Internalizing disorders: the whole is greater than the sum of the parts

The Hierarchical Taxonomy of Psychopathology (HiTOP) consortium is a group of investigators working to advance the empirical classification of psychopathology. In a previous issue of this journal they published a concise account of the work of their consortium1, and now they put forward a statement of intent and a summary of progress2.

Practitioners in the mental health field act as though each mental disorder is a discrete category – Mrs. Smith has panic disorder; Mr. Brown has major depressive disorder – and consider that treatment and future developments will naturally follow from the diagnosis. At one level this is appropriate and necessary for the orderly management of treatment for individual patients, but at a higher level this is not correct: the defining symptoms of each mental disorder exist on dimensions that extend from very mild and incomplete sets consistent with wellness to the very severe, complete sets that disable and distress and are incompatible with being well.

The classifications of mental disorders – DSM‐5 and ICD‐10 – are, at the simplest level, definitions of the threshold at which a set of symptoms becomes sufficiently complete, disabling or distressing to be of clinical concern, and an indicator of the need for treatment. The point on a dimension of increasing severity where a diagnosis is warranted is not indicated by any external measure such as a sudden change in pathophysiology or of distress or disability. The threshold for a diagnosis in each classification is made by experts convened to define it and hence is somewhat arbitrary. There is broad consensus that mental disorders exist on dimensions, not categories, and in 2008 two members who would later join the HiTOP consortium convened a meeting and edited a seminal book, Dimensional Approaches in Diagnostic Classification, as part of work on refining the research agenda for DSM‐53.

Multivariate research has indicated that a latent general liability – internalizing – accounts for higher‐than‐chance levels of mood and anxiety disorder comorbidity, a finding that has been replicated and extended many times in different data sets and cultures (note that half of people who meet criteria for an anxiety or depressive disorder have a second diagnosis, and a quarter meet criteria for three or more).

For example, within the HiTOP consortium, Eaton et al4 modelled seven internalizing disorders in a nationally representative sample of 43,093 individuals. The study used a structured diagnostic interview optimized to cover the DSM‐IV defining characteristics of these disorders. They found that a two‐dimensional (distress‐fear) liability structure for internalizing fit best and replicated across gender, assessment waves, and lifetime and 12‐month diagnoses. These internalizing liabilities, not the individual disorders, predicted future internalizing pathology, suicide attempts, angina, and ulcer.

Waszczuk et al5 conducted a study based on the Interview for Mood and Anxiety Symptoms that assessed, without the usual skip outs, DSM‐IV and ICD‐10 emotional disorder symptoms and other manifestations of emotional disorders such as hopelessness, desperation, loss of libido, social withdrawal, and self‐harm. In a series of analyses that ranged from symptom components to latent structures, they reported that dimensional components are better predictors of functioning than categorical DSM‐IV diagnoses, even though impairment is explicitly included in clinical diagnoses but is not part of those symptom components.

There are two implications from this body of work. First, that considering groups of disorders may be more informative than considering individual diagnoses. Second, that opening up research to include symptoms not presently included in classifications may point to new disorders or new arrangements of existing disorders and reduce the circularity of reanalyzing data from interviews designed to inform existing classifications.

There has been other work on classification independent of the HiTOP consortium that is relevant to the current Forum. As part of the work for DSM‐5 and ICD‐11, a working group6, including two members who would later join the HiTOP consortium, explored the feasibility of a meta‐structure based on eleven validating criteria comprising both clinical features and risk factors (i.e., shared genetic risk factors; familiarity; shared specific environmental risk factors; shared neural substrates; shared biomarkers; shared temperamental antecedents; shared abnormalities of cognitive or emotional processing; symptom similarity; high rates of comorbidity; course of illness; treatment response). DSM‐IV disorders were allocated to one of five clusters as a starting premise. Teams of experts then reviewed the literature to determine within‐cluster similarities on the eleven predetermined validating criteria and discovered that those similarities were consistently greater than between‐cluster similarities.

The five clusters were neurocognitive (identified principally by neural substrate abnormalities), neurodevelopmental (identified principally by early and continuing cognitive deficits), psychosis (identified principally by clinical features and biomarkers for information processing deficits), emotional/internalizing (identified principally by the temperamental antecedent of negative emotionality), and externalizing (identified principally by the temperamental antecedent of disinhibition). The working group considered that there could be advantages for clinical practice, public administration and principally from the adoption of such an organizing principle. The chapter order in DSM‐5 was changed to reflect this.

Computerized cognitive behavioural therapy (CBT) has a long history of focussing on the internalizing disorders as a group. Newby et al7 identified seventeen randomized controlled trials. Results showed that “transdiagnostic” computerized CBT outperformed control conditions on all outcome measures at post‐treatment, with large effect sizes for depression (g=0.84), and medium effect sizes for anxiety (g=0.78) and quality of life (g=0.48), comparable to the benefits seen in diagnosis specific studies8.

Lastly, and again using “transdiagnostic” computerized CBT, Mewton et al9 assessed changes in the internalizing construct using a longitudinal latent trait framework that compared internalizing factor means at pre‐ and post‐treatment. The standardized mean reduction in the internalizing construct with treatment was large (effect size 1.23, SE=0.09, p<0.001).

We conclude that treatment aimed at the internalizing construct is to be preferred to disorder specific treatment. In the internalizing disorders, whether one is investigating prognosis, impairment or response to treatment, the whole is greater than the sum of the parts.