Important Questions Remain to be Addressed Before Adopting a Dimensional Classification of Mental Disorders

Widiger and Trull (February–March 2007) raise important nosological issues that warrant serious consideration not only for the personality disorders, but for all mental disorders, as the Diagnostic and Statistical Manual of Mental Disorders (DSM) is revised during the next few years. As argued compellingly by the authors, dimensional classification may indeed offer substantial improvement over the present categorical system. Several questions remain to be addressed in evaluating whether this is the case and, if so, how dimensional classification can be implemented to best serve the disparate users of the DSM.

Adopting a conceptualization of personality disorders as continuous with normal personality traits would represent a step forward only if these disorders really are dimensional in nature. Despite strong historical preferences for dimensional models by psychologists and categorical models by psychiatrists, research into whether mental disorders represent discrete clinical entities has only recently begun in earnest. Results emerging from this research suggest that some disorders may best be understood as categories whereas others are better conceived as dimensions (see Ruscio, Haslam, & Ruscio, 2006, for a review). Such results underscore the importance of determining empirically the most valid representation for each disorder (Meehl, 1986; Ruscio & Ruscio, 2004). They also hint that an all-categorical or all-dimensional classification may ultimately prove to be less valid than one incorporating both categorical and dimensional elements.

For disorders that are truly dimensional, several questions will need to be considered to ensure that dimensional classification will be workable in practice:

Feasibility

Can the proposed system be implemented by practitioners in the limited time available for assessment and diagnosis? Clinicians do not routinely assess all symptoms of the 10 DSM-IV personality disorders. More often, they assess the (far smaller) subset of symptoms for disorders which seem likely, given the clinical picture. Consequently, clinician-administered personality measures and lengthy patient questionnaires may add significantly to current assessment burden. This may limit feasibility for some clinicians as well as acceptability for some clients, who may not see these assessments as relevant to the problems that brought them in for treatment. Evaluation of feasibility will be crucial to ensure that any increase in assessment burden does not have the unintended consequence of reducing attention to personality pathology.

Reliability

Can the system be applied reliably by users? An extensive literature suggests that human judges can simultaneously combine and weigh about a half-dozen variables reliably and validly (Cooksey, 1996). The task is even more challenging when the variables are correlated or must be combined in complex ways. It is noteworthy that all of the dimensional models described by Widiger and Trull include many more variables than this. In developing a dimensional classification, it seems important to consider known limits of human cognitive capacity in addition to content coverage. This includes showing that the multifactorial profiles can be interpreted consistently by diagnosticians, as well as considering ways of minimizing the number of elements that must be combined (as in the two- and three-point codes of the MMPI-2; Graham, 2000).

Communication

Does the system facilitate effective, efficient communication—not only with clients, but with other treatment providers, administrators, insurance companies, and policy makers? For all their shortcomings, categorical diagnostic labels (“borderline personality disorder”) efficiently convey a great deal of information and simplify record keeping. These shorthand labels are doubtlessly imprecise, especially for disorders with substantial intracategory heterogeneity. Nevertheless, they offer advantages that a more precise—but more cumbersome—system may have difficulty replicating. An important challenge for any dimensional system will be to balance precision with efficiency in communication.

Thresholds

Does the system incorporate defensible diagnostic thresholds? Widiger and Trull call persuasively for a thoughtful reevaluation of current thresholds. Establishing well-justified, empirically-based disorder thresholds is an important priority, whether the goal is to set the boundaries of discrete disorders or to locate appropriate cutoffs along dimensions for clinical, research, or administrative decisions. Of course, reasonable people may disagree about how and where thresholds should be located. Basing thresholds on the extent of functional impairment will be opposed by some who argue that impairment should not be used in defining mental disorders (Lehman et al., 2002) and disputed by others who disagree over the amount of impairment needed for diagnosis. In the meantime, judgments about insurance coverage, disability compensation, and other unavoidable decisions will need to be made. Defensible thresholds for these sorts of decisions may need to be established before dimensional classification replaces categorical classification in clinical settings.

Research

Does the system facilitate the accumulation of knowledge? As noted by the authors, a valuable benefit of dimensional classification is the ability to set different cutoffs for different clinical decisions in different settings. One downside to this flexibility, however, may be a reduction in communication and data-sharing across settings that employ different cutoffs for the same disorder. Similar problems may ensue if different researchers use different diagnostic cutoffs for inclusion in clinical studies. An important priority is to ensure sufficient standardization of cutoffs to allow knowledge to amass across clinics and laboratories.

Prediction

Does the system powerfully predict important outcomes? The ultimate test of a diagnosis may be its ability to facilitate prediction of prognosis and treatment outcome (cf. Hayes, Nelson, & Jarrett, 1987). The additional information inherent in a dimensional score suggests that prediction should be improved with a dimensional model. However, if a disorder really is categorical in nature, dimensional variation around the categorical boundary may add nothing but measurement error (Ruscio & Ruscio, 2002). Consequently, whether a categorical or dimensional (or hybrid) model optimizes clinical prediction is an empirical question. This question is best answered through side-by-side comparisons of the predictive validity of competing models with respect to a range of significant external variables.

Widiger and Trull are to be commended for raising awareness of this important debate and working to resolve obstacles that have prevented the adoption of dimensional classification in prior DSM editions. It is incumbent upon the field to address the challenging questions that remain in order to ensure that future classification systems are increasingly valid and useful.