Sudden transitions in symptom levels

In the past years, there has been renewed interest in the potential application of complex systems theory in the field of psychiatry (Heinzel et al., Reference Heinzel, Tominschek and Schiepek2014; Borsboom, Reference Borsboom2017; Haken and Tschacher, Reference Haken and Tschacher2017; Nelson et al., Reference Nelson, McGorry, Wichers, Wigman and Hartmann2017; Schiepek et al., Reference Schiepek, Viol, Aichhorn, Hutt, Sungler, Pincus and Scholler2017). In short, complex system theory entails that complex systems, ranging from ocean ecosystems to climate, financial markets or the evolutionary development of species, all have certain principles in common that predict their behaviour. These relate, for example, to the resilience of a system to remain in its present stable state. High resilience refers to a high level of stability of the system (deep basin of attraction) meaning that the system can easily face perturbations without being tipped out of its current equilibrium (Scheffer, Reference Scheffer2009) (see Fig. 1).

Fig. 1.

Two dimensional stability landscape. The ball represents the current state of a complex system and the line constitutes a surface showing the stability of the ball (or system state) in the current situation. In the left panel, the state of the system is in a deep basin of attraction, meaning that it will take a considerable perturbation of the system before psychopathology can develop. In the right panel, however, the situation is different. Here, only a small perturbation may already be enough for a transition towards a state of psychopathology.

In complex systems, this level of resilience may slowly diminish, even without noticeable signs. Once low, the system is highly instable and at this point even very minor contextual disturbances, also called perturbations, can push the system over a tipping point towards another basin of attraction (Scheffer, Reference Scheffer2009). This is why complex systems are characterized by sudden transitions, so-called phase transitions, that appear to emerge ‘out of the blue’. Similar transitions have been observed in psychiatry, as psychiatric symptoms sometimes (re)appear in a very abrupt way (Hayes et al., Reference Hayes, Laurenceau, Feldman, Strauss and Cardaciotto2007). Already at the end of the previous century the idea arose, that psychological phenomena might also behave according to the principles of complex systems (van der Maas and Molenaar, Reference van der Maas and Molenaar1992; Hayes and Strauss, Reference Hayes and Strauss1998; Beirle and Schiepek, Reference Beirle and Schiepek2002; Schiepek and Perlitz, Reference Schiepek, Perlitz and Meyers2009). If this is indeed the case, this is very relevant since the principles of complex systems teach us important things about the nature of psychopathology and how to understand and foresee sudden transitions in symptoms. One of the interesting consequences would be that the early identification of alterations in the level of instability of the system could reveal the proximity of the system's tipping point, or in other words, the likelihood that a sudden shift in symptoms occurs (Scheffer et al., Reference Scheffer, Bascompte, Brock, Brovkin, Carpenter, Dakos, Held, van Nes, Rietkerk and Sugihara2009, Reference Scheffer, Carpenter, Lenton, Bascompte, Brock, Dakos, van de Koppel, van de Leemput, Levin, van Nes, Pascual and Vandermeer2012). Some recently conducted studies were able to translate this idea into simulation studies and empirical designs that attempted to test this assumption in the field of psychiatry (Schiepek et al., Reference Schiepek, Tominschek, Karch, Lutz, Mulert, Meindl and Pogarell2009; van de Leemput et al., Reference van de Leemput, Wichers, Cramer, Borsboom, Tuerlinckx, Kuppens, van Nes, Viechtbauer, Giltay, Aggen, Derom, Jacobs, Kendler, van der Maas, Neale, Peeters, Thiery, Zachar and Scheffer2014; Cramer et al., Reference Cramer, van Borkulo, Giltay, van der Maas, Kendler, Scheffer and Borsboom2016; Wichers et al., Reference Wichers and Groot2016). Most of the focus in these studies has been on the possibility of foreseeing shifts in the levels of symptoms (increasing or decreasing levels). Yet, it would be important to not only foresee a shift in symptom level, but to also foresee the type of symptoms in which the transition occurs. For example, we want to foresee, if someone is approaching a symptom transition, whether that is a transition characterized by increasing manic, anxious or depressive symptoms. In this paper, we therefore want to explore whether we can extend the assumptions based on a complex systems perspective on psychopathology to foresee the type of these symptom shifts. Thus, rather than applying assumptions from the complex systems perspective on foreseeing level shifts in symptoms on a single dimension of psychopathology (in which the constituent symptoms can move from being absent to being present), it would be interesting to explore whether we could extend the impact of these assumptions to a multidimensional psychopathological space, in which we can foresee what types of symptoms are likely to develop.

The ability to detect the direction of transitions in psychopathology in high-risk individuals is highly relevant. During subclinical stages of psychopathology, people often experience a combination of symptoms that cross a wide range of psychopathological dimensions (Fusar-Poli et al., Reference Fusar-Poli, Yung, McGorry and van Os2014; McGorry and Nelson, Reference McGorry and Nelson2016). Therefore, it is often entirely unclear how and towards what type of symptoms psychopathology will develop in these high-risk individuals. This, however, is an urgent question as optimal clinical decision making in an early phase is important to reduce and prevent further development of psychopathology (McGorry et al., Reference McGorry, Hickie, Yung, Pantelis and Jackson2006; Cross et al., Reference Cross, Hermens, Scott, Ottavio, McGorry and Hickie2014). This urgency is expressed in the amount of resources that are invested in the development of adequate clinical staging and profiling techniques (Wigman et al., Reference Wigman, van Os, Thiery, Derom, Collip, Jacobs and Wichers2013; McGorry et al., Reference McGorry, Keshavan, Goldstone, Amminger, Allott, Berk, Lavoie, Pantelis, Yung, Wood and Hickie2014; Berk et al., Reference Berk, Post, Ratheesh, Gliddon, Singh, Vieta, Carvalho, Ashton, Berk, Cotton, McGorry, Fernandes, Yatham and Dodd2017). The complex systems perspective may contribute to these aims as it provides a complementary angle from which to understand the development of psychopathology and find solutions to improve personalized prediction.

This manuscript will provide an overview of ideas and hypotheses that follow from taking a complex systems perspective on psychopathology, focusing on foreseeing the type of symptoms that are most likely to show sudden transitions. First, we will describe in more detail what parts of complex systems theory, regarding early warning signals (EWS) and transitions, have already been related to the field of psychopathology. Second, we will explain what additional predictions may follow from this theory with respect to foreseeing and differentiating the type of sudden shifts in symptoms. Also, we will discuss what research designs would be needed to empirically test these predictions and the relevance of these ideas for clinical practice.

Support for a complex systems perspective on psychopathology

As mentioned above, there are reasons to assume that psychopathology behaves according to the principles of complex systems. First, sudden shifts in symptoms are observed in patients. Although psychopathology seems dimensional in nature in the sense that individuals can be anywhere on a continuum between having no symptoms and having severe symptoms, the road of symptom change within individuals can be much bumpier. Patients often report sudden relapses or sudden improvements in symptoms (Hayes et al., Reference Hayes, Laurenceau, Feldman, Strauss and Cardaciotto2007; Tang et al., Reference Tang, Derubeis, Hollon, Amsterdam and Shelton2007; Heinzel et al., Reference Heinzel, Tominschek and Schiepek2014). This has been confirmed by statistical analyses of symptom patterns over time in depressed patients, which revealed that most patients show a bimodal distribution in symptoms. In other words, they experience either low levels or high levels of symptoms (Hosenfeld et al., Reference Hosenfeld, Bos, Wardenaar, Conradi, van der Maas, Visser and de Jonge2015). This suggests that they experience sudden jumps in their symptom levels. Also, a recent (n = 1) double-blind time-series experiment (Wichers et al., Reference Wichers and Groot2016), in which levels of symptoms were weekly and prospectively monitored over 239 days, confirmed the presence of a sudden jump in depressive symptoms in this person. At this change point, the level of symptoms suddenly went up and seemed to stabilize afterwards at a higher point on the continuum of depression. Although not all symptom transitions may occur in an abrupt fashion, these observations at least suggest that abrupt symptom changes are quite common, which is in line with the expectations from complex system theory. Currently, extensive empirical research that has mapped symptom patterns in patients frequently and prospectively is lacking. More research is thus needed to confirm the assumption that symptom transitions often occur in an abrupt fashion.

Second, verbal descriptions of patients suggest that sudden and discontinuous changes in their symptom experience (Hayes et al., Reference Hayes, Laurenceau, Feldman, Strauss and Cardaciotto2007) may occur in the absence of an obvious, temporally proximal cause or reason. From a traditional approach, these unexpected symptom changes are difficult to explain as we know that external causes play an important role in symptom development. Logic dictates that changes in symptoms are directly preceded by changes in specific factors (such as in the social environment, stressful events or therapy). From a complex systems perspective, however, unexpected symptom change does make sense. This theory proposes that large shifts can also occur following minor, seemingly innocent stressors (Boeing, Reference Boeing2016). These shifts are most likely when a system's resilience to remain in its current basin of attraction is very low (Fig. 1), meaning that the system is in an unstable situation. From a complex systems perspective, such an unstable situation may result from the impact of a distal cause that happened some time ago, and that gradually led to the present loss of resilience. When resilience becomes very low, even minor seemingly unimportant disturbances can tip over the system to an alternative state (see Fig. 1). This can explain why patients may experience long intervals between potential environmental causes and the onset of their symptoms. Also, it can explain why sudden shifts in symptoms may occur in the absence of any obvious immediate trigger. Later we discuss how stability of the system can be empirically assessed.

Third, elements within complex systems are in a continuous and complex interplay with each other. In many complex systems, reinforcing feedbacks are present that, if strong enough, can push the system to another alternative state (Scheffer, Reference Scheffer2009). Such feedback loops are also likely to occur between mental states. Recent studies have confirmed the presence of feedback loops through network models, which showed that negative mental states, such as feeling down or irritated, are related to the occurrences of other negative mental states later in time. These effects may form vicious circles (Wichers, Reference Wichers2014). Findings from most studies supported the hypothesis that reinforcing feedback loops were more pronounced in people with either higher levels of psychopathology or at risk of psychopathology, compared with individuals in the general population (Pe et al., Reference Pe, Kircanski, Thompson, Bringmann, Tuerlinckx, Mestdagh, Mata, Jaeggi, Buschkuehl, Jonides, Kuppens and Gotlib2015; Wigman et al., Reference Wigman, van Os, Borsboom, Wardenaar, Epskamp, Klippel, Merge, Myin-Germeys and Wichers2015; Bringmann et al., Reference Bringmann, Pe, Vissers, Ceulemans, Borsboom, Vanpaemel, Tuerlinckx and Kuppens2016; Klippel et al., Reference Klippel, Viechtbauer, Reininghaus, Wigman, van Borkulo, Myin-Germeys and Wichers2017), although this was not confirmed by all studies (Eijlander et al., Reference Eijlander, Van Rooij, Bockting, Schiepek, Aas, Strunk, Aichhorn, Cramer and Lichtwarck-Aschoffn.d.; Groen et al., Reference Groen, Snippe, Bringmann, Simons, Hartmann, Bos and Wichersn.d.; de Vos et al., Reference de Vos, Wardenaar, Bos, Wit, Bouwmans and de Jonge2017). Moreover, a simulation study showed that networks with more strongly connected symptoms showed transitions to a depressed state more often compared with networks with weak connections (Cramer et al., Reference Cramer, van Borkulo, Giltay, van der Maas, Kendler, Scheffer and Borsboom2016). Furthermore, exposure to external stress resulted in sudden shifts in symptoms only in strongly connected networks. Within such networks, removal of the stressor after the phase transition occurred did not cause the system to shift back to its original state (Cramer et al.). The fact that recovery is not linearly related to the removal of the cause of the shift is called ‘hysteresis’. Such non-linearity is typical for complex systems. This phenomenon also has face validity for the field of psychiatry as it may explain why people remain stuck in a clinical state even after removal of certain provocative factors that had prompted the mental complaints.

Finally, the most direct support for the idea that symptom changes behave according to the principles observed in the complex system stems from empirical research showing that transitions in symptom levels can be anticipated by directly assessing changes in the stability of the system. From other fields of study (e.g. ecology and computer science), it is known that these changes in stability can be observed using certain ‘EWS’ (Tretyakov et al., Reference Tretyakov, Takayasu and Takayasu1998; van Nes and Scheffer, Reference van Nes and Scheffer2007; Dakos et al., Reference Dakos, Scheffer, van Nes, Brovkin, Petoukhov and Held2008). Such signals involve changes in the dynamics of important variables of a complex system, like increasing levels of autocorrelation (i.e. the current state of an element of the system becomes a better predictor for its future state), variance (i.e. elements of the system show greater amplitude changes in their intensity levels) or flickering (sudden changes in intensity levels). These reflect increasing instability of the system and have been shown to closely precede critical phase transitions in various sorts of complex systems. Considering the above, we expect that EWS (which involve more complex aspects of time-series dynamics than simple intensity changes of symptoms) may signal an increased likelihood of a phase transition and that such EWS can predict such transitions substantially earlier than simple changes in mean levels of these symptoms. This means that if psychopathology also behaves as a complex system, we may be able to find EWS that we can use to foresee important shifts in symptoms in an earlier phase and in a personalized manner. A few recent empirical studies already found support for this hypothesis (Schiepek et al., Reference Schiepek, Tominschek, Karch, Lutz, Mulert, Meindl and Pogarell2009; van de Leemput et al., Reference van de Leemput, Wichers, Cramer, Borsboom, Tuerlinckx, Kuppens, van Nes, Viechtbauer, Giltay, Aggen, Derom, Jacobs, Kendler, van der Maas, Neale, Peeters, Thiery, Zachar and Scheffer2014; Wichers et al., Reference Wichers and Groot2016). For example, in the time-series experiment in which a patient was followed over 239 days completing multiple measurements of mental states a day, EWS were observed in the sum score of all measured mental states. These EWS anticipated a subsequent phase transition in depressive symptoms (Wichers et al., Reference Wichers and Groot2016). Although these findings still await replication by a large-scale study in which EWS are followed over time within persons, support for the idea that psychopathology behaves as a complex system is accumulating. To conclude, many hypotheses have been formulated for the application of complex systems theory for the field of psychiatry and some empirical studies have been carried out suggesting we may be able to foresee transitions in levels of symptoms based on the system's changes in stability. We now want to further explore what this theoretical framework can do for foreseeing the type of symptom shifts that individuals may express in the near future. Before we move to the above mentioned theoretical explorations, however, it is important to first discuss the nature of psychopathology.

Redefining psychopathology as a multidimensional space

The idea that mental disorders are distinct and independent entities is not supported by empirical evidence (Carragher et al., Reference Carragher, Krueger, Eaton and Slade2015). All evidence points to the fact that diagnostic classifications are not independent and show huge overlap with one another (Kessler et al., Reference Kessler, Berglund, Demler, Jin, Merikangas and Walters2005; Merikangas et al., Reference Merikangas, He, Burstein, Swanson, Avenevoli, Cui, Benjet, Georgiades and Swendsen2010). Comorbidity is a rule rather than exception (Krueger and Eaton, Reference Krueger and Eaton2015). Also, patterns of symptoms within diagnoses are quite heterogeneous (Wardenaar and de Jonge, Reference Wardenaar and de Jonge2013). The human-made top-down boundaries and classifications established in diagnostic manuals may thus distort our view on the real structure of psychopathology. Clear boundaries between mental disorders seem absent and a patient's pattern of symptoms seems to spread across the various dimensions of psychopathology to form unique clusters of problems in each person. Despite such continuity, however, empirical studies show that some types of symptoms lump together more often while other combinations of symptoms hardly ever co-occur (Goekoop and Goekoop, Reference Goekoop and Goekoop2014; Boschloo et al., Reference Boschloo, van Borkulo, Rhemtulla, Keyes, Borsboom and Schoevers2015; Bekhuis et al., Reference Bekhuis, Schoevers, van Borkulo, Rosmalen and Boschloo2016). A common explanation for the co-occurrence of certain symptoms or mental states (e.g. feeling hopeless, guilty, down, fatigued, irritated) is that an underlying latent entity is responsible (e.g. depression). However, this view alone cannot explain that the precise combinations of symptoms differ between people and can be mixed with symptoms from other clusters. A more recent view is the network perspective, which does not necessarily assume an underlying causal entity, but theorizes that symptoms can also trigger each other and thereby form clusters of co-occurring symptoms in a self-organized, or bottom-up fashion (Cramer et al., Reference Cramer, Waldorp, van der Maas and Borsboom2010; Goekoop and Goekoop, Reference Goekoop and Goekoop2014; Borsboom, Reference Borsboom2017). We can then imagine a reality in which no absolute boundaries exist between diagnoses, and in which symptoms of psychopathology are all, to a varying extent, related to each other within one psychopathological space. However, some types of symptoms or mental states seem to lump together more often in this space while other combinations of symptoms hardly ever co-occur. From the network perspective, this may make sense, as it can be easily imagined that one mental state, for example, ‘feeling down’ easily triggers a certain other mental state, like ‘worrying’, but not so easily the mental state representing the feeling of being watched. Thus, certain mental states may be further apart from each other in space than other mental states. Similarly, certain groups of mental states may all be at a relatively close distance – meaning that they easily activate each other – making it likely that they are eventually ‘switched on’ together. Such groups may represent combinations of symptoms that have been frequently observed together and therefore received a group-identifying label such as ‘depression’ or ‘psychosis’.

Thus, symptoms and dimensions of psychopathology show preferential connections, which put constraints on the likelihood of certain psychopathological syndromes. Despite such global constrains, however, individual patients are known to differ in the wiring patterns of their psychopathology networks, which give rise to unique deviations from group-level connectivity. Thus, whereas in one person depressed feelings (such as sadness, guilt or shame) can easily trigger anxiety symptoms (such as worrying or inner tension) because they are close in the multidimensional space, this is not necessarily true in another person (see Fig. 2). Second, not only the connections between the clusters (lumps) of symptoms, but also the clusters themselves are likely to differ per individual with regard to their precise content: whereas for one person anxiety and depressive symptoms may form a single cluster or lump, this may not necessarily be the case for the next person. This implies that the structure of psychopathology and its supposed dimensions may differ per person and may not even be stable over time within a person. These theoretical ideas on the nature of psychopathology may explain the observed heterogeneity in symptom profiles (see Fig. 2).

Fig. 2.

Multidimensional space of psychopathology symptoms. The circles represent different symptoms and the distance between them represents the ease with which they can trigger their neighbour symptom. The coloured groups represent different clusters of symptoms. In scenario A, it is more likely that depressive symptoms may eventually trigger the manic cluster than in scenario B, while depressive symptoms in scenario B will more easily activate anxiety symptoms.