Hostname: page-component-78c5997874-94fs2 Total loading time: 0 Render date: 2024-11-14T07:25:33.326Z Has data issue: false hasContentIssue false

Longitudinal symptomatic interactions in long-standing schizophrenia: a novel five-point analysis based on directed acyclic graphs

Published online by Cambridge University Press:  05 August 2021

Giusi Moffa
Affiliation:
Department of Mathematics and Computer Science, University of Basel, Basel, Switzerland Division of Psychiatry, University College London, 149 Tottenham Court Road, London W1T 7NF, UK
Jack Kuipers
Affiliation:
D-BSSE, ETH Zurich, Basel, Switzerland
Giuseppe Carrà
Affiliation:
Division of Psychiatry, University College London, 149 Tottenham Court Road, London W1T 7NF, UK Department of Medicine and Surgery, University of Milano Bicocca, Via Cadore 48, Monza 20900, Italy
Cristina Crocamo
Affiliation:
Department of Medicine and Surgery, University of Milano Bicocca, Via Cadore 48, Monza 20900, Italy
Elizabeth Kuipers
Affiliation:
Department of Psychology, IoPPN, King's College London, London SE5 8AF, UK
Matthias Angermeyer
Affiliation:
Department of Psychiatry, University of Leipzig, Johannisallee 20, 04137 Leipzig, Germany
Traolach Brugha
Affiliation:
Department of Health Sciences, College of Life Sciences, University of Leicester, Centre for Medicine, University Road, Leicester LE1 7RH, UK
Mondher Toumi
Affiliation:
Laboratoire de Santé Publique, Université de la Méditerranée, Marseille, France
Paul Bebbington*
Affiliation:
Division of Psychiatry, University College London, 149 Tottenham Court Road, London W1T 7NF, UK
*
Author for correspondence: Paul Bebbington, E-mail: [email protected]
Rights & Permissions [Opens in a new window]

Abstract

Background

Recent network models propose that mutual interaction between symptoms has an important bearing on the onset of schizophrenic disorder. In particular, cross-sectional studies suggest that affective symptoms may influence the emergence of psychotic symptoms. However, longitudinal analysis offers a more compelling test for causation: the European Schizophrenia Cohort (EuroSC) provides data suitable for this purpose. We predicted that the persistence of psychotic symptoms would be driven by the continuing presence of affective disturbance.

Methods

EuroSC included 1208 patients randomly sampled from outpatient services in France, Germany and the UK. Initial measures of psychotic and affective symptoms were repeated four times at 6-month intervals, thereby furnishing five time-points. To examine interactions between symptoms both within and between time-slices, we adopted a novel technique for modelling longitudinal data in psychiatry. This was a form of Bayesian network analysis that involved learning dynamic directed acyclic graphs (DAGs).

Results

Our DAG analysis suggests that the main drivers of symptoms in this long-term sample were delusions and paranoid thinking. These led to affective disturbance, not vice versa as we initially predicted. The enduring relationship between symptoms was unaffected by whether patients were receiving first- or second-generation antipsychotic medication.

Conclusions

In this cohort of people with chronic schizophrenia treated with medication, symptoms were essentially stable over long periods. However, affective symptoms appeared driven by the persistence of delusions and persecutory thinking, a finding not previously reported. Although our findings as ever remain hostage to unmeasured confounders, these enduring psychotic symptoms might nevertheless be appropriate candidates for directly targeted psychological interventions.

Type
Original Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re- use, distribution and reproduction, provided the original article is properly cited.
Copyright
Copyright © The Author(s), 2021. Published by Cambridge University Press

Introduction

Early studies of severe mental disorders distinguished those in which abnormalities of perception and interpretation seemed of central importance, from those where the salient abnormalities reflected mood disturbance. This formed the basis of Kraepelin's dichotomy between manic depressive illness and dementia praecox (otherwise schizophrenia). However, cross-sectional and longitudinal associations of psychotic experiences with anxiety and depression are now well established (e.g. Ben-Zeev, Ellington, Swendsen, & Granholm, Reference Ben-Zeev, Ellington, Swendsen and Granholm2011; Krabbendam et al., Reference Krabbendam, Myin-Germeys, De Graaf, Vollebergh, Nolen, Iedema and Van Os2004; Quattrone et al., Reference Quattrone, Di Forti, Gayer-Anderson, Ferraro, Jongsma, Tripoli and Reininghaus2019; Sax et al., Reference Sax, Strakowski, Keck, Upadhyaya, West and McElroy1996; Scott et al., Reference Scott, Martin, Bor, Sawyer, Clark and McGrath2009; Smith et al., Reference Smith, Fowler, Freeman, Bebbington, Bashforth, Garety and Kuipers2006; Thewissen et al., Reference Thewissen, Bentall, Oorschot, Campo J, van Lierop, van Os and Myin-Germeys2011; Upthegrove, Marwaha, & Birchwood, Reference Upthegrove, Marwaha and Birchwood2017; Yung et al., Reference Yung, Buckby, Cosgrave, Killackey, Baker, Cotton and McGorry2007), whereas a smaller number of studies indicate similar findings for low self-esteem, worry and sleep disturbance (e.g. Freeman et al. Reference Freeman, Stahl, McManus, Meltzer, Brugha, Wiles and Bebbington2012; Hartley, Haddock, Vasconcelos E Sa, Emsley, & Barrowclough, Reference Hartley, Haddock, Vasconcelos E Sa, Emsley and Barrowclough2014; Sheaves et al. Reference Sheaves, Bebbington, Goodwin, Harrison, Espie, Foster and Freeman2016). These affective changes were initially explained as a secondary response to the impact of schizophrenia on social status and circumstances (Birchwood, Iqbal, & Upthegrove, Reference Birchwood, Iqbal and Upthegrove2005). However, they might equally reflect mediation of the emergence of psychotic phenomena subsequent to psychosocial trauma (Hardy et al., Reference Hardy, Emsley, Freeman, Bebbington, Garety, Kuipers and Fowler2016; Marwaha & Bebbington, Reference Marwaha and Bebbington2015).

Whatever the mechanisms, depression and anxiety regularly precede the earliest stages of psychosis (Bird, Evans, Waite, Loe, & Freeman, Reference Bird, Evans, Waite, Loe and Freeman2019; Freeman, Taylor, Molodynski, & Waite, Reference Freeman, Taylor, Molodynski and Waite2019; Fusar-Poli, Nelson, Valmaggia, Yung, & McGuire, Reference Fusar-Poli, Nelson, Valmaggia, Yung and McGuire2014) and a propensity to affective dysregulation may therefore be intrinsic to the schizophrenic process (Marwaha, Broome, Bebbington, Kuipers, & Freeman, Reference Marwaha, Broome, Bebbington, Kuipers and Freeman2014). Thus, symptoms of low self-esteem, worry, sleep disturbance, anxiety and depression may constitute contributory causes of specific psychotic symptoms, especially paranoia (e.g. Bebbington et al., Reference Bebbington, McBride, Steel, Kuipers, Radovanovič, Brugha and Freeman2013; Freeman, Reference Freeman2016; Kuipers et al., Reference Kuipers, Garety, Fowler, Freeman, Dunn and Bebbington2006). Damaging social experiences also seem to be implicated in the development of psychotic phenomena (e.g. Bebbington et al., Reference Bebbington, Jonas, Kuipers, King, Cooper, Brugha and Jenkins2011; Hailes, Ronggin, Danese, & Fazel, Reference Hailes, Ronggin, Danese and Fazel2019; Kelleher et al., Reference Kelleher, Keeley, Corcoran, Ramsay, Wasserman, Carli and Cannon2013; McGrath et al., Reference McGrath, Saha, Lim, Aguilar-Gaxiola, Alonso and Andrade2017). This relationship might in turn be mediated or moderated by non-psychotic symptoms (e.g. Gibson, Alloy, & Ellman, Reference Gibson, Alloy and Ellman2016; Gracie et al., Reference Gracie, Freeman, Green, Garety, Kuipers, Hardy and Fowler2007; Marwaha & Bebbington, Reference Marwaha and Bebbington2015). This has clinical relevance, as such symptoms might then serve as significant targets for psychological or pharmacological treatments. Finally, although psychoactive substances such as cannabis may elicit paranoia by directly inducing anomalous perceptual experiences such as hallucinations, they may also operate through these common affective processes (Freeman et al., Reference Freeman, Dunn, Murray, Evans, Lister, Antley and Morrison2015).

Such interactional models are difficult to analyse. However, recent advances in statistical methods allow stronger inferences about complex causal relationships. Applied to mental disorders, these treat the pattern of symptoms as a causal system of individual interacting variables (Borsboom & Cramer, Reference Borsboom and Cramer2013; Fried et al., Reference Fried, van Borkulo, Cramer, Boschloo, Schoevers and Borsboom2017; Isvoranu et al., Reference Isvoranu, van Borkulo, Boyette, Wigman, Vinkers and Borsboom2017; Isvoranu, Borsboom, van Os, & Guloksuz, Reference Isvoranu, Borsboom, van Os and Guloksuz2016; Klippel et al., Reference Klippel, Myin-Germeys, Chavez-Baldini, Preacher, Kempton, Valmaggia and Reininghaus2017; McNally, Reference McNally2016; McNally et al., Reference McNally, Robinaugh, Wu, Wang, Deserno and Borsboom2015; Murphy et al., Reference Murphy, Tully, Pyle, Gumley, Kingdon, Schwannauer and Morrison2017; Wigman, de Vos, Wichers, van Os, & Bartels-Velthuis, Reference Wigman, de Vos, Wichers, van Os and Bartels-Velthuis2017). Several methods are available for such analyses. They include association networks, partial correlation networks and relative importance networks (McNally, Reference McNally2016). All describe the strength and the relative importance of different covariates in the prediction, but not causation. We have therefore argued for the use in psychological research of probabilistic graphical models based on directed acyclic graphs (DAGs). These provide improved insights into putative causal links, and allow us to represent the full picture in one unique model. In this way, we can more easily appreciate the complexity of the mechanisms linking the variables to each other (Kuipers, Moffa, Kuipers, Freeman, & Bebbington, Reference Kuipers, Moffa, Kuipers, Freeman and Bebbington2019; Moffa et al., Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017).

DAGs are graphical structures underlying Bayesian networks modelling the overall dependence structure of multiple variables. The specific variables we wish to analyse are represented as nodes connected by directed edges, and these together form a network (see Moffa et al., Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017 for a more detailed description). Kuipers and Moffa (Reference Kuipers and Moffa2017) introduced a novel Bayesian method for learning DAG structures and have more recently extended it to achieve greater efficiency (Kuipers et al., 2018; https://arxiv.org/abs/1803.07859). The Bayesian approach to learning DAG structures coupled with Pearl's do calculus (Pearl, Reference Pearl2009) allows us to derive ranges of putative causal effects. We have previously used this strategy to demonstrate the role of affective change in mechanisms linking bullying with persecutory ideation and hallucinations (Moffa et al., Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017).

Bayesian networks can also be useful for modelling the overall dependence structure of multiple variables measured over different time-points. In such dynamic Bayesian networks (DBNs), we may refer to the individual waves of data collection as time-slices: there are thus nodes for each variable in each time-slice. Nodes may be connected within a time-slice, but may also be linked to variables in the follow-up time-slice, thereby forming a larger DAG structure incorporating measurements of all variables over time. We have previously applied this form of Bayesian network analysis to initial and 18-month follow-up data from the 2000 British National Psychiatric Morbidity survey in order to evaluate the interaction between non-psychotic and psychotic phenomena over time (Kuipers et al., Reference Kuipers, Moffa, Kuipers, Freeman and Bebbington2019).

Schizophrenia is classically a chronic condition, and even with modern clinical management and the consistent deployment of antipsychotic medication some sufferers certainly experience enduring and disabling symptoms. Establishing the mechanisms of symptom persistence may suggest new initiatives for treatment. An opportunity to do this is provided by the large European Schizophrenia Cohort (EuroSC) comprising people with longstanding difficulties in contact with services, and in receipt of prescribed medication (Bebbington et al., Reference Bebbington, Angermeyer, Azorin, Brugha, Kilian and Johnson2005). Detailed symptom information was obtained at five time-points over a 2-year period. The current analysis applies DAGs to this longitudinal dataset in a novel application of this methodology.

We hypothesised that, in line with mechanisms found to precede acute psychotic disorders, affective disturbance in the form of anxiety and depression would exert an important and consistent effect on the persistence of delusions and hallucinations.

Methods

Participants

EuroSC was a naturalistic 2-year follow-up of a cohort of people aged 18–64, with an established diagnosis of schizophrenia. They were recruited from secondary psychiatric services in nine community mental health catchment areas, three in France, four in Germany and two in the UK. Ethical approval for the study was obtained locally in each country. Their illness was longstanding, averaging 16 years in duration at intake. Ethical approval for the study was obtained locally in each jurisdiction. Eligible patients were aged 18–64 years at the time of enrolment, had a diagnosis of schizophrenia according to DSM-IV criteria, and had given signed informed consent. Only those remaining in the established programmes of psychiatric care were followed up, and treatment was centred on prescribed medication. The diagnosis of schizophrenia and the assessment of positive and negative symptoms were based on systematic semi-structured interviews. Recently hospitalised subjects were excluded, and the sample was consequently not acutely unwell. However, participants remained significantly symptomatic. The survey provides detailed information on clinical symptoms at five time-points. In total, 1208 people with schizophrenia entered the study, 288 in France, 302 in the UK and 618 in Germany.

However, the primary objective of EuroSC was to identify and describe the types of treatment and patterns of care seen in people with schizophrenia, and to relate these to clinical outcomes.

Measures

An extensive battery of instruments was used to collect information during face-to-face interviews. In the UK and Germany, SCAN (Schedules for Clinical Assessment in Neuropsychiatry-version 1.0; WHO, 1992) was used to evaluate the 4-week period before interview and the most significant period of earlier psychopathology. Its component algorithm then allowed the establishment of research diagnoses of schizophrenia. In the French centres, schizophrenia was identified using the Structured Clinical Interview for DSM-IV (SCID; Spitzer, Williams, Gibbon, and First, Reference Spitzer, Williams, Gibbon and First1992).

For the purpose of the current analyses, we selected items from the interviewer-administered Positive and Negative Syndrome Scale (PANSS) (Norman, Malla, Cortese, & Diaz, Reference Norman, Malla, Cortese and Diaz1996): P1, delusions; P3, hallucinations; P4, excitement; P5, grandiosity; P6, suspiciousness; P7, hostility; G2, anxiety; G11, poor attention and G14, poor impulse control. These were rated in relation to the preceding 72 h, and our analysis was based on items scored as being clearly present and of moderate severity (a rating of 4 or above). We also sourced items from the Calgary Depression Scale for Schizophrenia (CDSS) (Addington, Addington, & Schissel, Reference Addington, Addington and Schissel1990; Addington, Addington, Maticka-Tyndale, & Joyce, Reference Addington, Addington, Maticka-Tyndale and Joyce1992): item 1, depressed mood; item 2, hopelessness and item 3, self-depreciation. This choice of items was derived from our initial hypothesis that affective symptoms have a significant role in the emergence and maintenance of psychotic symptoms.

Procedures

Trained research assistants obtained informed consent from participants, while reassuring them about privacy protection. The current analysis uses data from all five time-points. We term the interval between adjacent time-points a time-slice.

Inevitably some participants were lost to follow-up, progressively so during the course of the study, such that only 810 were available for assessment by the fifth time-point. However, attrition analyses showed no significant differences in demographic factors or in any of the study variables between participants with data missing at any follow-up and the participants who remained in the study throughout (Carrà et al., Reference Carrà, Crocamo, Angermeyer, Brugha, Toumi and Bebbington2020). We were able to include in the analysis all participants with complete data across any two adjacent time-points. The data for the prior time-point then contribute to learning the internal structure of the time-slice. All participants with observations at two adjacent time-points contributed to learning the connections between time-slices, whereas those with available data for multiple pairs of adjacent time-points contributed to the analysis once for each pair.

For the sociodemographic characteristics of the sample, see Table 1. The scores on PANSS and the CDSS over the five datapoints are shown in Table 2, together with the exact number of participants included each time. Overall, the mean scores for the psychotic symptoms and depression measures showed statistically non-significant reductions with time (PANSS positive 12.4 to 11.3; PANSS negative 15.8 to 15.1; PANSS General 29.3 to 26.6; CDSS 2.9 to 2.1).

Table 1. Sociodemographic and clinical characteristics at baseline

s.d., standard deviation; FGA, first-generation antipsychotics; SGA: second-generation antipsychotics; PANSS, Positive And Negative Syndrome Scale; CDSS, Calgary Depression Scale for Schizophrenia.

Values in parentheses are percentages except as otherwise indicated.

Table 2. Mean (s.d.) for PANSS and CDSS total scores and participant numbers at all five time-points

PANSS, Positive And Negative Syndrome Scale; CDSS, Calgary Depression Scale for Schizophrenia.

Statistical methods

As the data provide repeated measurements at five successive time-points, we had access to the relationship between an antecedent data-point and its immediate successor 6 months later in relation to four separate intervals. Combining the measurements of all symptoms in one model allows us to capture the mechanisms linking the variables to each other in a single representation, and informs us about potential causal effects between a range of symptoms on a 6 month scale.

We accordingly used dynamic Bayesian networks to capture the time component and to describe the evolution of the variables (Friedman, Murphy, & Russell, Reference Friedman, Murphy and Russell1998). Such networks model the overall dependence structure of multiple variables, measured over different time-points, and can be visualised in DAGs (see online Supplementary Fig. S1 for a toy example). In their simplest form, DBNs assume that the connections within a time-slice are stationary over time and identical across time-slices, implying that the way symptoms relate to each other does not change. Essentially, the requirement for stationarity over time and identical connections across time-slices translate into the assumption that the way in which symptoms interact, with each other and over time, does not change substantially with disease progression (these assumptions receive tangential support from the previous attrition analysis: Carrà et al., Reference Carrà, Crocamo, Angermeyer, Brugha, Toumi and Bebbington2020).

The problem of inferring the network that describes the observed data is essentially reduced to learning a network of size equivalent to the number of variables considered at each time-point. For each variable in the follow-up time-slice, we also need to learn the set of parents from the antecedent time-slice [i.e. the variables from the previous time-point with a direct link (edge) to the variable under consideration, for which the data rejected all possible conditional independence relationships]. The computational cost of identifying the models characterising the data well does not increase significantly. Using DBNs allows us to characterise the networks describing the relationships within each time-slice, as well as those between time-slices. Analysing the connections between variables at one time-point, while assuming that relationships remain the same over time, allows us to characterise the networks within a time-slice more confidently, by pooling together data from different time-points. On the contrary, analysing the links between variables at adjacent time-points also enables us to discover the relationships between slices. Situations where we have a variable A at time T affecting a variable B at time T + 1, and B at time T affecting A at time T + 1 are suggestive of a feedback loop.

From a conceptual perspective, we learn a model by jumping through different networks and evaluating a score indicating how well each structure explains the data. However, rather than choosing a single model we follow a Bayesian approach where we account for all models in a measure that is proportional to their score. From each model, we then derive putative causal effects resulting in a distribution which accounts for the uncertainty in the network structure (an approach we first developed in Moffa et al., Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017). Finally, we also take into account the uncertainty in the parameter estimation of each DAG to derive the full posterior distribution of intervention effects (akin to Kuipers et al., Reference Kuipers, Moffa, Kuipers, Freeman and Bebbington2019).

To explore the space of DAG structures of DBNs in a Bayesian fashion in the current analysis, we relied on our partition Markov chain Monte Carlo method (Kuipers & Moffa, Reference Kuipers and Moffa2017) for sampling DAG structures, and used the implementation for DBNs in the R package BiDAG (https://CRAN.R-project.org/package=BiDAG). To predict the range of causal effects we also followed the Bayesian strategy implemented by Moffa et al. (Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017) and Kuipers et al. (Reference Kuipers, Moffa, Kuipers, Freeman and Bebbington2019) for binary variables, using the R package Bestie (https://CRAN.R-project.org/package=Bestie). The main underlying assumptions are that a causal link exists between the variables, that a causal DAG can describe their causal relationship and that all common causes are represented on the DAG (i.e. there are no unmeasured confounders). The method features two special characteristics. First, it informs both the strength and the direction of causal effects. Second, it captures the uncertainty in the inference that derives from the fact that several different graphical arrangements of variables may each be capable of explaining the data reasonably well, by sampling from the range of possible DAGs in proportion to their posterior distribution. In total, we drew a sample of 10 000 DAGs, in such a way that better fitting graphs appear more often.

Each of the DAGs sampled from the posterior distribution represents a model of the data. The individual DAGs allow us to calculate the effect of one variable on another. Specifically, we can derive an estimate of the probability of each variable taking the value 0 or 1 (depending on the state of its parents in the network), along with the posterior distribution of that probability. The value of one variable is set to 0, and the probability that each remaining variable is either 0 or 1 can then be calculated. The selected variable is then set to 1 and the process is repeated. In Pearl's (Reference Pearl2009) terms, the difference between the two sets of values constitutes the causal effect of the selected variable on the others. This procedure is followed in turn for each variable in the DAG. By collating the estimated effects over all the DAGs in the sample we can then obtain the overall effect of changing each variable on all the others. This enables us to identify the most plausible causal mechanisms.

Results

Figure 1 shows the distribution of causal effects of row labels on column labels: variables from the prior time-slice occupy the rows, whereas those from the subsequent time-slice occupy the columns. Figure 2 shows the DAG corresponding to these findings. The DAG analysis clearly demonstrates that the structure of relationships between symptoms is complex beyond the scope of multivariable logistic regression: the DAGs enable the direction of several effects to be displayed simultaneously.

Fig. 1. Distributions of causal effects between variables in adjacent time-slices. Each cell displays the distribution of causal effects of row labels (variables from the prior time point) on column labels (variables from the subsequent time point). For readability we truncated the distribution to the range −0.1 to +0.5. The red line in each box represents zero causal and the box is coloured if the 95% credible interval does not straddle the zero causal effect line, to highlight a statistically significant effect. The numerical value in each box indicates the posterior mean of the effect distribution, which gives an estimate for average causal effect (zero in the case of no effect). For example enforcing persecutory ideas would seem to increase the probability of delusion at the following time point by 0.32 on average, whereas hallucinations do not show consistent evidence of downstream effects on other variables at the next time point.

Fig. 2. DAG of relationships between variables at initial and subsequent time-slices. Dashed arrows represent the direct relationships between variables from one time-slice to the next. Continuous arrows indicate putative causal links within a time-slice. The density of the arrows represents the strength of the connections. Single-headed arrows correspond to causal links whose direction could be identified under the no unmeasured confounders assumption. Double-headed arrows on the other hand imply that the data are not sufficient to identify a causal direction. The presence of doubly directed links leads to the presence of bimodal peaks in the distribution of causal effect, as we can see in Fig. 1, for example for the effect of depressed mood at time T on hopelessness at time T + 1, where we can easily recognise two peaks. The larger peak corresponds to structures including the causal path going through the edge from depressed mood to self-depreciation, increasing the total effect. The smaller peak corresponds to the relatively fewer structures where we only have the reversed edge between self-depreciation and depressed mood, leading to a single causal path (depression at time T → depression at time T + 1 → hopelessness at time T + 1) and a smaller total effect.

It is apparent from these figures that most symptoms have momentum, that is, they self-predict over time. Indeed, this was the most consistent relationship, and the only symptoms not following this pattern were hopelessness and hostility, implying that these phenomena are short-lived. However, the most striking finding in Fig. 2 is the demonstration of the strong central role of delusions and persecutory ideas, whether by direct or indirect routes. Their persistence in subsequent time-slices enables them to influence the presence of hallucinations at follow-ups. Initial hopelessness influences the presence of later hopelessness, but does so indirectly by increasing anxiety, which in turn has depressive consequences. These then go on to elicit a sense of hopelessness.

Only delusions and hopelessness directly influence symptoms other than themselves over these 6-month intervals: delusions operating on persecutory ideas, and hopelessness on anxiety. Grandiosity is also of interest, as it is the end point of several causal chains: leading from delusions via poor attention and excitement, or from persecutory ideas, either directly via hostility and excitement, or via poor impulse control, hostility and excitement. In this cohort, hallucinations are either driven by previous hallucinations, or by delusions. There is no evidence that either of these classically psychotic symptoms is driven by affect in this cohort of people with longstanding schizophrenia.

Discussion

Participants in EuroSC were recruited from patients engaged in established programmes of treatment for schizophrenia, involving monitoring, support and the prescription of medication. They were only followed up in the EuroSC programme if they remained in treatment. Recently hospitalised subjects were excluded, so the sample was not acutely unwell. However, their illness was longstanding, averaging 16 years in duration. The diagnosis of schizophrenia and the assessment of positive and negative symptoms were based on systematic semi-structured interviews. Although (as noted above) symptoms of all types declined in frequency over the study period, the decline was modest, consistent with the fact that participants were not chosen on the basis of a current acute episode. This large and representative cohort, assessed over a period of 2 years at 6-month intervals, provides our study with particular strengths. Retention of participants was good, and distortion of the sample from attrition relatively limited.

Because study participants had generally been treated consistently with antipsychotic medication (Table 1) it is very difficult to discern the effects of medication on their symptoms, in particular because the positivity condition in causal inference requires that people not receiving treatment are included. This was confirmed by the fact that when we included drugs in the structure learning procedure, we found them disconnected from the network of symptoms. This implies further that decisions about treatment tend to be driven by factors unrelated to symptoms, such as side effects. For this reason, we did not pursue a version of DAG incorporating medications

Our most consistent results relate to the connection of symptomatic measures with themselves over a considerable period. It appears that, once formed, symptoms seem to have a degree of momentum that is appreciably independent of the persistence of other symptoms. This finding has not been reported before, and appears to be a significant mechanism underpinning the overall stability of the syndrome of schizophrenia in this sample. It should be emphasised that this occurred in the context of support from community mental health teams and the consistent provision of antipsychotic medication; however targeted psychological treatments were not generally available at the time of the survey.

The longitudinal design provides the basis for our DAG analysis, a novel application in this context. The concept of cause demands a temporal element, in that causes must be antecedent to effects. However, the appropriate period of effect is not intrinsically identifiable. The capacity to demonstrate cause in a longitudinal convenience sample is hostage to the assessment interval: if this is too long, effects may be masked, if too short, there may not have been time for the effect to occur. Experience sampling studies indicate that it is highly probable that causal interactions between psychological processes occur over periods of hours or days (Delespaul, deVries, & van Os, Reference Delespaul, deVries and van Os2002; Myin-Germeys, Delespaul, & van Os, Reference Myin-Germeys, Delespaul and van Os2005). Such short-term effects may also be cumulative, and remain apparent over longer periods (Fowler et al., Reference Fowler, Hodgekins, Garety, Freeman, Kuipers, Dunn and Bebbington2012), although they may be disguised in a combination of autocorrelation and cross-correlation.

Our five-point analysis of psychotic symptoms is based on the analysis of repeated presentations of the PANSS. This instrument does not provide a specific category of persecutory delusions: delusions are identified phenomenologically, whereas persecutory themes are assessed in a separate item. Some delusions may therefore have non-persecutory content. Equally, persecutory ideas are not necessarily delusional. Nevertheless, it is apparent from Fig. 2 that there is indeed a reassuringly close overall relationship between delusional processing and persecutory ideation. Time 1 delusions predict delusions and persecutory ideation at time 2, time 1 persecutory thinking predicts itself at time 2, and its persistence in turn predicts delusions at time 2. This relationship turns out to be important for the maintenance of two distinct groups of symptoms over the 6-month gap between time-slices. Thus, directly and indirectly, delusions and persecutory ideation result in anxiety, depressed mood, self-depreciation and hopelessness, while a second, derivative, group comprises poor attention, hostility, excitement and poor impulse control, through which grandiosity may be a further consequence. In these data hallucinations appear to be unrelated to other symptoms apart from delusions, and they seemed to be the consequence of delusions rather than the reverse.

Our initial hypothesis was based on the idea that affective disturbance would be causally related to psychotic symptoms, corresponding to their apparent role in acute psychotic episodes and in the development of psychosis in relation to early trauma (Gracie et al., Reference Gracie, Freeman, Green, Garety, Kuipers, Hardy and Fowler2007; Hardy et al., Reference Hardy, Emsley, Freeman, Bebbington, Garety, Kuipers and Fowler2016; Kelleher et al., Reference Kelleher, Keeley, Corcoran, Ramsay, Wasserman, Carli and Cannon2013; Marwaha & Bebbington, Reference Marwaha and Bebbington2015; Moffa et al., Reference Moffa, Catone, Kuipers, Kuipers, Freeman, Marwaha and Bebbington2017). However, counter to our predictions, we did not find this in the current sample. Psychotic symptoms predicted each other, whereas affect appeared to be downstream, in line with traditional views of the relationship between psychotic and affective symptoms. This suggests that, in this stable sample with longstanding disorder, the relationships between symptoms change, such that psychotic symptoms drive affective consequences, particularly depression and anxiety. We did not have in our sample people who were at the onset of their disorder, when affect may be a much more crucial driver of psychosis symptoms (Bebbington et al., Reference Bebbington, McBride, Steel, Kuipers, Radovanovič, Brugha and Freeman2013; Bird et al., Reference Bird, Evans, Waite, Loe and Freeman2019; Freeman, Reference Freeman2016; Freeman et al., Reference Freeman, Taylor, Molodynski and Waite2019; Hardy et al., Reference Hardy, Emsley, Freeman, Bebbington, Garety, Kuipers and Fowler2016; Marwaha et al., Reference Marwaha, Broome, Bebbington, Kuipers and Freeman2014). Although in the early phases of disorder there may be a precursor affective cascade of anxiety and depression, our results indicate that this is not so in the case of longstanding disorders.

These results suggest that delusional processes are key to the maintenance of the range of psychotic disturbance seen in this cohort of people with longstanding schizophrenia. It is consistent with the specific targeting of delusions in treatment. There is now a growing body of evidence that positive symptoms unresponsive to vigorous pharmacological treatment may be ameliorated by psychological treatments aimed at disrupting the processes involved in the maintenance of delusions (Garety et al., Reference Garety, Fowler, Freeman, Bebbington, Dunn and Kuipers2008; Kuipers, Yesufu-Udechuku, Taylor, & Kendall, Reference Kuipers, Yesufu-Udechuku, Taylor and Kendall2014). A recent study has shown that a blended (i.e. with combined therapist and digital input), targeted therapy for paranoid ideas was able to improve established delusions (Garety et al., Reference Garety, Ward, Emsley, Greenwood, Freeman, Fowler and Hardy2021).

Limitations

As noted, diagnosis was based on information obtained from instruments that differed marginally in their approach to the evaluation of symptoms: SCAN in the UK and Germany and SCID in France. The measures of affect, using the CDSS, were not elaborate. The fact that this was a long-term sample, recruited at a time of stability, may have minimised interactions between symptoms; the assumption of stationarity over time may also have constrained our interpretation of the results. Although the type of medication was identified in the survey, we subsequently excluded it from the analyses presented here because it did not interact with any of the symptoms, either affective or psychotic; this might suggest that by this stage medication effects on mental state had plateaued. Finally our findings inevitably remain hostage to unmeasured confounders.

Conclusion

In this longitudinal investigation of people with persistent psychotic and affective symptoms over a 2-year period, delusions seemed to be the main drivers of affect (depression, anxiety and grandiosity). Hallucinations appeared to be the consequence of delusions, with no separate affective drivers. Medication did not appear to modify symptom patterns. These results suggest, as has been argued elsewhere, that in longstanding psychosis, further improvement might perhaps be achieved through psychological interventions, such as those recommended in the UK by NICE (2014).

Supplementary material

The supplementary material for this article can be found at https://doi.org/10.1017/S0033291721002920

Footnotes

*

Joint first authors.

References

Addington, D., Addington, J., Maticka-Tyndale, E., & Joyce, J. (1992). Reliability and validity of a depression rating scale for schizophrenics. Schizophrenia Research, 6, 201208.CrossRefGoogle ScholarPubMed
Addington, D., Addington, J., & Schissel, B. (1990). A depression rating scale for schizophrenics. Schizophrenia Research, 3, 247251.CrossRefGoogle ScholarPubMed
Bebbington, P. E., Angermeyer, M., Azorin, J. M., Brugha, T., Kilian, R., Johnson, S., …, EuroSC Research Group (2005) The European Schizophrenia Cohort (EuroSC): A naturalistic prognostic and economic study. Social Psychiatry and Psychiatric Epidemiology, 40, 707717.CrossRefGoogle ScholarPubMed
Bebbington, P., Jonas, S., Kuipers, E., King, M., Cooper, C., Brugha, T., … Jenkins, R. (2011). Childhood sexual abuse and psychosis: Data from a cross-sectional national psychiatric survey in England. British Journal of Psychiatry, 199, 2937.CrossRefGoogle ScholarPubMed
Bebbington, P., McBride, O., Steel, C., Kuipers, E., Radovanovič, M., Brugha, T., … Freeman, D. (2013). The structure of paranoia in the general population. British Journal of Psychiatry, 202, 419427.CrossRefGoogle ScholarPubMed
Ben-Zeev, D., Ellington, K., Swendsen, J., & Granholm, E. (2011). Examining a cognitive model of persecutory ideation in the daily life of people with schizophrenia: A computerized experience sampling study. Schizophrenia Bulletin, 37, 12481256.CrossRefGoogle ScholarPubMed
Birchwood, M., Iqbal, Z., & Upthegrove, R. (2005). Psychological pathways to depression in schizophrenia: Studies in acute psychosis, post psychotic depression and auditory hallucinations. European Archives of Psychiatry and Clinical Neuroscience, 255, 202212.CrossRefGoogle ScholarPubMed
Bird, J. C., Evans, R., Waite, F., Loe, B. S., & Freeman, D. (2019). Adolescent paranoia: Prevalence, structure, and causal mechanisms. Schizophrenia Bulletin, 45(5), 11341142.CrossRefGoogle ScholarPubMed
Borsboom, D., & Cramer, A. O. (2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91121.CrossRefGoogle Scholar
Carrà, G., Crocamo, C., Angermeyer, M., Brugha, T., Toumi, M., & Bebbington, P. (2020). The mediating role of depression in pathways linking positive and negative symptoms in schizophrenia. A longitudinal analysis using latent variable structural equation modelling. Psychological Medicine, 50, 566574.CrossRefGoogle ScholarPubMed
Delespaul, P., deVries, M., & van Os, J. (2002). Determinants of occurrence and recovery from hallucinations in daily life. Social Psychiatry and Psychiatric Epidemiology, 37, 97104.CrossRefGoogle ScholarPubMed
Fowler, D., Hodgekins, J., Garety, P., Freeman, D., Kuipers, E., Dunn, G., … Bebbington, P. (2012). Negative cognition, depressed mood and paranoia: A longitudinal pathway analysis using structural equation modelling. Schizophrenia Bulletin, 38, 10631073.CrossRefGoogle Scholar
Freeman, D. (2016). Persecutory delusions: A cognitive perspective on understanding and treatment. The Lancet Psychiatry, 3, 685692.CrossRefGoogle ScholarPubMed
Freeman, D., Dunn, G., Murray, R. M., Evans, N., Lister, R., Antley, A., … Morrison, P. D. (2015). How cannabis causes paranoia: Using the intravenous administration of 9-tetrahydrocannabinol (THC) to identify key cognitive mechanisms leading to paranoia. Schizophrenia Bulletin, 41, 391399.CrossRefGoogle ScholarPubMed
Freeman, D., Stahl, D., McManus, S., Meltzer, H., Brugha, T., Wiles, N., & Bebbington, P. (2012). Insomnia, worry, anxiety and depression as predictors of the occurrence and persistence of paranoid thinking. Social Psychiatry and Psychiatric Epidemiology, 47, 11951203.CrossRefGoogle ScholarPubMed
Freeman, D., Taylor, K. M., Molodynski, A., & Waite, F. (2019). Treatable clinical intervention targets for patients with schizophrenia. Schizophrenia Research, 211, 4450.CrossRefGoogle ScholarPubMed
Fried, E. I., van Borkulo, C. D., Cramer, A. O., Boschloo, L., Schoevers, R. A., & Borsboom, D. (2017). Mental disorders as networks of problems: A review of recent insights. Social Psychiatry and Psychiatric Epidemiology, 52, 110.CrossRefGoogle ScholarPubMed
Friedman, N., Murphy, K., & Russell, S. (1998) Learning the structure of dynamic probabilistic networks. In Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence. San Francisco, CA: Morgan Kaufmann Publishers Inc., 139147.Google Scholar
Fusar-Poli, P., Nelson, B., Valmaggia, L., Yung, A. R., & McGuire, P. K. (2014). Comorbid depressive and anxiety disorders in 509 individuals with an at-risk mental state: Impact on psychopathology and transition to psychosis. Schizophrenia Bulletin, 40, 120131.CrossRefGoogle ScholarPubMed
Garety, P. A., Fowler, D. G., Freeman, D., Bebbington, P., Dunn, G., & Kuipers, E. (2008). Cognitive–behavioural therapy and family intervention for relapse prevention and symptom reduction in psychosis: Randomised controlled trial. British Journal of Psychiatry, 192(6), 412423.CrossRefGoogle ScholarPubMed
Garety, P., Ward, T., Emsley, R., Greenwood, K., Freeman, D., Fowler, D., Kuipers, E., Bebbington, P., Rus-Calafell, M., McGourty, A., Sacadura, C., Collett, N., James, K., Hardy, A. (2021). The effects on paranoia of a blended digital therapy, SlowMo, which targets reasoning: A randomised controlled trial. JAMA Psychiatry, 78, 714725.CrossRefGoogle Scholar
Gibson, L. E., Alloy, L. B., & Ellman, L. M. (2016). Trauma and the psychosis spectrum: A review of symptom specificity and explanatory mechanisms. Clinical Psychology Review, 49, 92105.CrossRefGoogle ScholarPubMed
Gracie, A., Freeman, D., Green, S., Garety, P. A., Kuipers, E., Hardy, A., … Fowler, D. (2007). The association between traumatic experience, paranoia and hallucinations: A test of psychological models. Acta Psychiatrica Scandinavica, 116, 280289.CrossRefGoogle ScholarPubMed
Hailes, H. P., Ronggin, Y., Danese, A., & Fazel, S. (2019). Long-term outcomes of childhood sexual abuse: An umbrella review. The Lancet. Psychiatry, 6, 830839.CrossRefGoogle ScholarPubMed
Hardy, A., Emsley, R., Freeman, D., Bebbington, P., Garety, P. A., Kuipers, E., … Fowler, D. (2016). Psychological mechanisms mediating effects between trauma and psychotic symptoms: The role of affect regulation, intrusive trauma memory, beliefs and depression. Schizophrenia Bulletin, 42(Suppl 1), S34S43.CrossRefGoogle ScholarPubMed
Hartley, S., Haddock, G., Vasconcelos E Sa, D., Emsley, R., & Barrowclough, C. (2014). An experience sampling study of worry and rumination in psychosis. Psychological Medicine, 44, 16051614.CrossRefGoogle ScholarPubMed
Isvoranu, A. M., Borsboom, D., van Os, J., & Guloksuz, S. (2016). A network approach to environmental impact in psychotic disorder: Brief theoretical framework. Schizophrenia Bulletin, 42, 870873.CrossRefGoogle ScholarPubMed
Isvoranu, A. M., van Borkulo, C. D., Boyette, L. L., Wigman, J. T., Vinkers, C. H., & Borsboom, D. (2017). Group investigators. A network approach to psychosis: Pathways between childhood trauma and psychotic symptoms. Schizophrenia Bulletin, 43, 187196.CrossRefGoogle Scholar
Kelleher, I., Keeley, H., Corcoran, P., Ramsay, H., Wasserman, C., Carli, V., … Cannon, M. (2013). Childhood trauma and psychosis in a prospective cohort study: Cause, effect, and directionality. American Journal of Psychiatry, 170, 734741.CrossRefGoogle Scholar
Klippel, A., Myin-Germeys, I., Chavez-Baldini, U., Preacher, K. J., Kempton, M., Valmaggia, L., … Reininghaus, U. (2017). Modeling the interplay between psychological processes and adverse, stressful contexts and experiences in pathways to psychosis: An experience sampling study. Schizophrenia Bulletin, 43, 302315.CrossRefGoogle ScholarPubMed
Krabbendam, L., Myin-Germeys, I., De Graaf, R., Vollebergh, W., Nolen, W. A., Iedema, J., & Van Os, J. (2004). Dimensions of depression, mania and psychosis in the general population. Psychological Medicine, 34, 11771186.CrossRefGoogle ScholarPubMed
Kuipers, E., Garety, P., Fowler, D., Freeman, D., Dunn, G., & Bebbington, P. (2006). Cognitive, emotional and social processes in psychosis: Refining cognitive behavioural therapy for persistent positive symptoms. Schizophrenia Bulletin, 32, s24s31.CrossRefGoogle ScholarPubMed
Kuipers, J., & Moffa, G. (2017). Partition MCMC for inference on acyclic digraphs. Journal of the American Statistical Association, 112, 282299.CrossRefGoogle Scholar
Kuipers, J., Moffa, G., Kuipers, E., Freeman, D., & Bebbington, P. (2019). Links between psychotic and neurotic symptoms in the general population: An analysis of longitudinal British national survey data using directed acyclic graphs. Psychological Medicine, 49, 388395.CrossRefGoogle ScholarPubMed
Kuipers, E., Yesufu-Udechuku, A., Taylor, C., & Kendall, T. (2014). Management of psychosis and schizophrenia in adults: Summary of updated NICE guidance. British Medical Journal, 348, g1173. doi: 10.1136/bmj.g1173CrossRefGoogle ScholarPubMed
Marwaha, S., & Bebbington, P. (2015). Mood as a mediator of the link between child sexual abuse and psychosis. Social Psychiatry and Psychiatric Epidemiology, 50, 661663.CrossRefGoogle ScholarPubMed
Marwaha, S., Broome, M. R., Bebbington, P. E., Kuipers, E., & Freeman, D. (2014). Mood instability and psychosis: Analyses of British national survey data. Schizophrenia Bulletin, 40, 269277.CrossRefGoogle ScholarPubMed
McGrath, J. J., Saha, S., Lim, C. C. W., Aguilar-Gaxiola, S., Alonso, J., Andrade, L. H., … WHO World Mental Health Survey Collaborators. (2017). Trauma and psychotic experiences: Transnational data from the world mental health survey. British Journal of Psychiatry, 211, 373380.CrossRefGoogle ScholarPubMed
McNally, R. J. (2016). Can network analysis transform psychopathology? Behaviour Research and Therapy, 86, 95104.CrossRefGoogle ScholarPubMed
McNally, R. J., Robinaugh, D. J., Wu, G. W. Y., Wang, L., Deserno, M., & Borsboom, D. (2015). Mental disorders as causal systems: A network approach to posttraumatic stress disorder. Clinical Psychological Science, 3, 836e849.CrossRefGoogle Scholar
Moffa, G., Catone, G., Kuipers, J., Kuipers, E., Freeman, D., Marwaha, S., … Bebbington, P. (2017). Using directed acyclic graphs in epidemiological research in psychosis: An analysis of the role of bullying in psychosis. Schizophrenia Bulletin, 43, 12731279.CrossRefGoogle ScholarPubMed
Murphy, E. K., Tully, S., Pyle, M., Gumley, A. I., Kingdon, D., Schwannauer, M., … Morrison, A. P. (2017). The beliefs about paranoia scale: Confirmatory factor analysis and tests of a metacognitive model of paranoia in a clinical sample. Psychiatry Research, 248, 8794.CrossRefGoogle Scholar
Myin-Germeys, I., Delespaul, P., & van Os, J. (2005). Behavioural sensitization to daily life stress in psychosis. Psychological Medicine, 35, 733741.CrossRefGoogle ScholarPubMed
Norman, R. M. G., Malla, A. K., Cortese, L., & Diaz, F. (1996). A study of the interrelationship between and comparative interrater reliability of the SAPS, SANS and PANSS. Schizophrenia Research, 19(1), 7385.CrossRefGoogle ScholarPubMed
Pearl, J. (2009). Causality: Models, reasoning, and inference (2nd ed.). Cambridge, UK: Cambridge University Press.CrossRefGoogle Scholar
Quattrone, D., Di Forti, M., Gayer-Anderson, C., Ferraro, L., Jongsma, H. E., Tripoli, G., … Reininghaus, U. (2019). Transdiagnostic dimensions of psychopathology at first episode psychosis: Findings from the multinational EU-GEI study. Psychological Medicine, 49, 13781391.CrossRefGoogle Scholar
Sax, K. W., Strakowski, S. M., Keck, P. E. Jr., Upadhyaya, V. H., West, S. A., & McElroy, S. L. (1996). Relationships among negative, positive, and depressive symptoms in schizophrenia and psychotic depression. British Journal of Psychiatry, 168, 6871.CrossRefGoogle ScholarPubMed
Scott, J., Martin, G., Bor, W., Sawyer, M., Clark, J., & McGrath, J. (2009). The prevalence and correlates of hallucinations in Australian adolescents: Results from a national survey. Schizophrenia Research, 107, 179185.CrossRefGoogle ScholarPubMed
Sheaves, B., Bebbington, P. E., Goodwin, G. M., Harrison, P. J., Espie, C. A., Foster, R. G., & Freeman, D. (2016). Insomnia and hallucinations in the general population: Finding from the 2000 and 2007 British psychiatric morbidity surveys. Psychiatry Research, 241, 141146.CrossRefGoogle ScholarPubMed
Smith, B., Fowler, D. G., Freeman, D., Bebbington, P., Bashforth, H., Garety, P., … Kuipers, E. (2006). Emotion and psychosis: Links between depression, self-esteem, negative schematic beliefs and delusions and hallucinations. Schizophrenia Research, 86, 181188.CrossRefGoogle ScholarPubMed
Spitzer, R. L., Williams, J. B., Gibbon, M., & First, M. B. (1992). The structured clinical interview for DSM-III-R (SCID). I: History, rationale, and description. Archives of General Psychiatry, 49(8), 624629.CrossRefGoogle ScholarPubMed
Thewissen, V., Bentall, R. P., Oorschot, M., Campo J, A., van Lierop, T., van Os, J., & Myin-Germeys, I. (2011). Emotions, self-esteem, and paranoid episodes: An experience sampling study. British Journal of Clinical Psychology, 50(2), 178–95.CrossRefGoogle ScholarPubMed
Upthegrove, R., Marwaha, S., & Birchwood, M. (2017). Depression and schizophrenia: Cause, consequence, or trans-diagnostic issue? Schizophrenia Bulletin, 43, 240244.Google ScholarPubMed
WHO-World Health Organization. (1992). SCAN: Schedule for clinical assessment in neuropsychiatry. Geneva: WHO Press.Google Scholar
Wigman, J. T., de Vos, S., Wichers, M., van Os, J., & Bartels-Velthuis, A. A. (2017). A transdiagnostic network approach to psychosis. Schizophrenia Bulletin, 43, 122132.CrossRefGoogle ScholarPubMed
Yung, A. R., Buckby, J. A., Cosgrave, E. M., Killackey, E. J., Baker, K., Cotton, S. M., & McGorry, P. D. (2007). Association between psychotic experiences and depression in a clinical sample over 6 months. Schizophrenia Research, 91, 246253.CrossRefGoogle Scholar
Figure 0

Table 1. Sociodemographic and clinical characteristics at baseline

Figure 1

Table 2. Mean (s.d.) for PANSS and CDSS total scores and participant numbers at all five time-points

Figure 2

Fig. 1. Distributions of causal effects between variables in adjacent time-slices. Each cell displays the distribution of causal effects of row labels (variables from the prior time point) on column labels (variables from the subsequent time point). For readability we truncated the distribution to the range −0.1 to +0.5. The red line in each box represents zero causal and the box is coloured if the 95% credible interval does not straddle the zero causal effect line, to highlight a statistically significant effect. The numerical value in each box indicates the posterior mean of the effect distribution, which gives an estimate for average causal effect (zero in the case of no effect). For example enforcing persecutory ideas would seem to increase the probability of delusion at the following time point by 0.32 on average, whereas hallucinations do not show consistent evidence of downstream effects on other variables at the next time point.

Figure 3

Fig. 2. DAG of relationships between variables at initial and subsequent time-slices. Dashed arrows represent the direct relationships between variables from one time-slice to the next. Continuous arrows indicate putative causal links within a time-slice. The density of the arrows represents the strength of the connections. Single-headed arrows correspond to causal links whose direction could be identified under the no unmeasured confounders assumption. Double-headed arrows on the other hand imply that the data are not sufficient to identify a causal direction. The presence of doubly directed links leads to the presence of bimodal peaks in the distribution of causal effect, as we can see in Fig. 1, for example for the effect of depressed mood at time T on hopelessness at time T + 1, where we can easily recognise two peaks. The larger peak corresponds to structures including the causal path going through the edge from depressed mood to self-depreciation, increasing the total effect. The smaller peak corresponds to the relatively fewer structures where we only have the reversed edge between self-depreciation and depressed mood, leading to a single causal path (depression at time T → depression at time T + 1 → hopelessness at time T + 1) and a smaller total effect.

Supplementary material: PDF

Moffa et al. supplementary material

Moffa et al. supplementary material

Download Moffa et al. supplementary material(PDF)
PDF 300.7 KB