The Structure of the DSM

Click here for the dataviz.

The DSM-5 is a controversial Document. This has many reasons, not all of which are its own fault. As Louis Menand puts it:

Mental disorders sit at the intersection of three distinct fields. They are biological conditions, since they correspond to changes in the body. They are also psychological conditions, since they are experienced cognitively and emotionally—they are part of our conscious life. And they have moral significance, since they involve us in matters such as personal agency and responsibility, social norms and values, and character, and these all vary as cultures vary. (in the New Yorker)

Trying to integrate these contradictory sets of expectations guarantees a controversial result. But we can make also structural critiques of the DSM-5, which can, at least to a certain extent be raised on an a priori basis. One that is frequently uttered goes somewhat like this:

The diagnoses that the DSM-5 offers are mostly redundant: The diagnostic criteria are phrased in a way that makes it impossible to differentiate between disorders, which must be differentiated e.g. on biological grounds, and even worse, it makes it impossible to conduct reliable diagnosis on its basis.

To evaluate how this argument holds up, I have produced the above graphic. By using textmining techniques, e.g. the cosine-similarity measure for word vectors, I have mapped out which diagnoses in the DSM 5 that are most similar in their textual expression. The thickness of the links codes textual similarity. The size of the bubbles codes the yearly prevalence, if and only if the DSM-5 provides it, otherwise it is sized according to the mean. Please note that the similarity between some substance-disorders has not been analyzed, as I had quite limited calculation-powers and they are redundant anyway. Some of the links are not surprising at all, for example the strong links between bipolar disorders and different forms of depression. Cut-off (lowest depicted similarity) was a similarity-value of 0.15, the maximium value was 0.85, (mean = 0.3). This yields 325 links. Some links I expected are missing, for example between borderline and bipolar disorders, which are commonly considered to be hard to tell apart in clinical praxis. This seems to be a good point in case for those who wish to argue that the DSM-5 gives valuable differentiating information.