Moving from DSM-5 to ICD-11: A joint problem?

ICD Insights

The Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) was published in May 2013 and received widespread attention, not just in medical and scientific circles, but also in the media more generally. In part, the high interest was a result of the high profile of some of the DSM-5 critics who were not just the usual anti-psychiatry suspects. They included Allen Frances, the chair of the task force that had produced DSM-IV (Frances and Nardo, 2013) and Tom Insel, director of the US National Institute of Mental Health (NIMH) (www.nimh.nih.gov/about/director/2013/transforming-diagnosis.shtml). One of the major criticisms of Frances and his colleagues was that the authors of DSM-5 criteria seemed more concerned about avoiding false negatives than creating false positives and hence have turned ‘the existing diagnostic inflation in psychiatry into reckless hyperinflation’ (Frances and Nardo, 2013). The main concern of Insel was that DSM-5 is ‘based on a consensus about clusters of clinical symptoms, not [on] any objective laboratory measure’. He continued that psychiatric patients ‘deserve better’ and advocated the NIMH Research Domain Criteria as a solution. These aim to anchor diagnoses in a biological substrate of, for example, genetics neuroimaging and cognitive neuroscience. While we broadly agreed with both of these concerns, we see the issues of where to place the threshold between normality and disease and of how to establish biological validity as being interrelated. Here we will consider how the International Classification of Diseases, 11th Revision (ICD-11), in its section on psychiatric disorders, might attempt to do better than DSM-5 regarding these two issues and we will introduce a third concern – which is to do with the entire manner in which DSM diagnoses have been constructed ever since the inception of DSM-III.

Where to place the threshold

In a classic paper published nearly 40 years ago, Kendell (1975) grappled with the fundamental problem of how we should best define disease. He pointed out that often in medicine disorders can be regarded as at the extreme end of the continuum between disease and normality. Any attempt to position the threshold purely on statistical grounds is bound to be arbitrary and unsatisfactory. Kendell therefore invoked the notion of biological disadvantage and attempted to refine this concept, originally put forward by Scadding (1967). Kendell’s view was that biological disadvantage must entail either shortening of life or reduction in fertility. While Kendell’s definition is not perfect, there are many medical conditions that are unpleasant and disabling that do not affect lifespan or fecundity, it does work well for many psychiatric disorders at the severe end of the spectrum. For example, schizophrenia and severe depression are both associated with an increased risk of death by suicide and, especially in men, having fewer than average offspring. (Interestingly, there is some evidence of a ‘balancing’ effect in depression such that the unaffected siblings of depressed subjects have higher than average fecundity (Power et al., 2013).) There is also mounting evidence that both schizophrenia and recurrent depression are associated with increased rates of life-shortening physical disorders such as heart disease and hypertension (Farmer et al., 2008). While none of these facts tell us where ICD-11 should place its thresholds, they do suggest that a narrow, specificity-before-sensitivity philosophy is likely to take us closer to biologically definable entities.

Genetics and other biological substrata

The idea that genetic studies may be used to explore validity has been around for some time. Back in the 1980s we used a twin study approach to scrutinize the DSM-III definition of schizophrenia and found that by some measures it did rather well (Farmer et al., 1987). DSM-III defined a condition that was about 80% heritable but broadening the criteria for twin concordance to include other DSM disorders, including mood disorders, diminished the genetic contribution. However, subsequently using a large dataset derived from the same twin register we found that there was a substantial genetic overlap between DSM-IV schizophrenia and bipolar disorder (Cardno et al., 2002). This seemed controversial at the time but has subsequently been supported by molecular genetic studies. Indeed, analysis of data on enormous numbers of patients included in genome-wide association studies (GWAS) suggest degrees of aetiological overlap between schizophrenia, bipolar disorder, depression, autism and attention deficit hyperactivity disorder (Smoller et al., 2013). Neuroimaging studies also point not to characteristic lesions but to some rather subtle changes such as decreases in hippocampal volume that are found both in schizophrenia and depression (Cole et al., 2011). In short, while genetics, imaging and other biological approaches indicate that biological substrata exist, the methods are not yet sufficiently sharp to be used to carve Nature at the joints.

The structure of DSM criteria

It is now often forgotten that the way in which DSM-III criteria were structured was in 1980 revolutionary. For the first time an entire ‘official’ system of diagnosis was set out as explicit, operational criteria (McGuffin et al., 1991). This provided a huge advantage over the then conventional clinical descriptive approach of providing much enhanced diagnostic reliability. High reliability is now accepted as a prerequisite for research and this could in part explain why since DSM-III the American Psychiatric Association’s criteria have overtaken ICD as the more favoured system of classification used in research papers around the world. However, there are real drawbacks to applying operational criteria in purely clinical settings, the main one being that clinicians tend to work by pattern recognition whereas diagnostic criteria as set out in DSM-III and its successors require algorithm solving. Solving algorithms is of course easy for a computer but often tedious and difficult for human beings. Some of the algorithms embodied in DSM-5 have become quite fiendish; for example, the distinction of mild, moderate and severe depressive episodes now requires not just a symptom count but complicated combinations from two lists of symptoms plus a judgement on impairment across six possible domains. The DSM has often been referred to as the psychiatrist’s bible, but very few clinicians are going to be capable of observing its rules religiously.

What ICD-11 should do

While the authors of the DSM and the ICD mental disorders section have expressed the wish that the two classifications should be harmonized, there are certain differences that ICD-11, due to be published in 2015, should maintain or introduce. The first is that there should be a preservation of stringency of diagnosis. This means not just maintaining strict thresholds for current diagnostic categories but avoidance of the new single symptom ‘disorders’ introduced by DSM-5 to explain hoarding, binging, ‘mood dysregulation’ or mild forgetfulness. The second is that consideration should be given to introducing an NIMH-style research domain approach alongside the clinical criteria. Perhaps current categories mean that we are looking in the wrong places for Nature’s joints. The third and related point is that the distinction between research criteria (operational) and clinical criteria (descriptive) that already exists in ICD-10 should be maintained.

There is, however, a fourth point to consider, which is whether we really need ICD-11 at all. The attempt to improve on DSM-IV based on new research has not been a success and it is unlikely that the coming year will result in any breakthroughs or paradigm shifts that will make ICD-11 a real advance on ICD-10. So perhaps psychiatry needs to patiently persist with the current flawed but serviceable working hypotheses of ICD-10 and carry on with the task of discovery of whether and where the joints of Nature exist.