Would the use of dimensional measures improve the utility of psychiatric diagnoses?

Accepting A. Jablensky's1 assessment that establishing the validity of either discrete categorical psychiatric diagnoses or dimensional measures of clusters of psychiatric symptoms is a very long‐term (or, possibly, unachievable) objective, what should nosologists and diagnosticians be doing to improve the utility of the categories we are currently using?

One issue debated at length during the deliberations for DSM‐5 was the use of dimensional measures to supplement the standard categorical diagnoses (e.g., for schizophrenia) or, possibly, as a replacement for the categorical diagnoses (e.g., personality disorders). In the end, the final version of DSM‐5 retained the categorical diagnostic structure of previous classifications, largely relegating the dimensional measures to the Emerging Measures and Models section (Section III) of the volume.

But there is continuing debate about the potential clinical utility of converting the current categorical diagnostic system to a dimensional system of classification which would be closer to the observed continuous nature of the severity, duration, and disability associated with psychiatric symptoms2, 3. To achieve this long‐term goal, the Research Domain Criteria (RDoC) project of the U.S. National Institute of Mental Health specifically aims to selectively fund research that will replace current psychiatric diagnostic systems based on descriptive phenomenology with “new ways of classifying mental disorders based on dimensions of observable behaviour and neurobiological measures”3, 4.

Theoretically, dimensional measures could either be used to directly determine different diagnoses within a new dimensional classification network, or as adjunctive measures to classify distinct subtypes of the psychiatric disorders in current categorical diagnostic systems (ICD or DSM). If dimensional measures could help to identify distinct clusters of symptoms with different clinical courses and responses to treatment, the use of such measures could increase the utility of diagnostic classifications. But is it realistic to think that they can be used in this way?

There are several problems with using dimensional scores to directly assign diagnoses. Many currently available dimensional measures are highly correlated, so to achieve the goal of a diagnostic system with improved utility, current dimensional measures would either need to be substantially revised or diagnoses would need to be defined as specific patterns of dimensional scores.

Neither of these tasks is simple. Using a cut‐off score of a dimensional measure to assign a diagnosis would collapse that measure into the traditional dichotomized diagnostic labels – the main problem dimensional measures are supposed to resolve. Moreover, the scores of most dimensional measures change frequently either in response to treatment or as part of the natural course of the condition, so diagnoses based on dimensional scores would need to change frequently.

Let's assume that it is possible: a) to develop relatively independent dimensional measures, b) to identify points (or ranges) of rarity on the continuous dimensional measures that would justify specifying a score above which symptoms are to be considered “diagnostically relevant”, and c) to indicate the time(s) in the course of the condition when the dimensional scores would be used to determine a diagnosis (e.g., prior to initiating treatment). Even in that case, assigning diagnoses for all dimensions for which the dimensional score exceeds a specified diagnostically relevant level would result in an unmanageable number of diagnostic categories. Assuming only 10 symptom/diagnostic dimensions, there would be 10 single‐dimension disorders, 45 dual‐dimension disorders, 360 triple‐dimension disorders, 2520 quadruple‐dimension disorders, and so forth.

Many of the cells in this matrix of dimension‐based diagnoses would be empty, but determining the course and preferred treatment for each of the large number of cells with a substantial number of cases would require studies several orders of magnitude larger than the largest current studies. For individuals with diagnostically relevant scores in more than one dimension, there would also be the difficulty of prioritizing the various conditions and deciding whether to administer relevant treatments simultaneously or sequentially.

The use of dimensional scores to classify subtypes of current criteria‐based categorical diagnoses would require resolving several additional problems. Would there be a universal set of dimensions used for all patients, a menu of dimensions among which a specific subgroup would be used for each diagnosis, or diagnosis‐specific dimensions? Would the diagnostic subtypes change as the dimensional scores change? And for diagnoses that consider four or more symptom‐based dimensions, the potential number of subtypes (based on the number of dimensions for which the score exceeds a diagnostically relevant level) would be unmanageable. The complex diagnostic algorithms needed to address these issues would likely make the diagnostic criteria unusable in routine clinical practice.

One other potential use of dimensional measures would be to directly determine treatment. If treatments in psychiatry were immediately effective (like antihypertensives) and available treatments were uniquely targeted on specific symptom clusters, then it would be reasonable to regularly change treatments based on patients’ current symptom profiles as assessed by narrowly defined dimensional measures. But, for the foreseeable future, neither of these conditions is satisfied, so dimensional measures can only be used to assess the effectiveness of different treatment strategies, not to select specific treatment strategies.

The use of dimensional measures of symptom severity to monitor changes in symptoms over time is clinically useful, because they provide a reasonably accurate assessment of the current clinical state of patients and of the effectiveness of specific treatments. But these measures are of limited use for predicting the course of a condition or for predicting the likely effectiveness of specific treatments – essential roles they would need to play if they were to be used to classify diagnoses or subtypes of categorical diagnoses.

The RDoC initiative to identify diagnostically relevant dimensions is intellectually appealing to clinicians and researchers who are frustrated by the inability of categorical diagnostic systems to reflect the complex reality they see every time they interact with patients. But current dimensional measures of behavioural, emotional or neurobiological processes (and the new dimensional measures that will emerge from the RDoC initiative) are correlated with each other and variable over time5. Their use will neither improve the validity nor the utility of the diagnoses we employ.

Michael R. Phillips
Shanghai Mental Health Center, Shanghai Jiao Tong University School of Medicine, Shanghai, China; Departments of Psychiatry and Global Health, Emory University, Atlanta, GA, USA