Network Analysis: An Integrative Approach to the Structure of Psychopathology - Borsboom & Cramer - 2013 - Article

Introduction

There appear to be no simple answers to questions such as: "Why is it that some people are vulnerable to developing mental disorders, while others are not?" and "Why are researchers unable to identify the psychological, neurobiological or genetic essential characteristics of mental disorders?". Psychopathology research tried to reduce mental disorders to simple psychological, environmental or biological dysfunctions or predispositions, but this has failed. This is due to the high multi-factorial etiology of these disorders: these disorders are not simply caused by one factor. Even though some disorders are heritable, the effects of particular genes on the risk for developing particular mental disorders are small and often not specific to the disorder. 

However, even though that disorders have high multi-factor etiology, the symptoms do covariate. In other words: a symptoms of for example major depression (MD) does hang together more with another symptom of MD instead of with a symptom of, for example, panic disorder. So, the authors state, there must be something that makes these symptoms correlate. The question then is, what is that thing?

The authors state that the most common answer to that question is: "the disorder". According to the disease model, which is based on the paradigm in modern Western medicine, the problems that people encounter (symptoms), are causes of a small set of underlying disorders. An analogy is a lung tumor (disorder) which causes shortness of breath, chest pain and coughing up blood (symptoms). So, according to the disease model, the symptoms of major depression correlate, because they are caused by the same underlying disorder, namely major depression! 

Because of this thinking, many researchers looked for psychological/environmental/biological essences of mental disorders. However, as noted before, they have not been able to find these essences. The authors of this article use this as proof for that the use of the disease model for psychopathology is not correct. So, they falsificate the disease model in psychopathology.

There are other researchers who also question the use of the disease model for psychopathology. However, at the moment, it is not clear which alternative should be used. The author states that modern psychometric approaches are the mirror of the disease model. So, clinical symptoms are explained with the use of a small set of latent variables. For example, when panic disorder is a latent variable (an unobservable disorder) that causes observable symptoms, such as experiencing panic attacks, then from this psychometric perspective, symptoms are regarded as measurements of a disorder. Therefore, the symptoms are aggregated in a total score that reflect a person's stance on that variable. For a long time, this was the only psychometric way for clinical psychology.

However, the authors of this article introduce a new psychometric game: network analysis! They state that complex network approaches have the potential to provide a way of thinking about disorders that is in line with their complex organization. In these approaches, disorders are conceptualized as systems of causually connected symptoms rather than as effects of a latent disorder. With the use of network analysis, these systems can be represented, analyzed and studied in their full complexity. The use of network analysis also drops the idea that symptoms of a disorder have a single causal background. In this article, the authors explain the basic premises of the network approach and offer a practical guide for how to collect and analyze psychopathological data with a network model. 

They start by discussing the "erroneous influential paradigm" of current psychopathology research that symptoms are caused by disorders.

Symptoms and disorders in psychopathology

What is known, is that people suffer from symptoms and that these symptoms cluster together. The symptoms of disorders are the only things that we can empirically test. We can not empirically test the mental disorders: there is no test for schizophrenia as compared to the Down syndrome (the presence of a third copy of chromosome 21).

The authors contrast the situation in psychopathology with that in medicine. For example, when someone suffers from headaches, forgetfulness and foggy eyesight, these may be the result of a brain tumor. This tumor would then be identifiable and would be conceptually separate from its symptoms. So, this means that someone may have: a) headaches without a brain tumor; b) a brain tumor without headaches; c) headaches and a brain tumor, and d) the headaches would not have been present without the tumor. In medicine, researchers can separate the medical condition from its symptoms (so, one can occur without the other) and they can identify the medical condition as the root cause of the symptoms. Therefore, it is good to treat the root cause of the symptoms (the tumor). So, in medicine, the disease model works well.

However, in psychopathology, this model is not correct. Clients are often diagnosed with a disorder because of the symptoms that they experience. After this diagnosis, a treatment protocol is chosen. So, this would reflect that there is a 'root cause'. However, it has been impossible to identify these root causes empirically (so, "major depression" can not be found). The authors state that this will not change, not even with advances in technology. They explain this by stating that for a disease model to hold, it should be possible to conceptually separate conditions from symptoms. They state that this is unlikely for mental disorders: if major depression existed independently of its symptoms, then this would mean that it is possible to have major depression without experiencing the symptoms of it. Also, if substance use was a disorder that was independent, then this would mean that it is possible to have the disorder without using a substance. Therefore, the authors accept the assumption that mental disorders cannot be separated from their symptoms. Therefore, the disorders cannot be causes of these symptoms: at least not in the same way as a tumor or cancer in medicine. Therefore, the treatment of disorders as being causes that exist independently of their symptoms as is done currently, is faulty.

The authors state that the relation between symptoms and disorders has to be conceptualized in a different way. Arguments for this statement are that mental disorders are not identifiable as separate disease entities, and that there are also many direct causal relations between symptoms.

In medicine, the relation between symptoms and disease is asymmetric: so, the tumor causes foggy eyesight and not the other way around. In psychopathology, research suggests that mental disorders may be caused by the direct activation of symptoms through, for example, adverse life events. For example, when someone loses a significant other, then this might lead to insomnia. Insomnia, in turn may lead to an episode of major depression. The authors state that the mechanism that might cause this is a symptom-symptom causation. For example, consider this chain: chronic stress -> depressed mood -> self-reproach -> insomnia -> fatigue -> concentration problems. This chain includes five symptoms of major depression and lead to a diagnosis of an episode of major depression. However, these paths may differ between people. For example, someone who develops major depression after health problems, follows a different path compared to that described. Therefore, the authors state, it is very unlikely that symptoms are caused by mental disorders. They end with stating that there do not only exist causal and meaningful relations between symptoms, but that these relationships are the very thing of which mental disorders are made!

Complex psychopathology networks

In network analysis, symptoms are not interpreted as a function of a set of underlying or latent disorders. Instead, symptoms are seen as mutually interacting, often reciprocally reinforcing, elements of a complex network. So, rather than interpreting symptoms as measurements of a latent disorder, symptoms are viewed as part of a causal system.

So, in network analysis, the symptoms are seen as the active ingredients of the mental disorders. There are two assumptions: a) given the current evidence, we should forestall the conclusion that symptoms of the same disorder are uniformly caused by a single psychological or biological condition and b) psychopathology symptoms influence each other. Even though many researchers could easily accept these assumptions, this acceptance would be potentially radical. An explanation for this is that, if it is indeed the case that there are direct and reciprocal interactions between systems, then it becomes unclear whether the disorder itself is even required to make sense of the symptoms. For example, decreased appetite and losing weight do not correlate highly because they are caused by the same disorder (major depression), but because they are causally related (decreased appetite -> losing weight). The second consequence of accepting the assumptions or hypothesis is that comorbidity can no longer be meaningfully explained as a correlation between two disorders, or as a result of a common underlying neurobiological dysfunction or even "super disorder". Instead, in network analysis, the causal relations between symptoms are seen as pathways that can connect different disorders, for example via bridge symptoms (symptoms that are part of both disorders).

In network analysis, boundaries between disorders are fuzzy, because multiple pathways may exist in such a way that there is no objective or "true" point at which to carve the symptom network in two (with each representing a separate disorder). So, in network analysis, the boundaries are not fuzzy because of methodologicall limitations, but because of the intrinsic structure of disorders. In summary: boundaries between disorders are fuzzy, because there are no true boundaries. In the network approach, we can define disorders as sets of more densely connected symptoms (like a school of fish or a flock of birds), but these disorders are literally intertwined with one another and cannot be neatly separated. The final consequence of accepting the assumptions is that the target of therapeutic interventions may change. So, if major depression does not exist as an entity independent of its symptoms, then attempting to treat it analogous to the way medical conditions are treated (removing the tumor), is faulty. So, in network approach, interventions are optimally targeted at the symptoms themselves or at the causal relationships that connect them. According to the authors, this view fits well with many already existing therapeutic interventions.

Constructing and analyzing psychopathology networks

Network analysis has its roots in physics and mathematics. A network is a set of elements (nodes) which are connected through a set of relations. These elements and nodes can be anything. For example, nodes can be neurons and the relations can be the number of times any two neurons fire at the same time. Therefore, using network analysis is relatively easy, because the use of network models do not require extensive prior knowledge, in contrast to other methodologies. Instead, one only needs a set of elements and an idea of these elements are connected. 

Constructing psychopathology networks can be done in different ways. First, one can use the information in the diagnostic systems (such as the DSM), because these systems often contain clues about causal relationships in disorders. Second, one can use the assessment of causal relations between symptoms, which are rated by clinicians or patients. Third, one may use data on symptom endorsement frequencies to extract empirical patterns of association that can serve as input for network structures (odds ratios, partial correlations or pathways that are detected through causal search algorithms).

A small world in the network analysis literature refers to that there is a large group of nodes that are all connected to one another, either directly or via intermediary nodes. So, on average, paths from one node to another are short and there is a large degree of clustering. Milgram was the first to show the small world phenomenon: he famously instructed people to send letters to other people that they did not know by giving the letters to acquaintances they felt might know the target. The people who received the letter did the same. He found that, on average, it took six steps to reach the target. This finding became known as: "six degrees of separation". So, a small world refers to that even though a network may be big and consist of clusters, any node can be reached from another node within just a few (six) steps!

The many roads to disorders: individual networks

When performing between-subjects psychology network analysis, this can yield information about the general structure of psychiatric disorders. However, it does not tell us a lot about individual differences. For example, why did Bob develop an episode of major depression, while Susan developed a panic disorder? To answer these questions, one needs to study the networks of individuals. Each individual has his or her own network, with each specific vulnerabilities (or risk factors). 

To study these individual networks, the use of cross-sectional data will be useless. Instead, the authors state that time-series is a good alternative. In time-series, researchers ask individuals to report on various aspects of their physiological and psychological well-being at least once a day, for many days. For example, in the experience sampling method, people are asked to report, during their daily life, their thoughts, feelings, and symptoms as well as the context in which these thoughts/feelings/symptoms take place. An advantage of using this method is that researchers are able to collect time-intensive data and also a) data on the relationship between events happening in a person's life and the subsequent ripple effects of that event in the symptoms that the person experiences, and b) data from people without psychopathology who might be progressing toward developing a mental disorder. The authors state that one can also learn about intraindividual behavior by simulating time-intensive intraindividual data. With the use of this simulated data, they can answer interesting questions in psychopathology.

Conclusion

So, for mental disorders, there is no evidence for that essentialism is appropriate. The authors have explained this by contrasting psychopathology to medicine. For example, in medicine, a tumor is identifiable, but in psychopathology, major depression for example, is not identifiable. Network analysis acknowledges that there is something real about mental disorders, but not in the same way as in the current paradigm. Instead, they propose that causal networks of thoughts, feelings, behaviors, and physiological phenomena interact with each other and may lead to mental disorders. So, in network perspectives, mental disorders are conceptualized as clusters of mechanically connected properties.