Positive and negative symptoms in schizophrenia: A longitudinal analysis using latent variable structural equation modelling

Abstract

Background

Recent network models of schizophrenia propose it is the consequence of mutual interaction between its symptoms. While cross-sectional associations between negative and positive symptoms are consistent with this idea, they may merely reflect their involvement in the diagnostic process. Longitudinal analyses however may allow the identification of possible causal relationships. The European Schizophrenia Cohort (EuroSC) provides data suitable for this purpose.

Methods

EuroSC includes 1208 patients randomly sampled from outpatient services in France, Germany and the UK. Initial measures were repeated after 12 and 24 months. Latent variable structural equation modelling was used to investigate the direction of effect between positive and negative symptoms assessed with the Positive and Negative Syndrome Scale, controlling for the effects of depressed mood and antipsychotic medication.

Results

The structural model provided acceptable overall fit [χ² (953) = 2444.32, P < 0.001; CFI = 0.909; RMSEA = 0.046 (90% CI: 0.043, 0.048); SRMR = 0.052]. Both positive and negative symptoms were persistent, and strongly auto-correlated. There were also persistent cross-sectional associations between positive and negative symptoms. While the path from latent positive to negative symptoms from T1 to T2 approached conventional levels of statistical significance (P = 0.051), that from T2 to T3 did not (P = 0.546). Pathways in the reverse direction were uniformly non-significant.

Conclusions

There was no evidence that negative symptoms predict later positive symptoms. The prediction of negative symptoms by positive symptoms was ambiguous. We discuss implications for conceptualization of schizophrenic processes.

Introduction

The origins and course of schizophrenia are acknowledged to be complex. For over a century it has been conceived and diagnosed on the basis of distinctively characteristic symptoms, in particular delusions and hallucinations. In addition, since Kraepelin and Quen, 1990 a conceptual and clinical contrast has been made between positive and negative symptoms. This distinction was strongly reasserted nearly 40 years ago (Strauss et al., 1974; Crow, 1980; Andreasen and Olsen, 1982), and negative symptoms have been included among the diagnostic criteria of schizophrenia since DSM-II (APA, 1968), remaining one of the five core dimensions in DSM-5 (APA, 2013).

The conceptual differentiation between the two symptom types has face validity, given that negative symptoms consistently load on at least two factors (diminished expression and anhedonia/asociality) that distinguish them from positive symptoms (Blanchard and Cohen, 2006). However, the relative diagnostic weight accorded to them has varied over time (Tandon et al., 2013), thereby affecting the propensity of the symptom types to be associated within the diagnostic category of schizophrenia (Galderisi et al., 2017). Notwithstanding, positive and negative symptoms are clearly associated, and the standard psychiatric view attributes this association to the expression of an underlying neuropathological diathesis (Walker, 1994; Seidman et al., 2016). In other words, negative symptoms may owe their origin to the same processes that produce positive symptoms (Pogue-Geile and Harrow, 1984). However, a diathesis account must deal with some inconvenient findings. First, schizophrenia characterized by salient negative symptoms (deficit schizophrenia) is more strongly correlated with neuro-anatomical abnormalities than non-deficit schizophrenia (Rowland et al., 2009). Secondly, both sorts of schizophrenia are thought to be associated with abnormalities of dopamine metabolism, but positive symptoms are more sensitive to medication targeting the dopamine system (Leucht et al., 2009, Leucht et al., 2017). Third is the occurrence of positive and negative symptoms both across a range of clinical conditions and below the threshold of a diagnosable disorder (van Os and Reininghaus, 2016; Waters and Fernyhough, 2017; Bebbington and Freeman, 2017; Strauss and Cohen, 2017). Fourthly, when assessed in the early phases of illness, positive and negative symptoms show more than chance association, being roughly equally predictive of each other (Dominguez et al., 2010). Finally, the putative shared disease process must be variably expressed, given that symptoms emerge de novo and then fluctuate (Pogue-Geile and Harrow, 1984). Although this variability might be the consequence of an inherent biological rhythmicity, it seems equally likely to be due to variations in the psychosocial and pharmacological environment (Buchanan, 2007). Taken together, these findings imply that the causes of the two symptom types may not map perfectly onto each other. There are, however, alternatives to the diathesis account. Several authors have recently advocated a network model of psychopathology, whereby symptoms co-vary as a result of causal interactions between them, with no requirement for an underlying disease process (Borsboom and Cramer, 2013; McNally, 2016; Moffa et al., 2017). Symptoms may cause each other through a number of possible mechanisms including cumulative processes and feedback loops. For example, particular symptoms of schizophrenia may affect the social environment in ways that facilitate the expression of other symptoms. Thus, negative symptoms may encourage social withdrawal so that hallucinations and paranoid ideation become more pronounced (Wickham et al., 2014). Likewise, positive symptoms, for example extreme paranoia, may induce asociality and other facets of negative symptoms (Messinger et al., 2011). In addition, depressed mood may have a direct effect on both positive and negative symptoms (Fowler et al., 2012; Millan et al., 2014).

However, the possibility that negative symptoms are at least sometimes secondary has been conceded, since a basic distinction has been made between primary and secondary negative symptoms: it is the former that are thought to be the direct result of a putative disease process (Carpenter et al., 1988). Hence, secondary negative symptoms may be the consequence of unrelieved positive symptoms, the adverse effects of antipsychotics, or the social distortions brought about by the experience of having schizophrenia (Kirkpatrick et al., 2006). Nonetheless, apart from the sensitivity of social impoverishment to environmental improvement (Brown and Wing, 1970), changes in the level of these putative causes of secondary negative symptoms have little apparent influence on their subsequent course (Savill et al., 2015).

The possibility of mutual interactions is more meaningfully supported by evidence showing that primary negative symptoms are more common in people who have experienced positive symptoms at some point, resulting from the cumulative effect of positive symptoms. In addition, the two symptom types are less likely to be correlated when assessed cross-sectionally (Pogue-Geile and Zubin, 1987), than longitudinally (Tandon et al., 2013).

Investigation of the covariation of symptoms may therefore expand our understanding of the aetiology and course of schizophrenia. Unfortunately, such longitudinal studies have been rare, despite the long interval since the first reports on this topic, and their results have been inconsistent. One small study found that negative and positive symptoms were correlated at both the acute and six-month remission phases (Addington and Addington, 1991). Another reported that avolition and anhedonia predicted hallucinations, delusions and bizarre behaviour at 30-month follow-up (Kulhara and Chandiramani, 1990). In one 3-year follow-up investigation positive and negative symptoms appeared to increase together over time (Dollfus and Petit, 1995). Some of the inconsistency may relate to variable length of follow up: a more extended (10-year) follow-up study suggested that positive and negative symptom clusters are longitudinally independent (Eaton et al., 1995). Interestingly, a longitudinal study on the emergence of psychotic symptom types in general populations has shown that negative symptoms increased only after the emergence of positive symptoms (Werbeloff et al., 2015).

In addition, heterogeneity may play a key-role. In one study, changes in negative and positive symptoms were in general positively related for most of the combined latent trajectory classes identified; however, in some participants (who lacked readily identifiable features) positive subscale scores became increasingly predominant over negative scores (Chen et al., 2013).

It is clear that larger longitudinal studies are needed, consistent with the NIMH-MATRICS Consensus Statement on Negative Symptoms recommendation of more research on patterns of persistence of negative symptoms (Kirkpatrick and Fischer, 2006). The European Schizophrenia Cohort (EuroSC) was specifically set up to compare the attributes and correlates of diagnosed schizophrenia in large and representative cohorts from three European countries, France, Germany and the UK (Bebbington et al., 2005). Data were collected over a period of three years. It therefore provided a valuable opportunity to study the interplay between symptoms over time by applying latent variable (cross-lagged) structural equation modelling to three waves of assessment.

In this paper, we contrast four theoretical possibilities that might account for relationships between positive and negative symptoms of psychosis. The first is that negative symptoms are the consequence of positive symptoms, the second that the relationship operates in the reverse direction. The third proposes that the relationship is continuing and reciprocal i.e., that feedback loops exist between positive and negative symptoms. The final possibility is that the symptoms are associated because they have a common cause, the obvious candidate being an underlying pathophysiological diathesis. Longitudinal data may be capable of distinguishing between these four accounts. Consider data on negative and positive symptoms at three data-points. On their own, cross-sectional relationships between the symptom-types at each data point are uninformative about causal direction. However, cross-lagged relationships across either interval may indicate reciprocal effects or unique causal directions between the variables. Finally, auto-correlational relationships within symptom types assessed at successive data-points is consistent with persisting shared-cause models.