Rethinking how psychiatric presentations are understood: an introduction to mental disorders as complex systems

Joel Schamroth & Salman Razzaki
11 min readJul 5, 2022

--

Written by Iain Jordan, Joel Schamroth and Salman Razzaki

  • Current reductionist approaches to understanding mental ill-health have led to little progress in the field in recent decades. The practical and theoretical limitations of existing psychiatric classification systems are well established.
  • Mental disorders are dynamic, complex and heterogeneous. Tools from complexity science can model mental disorders as causal networks of symptoms, psychological processes, and contextual factors.
  • These emerging approaches hold promise for enabling a more complete and useful understanding of the nature of mental disorders and can help aid the improvements in diagnosis and treatment.

Psychiatry & the latent disease model

Existing psychiatric classification systems adopt the assumptions of the latent disease model. According to this model, symptoms emerge from an underlying disease entity, and symptoms guide clinicians towards the unifying disease process from which they originate. For example, lung cancer may present in different patients with shortness of breath, a neuropsychiatric syndrome or weight loss, but what unites these symptoms is that they stem from the same cause, the same latent disease.

Mainstream psychiatric practice and research has long operated under these assumptions, with psychiatric symptoms being understood to result from an underlying disease entity or ‘brain disorder’.(1) In the early years of modern psychiatry and neurology, this approach led to important discoveries such as niacin deficiency as the underlying causes of ‘general paralysis of the insane’ and pellagra respectively — once common causes of mental illness. However, these early victories can now be understood as exceptions to the rule.

Mental disorders do not fit the latent disease model in the way that diseases of the heart or lungs do.(2) Despite advances in genetics and neuroimaging in other medical specialties, attempts to reduce mental disorders to discrete biological processes (biological reductionism) have largely been unsuccessful. The latent disease model has yielded little progress in our understanding of the causes of mental ill health or progress in therapeutics over the past 50 years.(3)

The shortcomings of the DSM

The Diagnostic and Statistical Manual of Mental Disorders (DSM)(4) and similar classification systems implicitly adhere to the latent disease model. While these classification systems may be useful heuristics for guiding treatment, critics argue that they fundamentally misrepresent the real world.(5)

The limitations of the DSM/ICD are well documented. Inter-rater reliability is often unacceptably low, meaning that for very common conditions like major depressive disorder and generalized anxiety disorder, two clinicians seeing the exact same patient are likely to reach differing DSM diagnoses.(6) People with the same diagnosis may not share any symptoms and at the same time, within a single diagnosis there may be thousands of symptom profiles.(7)

Comorbidity is pervasive amongst those diagnosed with psychiatric disorders and yet this becomes difficult to interpret when the boundaries between disorders are arbitrary and largely based on history and convention.(8,9) Multiple risk factors may cause a single mental disorder (equifinality) and a single risk factor may cause multiple different disorders (multifinality).(10) Many patients with clinical-level problems may fall through the cracks and not satisfy criteria for any disorder.(11)

Despite decades of research based on reductionist models, we still do not know what types of phenomena mental disorders are in the way we understand, for example, the nature of cancer.(12) Given these shortcomings, there is a compelling case to be made that such classification systems are holding back research in psychiatry and must be abandoned in order to make real advances into the insights of the eitiology of mental disorders.(2)

Rethinking psychiatry and embracing complexity

Recent years have seen new initiatives using data-driven approaches to understanding mental ill health with approaches that move beyond classification to instead study underlying processes. These can be understood as ‘transdiagnostic’ efforts to move beyond existing categorical diagnoses in order to advance our understanding of mental ill-health.(13)

The Hierarchical Taxonomy of Psychopathology (HiToP)(14) and the Research Domain Criteria (RDoC)(15) aim to explore the underlying latent phenomenological structures and neural processes which give rise to mental disorders. Various computational approaches such as Bayesian inference and reinforcement learning have become available to aid researchers’ understanding of the nature of mental ill-health, and potential new treatments.(16)

Transcending the reductionist latent disease model means seeing psychiatric conditions as complex phenomena that are multicausal, heterogenous and dynamic.(17) Complexity science (specifically, the study of networks and complex dynamical systems) offers a novel way to conceive of psychiatric conditions.

Complex systems science is a transdisciplinary approach to understanding natural systems, and the scientific successor to reductionism.18 While linear systems may be complicated and have many moving parts, their internal dynamics are characterised by the reliability and stability of these systems to produce well-defined, predictable outputs. By contrast, complex systems exhibit dynamics which are difficult to understand even when their component parts are well-understood. They are characterised by ‘emergence’, the phenomena whereby the properties of the whole cannot easily be understood or predicted from the knowledge of its components. They display nonlinear dynamics, meaning that changes in inputs lead to unpredictable changes in outputs. Mental ill-health emerges from complex biological systems.

Mental disorders as networks of symptoms

Among the fields of complexity, network theory has emerged as the most empirically based and perhaps the most useful in studying complex systems. (19) It is also the field of study which appears to lend itself best to studying mental disorders. Network theory emerged from the study of the nature and dynamics of relationships between phenomena such as groups of people (social networks) or computers (the internet). In the past decade, increasingly sophisticated mathematical and computational tools have allowed researchers to model the dynamic relationship between networks of psychiatric symptoms (and other causal factors such as poverty and trauma).

In a network approach, it is the symptoms themselves, interacting with, triggering, and affecting each other within this network, that create emergent psychiatric phenotypes.(2,20) Networks of interacting symptoms are represented graphically by nodes and edges. A node can represent any phenomenon (in this case, a symptom) and an edge represents the association between two nodes. It is evident that many of the symptoms in the DSM and other classifications are causally related. For instance in Major Depressive Disorder (MDD), loss of appetite causes weight loss. Relationships (including casual relationships) between symptoms can thus be modelled statistically and represented graphically by diagrams such as the ones below (Figure A).

Figure A

Figure B

Symptoms that commonly co-occur are closely interconnected in the network, forming local clusters. The DSM’s arbitrary lists of diagnostic criteria may approximately reflect some clusters or groups of symptoms in the network, but network theory offers an explanatory model of symptom interaction. For example, the symptoms that we would associate with the DSM criteria for depression are not merely a list but rather interact causally (rumination can lead to insomnia, which leads to tiredness, and eventually depressed mood).

In a cluster, like the one for MDD above, the strength of the statistical relationship between symptoms (the strength of the association between nodes) can be represented by the thickness of a line (figure B). The centrality of a node refers to a node’s connectivity with other nodes (ie how many other nodes it is connected to or the strength of their association).2 Highly central symptoms are causally related to many other symptoms. When a symptom is present in two different disorders, it can be thought of as a bridge symptom. It is then easy to see how a person with symptoms of (or at risk for) one disorder can develop another disorder. This bridge symptom can spread activation from one network cluster to another (see B2,B3,B4 in figure C). Comorbidity is therefore best understood as reflecting not only indistinct boundaries between ‘disorders’ but also seems to be a fundamental characteristic of classification systems.(21,22)

Figure C. from Fried et al 2016. Comorbidity is demonstrated between symptom cluster X and symptom cluster Y, each representing a specific psychiatric condition. B1,B2,B3 are bridge symptoms that are present in both diagnoses. Here E represents an external stressor (E) for example a physical illness or distressing life event that causes symptom X3.

Complex systems tend towards stable states termed attractor states.(23) Both health and illness may be thought of as attractor states. A phase transition occurs when a shift occurs from one attractor state to another.(20,23) Figure D illustrates how networks with different properties may behave differently in response to an influence such as stress. Here, attractor states are visualised as depressions in a landscape. For Bob, symptoms in his network are tightly connected. His symptoms activate each other until a tipping point is reached, and Bob transitions from stable state A (healthy) to stable state B (depression).

By comparison, the relationship between symptoms in Alice’s network is weaker and she never develops this causal cascade and her symptoms do not harden into a self perpetuating syndrome. After a period of stress she returns to a healthy, stable state.(20,24)

Figure D. From Borsboom et al 2016.

A familiar concept such as resilience can this be defined as the tendency of a network to remain in a healthy stable state. By comparison, the concept of vulnerability to illness can be thought of as the tendency of networks to transition into an unhealthy stable state. Thus, for people vulnerable to mental ill-health, networks of symptoms (and other causal factors) harden into self-reinforcing syndromes (vicious cycle), and in those who are not, protective factors form a mutually reinforcing network (virtuous cycle), or symptoms are weakly causally connected and thus do not tend towards an illness state.

Certain characteristics of networks — such as strong connectivity — also mean greater potential for acute phase transitions from health to illness. By contrast, in weakly connected networks, symptoms will increase gradually prior to an eventual phase transition.(20) Such variation in network dynamics illuminates longstanding debate about whether psychiatric conditions should be viewed as dimensional or belonging to discrete categories (23,24). Depending on the dynamics of a given network, the appearance may be of a clear differentiation between health and illness, while for another patient or disorder, the phenotype may appear to be more continuous.(23)

From complexity science to clinical practice

In the coming years, network science will increasingly be used to generate testable hypotheses about the nature of mental disorders by using mathematical and computational tools originally developed by physicists and computer scientists. For instance, Robinaugh et al. have recently developed a computational model of panic disorder, the first ever formal mathematical model of a mental disorder.(25) In future, it may be commonplace for models of individual symptom networks to inform fine-grained deployment of different psychotherapeutic and non-psychotherapeutic treatment modalities — for instance, by tailoring therapeutic approaches to individual patients, symptoms (nodes) with high centrality may be excellent targets for treatment. (26,27,28)

Currently, there is considerable treatment effect heterogeneity in trials of psychotherapeutic treatments for depression.(29) This presents opportunities for using dynamic network approaches to examine transdiagnostic, process based therapeutic approaches for common mental disorders.(30) Increasingly precise modelling could pave the way for testing of theories and treatments for groups or individual patients without risk by modelling effects in a virtual environment. Crucially such models must move beyond static, cross sectional networks of symptoms and start to take into account dynamics over time and the influence of sociocultural context.

Understanding mental disorders as networks of interacting phenomena can be useful heuristics for clinicians and patients which aid case conceptualisation, selection of treatment strategies, and follow-up. It is likely to be the case that experienced clinicians already think in terms of causal relationships between symptoms, explicitly so in the case of modern psychotherapeutic formulation. As the research described herein develops, formal theories of psychopathology will hopefully be more intuitive and useful than the crude models of current classification systems.

It will take time as theory and model mutually reinforce each other to develop an increasingly sophisticated understanding of mental ill-health — the field continues to expand apace. While research is preliminary and yet to be translated to produce meaningful bedside results, in part because the nature of messy real-world data collected leads to inconsistent and difficult to reproduce networks (31), mathematical and modelling tools are being refined to better reflect reality. One day, formal modelling may well be as normal a part of assessment and treatment as classification systems are today.

REFERENCES:

  1. Insel T, Cuthbert B. Brain disorders? Precisely. Science. 2015;348(6234):499–500.
  2. Hofmann S, Curtiss J. A complex network approach to clinical science. European Journal of Clinical Investigation. 2018;48(8):e12986.
  3. Kendler, K. S. The dappled nature of causes of psychiatric illness: Replacing the organic–functional/hardware–software dichotomy with empirically based pluralism. Molecular psychiatry. 2012;17(4),377–388.
  4. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, 5th ed. American Psychiatric Publishing, Washington D.C. 2013.
  5. Fried E. Studying Mental Disorders as Systems, Not Syndromes. PsyArXiv. 8 Aug. 2021. Web. doi:10.31234/osf.io/k4mhv
  6. Regier D, Narrow W, Clarke D, Kraemer H, Kuramoto S, Kuhl E et al. DSM-5 Field Trials in the United States and Canada, Part II: Test-Retest Reliability of Selected Categorical Diagnoses. American Journal of Psychiatry. 2013;170(1):59–70.
  7. Fried E, Nesse R. Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders. 2015;172:96–102.
  8. Plana-Ripoll O, Pedersen C, Holtz Y, Benros M, Dalsgaard S, de Jonge P et al. Exploring Comorbidity Within Mental Disorders Among a Danish National Population. JAMA Psychiatry. 2019;76(3):259.
  9. Hyman S. The Diagnosis of Mental Disorders: The Problem of Reification. Annual Review of Clinical Psychology. 2010’6:1,155–179
  10. Cicchetti, D., & Rogosch, F. A. Equifinality and multifinality in developmental psychopathology. Development and Psychopathology. 1996;8(4), 597–600
  11. Newson J, Thiagarajan T. Assessment of Population Well-Being With the Mental Health Quotient (MHQ): Development and Usability Study. JMIR Mental Health. 2020;7(7):e17935.
  12. Kendler K, Zachar P, Craver C. What kinds of things are psychiatric disorders?. Psychological Medicine. 2010;41(6):1143–1150.
  13. Fusar-Poli P, Solmi M, Brondino N, Davies C, Chae C, Politi P, Borgwardt S, Lawrie SM, Parnas J, McGuire P. Transdiagnostic psychiatry: a systematic review. World Psychiatry. 2019;18(2):192–207.
  14. Kotov R., Krueger, R. F. Watson et al. The Hierarchical Taxonomy of Psychopathology (HiTOP): A dimensional alternative to traditional nosologies. Journal of abnormal psychology. 2017;126(4), 454.
  15. Insel T., Cuthbert B, Garvey, M et al. Research domain criteria (RDoC): toward a new classification framework for research on mental disorders. American Journal of psychiatry. 2010;167(7),748–751.
  16. Huys Q, Browning M, Paulus M, Frank M. Advances in the computational understanding of mental illness. Neuropsychopharmacology. 2020;46(1):3–19
  17. Fried E, Robinaugh D. Complexity in mental health research: theory, method, and empirical contributions. 2021. [online] Available at: https://www.biomedcentral.com/collections/mental-health-research.
  18. Dent, E. B. Complexity science: A worldview shift. Emergence. 1999; 1(4), 5–19.
  19. Barabási A. The network takeover. Nature Physics. 2011;8(1):14–16.
  20. Borsboom D. A Network Theory of Mental Disorders. World Psychiatry. 2017; 16,1,26 5–13.
  21. Borsboom D, Cramer A. Network Analysis: An Integrative Approach to the Structure of Psychopathology. Annual Review of Clinical Psychology. 2013;9(1):91–121.
  22. Borsboom D. The Structure of the DSM. Archives of General Psychiatry. 2002;59(6):569–570.
  23. Borsboom D, Rhemtulla M, Cramer A et al. Kinds versus continua: a review of psychometric approaches to uncover the structure of psychiatric constructs. Psychological Medicine. 2016;46(8):1567–1579.
  24. Fried E, van Borkulo C, Cramer A et al. Mental disorders as networks of problems: a review of recent insights. Social Psychiatry and Psychiatric Epidemiology. 2016;52(1):1–10.
  25. Robinaugh, Donald, et al. Advancing the Network Theory of Mental Disorders: A Computational Model of Panic Disorder. PsyArXiv, 29 May 2019. Web: doi:10.31234/osf.io/km37w.
  26. Fried E. Lack of Theory Building and Testing Impedes Progress in The Factor and Network Literature. Psychological Inquiry. 2020;31(4):271–288.
  27. Robinaugh D, Millner A, McNally R. Identifying highly influential nodes in the complicated grief network. Journal of Abnormal Psychology. 2016;125(6):747–757.
  28. Hagan K, Matheson B, Datta N et al. Understanding outcomes in family-based treatment for adolescent anorexia nervosa: a network approach. Psychological Medicine. 2021;1–12.
  29. Kaiser, T, Volkmann C, Volkmann A. et al. Heterogeneity of Treatment Effects in Trials on Psychotherapy of Depression. PsyArXiv, 23 Nov. 2020. Web.doi:10.31234/osf.io/mqzvy.
  30. Hofmann S, Curtiss J, Hayes S. Beyond linear mediation: Toward a dynamic network approach to study treatment processes. Clinical Psychology Review. 2020;76:101824.
  31. Carlucci, L. Saggino, A. & Balsamo, M. On the efficacy of the unified protocol for transdiagnostic treatment of emotional disorders: A systematic review and meta-analysis. Clinical psychology review. 2021;87,101999.

--

--

Joel Schamroth & Salman Razzaki

Thoughts on the intersection between health, technology and society