A provocative but problematic paper just out offers a new perspective on psychiatric symptoms.
The basic idea is that rather than psychiatric disorders being entities, they are just bundles of symptoms which cause each other:
...symptoms are unlikely to be merely passive psychometric indicators of latent conditions; rather, they indicate properties with autonomous causal relevance. That is, when symptoms arise, they can cause other symptoms on their own. For instance, among the symptoms of MDE we find sleep deprivation and concentration problems, while GAD (generalized anxiety disorder) comprises irritability and fatigue. It is feasible that comorbidity between MDE and GAD arises from causal chains of directly related symptoms; e.g., sleep deprivation (MDE) -> fatigue (MDE) -> concentration problems (GAD) -> irritability (GAD).
The authors seem to have mixed up their labels in the middle there, but you see the drift. This symptom-based approach stands in contrast to the idea that psychiatric illnesses are underlying things which lead to some symptoms. So it's a challenge to the notion of underlying biological dysfunction (except maybe for specific symptoms), but it's equally incompatible with any theory of underlying psychological causes - there's no room for Freudian unconscious "complexes" here. So there's something very straightforward and un-mysterious about this model, which will either make it attractive or suspect, depending on whether you think human life is mysterious or not. What's the evidence? First, the authors do an analysis of the DSM-IV diagnostic manual in terms of symptoms. They take every symptom which is mentioned in at least one diagnosis. They found 439 symptoms in total, over 201 disorders, with many symptoms, such as insomnia, shared between lots of different "disorders". They then used network analysis to create a kind of graph where the "distance" between the nodes (symptoms) is based on the number of shared diagnoses. The authors found that while some symptoms are unique to just one disorder, there's a core of highly shared symptoms which form a "giant component"
It's a very clever approach, but I wonder what it really tells us. The DSM-IV is not a set of data about mental illness. It's at best some data about what psychiatrists think about mental illness. Actually, it's not even that: it's data about what a particular set of psychiatrists, at a particular time, were able to agree upon. DSM-V is coming soon, and before that we had DSM's I, II and III. What about them? Do they have a different network structure? I'd have thought they would, but we don't know. We've already seen the kinds of politics that lie behind the decision to include or exclude a diagnosis in the DSM. In the upcoming DSM-V they're seriously proposing to add a new diagnosis ("TDDD"), purely in order to stop people getting another diagnosis (childhood "bipolar"). Now, there is a lot of symptom overlap between TDDD and bipolar disorder - for the very simple reason that one was designed for the purpose of diverting patients from the other. But that overlap doesn't tell us anything about real people with real symptoms. This is an extreme example, and to be fair to the authors they do acknowledge some of these problems with the DSM, but still. The authors go on to show that the symptomatic closeness between DSM-IV disorders predicts the rates of comorbidity between those disorders, as measured in the American population survey the NCS-R. This is true even of disorders which don't share any common symptoms, but which are connected indirectly, by a mutual friendship as it were. Finally they show that a statistical model based on interacting symptoms can predict the prevalence of depression (10% per year according to the NCS-R survey) and GAD (3% per year). It does so much better than a random model in which symptoms randomly interact. However, I'm not convinced that all these show us that the symptom-network approach is the best model to explain the occurence of these disorders. It only shows us that it's a model that works better than a rather crazy random model. I'm also not sure that being able to model the NCS-R data is necessarily a good thing, since these data are themselvesof questionable validity. But it's a genuinely interesting approach and well worth following up.
Borsboom D, Cramer AO, Schmittmann VD, Epskamp S, and Waldorp LJ (2011). The small world of psychopathology. PloS one, 6 (11) PMID: 22114671
