Categories vs Dimensions: Wave or Particle?

CONCEPTS IN PSYCHIATRY

There is a common view among critics of psychiatric diagnosis—especially the DSM system—that its central flaw is being categorical, and that the solution is to make it dimensional. This is the premise behind the Hierarchical Taxonomy of Psychopathology (HiTOP) alternative and similar proposals: that DSM has carved psychiatric illness into some 400 discrete categories when, in reality, psychopathology consists of only a handful of underlying dimensions.

The perspective offered here is different. The problem with DSM is not that it is categorical. The problem is that it is unscientific. Science, properly applied, might support a categorical diagnosis, or it might support a dimensional one—but this cannot be presumed in advance, on either side.

Categories and Dimensions Both Exist in Medicine

Height, weight, blood pressure, and cholesterol are dimensional. So are personality traits. This is not an anti-categorical stance; it is simply an empirical fact about how these variables are distributed in nature. On the other hand, Tay-Sachs disease and Down syndrome are categorical—also simply a matter of fact, not theoretical preference.

It is true that classic categorical conditions, such as Mendelian genetic diseases, are less common than dimensional medical problems like chronic pain, glycemic control in diabetes, or the severity of coronary artery disease. But categorical diseases clearly exist in medicine, which means there is no a priori reason psychiatric disease must be noncategorical.

Schizophrenia and Manic-Depressive Illness as Categorical Diseases

Two major psychiatric illnesses—schizophrenia and manic-depressive illness (bipolar disorder)—are, based on twin-study heritability research, almost entirely genetic in origin. These are genetic, categorical diseases. That does not mean every patient with either condition looks identical. Nor does it mean that the diseases themselves do not vary in severity, with some patients having milder and others more severe forms. No genotype produces a phenotype with 100% uniformity—even Down syndrome, unambiguously caused by trisomy 21, shows some phenotypic variation.

Biological variability in phenotype is simply a fact of nature, present in disease and nondisease states alike. This variability does not, by itself, make a condition dimensional rather than categorical. The real question is whether that normal variability overlaps extensively with other conditions, or only marginally.

Please Stop Saying “Carving Nature at its Joints”

A common phrase, repeated ad nauseum, is that we do not know how to “carve nature at its joints.” Those who repeat this tired metaphor, we are sure, have no idea where it comes from.Do you? Pause…..It is from Plato (1997/ca. 370 BCE),¹ where he is discussing the nature of knowledge. His analogy, which also is found in Chinese philosophy, is that the knower is like a butcher, who knows where to carve the animal, at its joints, as opposed to just cutting anywhere.

Readers and fellow psychiatric clinicians, there is not only one metaphor in the world. Metaphors are useful for thought, but they also constrain thought. If you only have one way of thinking about a topic, and you come to some conclusion, your judgment may not be true, simply because you were using the wrong metaphor, or you were not thinking about the topic in enough different ways so that you could understand it better.

Let’s give up the butcher metaphor please. It is not the only way to think about psychopathology. Philosophers use the term “thought experiment” for the idea that you should look at a problem from different angles, different ways of thinking, not just one. Plato was a great philosopher, but he was not the only one.

Here’s a different thought experiment.

The Overlap Fallacy

A common mistake among critics of categorical diagnosis is to treat any degree of phenotypic overlap between 2 conditions as proof that they represent a single dimension. If Disease A overlaps with Disease B by 0.1%, and does not overlap on the other 99.9%, that is still overwhelmingly 2 distinct conditions—not 1 dimension. The same logic holds at 1%, 5%, or 10% overlap: none of these approach the 90% or greater overlap that would suggest true dimensional unity.

Critics often assume that because diagnostic boundaries are fuzzy, diagnostic categories are therefore invalid. But as Roth and Kroll (1986) observed, “The observation that a concept has indefinite boundaries... merely reflects the absence of sharp lines of demarcation in nature. Furthermore, a concept which has uncertain application in borderline cases may have a wide range of cases for which application is entirely beyond doubt.”² This is the exact mistake made when overlap between conditions is taken as evidence that categories do not exist.

Consider a medical analogy: gastric ulcers and coronary artery disease overlap in producing chest or epigastric pain. No one concludes from this that ulcer disease and heart disease are 1 dimension.

The same reasoning applies to schizophrenia and manic-depressive illness, which overlap in roughly one-third of associated genes but diverge on the other two-thirds. They overlap in some symptoms but diverge in many others: all patients with schizophrenia have psychosis, while roughly half of patients with bipolar illness never experience psychotic symptoms at all—not to mention the further differences between manic and depressive presentations. To claim that schizophrenia and manic-depressive illness represent a single “psychotic dimension” ignores the fact that their overlap, however real, involves a minority of traits rather than the majority.

More or Less, Not Either/Or

Categorical and dimensional models should not be treated as an all-or-nothing choice. The relevant question is whether the majority of traits and features studied show high overlap between putative categories (favoring a dimensional model), or whether overlap is present but limited, with most traits diverging (favoring a categorical model, despite some dimensionality at the margins).

This is a matter of degree, not an either/or proposition—but that does not make it dimensional by default. As Wittgenstein (1958) famously remarked, “A fuzzy beam of light is just as real as a sharply focused one.”³ Diagnostic boundaries need not be perfectly sharp to correspond to real disease entities. If 2 conditions fail to overlap in 70%, 80%, 90% or more of their traits, they are, for practical and biological purposes, different diseases with some shared features that may be biologically or physiologically incidental to the core distinction. Ulcer disease and coronary artery disease remain two different diseases despite their shared symptom of chest pain.

A Physics Analogy

Light can be understood as a particle or a wave, depending on the lens applied. Categorical medical diseases are similarly distinct categories when viewed one way, yet reveal dimensionality when examined through genetic or phenotypic variation. Both descriptions can be true depending on the level of analysis—but this does not mean everything is always dimensional, nor does it mean disease categories are always absolute. The eclectic, postmodern conclusion that conditions are equally both, all the time, is equally mistaken.

The governing rule is this: if the majority of evidence supports a categorical distinction, the condition is a category, regardless of some residual dimensionality. If the majority of evidence supports dimensionality, the condition is a dimension.

Applying This to Psychiatric Diagnosis

The diagnostic question for psychiatry, then, is not whether we should prefer categories or dimensions as a matter of ideology. It is what the scientific research actually supports for a given condition.

Paul McHugh and Phillip Slavney, at Johns Hopkins, proposed a version of this view: psychiatry deals with diseases (categories) in some cases and dimensions in others.⁴ The evidence to date tends to support disease models—categorical models—for conditions like schizophrenia and manic-depressive illness, and dimensional models for personality traits. This has direct implications for current nosology. In our view, borderline personality disorder and antisocial personality disorder are the principal personality disorders for which the available evidence supports retention as categorical syndromes, whereas the remaining personality disorders are more appropriately dimensionalized.

Such an approach argues against retaining invalid personality disorder categories (such as narcissistic and dependent), and against dimensional models of psychotic illness.

The principal limitation of HiTOP and similar dimensional systems is therefore methodological rather than merely taxonomic. By focusing almost exclusively on patterns of symptom covariance, it privileges a single source of evidence while largely setting aside the other scientific validators that have traditionally guided psychiatric diagnosis, including course, genetics, biomarkers, and characteristic phenomenology.⁵ These validators may support dimensional models for some forms of psychopathology, but they may equally support the existence of distinct disease entities for others.

Dimensional approaches also begin from the implicit assumption that psychopathology is fundamentally dimensional, rather than allowing the evidence to determine whether a given condition is best understood as a dimension or a category. Scientific classification should not begin with theoretical commitments about the structure of mental illness. It should begin with the totality of the empirical evidence and allow that evidence to determine whether a condition is categorical or dimensional.

The choice between categories and dimensions is therefore not a theoretical one but an empirical question that can only be answered by science, ie, experimentally or observationally obtained data.

Dr Ghaemi is a lecturer on psychiatry at Harvard Medical School, Cambridge Health Alliance, and is employed by Bristol Myers Squibb. The views expressed in this article are solely those of the authors and do not necessarily reflect the official policy or position of their employers.

Dr Ruffalo is an assistant professor of psychiatry at the University of Central Florida College of Medicine in Orlando, where he is director of psychotherapy training in the adult psychiatry residency program. He is also an adjunct assistant professor of psychiatry at Tufts University School of Medicine in Boston, Massachusetts, and on the faculty of The New Jersey Institute for Training in Psychoanalysis in Teaneck, New Jersey.

References

1. Plato. Phaedrus (A. Nehamas & P. Woodruff, Trans.). In: Cooper JM, ed. Plato: Complete works. Hackett Publishing; 1997:506-556.

2. Roth M, Kroll J. The Reality of Mental Illness. Cambridge University Press; 1986.

3. Wittgenstein L. Philosophical Investigations. 3rd ed., G. E. M. Anscombe, Trans. Blackwell; 1958.

4. McHugh PR, Slavney PR. The Perspectives of Psychiatry. Johns Hopkins University Press; 1983.

5. Robins E, Guze SB. Establishment of diagnostic validity in psychiatric illness: Its application to schizophrenia. Am J Psychiatry. 1970;126(7):983-987.