), symptom-specific effects (since core PTSD and DSOs were entered as two composite scores), and directionality of effects (since models were tested unidirectionally). By considering each symptom as an individual node, testing relationships between symptoms controlling for others in the network, and estimating networks over different timeframes, temporal network analysis is well-placed to address these queries and comprehensively assess cPTSD–psychosis symptom dynamics.
In summary, there have been no attempts to scrutinize the interrelations between post-traumatic sequelae and psychotic symptoms using temporal networks, and networks of this kind could clarify uncertainties around the potentially maintaining effect of cPTSD in trauma-exposed people with psychosis. This study therefore applied temporal network analysis to an ESM dataset collected via a clinical trial to establish the temporal dynamics of relationships between symptoms of cPTSD and psychosis. Specifically, we aimed to determine the persistence of relationships across different networks reflecting disparate timeframes as well as the potential temporal order of relationships between symptoms.
Method
Design
An intense-longitudinal ESM design was used (Myin-Germeys et al., Reference Myin-Germeys, Kasanova, Vaessen, Vachon, Kirtley, Viechtbauer and Reininghaus2018), wherein repeated measures of positive symptoms of psychosis (i.e. paranoia, voices, and visions), core PTSD (intrusive memories, avoidance, hyperarousal), and DSOs (emotional dysregulation, negative self-concept, and interpersonal difficulties) were taken up to 10 times a day over six consecutive days using a mobile app. A full description of the ESM protocol is described in Panayi et al. (Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024).
Participants
Data for this study were collected via the Study of Trauma and Recovery (STAR) trial, a randomized controlled trial testing the efficacy of and mechanisms underlying trauma-focused cognitive-behavioral therapy for psychosis (Peters et al., Reference Peters, Hardy, Dudley, Varese, Greenwood, Steel and Morrison2022). STAR trial participants were mental health service users recruited from five sites across the United Kingdom. A subsample of 153 participants who consented to additional ESM procedures prior to randomization were included in this study. Demographic and clinical characteristics are listed in Table 1; comparisons with the wider STAR sample are described in Panayi et al. (Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024).
Demographic and clinical characteristics (N = 153)
a Based on baseline PSYRATS data
b Based on baseline TALE data
Note: M, mean; SD, standard deviation; PTSD, post-traumatic stress disorder; DSOs, disturbances of self-organization; ITQ, International Trauma Questionnaire Cloitre et al. (Reference Cloitre, Shevlin, Brewin, Bisson, Roberts, Maercker and Hyland2018); GPTS, Green et al. Paranoid Thoughts Scale Freeman et al. (Reference Freeman, Loe, Kingdon, Startup, Molodynski, Rosebrock and Bird2021); PSYRATS, Psychotic Symptom Rating Scales Haddock et al.(Reference Haddock, McCarron, Tarrier and Faragher1999); TALE, Trauma and Life Events scale Carr et al. (Reference Carr, Hardy and Fornells-Ambrojo2018).
Inclusion criteria
All STAR trial participants met ICD-10 criteria for schizophrenia-spectrum diagnoses (F20–29) ascertained from the ICD-10 checklist by the research team, following clinical notes review and consultation with the care team, as appropriate, and scored ≥2 (‘moderate’ intensity) on the distress item of at least one psychotic symptom on the Psychotic Symptom Rating Scales (Haddock, McCarron, Tarrier, & Faragher, Reference Haddock, McCarron, Tarrier and Faragher1999). Participants also endorsed at least one traumatic life event on the Trauma And Life Events checklist (Carr, Hardy, & Fornells-Ambrojo, Reference Carr, Hardy and Fornells-Ambrojo2018) and met criteria for PTSD on the Clinician-Administered PTSD Scale for DSM-5 (Weathers et al., Reference Weathers, Bovin, Lee, Sloan, Schnurr, Kaloupek and Marx2018). Participants were not eligible for the STAR trial if their psychotic or PTSD symptoms were primarily organic in etiology, had a primary substance misuse diagnosis, required an interpreter to engage with the trial, or (within the prior 3 months) had a major medication change or received trauma-focused therapies. There were no additional criteria for taking part in the additional ESM study other than provision of informed consent for this additional ESM study. Of note, 31% of participants were recruited from Early Intervention for Psychosis (EIP) services in England, where best practice guidance advises against the use of potentially stigmatizing labels (e.g. schizophrenia, schizoaffective disorder) among those experiencing first-episode psychosis. Instead, ICD-10 F28 (other nonorganic psychotic disorder) or F29 (unspecified psychotic disorder) categories are routinely applied.
Baseline clinical measures
Specific baseline measures administered as part of the STAR trial assessment battery were used here for descriptive purposes. These include the Trauma and Life Events Checklist (Carr et al., Reference Carr, Hardy and Fornells-Ambrojo2018) to assess the occurrence of difficult life experiences, the International Trauma Questionnaire (Cloitre et al., Reference Cloitre, Shevlin, Brewin, Bisson, Roberts, Maercker and Hyland2018) to assess the presence and severity of PTSD and DSOs, Green et al. Paranoid Thoughts Scale-Revised (Freeman et al., Reference Freeman, Loe, Kingdon, Startup, Molodynski, Rosebrock and Bird2021) to measure paranoia and Psychotic Symptom Rating Scales (PSYRATS; Haddock et al., Reference Haddock, McCarron, Tarrier and Faragher1999) to measure voice hearing. The STAR trial added an adapted PSYRATs version capturing the presence of hallucinations in other modalities, used here to measure vision seeing (Tsang, Reference Tsangn.d.). A full description of the measures used in the STAR assessment battery is described in Peters et al. (Reference Peters, Hardy, Dudley, Varese, Greenwood, Steel and Morrison2022) and their application to the current study in Panayi et al. (Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024).
ESM measures
The ESM questionnaire involved 29 questions assessing nine domains; items used for this study are documented in Table 2. All items used in the analyses were scored on a 7-point Likert scale from 1 (‘Not at all’) to 7 (‘Very much so’). Items were based on previous studies of similar populations (Chun, Reference Chun2016; Kimhy et al., Reference Kimhy, Delespaul, Corcoran, Ahn, Yale and Malaspina2006), amended in consultation with stakeholders, including people with lived experience of trauma and psychosis, clinicians, and researchers. For an explanation of item derivation and validation, see Panayi et al. (Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024).
Breakdown of variables analyzed in this study and corresponding ESM items
a Reverse scored.
Note: PTSD, Post-traumatic stress disorder; DSOs, disturbances of self-organization.
Procedure
Participants were recruited from five National Health Service (NHS) Mental Health Trusts across the UK (NHS Research Ethics Committee ref: 20/LO/0853). The study involved an intensive longitudinal, time-contingent ESM design, consisting of up to 10 observations per day over a 6-day period. Observations were scheduled in a quasi-random nature, delivered randomly within consecutive 90-minute blocks adjusted to suit the waking hours of individual participants, in accordance with ESM best practice (Dejonckheere & Erbas, Reference Dejonckheere, Erbas, Myin-Germeys and Kuppens2022).
After providing informed consent for the ESM study, participants were supported by a researcher to download a mobile app, m-Path (Mestdagh et al., Reference Mestdagh, Verdonck, Piot, Niemeijer, Kilani, Tuerlinckx and Dejonckheere2023), onto their personal mobile phone or were provided a study phone. After completing a practice questionnaire to clarify understanding of the questions and resolve any queries, the researcher scheduled the notifications during the participants’ typical waking hours. Participants were then invited to complete ESM questionnaires administered via the app for six consecutive days, starting the following day. To address potential technical issues or promote procedural compliance, participants received a phone call on the second day of the study, and the response rate was monitored throughout the 6-day data collection period. All participants were debriefed and reimbursed at the end of the study.
Data analysis
The analyses were conducted in R 4.2.1 (R Core Team, 2023); the R code is presented in Supplementary Material 1. The mlVAR package (version 0.5.1; Epskamp, Deserno, et al., Reference Epskamp, Deserno and Bringmann2023) was used to estimate the temporal network models of cPTSD and psychosis symptoms. Shapiro–Wilk tests were used to assess normality in each variable. The assumption of stationarity (i.e. that variable means and standard deviations remain constant over time; Aalbers et al., Reference Aalbers, McNally, Heeren, de Wit and Fried2019; Bringmann et al., Reference Bringmann, Pe, Vissers, Ceulemans, Borsboom, Vanpaemel and Kuppens2016) was tested via Kwiatkowski–Phillips–Schmidt–Shin (KPSS) unit root tests using the R package tseries (version 0.54; Trapletti & Hornik, Reference Trapletti and Hornik2023).
To establish the temporal dynamics of cPTSD and psychotic symptoms, mlVAR was used to generate three complementary networks reflecting three timeframes: (1) a between-subjects network, i.e. Gaussian Graphical Model (GGM) composed of regularized partial correlations (after taking into account the remaining variables in the network) between participants’ means during the 6-day study duration period; (2) a contemporaneous network based on a GGM comprising edges that depict multilevel partial correlations between nodes in a single ESM measurement point (referred to as ‘moments’), after controlling for other contemporaneous and temporal associations; (3) a temporal network based on regression coefficients of lagged relationships between nodes from one timepoint to the next after controlling for all other nodes at the previous timepoint (Epskamp, Borsboom, et al., Reference Epskamp, Borsboom and Fried2018; Epskamp, Waldorp, et al., Reference Epskamp, Waldorp, Mõttus and Borsboom2018).
Model coefficients were plotted as graphical networks by mlVAR using the qgraph package (version 1.9.5; Epskamp, Costantini, et al., Reference Epskamp, Costantini, Haslbeck, Cramer, Waldorp and Schmittmann2023). Contemporaneous and between-subjects models were estimated conservatively, using the ‘AND-rule’ approach, which retains edges if both regressions on which the edge is based are significant (α = .05). Clustering metrics were applied to the between-subjects network using exploratory graph analysis (EGA) via the EGAnet package (Golino & Christensen, Reference Golino and Christensen2022) to avoid temporal interference while maintaining fullness of the data.
Results
Descriptive statistics
Means and standard deviations of within-participant means and standard deviations for each variable are presented in Table 3.
Means and standard deviations of within-participant means and within-participant standard deviations for all ESM variables
Note: M, mean; SD, standard deviation; PTSD, post-traumatic stress disorder; DSOs, disturbances of self-organization. All variables were measured using a 7-point Likert scale (range 1–7).
mlVAR estimation typically requires a minimum of 20 data points per participant, to prevent bias in within-person centering (Jordan, Winer, & Salem, Reference Jordan, Winer and Salem2020). Sensitivity analyses were conducted excluding participants with fewer than 20 observations (n = 32), but this did not affect the outcomes of the analyses. As such, the following results are presented from the full dataset (N = 153). Other indicators of adherence to the ESM protocol are listed in Panayi et al. (Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024).
Assumptions
Shapiro–Wilk tests indicated that memory intrusions, hyperarousal, emotional dysregulation, and negative self-concept were normally distributed, whereas avoidance, interpersonal difficulties, voices, visions, and paranoia were not (see Supplementary Table 1 for normality values and Supplementary Figures 1 and 2 for histograms). The size of partial correlation coefficients suggested no variables were multicollinear in any network (Epskamp & Fried, Reference Epskamp and Fried2018). KPSS tests suggested the data were stationary for all variables, except the DSO-emotional dysregulation negative self-concept subscales (see Supplementary Table 2 for KPSS test values). Multilevel regressions indicated extremely small increases (both b’s < .01) in these subscale scores as the measurement week progressed. Considering these effect sizes in combination with the KPSS probability values approaching nonsignificance (all p’s = .04) and the high Type I error of KPSS tests (Jordan et al., Reference Jordan, Winer and Salem2020), supplementary augmented Dickey–Fuller (ADF) tests were used to confirm these results, in line with best statistical practice for temporal network analysis (Blanchard et al., Reference Blanchard, Contreras, Kalkan and Heeren2022). ADF tests confirmed trend and level stationarity of these subscales in all participants (all p’s < .01). Prior research suggests temporal networks may be robust against violations of these assumptions, but this remains unclear (Aalbers et al., Reference Aalbers, McNally, Heeren, de Wit and Fried2019; Blanchard et al., Reference Blanchard, Contreras, Kalkan and Heeren2022; Faelens, de Putte, Hoorelbeke, de Raedt, & Koster, Reference Faelens, de Putte, Hoorelbeke, de Raedt and Koster2021). These assumptions are therefore reported here in the interest of fullness and supporting efforts to establish the robustness of temporal networks.
Network estimation and visualization
Between-subjects network
The between-subjects network is illustrated in Figure 1a, depicting intraindividual correlations between mean levels of each node across the ESM testing period. This network suggests, on average throughout the week, paranoia correlated positively with intrusive memories and hyperarousal, whereas visions associated with the former and voices the latter. Regarding DSOs, the network indicates that paranoia correlated positively with interpersonal difficulties, and voices with emotional dysregulation, on average throughout the week. Network coefficients are presented in Supplementary Table 3.
Graphical models representing (a) between-subjects network depicting average symptom relationships over the course of the testing period; (b) contemporaneous network depicting symptom relationships within a single moment, controlling for all other relationships in the network; (c) temporal network depicting symptom relationships between each moment, controlling for all relationships at the previous moment. Blue edges denote positive relationships; thicker edges denote stronger relationships; arrows denote the direction of prediction. Node colors represent symptom groups, not outcomes of clustering analysis.
Contemporaneous network
The contemporaneous network – i.e. node associations from a single moment after controlling for other contemporaneous and temporal associations – is illustrated in Figure 1b. In this timeframe, relationships persisted for paranoia with hyperarousal and interpersonal difficulties, and (weakly) between visions and intrusive memories, and voices and emotional dysregulation. Voices, visions, and (most strongly) paranoia all newly related to negative self-concept in this network. New relationships also emerged between paranoia and emotional dysregulation, and between voices and intrusive memories. Network coefficients are presented in Supplementary Table 4.
Temporal network
The temporal network – i.e. the predictive strength of each node from one moment (t) to the next (t + 1) – is illustrated in Figure 1c. Autoregressive loops depict the degree to which nodes predicted themselves in the interval between measurements; particularly strong loops for emotional dysregulation, negative self-concept, hyperarousal, and paranoia suggest these experiences may be self-perpetuating and as such remain relatively stable over time. Paranoia showed positive, reciprocal momentary relationships with emotional dysregulation, negative self-concept, and (most strongly) hyperarousal. Neither voices nor visions demonstrated reciprocal temporal relationships with other nodes in the network; visions solely predicted negative self-concept while voices were solely predicted by paranoia. Network coefficients are presented in Supplementary Table 5.
Centrality metrics
Measures of centrality (i.e. the degree to which nodes predict or are predicted by other nodes in the network) were derived from the temporal network matrix. All nodes exerted some influence on the network, except voices and avoidance. Similarly, all nodes received some influence from the network, except visions. Paranoia demonstrated the highest in- and outstrength centrality in the network. Emotional dysregulation demonstrated the second-highest outstrength, while hyperarousal showed the second-highest in-strength. Centrality plots are provided in Supplementary Figure 3.
Discussion
In a large ESM sample of 153 participants with comorbid PTSD and psychosis, we applied temporal network analysis to establish the temporal dynamics of relationships between symptoms of cPTSD and psychosis in the flow of daily life. Our between-subjects network suggested that, on average over the testing period, all cPTSD symptoms (aside from avoidance and negative self-concept) related to at least one positive psychotic symptom. Many average relationships persist at the momentary level, indicating symptoms of cPTSD and psychosis may co-occur, especially paranoia with hyperarousal and negative self-concept where associations are strongest. Associations between negative self-concept and paranoia, voices, and visions; between paranoia and emotional dysregulation; and between voices and intrusive memories were absent in the between-subjects network, suggesting these relationships may only manifest over shorter timeframes. The temporal network suggested that paranoia reciprocally predicted, and was predicted by, hyperarousal, negative self-concept and emotional dysregulation from moment to moment, as opposed to temporally unidirectional relationships. In contrast, cPTSD symptoms did not relate to voices in the temporal network, indicating these may co-occur, but not temporally predict one another. Likewise, visions were associated with intrusive memories in the between subjects, but not temporal, network, suggesting a temporal relationship may manifest over longer timeframes.
There is a consistent, strong relationship between paranoia and hyperarousal in all three networks, suggesting these experiences are closely interlinked in our sample. Paranoia and hypervigilance often co-occur, as individuals are likely to be on edge or on guard under interpersonal threat (Alsawy, Wood, Taylor, & Morrison, Reference Alsawy, Wood, Taylor and Morrison2015; Freeman et al., Reference Freeman, Thompson, Vorontsova, Dunn, Carter, Garety and Ehlers2013). Though, safety behaviors in response to threat, including hypervigilance, have been shown to counterintuitively maintain symptoms of both PTSD and psychosis (Beierl, Böllinghaus, Clark, Glucksman, & Ehlers, Reference Beierl, Böllinghaus, Clark, Glucksman and Ehlers2020; Greenburgh et al., Reference Greenburgh, Barnby, Delpech, Kenny, Bell and Raihani2021). Consistently, a prior cross-sectional network analysis demonstrated the bridging role of hyperarousal between PTSD and psychosis (Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021). Our findings demonstrate temporality of these effects, indicating a reciprocal relationship between hyperarousal and paranoia in people with co-occurring cPTSD and psychosis symptoms that – according to centrality indices – may maintain the comorbidity. The maintaining role of paranoia/hyperarousal in this comorbidity somewhat contrasts with historic proposals that post-traumatic sequelae and psychosis overlap via intrusive trauma memories (Morrison et al., Reference Morrison, Frame and Larkin2003). That said, our findings provide temporal evidence for the stipulated emphasis on current threat in cognitive models of PTSD and psychotic symptoms (Ehlers & Clark, Reference Ehlers and Clark2000; Freeman, Reference Freeman2007; Hardy, Reference Hardy2017). Notably, positive psychotic symptoms insufficient to warrant a schizophrenia-spectrum diagnosis have been repeatedly documented among people with PTSD (Braakman, Kortmann, & Van Den Brink, Reference Braakman, Kortmann and Van Den Brink2009; Shevlin, Armour, Murphy, Houston, & Adamson, Reference Shevlin, Armour, Murphy, Houston and Adamson2011). Whether our findings apply to models of psychosis in PTSD requires investigation, to indicate the need for adaptation of trauma interventions to target these experiences.
All three psychosis symptoms related to negative self-concept in the contemporaneous network. Prior networks demonstrated the bridging role of trauma-related beliefs (including negative self-concept) between PTSD and psychosis (Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023; Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021). We extend these findings by demonstrating temporal dynamics of this relationship; namely, that psychological mechanisms at play between trauma-related beliefs and psychosis may operate quickly, hence their presence only in the contemporaneous network. The temporal network suggests that, over time, negative self-concept may impact voices indirectly via paranoia. Consistently, prior cross-sectional analyses demonstrate delusional beliefs effectively mediate the relationship between PTSD and hallucinations (Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023; Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021) and with a previous ESM study that found that increases in delusional thinking preceded both visual and auditory hallucinations (Oorschot et al., Reference Oorschot, Lataster, Thewissen, Bentall, Delespaul and Myin-Germeys2012). For some, paranoia may be an expression of negative self-beliefs that prevents disconfirmation of (and thus maintains) said beliefs (Bentall, Corcoran, Howard, Blackwood, & Kinderman, Reference Bentall, Corcoran, Howard, Blackwood and Kinderman2001; Kesting, Bredenpohl, Klenke, Westermann, & Lincoln, Reference Kesting, Bredenpohl, Klenke, Westermann and Lincoln2013). The reciprocal relationship observed in the temporal network provides temporal evidence of this mutually maintaining effect in people with comorbid cPTSD and psychosis.
In contrast, the temporal network suggests that visions may unidirectionally precede negative self-concept from moment to moment. The potentially activating effect of visions on negative trauma-related beliefs about the self is uncontroversial given their thematic (if not direct) trauma-relatedness (Van Den Berg et al., Reference Van Den Berg, Tolmeijer, Jongeneel, Staring, Palstra, Van Der Gaag and Hardy2023). The findings of the temporal network therefore offer support to modern conceptualizations of visions as disintegrated trauma memories in trauma-exposed people with psychosis (Hardy, Reference Hardy2017; Morrison et al., Reference Morrison, Frame and Larkin2003). Accordingly, visions and memory intrusions did (at least weakly) co-occur in the contemporaneous network, with potential temporal relationships manifesting over longer periods, as these nodes were related in the between-subjects, but not temporal, network.
In addition to negative self-concept, our temporal network also implicates a temporal relationship between paranoia and emotional dysregulation. ESM studies have previously uncovered predictive effects of emotional dysregulation on paranoia (Kimhy et al., Reference Kimhy, Lister, Liu, Vakhrusheva, Delespaul, Malaspina and Wang2020; Panayi et al., Reference Panayi, Peters, Bentall, Hardy, Berry, Sellwood and Varese2024); our network suggests this is reciprocal. As above, this demonstrates the dynamic nature of cPTSD–psychosis symptom relationships: unhelpful emotion regulation strategies in response to paranoid ideation may paradoxically maintain distressing beliefs, and vice versa (Lim, Gleeson, Alvarez-Jimenez, & Penn, Reference Lim, Gleeson, Alvarez-Jimenez and Penn2018). A cross-sectional network of cPTSD and psychosis in this sample did not show a direct relationship between emotional dysregulation and delusions, potentially indicating our findings are specific to paranoia, or the timeframes captured by temporal networks (Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023). That said, a prior temporal network analysis did not demonstrate momentary relationships of paranoia with emotional dysregulation or self-esteem (closely related to negative self-concept) (Contreras, Valiente, Heeren, & Bentall, Reference Contreras, Valiente, Heeren and Bentall2020). This disparity is likely due to the subclinical nature of Contreras and colleagues’ sample, where additional protective factors may be present that prevent the development of clinically significant difficulties (perhaps mechanistically via the prevention of interrelationships between DSOs and paranoia, in light of our findings).
Like the aforementioned relationship between visions and intrusive memories, voices related to emotional dysregulation in the between-subjects and contemporaneous, but not temporal, network. Their co-occurrence aside, temporal relationships between these nodes may therefore manifest over longer periods, consistent with their relationship in meta-analyses and cross-sectional networks (Bloomfield et al., Reference Bloomfield, Chang, Woodl, Lyons, Cheng, Bauer-Staeb and Lewis2021; Frost et al., Reference Frost, O’Driscoll, Peters and Hardy2023). Emotion regulation moderates stress reactivity, especially among people with PTSD (Badour & Feldner, Reference Badour and Feldner2013; Shapero, Abramson, & Alloy, Reference Shapero, Abramson and Alloy2016). As such, the relationship between emotional dysregulation and voices in this study is consistent with the affective pathway to psychosis that suggests stress reactivity precipitates psychotic symptoms (Collip et al., Reference Collip, Wigman, Myin-Germeys, Jacobs, Derom, Thiery and van Os2013; Myin-Germeys & van Os, Reference Myin-Germeys and van Os2007). Our findings indicate this pathway may be especially reflective of people with comorbid cPTSD and psychosis.
An important consideration contextualizing the findings is the sample. The use of a clinical sample is a strength, given the relative centrality of paranoia emergent here that was not evident in prior network analyses of PTSD and psychosis in the general population (Astill Wright et al., Reference Astill Wright, McElroy, Barawi, Roberts, Simon, Zammit and Bisson2023; Jin et al., Reference Jin, Xu, Wang, Li, Wang, Sun and Wang2022). That said, the clinical trial through which the data for this study were collected may have introduced sample bias, since inclusion in the trial (principally) required participants to meet criteria for DSM-5 PTSD alongside meeting diagnostic criteria for an ICD-10 schizophrenia spectrum disorder and current, distressing psychotic symptoms (Peters et al., Reference Peters, Hardy, Dudley, Varese, Greenwood, Steel and Morrison2022). The sample may therefore be particularly affected by their traumatic life experiences in ways that are not representative of all people experiencing psychosis. Indeed, in contrast to the present study, paranoia was not highly central in a cross-sectional network of PTSD and psychosis in another clinical sample (Hardy et al., Reference Hardy, O’Driscoll, Steel, Van Der Gaag and Van Den Berg2021). Our findings may instead be particularly relevant to psychological interventions offered in EIP services, from which approximately one-third of participants were recruited.
Our networks should be considered in light of the interval phrasing of ESM items (i.e. ‘Since the last beep…’) used to generate specific PTSD and DSO nodes. A notable advantage of the study, this wording allowed us to better capture the potential infrequency of PTSD symptoms and psychologically reflective nature of DSOs that are likely to be missed by momentary (i.e. ‘Right now…’) wording, as is standard in ESM questionnaire design (Eisele, Kasanova, & Houben, Reference Eisele, Kasanova, Houben, Myin-Germeys and Kuppens2022). Strong relationships between items captured within the same timeframe may therefore reflect their temporal grouping (e.g. hallucination nodes being more strongly connected to each other than DSO nodes since they are both momentarily measured) as opposed to specific psychological mechanisms we propose. Additionally, the combination of momentary and interval items obfuscates the interpretation of temporal relationships, for instance, those in the contemporaneous network may reflect relationships in the interval between measurements as opposed to those within a single measurement point. Given this trade-off between measurement against capturing of infrequent but clinically important experiences, it will be important for future work to elucidate whether different timeframes impact network connection estimates. With temporal network analysis in its infancy, it remains unclear whether and how interval items affect outcomes, and we therefore encourage future research to estimate these networks using data across different timeframes to elucidate long-term relationships and psychological mechanisms.
Formal statistical power calculations for network analysis have yet to be devised, though a minimum of 20 observations per participant per node is considered ideal (Epskamp, Borsboom, et al., Reference Epskamp, Borsboom and Fried2018). As such, our sample size would not have allowed for the reliable estimation of a more complex network including potentially important additional nodes that could facilitate the identification of mechanistic pathways between cPTSD and psychosis. For instance, whether DSOs interact with paranoia via appraisals and distress (Garety, Kuipers, Fowler, Freeman, & Bebbington, Reference Garety, Kuipers, Fowler, Freeman and Bebbington2001; Morrison, Reference Morrison2001) or maladaptive coping strategies (e.g. substance use, dissociation; Goldstein et al., Reference Goldstein, Smith, Chou, Saha, Jung, Zhang, Pickering, Ruan, Huang and Grant2016; Hyland et al., Reference Hyland, Shevlin, Fyvie, Cloitre and Karatzias2020; Pilton et al., Reference Pilton, Varese, Berry and Bucci2015; Swendsen et al., Reference Swendsen, Ben-Zeev and Granholm2011). The exclusive focus on cPTSD in our study provides a foundation for future research that may test such pathways, consistent with the hypothesis-generating nature of network analysis (Epskamp, Maris, Waldorp, & Borsboom, Reference Epskamp, Maris, Waldorp and Borsboom2018).
As the first study exploring cPTSD and psychosis using complex research and statistical methodologies (i.e. ESM and network analysis) in a clinical population, our study has important theoretical and clinical implications. Our findings indicate that models of PTSD in psychosis apply to people with cPTSD, and provide temporal evidence for multifactorial, reciprocal pathways between post-traumatic sequelae and psychosis (e.g. sense of current threat; disintegrated trauma memories) (Berry, Varese, & Bucci, Reference Berry, Varese and Bucci2017; Gumley & MacBeth, Reference Gumley and MacBeth2007; Hardy, Reference Hardy2017). The central role of threat is not surprising given the heightened exposure to interpersonal victimization in people with psychosis, and etiological role of threat perception in psychosis (Heriot-Maitland, Wykes, & Peters, Reference Heriot-Maitland, Wykes and Peters2022; Trauelsen et al., Reference Trauelsen, Bendall, Jansen, Nielsen, Pedersen, Trier and Simonsen2015). There is ongoing debate as to the need for stabilization phases in trauma therapies that promote safety and develop helpful emotion regulation strategies (Hoeboer et al., Reference Hoeboer, de Kleine, Oprel, Schoorl, van der Does and van Minnen2021; Ross, Sharma-Patel, Brown, Huntt, & Chaplin, Reference Ross, Sharma-Patel, Brown, Huntt and Chaplin2021). Insofar as stabilization phases address paranoia and hyperarousal, our findings indicate the importance thereof alongside memory processing in trauma-focused psychosis interventions, as is the case with typical trauma therapies (e.g. trauma-focused cognitive behavioral therapy; eye-movement desensitization and reprocessing therapy [EMDR]) (Hardy et al., Reference Hardy, Keen, van den Berg, Varese, Longden, Ward and Brand2023; Keen, Hunter, & Peters, Reference Keen, Hunter and Peters2017; Varese et al., Reference Varese, Sellwood, Pulford, Awenat, Bird, Bhutani and Bentall2023). Ongoing clinical trials of trauma-focused interventions may confirm whether DSOs confer important treatment targets in psychosis (Burger et al., Reference Burger, van der Linden, Hardy, de Bont, van der Vleugel, Staring and van den Berg2022; Karatzias, Reference Karatzias2022; Peters et al., Reference Peters, Hardy, Dudley, Varese, Greenwood, Steel and Morrison2022). Our findings also indicate that novel interventions aimed specifically at promoting a sense of safety may be especially helpful in people with psychosis and comorbid cPTSD (e.g. Freeman et al., Reference Freeman, Waite, Isham, Freeman, Kabir, Kenny and Leal2024). Indeed, emerging evidence indicates DSOs are responsive to EMDR in early psychosis presentations (Varese et al., Reference Varese, Sellwood, Pulford, Awenat, Bird, Bhutani and Bentall2023).
Conclusions
Cross-sectional studies have established the mediating role of PTSD in the relationship between traumatic life experiences and psychotic symptoms. Prior network analyses have found specific paths between PTSD and psychosis via trauma-related beliefs and hypervigilance. Our study extends these findings by demonstrating the additional influence of DSOs on psychosis and adds directionality to prior networks that suggest a pivotal role of paranoia in maintaining cPTSD and psychosis. This has important implications for future studies investigating the cPTSD–psychosis comorbidity and the ongoing development of trauma-focused psychosis interventions.
Supplementary material
To view supplementary material for this article, please visit http://doi.org/10.1017/S0033291725000030.
Acknowledgments
The authors would like to thank the ESRC and NIHR for funding this study. Thank you also to all participants of the STAR trial for their participation.
The authors would also like to thank the following STAR trial collaborators for their contributions:
Affiliated with South London and Maudsley NHS Foundation Trust
Lucy Ring
Christopher Shoulder
Morwennna Rickard
Firat Denizcaglar
Kavita Bains
Freya Meyer
Leilia Jameel
Majella Byrne
Nadine Keen
Sarah Swan
Rachel Frost
Ze Freeman
Gary Ngai
Jess Bird
Constance Harvey
Edward Bickers
Kaisha Santosa
Samantha Mansell
Siamara Goddard
Molly Bird
Harriet Mellotte
Rhea Kohli
Katja Shultz
Antonia Obele
Elizabeth Leahy
Max Dormitontov
Faiza Abbaz
Ho Chak (Milton) Suen
Affiliated with King’s College London
Margaret Heslin
Emma Tassie
Sarah Byford
Elizabeth Kuipers
Laura Potts
Inez Verdaasdonk
Affiliated with Greater Manchester Mental Health NHS Foundation Trust
Tony Morrison
Wendy Jones
Eleanor Longden
Aqsa Choudary
Marina Sandys
Kim Towey-Swift
Elizabeth Murphy
Vicky Brooks
Samantha Bowe
Alice Newton-Braithwaite
Elliot Brewer
Leah Orme
Affiliated with University of Manchester
Tony Morrison
Affiliated with Cumbria, Northumberland, Tyne, and Wear NHS Foundation Trust
Rebecca Miskin
Laura McCartney
Marsha Cochrane
Antonia Newman
Sarah White
Nina Cioroboiu
Louise Prentice
Jane Mitchell
Doug Turkington
Kevin Meares
Libby Oakes
Affiliated with Oxford Health NHS Foundation Trust
Amy Langman
Emily Smyth
Georgina Kirtland
Natalie Barnes
Kate Costello
Collins Larrys
Iulia Grabovschi
Charlotte Sagnay de la Bastida
Affiliated with Sussex Partnership NHS Foundation Trust
Kathryn Greenwood
Cat Sacadura
Sarah Mansfield
Ann Steele
Lauren Edwards
Dan Elton
Angie Culham
Nick Grey
Molly Heeger
Lucy Fisher
Krutika Sharma
Chris Scane
Lisa Wood
Joseph Sherborne
Lauren Mose
Guy Emery
Aparajita Pandey
Affiliated with University of Sussex
Kathryn Greenwood
David Fowler
Funding statement
This work was supported by the Economic and Social Research Council (2488414 to P.P.) and the National Institute of Health Research (NIHR) (NIHR128623 and NIHR130971). A.H. was funded in part by the NIHR Maudsley Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London.
Competing interest
E.P., A.H., F.V., C.S., and R.U. provide psychological therapies for individuals with psychosis and/or PTSD in NHS settings, and E.P. is the Director of a psychological therapies specialist service for psychosis (PICuP). C.S. has written manuals for psychological therapies for psychosis and psychological formulation for which they receive book royalties (from APPI; Guildford Press; Wiley; Routledge; New Harbinger). E.P., A.H., F.V., and C.S. are employed to provide training and/or receive fees (or generate fees for their clinics or research units) for workshops and presentations on psychological therapies for psychosis and/or PTSD; E.P., A.H., F.V., and C.S. hold or have held grants to carry out trials of psychological therapy for individuals with psychosis.
Ethical standard
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
