Article Text

PDF

Cluster’s last stand
  1. S C Howell1,
  2. G D Eslick2,
  3. N J Talley3
  1. 1School of Public Health, The University of Sydney, Nepean Hospital, Penrith, New South Wales, Australia
  2. 2Department of Medicine, The University of Sydney, Nepean Hospital, Penrith, New South Wales, Australia
  3. 3Center for Enteric Neurosciences and Translational Epidemiological Research (CENTER), Mayo Clinic, Rochester, Minnesota, USA
  1. Correspondence to:
    Mr S C Howell
    School of Public Health, The University of Sydney, Nepean Hospital, Level 5, South Block, PO Box 63, Penrith, NSW 2751, Australia; HowellSwahs.nsw.gov.au

Statistics from Altmetric.com

Guthrie and colleagues (Gut 2003;52:1616–22) described the results of cluster analysis in a patient sample with severe irritable bowel syndrome (IBS). Their analysis investigated a broad range of factors in addition to symptoms; these included psychosocial measures (psychiatric involvement, health service encounters, quality of life) and physiological parameters (rectal thresholds). The authors have demonstrated that severe IBS can be classified according to non-symptom characteristics and, in particular, according to the level of psychological distress, service encounters, and rectal sensitivity. They describe three groups which they labelled “distressed high utilisers,” “distressed low utilisers”, and “tolerant low utilisers.” The authors defend their analysis on clinical grounds and point to treatment implications for each of these groups.

We feel that there are some fundamental points about the nature of cluster analysis that readers of this paper should not overlook. Cluster analysis was initially developed to create and/or evaluate classifications.1 Its application to gastrointestinal research has followed this approach. In recent years, clustering techniques have been applied to confirm that IBS and functional dyspepsia exist as separate clinical entities, and to evaluate specific syndrome subgroups, as described by the current Rome criteria.2–4 Following traditional clinical approaches, cluster solutions have generally been derived from symptom based parameters, including frequency, severity, and predominant complaint.

The term “cluster analysis” describes a range of procedures which use empirical methods to form groups of highly similar entities. While the notion that cluster analysis is solution seeking, operation of these techniques is essentially solution imposing; that is, clustering methods will always place objects into groups. Furthermore, as there are no formal statistical procedures to evaluate the resulting solution, the reasonableness of any solution is determined only on the basis of personal judgement. This is a problem. Indeed, critics of the approach have argued that cluster analysis encourages “naive empiricism”—that is, inclusion of as many variables as possible in the hope that a meaningful structure will come out.1,5 However, proponents of cluster analysis have suggested that careful selection of variables on theoretical grounds can overcome this limitation.1

It is intuitively obvious that any single entity can be classified according to a broad range of dimensions, and Guthrie et al have certainly demonstrated this with respect to IBS. However, we rarely classify any object or entity according to all possible dimensions simultaneously; this would lead to a complex set of descriptors which may be unwieldy and contain many redundancies. Rather, we tend to select out a subset of meaningful dimensions that best suit our purposes in forming a classification.

There are certainly theoretical grounds for considering psychological involvement when evaluating patients with IBS. The association of IBS with psychological disturbance and health care seeking has been well described, and our understanding of these factors has contributed greatly to current therapeutic approaches. However, we challenge the proposed classification of IBS according to psychological involvement on two grounds. Firstly, we view this as a step towards naïve empiricism; other researchers may be encouraged to replicate these analyses across a broader and even more diverse (yet irrelevant) range of dimensions. This is not likely to produce a parsimonious classification scheme that is useful in either clinical practice or the research setting. Secondly, classification of IBS according to psychological and/or psychiatric involvement may stigmatise some patients with this complaint; one of the unfortunate consequences of classification is the tendency to attach labels to the subgroups that emerge.

References

View Abstract

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.