Categorical versus dimensional approaches to diagnosis: methodological challenges
Introduction
The argument pitting categorical versus dimensional approaches to psychiatric diagnosis have been long ongoing with little sign of imminent resolution (Donner and Eliasziw, 1994, MacCallum et al., 2002). It was proposed that the Diagnostic and Statistical Manual of mental disorders classification, 4th edition (DSM-IV) be organized following a dimensional model rather than the categorical model used in DSM-III-Revised. A dimensional system classifies clinical presentations based on quantification of attributes rather than the assignment to categories and works best in describing phenomena that are distributed continuously and that do not have clear boundaries. Although dimensional systems increase reliability, others suggest that such an approach may be less useful than categorical systems for clinical practice and application (APA, 1994). In reality, many DSM-IV criteria still follow a categorical model. An example of a categorical approach to the diagnosis of a disorder would be the DSM-IV criteria for Major Depressive Disorder (APA, 1994), while the Hamilton depression scale (HAM-D; Hamilton, 1960), would represent a dimensional approach. To a great extent, the issue of which approach is best seems engendered by the belief that one or the other is right. In fact, advocates of both approaches may well be right, but in different circumstances. This paper tries to clarify the conditions under which one should utilize one approach over the other, to demonstrate how these two approaches necessarily interdigitate. We will argue that failure to utilize the correct approach under the appropriate circumstances can result in impaired clinical and research decision-making. Clearly a discussion of the methodological and statistical issues related to this problem cannot resolve the argument for any specific psychiatric diagnosis. What is intended here is to alert psychiatric clinicians and researchers to issues in the methodology literature that might facilitate their considerations.
Section snippets
Definitions and background
It is necessary first to establish terminology and to recall basic principles. For the purposes of this discussion a disorder will be defined generically as something actually wrong with a subject (Kraemer, 1992). It might be a disease, a condition, an abnormality, a sign, a symptom, a syndrome, etc. A diagnosis of a disorder, on the other hand, indicates an expert opinion that the specific disorder is to some extent present in a individual.
What links the two is the quality of a diagnosis
The fundamental equivalence of categorical and dimensional approaches
Categorical and dimensional approaches are fundamentally equivalent in that any categorical approach can be converted to a dimensional one, and vice versa. This is important for underscoring how these approaches interdigitate. Every dimensional D can be converted to a categorical diagnosis, simply by setting some cutpoint, d*. If D⩾d*, the categorical diagnosis is D+, and if D<d*, D−. A categorical diagnosis derived from the Hamilton Depression Score (HAM-D) score might be presence of
Researchers prefer dimensional approaches for hypothesis testing
It has long been known that when faced with a choice between an ordinal measure (here a dimensional diagnosis) and a dichotomization of that measure (here the corresponding categorical diagnosis), power of hypothesis testing is virtually always sacrificed in using the categorical diagnosis (Cohen, 1983, Veiel, 1988, MacCallum et al., 2002). How much power is sacrificed varies according to the chosen cutpoint. Even worse, conflicting research conclusions may be drawn from the same data depending
Clinical researchers must occasionally use categorical approaches
Testing null hypotheses for random differences between groups is far from the only thing researchers do. Thus, no matter how adamant a researcher might be about using dimensional diagnoses as outcome measures, researchers have to make decisions that require them to use a categorical approach to diagnosis. For example, in an RCT, one must decide whom to include or exclude from a study. Ethically, one cannot impose treatment for a disorder on someone who does not have that disorder, particularly
Clinicians and clinical researchers generally require a categorical approach
Whenever a diagnosis is to be used to make a decision about an individual, a categorical diagnosis is necessary. Clinicians who must decide whether to treat or not treat a patient, to hospitalize or not, to treat a patient with drug or with psychotherapy, to use this drug or that drug, this type of psychotherapy or that type, must inevitably use a categorical approach to diagnosis. The problem is not whether to use a categorical approach or not, but rather which categorical approach to use. In
Monitoring patient response—longitudinal studies
To this point, we have focussed on the choice of cutpoint (diagnoses) meant to distinguish those with and without a disorder. However, the choice of cutpoint is also very important when monitoring individual patients over time, to see, study or to deal with, onset, episodes, remission, recurrence, relapse, etc. This represents an important aspect of both research and clinical applications (Frank et al., 1991).
In Fig. 2. a hypothetical patient is followed continuously over time using a
Finding the optimal categorical approach: ROC analysis
In Fig. 1, the choice of the cutpoint to maximize power dictated one choice of cutpoint as optimal; in Table 1, another; in Fig. 2 yet others. The logic here is simple but poses a major challenge for addressing specific research or clinical diagnostic questions. The choice of optimal cutpoint of a dimensional diagnosis depends completely on the purpose to which that categorical diagnosis is to be put. If it is to differentiate those in the general adult population who are impaired by the
The logical circularity or the logical spiral of diagnosis?
In 1989, Kupfer and Thase commented on a certain “logical circularity” that exists in the process of development and validation of a diagnosis. A clinical (categorical) diagnosis of a disorder is used to validate a dimensional diagnosis of that disorder to be used in research to shed light on the disorder of interest. Such research produces insights that can then be used to improve the quality (reliability, validity) of dimensional diagnosis, and to identify optimal cutpoints for different
Summary and conclusions
All diagnoses can be done using a categorical approach or using a dimensional approach, and what is more, any specific categorical classification can be converted to a dimensional one and vice versa. Neither approach, in and of itself, is better than the other. However, as we have illustrated in this paper, it is very important to recognize that for certain tasks a categorical approach is best and for others a dimensional approach is best. When hypotheses are being tested, dimensional
References (17)
- et al.
Random assignment in clinical trialsissues in planning (Infant Health and Development Program)
Journal of Clinical Epidemiology
(1990) Diagnostic and statistical manual of mental disorders
(1983)The cost of dichotomization
Applied Psychological Measurement
(1983)- et al.
Statistical implications of the choice between a dichotomous or continuous trait in studies of interobserver agreement
Biometric
(1994) - et al.
Conceptualization and rationale for consensus definitions of terms in major depressive disorder: remission, recovery, relapse, and recurrence
Archives of General Psychiatry
(1991) - et al.
Helping low birth weight, premature babies
(1997) A rating scale for depression
Journal of Neurology, Neurosurgery, and Psychiatry
(1960)Infant health and development program: enhancing the outcomes of low birth weight, premature infants: a multisite randomized trial
Journal of the American Medical Association
(1990)
Cited by (219)
Transdiagnostic Fear and Anxiety: Prospective Prediction Using the No-Threat, Predictable Threat, and Unpredictable Threat Task
2023, Biological Psychiatry Global Open ScienceFactor analysis as a tool to explore the heterogeneity of type 2 diabetes: A feasibility study
2022, Diabetes Research and Clinical PracticeCitation Excerpt :An alternative view on heterogeneity could be provided by a dimensional, rather than categorial, approach. Clinicians generally prefer to classify patients into clearly defined diagnostic categories, facilitating clinical decisions [5]. On the other hand, many of the diagnostic categories used in clinical practice are the result of the application of a questionable threshold on a continuous variable (e.g., hypercholesterolemia, hypertension, obesity, or diabetes).
“Clinically significant distress” in internet gaming disorder: An individual participant meta-analysis
2022, Computers in Human BehaviorIntroduction to the hierarchical taxonomy of psychopathology
2022, Encephale