The organizational principle underlying the majority of diagnostic groupings in DSM-IV and ICD-10 is shared clinical phenomenology. In the accompanying series of seven papers, Andrews and colleagues explore the feasibility of going beyond the DSM-IV and ICD-10 presenting symptom approach and instead propose grouping disorders together based on shared ‘risk factors and clinical factors’ such as genetic risk factors, neural substrates, temperamental antecedents, abnormalities of cognitive or emotional processing, and high rates of co-morbidity. They conclude that it is ‘feasible’ to regroup the DSM-IV and ICD-10 disorders on this basis, proposing five clusters: a neurocognitive cluster identified primarily by neural substrate abnormalities, a neurodevelopmental cluster identified primarily by early and continuing cognitive deficits, a psychosis cluster identified primarily by clinical features and information-processing deficits, an emotional cluster identified primarily by the antecedent of negative emotionality, and an externalizing cluster identified by the temperamental antecedent of disinhibition. (A sixth unnamed grouping, analogous to the Not Otherwise Specified category in DSM-IV, is proposed for those disorders that cannot be assigned to any of the proposed five clusters.) The question, of course, is not whether it is possible to regroup disorders based on a principle other than predominant symptomatic presentation but whether such a regrouping would be an improvement in terms of clinical utility, research utility, educational utility, and administrative utility.

It should first be stated that any significant change in the diagnostic system inevitably comes with costs. Although one might think that rearranging disorder groupings would have only a minor impact compared to more significant changes such as adding new disorders or changing diagnostic criteria sets, the organization of the DSM diagnostic groupings has had a significant impact on the field. Psychiatry texts and other books that cover the full range of DSM disorders (e.g. the American Psychiatric Association's Handbook of Psychiatric Measures; Rush et al. Reference Rush, First and Blacker2008) mirror the DSM organizational plan in order to facilitate finding material relevant to the various disorders. The diagnostic domains of various professional and advocacy groups also conform to DSM diagnostic groupings. For example, the mission statement of the Anxiety Disorders Association of America (ADAA) states ‘the association is made up of professionals who conduct research and treat anxiety disorders and people who have a personal or general interest in learning more about anxiety disorders: obsessive–compulsive disorder (OCD), panic disorder (panic attack), social anxiety disorder (social phobia), post-traumatic stress disorder (PTSD), specific phobias, and generalized anxiety disorder (GAD) ’ (ADAA, 2009). Similarly, a number of subspecialty journals have sprung up that focus on particular DSM diagnostic groupings (e.g. Journal of Affective Disorders, Journal of Anxiety Disorders, Journal of Personality Disorders). These examples illustrate the tendency for DSM decisions to become reified and thus are indicative of the potential impact that such a reorganization can have on the field. Making changes is not something that should be undertaken lightly.

The costs involved in making changes to the DSM should not in and of themselves stand in the way of change if the benefits significantly outweigh the costs. However, before we can determine whether rearranging the diagnostic groupings is likely to improve the classification, we must first consider the function of the diagnostic groupings in the DSM and ICD classification systems. There are thousands of diseases and disorders listed in the International Classification of Diseases, Tenth Edition (ICD-10). In order to facilitate finding a particular disorder in the ICD, the classification is divided into major diagnostic divisions generally based on organ system (e.g. diseases of the digestive system), anatomic location (e.g. diseases of the ear and mastoid process), common pathophysiological process (e.g. infectious diseases, neoplasms) or medical specialty [e.g. separating diseases of the nervous system (Neurology) from mental and behavioral disorders (Psychiatry) ]. Within each of these major divisions are groupings that further subdivide the domains into smaller chunks of disorders. The organizational principles guiding these subdivisions vary according to the nature of the diagnostic domain (see Table 1). For example, infectious diseases are grouped by anatomical location (e.g. intestinal infectious diseases), type of organism (e.g. protozoal diseases), and mode of transmission (e.g. infections with a predominantly viral mode of transmission) and diseases of the circulatory system are divided according to pathophysiology (e.g. ischemic heart diseases) and anatomy (e.g. diseases of arteries, arterioles and capillaries). Throughout the ICD, for those subdomains in which the etiology and pathophysiology is unknown (e.g. headaches, connective tissues diseases), shared clinical presentation is the guiding principle.

Table 1. ICD-10 organizational principles by section

As with the other medical domains, several different organizing principles govern the diagnostic groupings in the DSM and the mental disorders section of the ICD. The first DSM-IV grouping, ‘Disorders Usually First Diagnosed in Infancy, Childhood, or Adolescence’ is for ‘convenience only’ and is intended to make it easier for mental health practitioners who specialize in children to locate the diagnoses they most commonly treat. The next three sections (i.e. Delirium, Dementia, Amnestic and Other Cognitive Disorders, Mental Disorders due to a General Medical Condition, and Substance-Related Disorders) contain disorders whose etiology is by definition known, i.e. each of these disorders is caused by either the direct physiological effects on the central nervous system of a substance or general medical condition. The remaining disorders in DSM-IV are grouped based on shared phenomenological features, reflecting the DSM-III descriptive approach in which etiological theories were eschewed in favor of descriptions of disorders' symptomatic manifestations. With only a few historical exceptions (e.g. Schizophrenia, Asperger's disorder), the names of most DSM disorders include elements of their presenting symptomatology (e.g. Major Depressive Disorder, Specific Phobia, Primary Insomnia). Basing diagnostic groupings on presenting symptomatology is intuitive and greatly facilitates finding disorders in the classification (e.g. Major Depressive Disorder in the Mood Disorders grouping, Specific Phobia in the Anxiety Disorders grouping, Primary Insomnia in the Sleep Disorders section). Furthermore, grouping disorders around predominant presenting symptoms facilitates differential diagnosis, which is one of the core functions of the diagnostic assessment process. Patients typically present for clinical attention complaining of one or more clinically significant symptoms and it is the clinician's task to consider the wide range of DSM disorders that could account for that symptom and cull the list down to the one or two disorders that are the most likely ‘contenders’ (First et al. Reference First, Frances and Pincus2004). Thus, the DSM-IV Mood Disorders grouping includes those disorders that should be considered by the clinician in the differential diagnosis of disordered mood (i.e. Major Depressive Disorder, Dysthymic Disorder, Bipolar Disorder, Cyclothymic Disorder, Mood Disorder due to a General Medical Condition, and Substance-induced Mood Disorder).

Leaving aside for the time being the issue of whether Andrews' proposed groupings are sufficiently empirically robust to serve as the basis for a major reorganization of the diagnostic groupings in DSM-V and ICD-11, how likely are these groupings to improve the clinical utility of the classification? Table 2 depicts the DSM-IV classification according to Andrews' proposed five-cluster meta-structure with the sixth grouping for those disorders that could not be assigned to any of the five clusters. One immediately evident problem is that the unassigned NOS-like grouping is much larger than any of the five clusters, containing 51% of the total number of disorders. The largest cluster in the meta-structure, the emotional disorders contains only 14% of the disorders, whereas the neurodevelopmental, neurocognitive, externalizing, and psychosis clusters contain 11%, 10%, 9%, and 5% of the disorders respectively. Andrews and colleagues (2009) claim that ‘the findings of the … six papers support a more parsimonious organization of the forthcoming classifications that could incorporate the shared risk and clinical characteristics for the majority of the DSM-IV and ICD-10 disorders’. In actuality, their proposed meta-structure can hardly be considered ‘parsimonious’ given that it leaves more than half of the disorders in the current classification ‘unassigned’ Furthermore, while it might be relatively straightforward for clinicians to figure out which disorders are contained in the neurodevelopmental, neurocognitive, and psychosis clusters because of their close resemblance to the DSM-III/DSM-IV groupings of ‘Specific and Pervasive Developmental Disorders’, ‘Delirium, Dementia, Amnestic, and Other Cognitive Disorders’ and ‘Schizophrenia and Other Psychotic Disorders’, discerning which disorders are contained in the ‘Emotional’ and ‘Externalizing’ clusters is likely to be more challenging. This is not simply because using these terms in this way would be unfamiliar to clinicians but because the descriptive names given to these clusters do little to convey that the underlying construct tying the disorders together is antecedent temperament. Another point of potential confusion that might have a negative impact on clinical utility relates to the variability in the organizational principles underlying the groupings. Some of the clusters, namely the neurocognitive and psychosis clusters would continue to be useful in facilitating differential diagnosis since, with the exception of superimposed delirium, disorders in these clusters are mutually exclusive. The opposite situation would apply to the emotional and externalizing clusters where co-morbidity within the cluster is one of the organizing principles so that rather than indicating those disorders that should be considered in the differential diagnosis, these clusters indicate disorders that are more likely to be co-morbid.

Table 2. DSM classification according to Andrews' meta-structure groupings

Although Andrews and colleagues (2009) acknowledge that ‘whether these clusters could be useful to the field is yet to be tested’, they do state that ‘a number of advantages could arguably occur from recognizing these clusters’. However, given the practical disadvantages of their proposed meta-structure, it is far from clear that the potential advantages they cite outweigh the costs or even whether their proposed meta-structure is the best method for actualizing their hoped-for benefits. In discussing the ‘clinical advantage of a cluster approach’, they note that ‘when a disorder belongs to a cluster, clinicians should treat the symptoms and ensure that the risk factors characteristic of that cluster are modified to reduce their potential impact’. Even if one were to assume that modification of risk factors is an achievable goal, would it not be simpler and more useful to describe the risk factors for each disorder in the DSM-V text rather than having the clinician look up the risk factors for the entire cluster, especially given the general lack of specificity of risk factors across the different clusters? Andrews and colleagues (2009) also claim that ‘identification of clusters may enhance clinical utility in primary care as well as in specialist psychiatric care … [by simplifying] an otherwise confusing system, and encourag[ing] clinicians, for instance, to assess anxious and depressive symptoms whenever they are faced with a patient with psychosomatic symptoms’. How do the authors think that simply grouping anxiety, depressive, and somatoform disorders together in the same section of the classification would achieve this goal in primary-care settings? A likely more effective method would be to design assessment tools for primary-care settings that routinely screened for mood, anxiety, and somatoform symptoms.

Perhaps the most important question raised by Andrews and colleagues at the beginning of the paper is ‘are there now sufficient data from neuroscience, genetics, epidemiology and therapeutics to identify groups of disorders […] identified by shared external validating factors rather than by symptom pictures alone?’ This question gets to the heart of the effort behind this paper. Both researchers (Charney et al. Reference Charney, Barlow, Botteron, Cohen, Goldman, Gur, Lin, Lopez, Meador-Woodruff, Moldin, Nestler, Watson, Zalcman, Kupfer, First and Regier2002; Kupfer et al. Reference Kupfer, First, Regier, Kupfer, First and Regier2002; Hyman, Reference Hyman2007) and clinicians (McHugh, Reference McHugh2005) have expressed substantial frustration with the superficial nature of the DSM-IV descriptive paradigm and with DSM-V on the horizon, there has been great interest in the possibility of making changes to the classification that reflect ‘the considerable advances in psychiatry since the publication of DSM-IV in 1994’. Although this set of papers demonstrates that information about shared risk factors can be used as the basis for regrouping disorders, are the findings robust enough to justify such a reorganization, especially given the likelihood that such groupings will be reified by both the clinical and research communities? The history of psychiatric research is replete with examples of promising findings that seem to herald major changes in our understanding of the etiology of disorders but that are unable to be consistently replicated (Risch et al. Reference Risch, Herrell, Lehner, Liang, Eaves, Hoh, Griem, Kovacs, Ott and Merikangas2009). Thus, it is important that the empirical data used as the basis for changes in the classification groupings be particularly solid and robust and likely to stand the test of time.

Examined under this light, the empirical foundation for these groupings seems rather shaky. First of all, the a priori decision to create five clusters, as opposed to having a fewer or greater number of clusters, seems arbitrary and appears to have been made entirely by the consensus of a small group of experts (i.e. ‘first we identified possible clusters and then examined the internal coherence of these clusters using external validating criteria’) rather than arising out of some empirically based methodology. Moreover, as noted by Andrews et al. (Reference Andrews, Goldberg, Krueger, Carpenter, Hyman, Sachdev and Pine2009), their meta-structure ‘is not based on systematic reviews. To do such a review for each disorder would have been a Herculean task even if appropriate data – disorder versus control versus all other disorder within cluster versus all disorders in other clusters – were available for all disorders. They are not.’ Furthermore, throughout the rest of the paper, there are numerous statements indicative of the many gaps in the empirical data base. For example, in their analyses of within-cluster membership in the ‘results’ section of Andrews and colleagues' paper, they repeatedly note that the limited data applied to only a subset of the disorders proposed for the grouping (e.g. for the neurocognitive cluster, ‘there is a paucity of data concerning delirium and the other cognitive disorders’, for the psychoses cluster, ‘there were little data for comparison among the psychoses grouped in DSM-IV-TR’, for the emotional cluster ‘there was a paucity of data concerning body dysmorphic disorder, adjustment disorder, and avoidant personality disorders and they are not further discussed’, etc.). Finally, the predominance of the large residual group of ‘disorders not yet assigned’ reflects the fact that there was ‘insufficient data on the Study Group criteria’.

Andrews and colleagues (2009) describe these clusters as emphasizing ‘the core features of the disorders, rather than emphasizing a detailed list of diagnostic criteria’. However, they skirt around the issue of what the core features actually are given the field's ignorance of the underlying etiology and pathophysiology of mental disorders. Although it is certainly credible that, as claimed by Andrews and colleagues (Andrews et al. Reference Andrews, Stewart, Morris-Yeates, Holt and Henderson1990, Reference Andrews, Slade and Issakidis2002) and Krueger (Reference Krueger1999) that that existence of high rates of comorbidity between groups of disorders suggests the possibility of ‘the action of some common etiological agent’ the precise nature of that ‘agent’ remains speculative at best. The absence of understanding about what these etiological factors might be is largely indicated by the fact that, with the exception of the term ‘neurodevelopmental’ which is at least vaguely indicative of a process, the names of the rest of the groupings (i.e. neurocognitive, psychoses, emotional, and externalizing) are descriptive rather than mechanistic names.

In summary, given the limitations and gaps in the empirical database informing our understanding of shared risk factors and clinical factors, the meager benefits afforded by adopting such a meta-structure as compared to the clear-cut costs in terms of loss of clinical utility and additional burden on the field, and the fact that more than half of the disorders in DSM-IV could not be assigned to one of the five clusters in the meta-structure, the authors have failed to establish that it is feasible at this point to regroup the disorders in DSM-V and ICD-11 ‘on the basis of features not confined to clinical picture’. Given the qualified language used throughout these papers [e.g. ‘the present set of papers is an exercise’, and ‘a classification based on the features of the DSM-V Task Force Study Group suggests the possibility of a classification based on etiological risk factors’ (emphasis added) ], one might wonder whether this meta-structure represents a serious proposal for DSM-V and ICD-11 or is merely a thought piece designed to stimulate further discussion. The conclusion of the paper suggests the former: Andrews and colleagues recommend that the DSM-V literature reviews, criteria revisions, and field trials be utilized to ‘fine tune the clusters and the disorders included in them’. A careful balancing of the benefits and costs suggests that adopting a meta-structure along the lines proposed by the authors should be deferred until we have a significantly deeper understanding about the pathophysiology and etiology of mental disorders.