Abstract
Current psychiatric classification adopts a disorder-focused diagnostic approach, as exemplified within ICD-11 and DSM-V. Although this approach has improved reliability of categorization, its validity and utility has been questioned (Harvey, Watkins, Mansell, & Shafran, 2004; Insel et al., 2009; Sanislow et al., 2010). Limitations include high comorbidity between supposedly distinct disorders; heterogeneity within diagnoses; limited treatment efficacy; and similarities across disorders in aetiology, latent symptom structure, and underlying biology. There is also evidence of transdiagnostic cognitive-behavioural processes (Harvey et al., 2004). An alternative approach is therefore to focus on fundamental underlying mechanisms of psychopathology rather than observed symptom clusters. This article considers the possible benefits, hurdles, and steps towards implementation of this transdiagnostic mechanistic approach, using the example of repetitive negative thought.
A central question for psychiatry and clinical psychology is how we best conceptualize and classify emotional disorders to optimize our understanding and treatments. Attempts to better classify psychiatric disorders underpinned the development of diagnostic approaches, such as the Diagnostic and Statistical Manual (DSM; e.g., Spitzer, Williams, Gibbon, & First, 1992) and the International Classification of Disorders (ICD). I consider the advantages and disadvantages of this disorder-focused diagnostic approach, before suggesting an alternative “mechanism-focused” approach. I review psychological mechanisms research, which indicates that there are common processes involved in both normal and pathological functioning (i.e., a dimensional approach) and that there are shared mechanisms across multiple disorders (i.e., “transdiagnostic” mechanisms). Together these observations argue for the value of focusing on mechanisms rather than diagnostic categories. I tentatively suggest how the field may incorporate this knowledge into a more effective conceptualization of affective disorders.
Classification and categorization that reflect commonalities and abstractions across a population is an important tool in scientific research: It enables important demarcations across a sample, organizes an area, identifies general principles, and makes it more tractable, by aggregating data across individuals. In clinical populations, diagnosis seeks to improve the validity and replicability of findings, through the classification of reasonably homogeneous groups with a common set of symptoms applied consistently by different researchers and clinicians.
Diagnosis takes a “syndromal approach,” in which groups of signs (what clinician sees) and symptoms (what the patient reports) are used to identify disease entities. In this disorder-focused approach, signs and symptoms that cluster together are conceived to act as pointers to a hypothesized underlying disease with a known aetiology, coherent course, and response to treatment. In psychiatry, classification has been based around distinct diagnostic categories (e.g., major depression, generalized anxiety disorder), each defined to a precise set of atheoretical criteria as exemplified by the DSM and ICD systems. Each syndrome is both theoretically constructed and empirically determined (e.g., can it be reliably diagnosed? Is it of clinical utility?). Diagnosis should therefore be a tool improving understanding of the aetiology of a presenting problem, prognosis for that individual if left untreated, and which treatment will be most helpful. Diagnosis has the potential clinical benefits of identifying distinct syndromes that respond differentially to alternative treatments and providing a common clinical language.
However, the diagnostic approach has potential disadvantages, especially for psychiatric disorders. Putting individuals into diagnostic boxes may minimize complexity and lose important personal and contextual information. In psychiatry, classification has become overly complicated, with over 350 disorders in DSM-IV. Furthermore, the categorical system specifies a cut-off that is often at an arbitrarily defined point, based on limited evidence, for example, the stipulation of 2 weeks of symptoms for a diagnosis of major depression.
A key benefit of diagnosis within the DSM and ICD systems is that detailed operationalization of symptoms has substantively improved the reliability of diagnosis. However, questions remain about the validity and utility of these diagnostic classifications (Harvey et al., 2004; Horwitz & Wakefield, 2007; Insel et al., 2009; Sanislow et al., 2010; Wakefield, 1992), especially concerning high rates of comorbidity across psychiatric conditions, heterogeneity within diagnostic classifications, and relatively poor efficacy of interventions. First, the 12-month rates of comorbid anxiety and/or depression are estimated at 40–80% (Kessler, Chiu, Demler, Merikangas, & Walters, 2005), raising questions about the core unique features of any specific diagnosis, and about the ability to discriminate between supposedly different disorders. Second, despite attempting to identify homogenous groups, heterogeneity within diagnoses is high, for example, within the diagnosis of major depression, two individuals could both meet criteria for major depression but only share a single common symptom, and have vastly different presentations. Third, even our best psychological and biological treatments for depression have high rates (20–60%) of suboptimal or even negligible responses and rates of relapse and recurrence are high for successful acute treatment (50–80%; Judd, 1997). It may be that focus on observed symptom clusters rather than on potential underlying mechanisms limits treatment efficacy (Harvey et al., 2004; Insel et al., 2009; Sanislow et al., 2010).
A further challenge to the disorder-focused approach is convergent evidence indicating similarities across different diagnoses. First, there is commonality of aetiology. Stress, early abuse, negative life events, trauma, having a parent who has a mental health problem, all predict future risk for a range of mental health problems, without specifically predicting what disorder(s) the individual will develop (Nolen-Hoeksema & Watkins, 2011). Second, analysis of the latent structure of symptoms across disorders reveals higher order dimensions that are shared across multiple diagnoses, for example, elevated negative affectivity (neuroticism) accounts for symptom variance across depression, generalized anxiety disorders, and panic disorder (Brown & Barlow, 2009; Brown, Chorpita, & Barlow, 1998). Paralleling this, a common underlying biology is found across different diagnoses, for example, increased activation of amygdala-based fear circuitry is shared across multiple disorders (Hyman, 2007). Because the current diagnostic system precedes contemporary neuroscience research, it is not informed by recent breakthroughs in genetics, or molecular, cellular, and systems neuroscience.
In parallel, clinical psychology research has investigated the hypothesis that there are cognitive-behavioural, interpersonal, and biological processes that (a) are shared across multiple disorders and (b) causally contribute to the onset, maintenance, recurrence, or recovery from disorders (Harvey et al., 2004). These cross-cutting mechanisms have been labelled “transdiagnostic.” Extensive syntheses of the experimental psychopathology literature indicate that several psychological processes are strong candidates to be transdiagnostic mechanisms because selective biases in these processes are found in multiple psychiatric disorders, with experimental and prospective evidence indicating that they causally contribute to the onset and maintenance of symptoms (Ehring & Watkins, 2008; Harvey al., 2004; Nolen-Hoeksema & Watkins, 2011; Watkins, 2008). Putative transdiagnostic mechanisms across affective disorders include selective attention to internal and external stimuli, avoidance, repetitive negative thinking, selective memory, and biased reasoning processes.
These mechanisms are shared across disorders because they also operate in healthy individuals, that is, because they are universal and normal human mechanisms. For example, selective orientation of attention to novel, feared, or personally salient stimuli is a normal and adaptive process found in healthy controls (Harvey et al., 2004). Such selective attention can be functional and adaptive in identifying and responding effectively to threat. However, across the anxiety disorders, selective bias of attention is found to the particular negative concerns of that disorder; for example, towards internal sensations such as heart rate in panic disorder; towards spiders in spider phobics. Such elevated attentional bias towards negative information causally contributes to increased symptoms of anxiety (Hertel & Mathews, 2011). Nonetheless, this mechanism reflects the normal process of attention orienting to salient personal concerns, similar to that found in individuals without anxiety disorders. Psychopathology can thus develop as a consequence of one or more of the following: (a) a mechanism becoming dysfunctional (Wakefield, 1992), oversensitive, or underregulated (e.g., exaggerated attentional sensitivity); (b) a mechanism operating as normal but applied to inappropriate stimuli or contexts, perhaps as a function of prior learning or holding extreme personal concerns (e.g., that increased heart rate signals danger); (c) exposure to extreme situations such as stress or trauma. In sum, psychological research provides evidence for the conceptualization of disorders on a continuum-dimensional approach, rather than a categorical one.
This evidence challenges the validity and utility of a solely disorder-focused diagnostic approach in psychiatric disorders, and indicates the need for alternative approaches. One alternative is a mechanism-focused approach to classification, focused on fundamental mechanisms, processes, or functions.
One example is the Research Domain Criteria Initiative from the U.S. National Institute of Mental Health (RDoC; Insel et al., 2009; Sanislow et al., 2010). Sanislow and colleagues (2010, p. 631) proposed that “uncoupling research efforts from clinically familiar categories to focus directly on fundamental mechanisms of psychopathology” has the potential to transform our knowledge of emotional disorders and our treatment efficacy. RDoC aims to “develop new ways of classifying disorders based on dimensions of observable behaviors and brain functions.” RDoC adopts both a transdiagnostic and a dimensional approach, reflecting processes that are shared across traditional diagnostic categories and span the range from normal to abnormal. The goal is to define the basic dimensions of functioning to be studied across multiple units of analysis, from genes to neural circuits to behaviours.
The RDoC approach intends to generate new constructs built around mechanisms integrated across genetic, neurobiological, and psychological levels of analysis. For example, one putative construct is loss, which draws together molecular mechanisms such as inflammatory processes and downregulation of glucocorticoid receptors; neural circuits including sustained amygdala reactivity and decreased dorsolateral prefrontal cortex activity; physiological mechanisms such as hypothalamic-pituitary-adrenal axis dysregulation; behaviours such as rumination, withdrawal, crying, sadness, attentional bias to negative valenced information, and self-reported hopelessness. The Loss construct therefore reflects mechanisms that repeatedly co-occur together as a shared cluster and that tend to change together, and which are implicated across several affective disorders, including depression, although, to date, much of the evidence linking these levels of analysis together is indirect.
In turn, these constructs are grouped into higher order domains of functioning. Putative domains suggested are negative valence systems; positive valence systems; cognitive systems; systems for social processes; and arousal/regulatory systems. RDoC aims to classify on homogeneity in the presentation and clustering of biological and psychological mechanisms, and to ascertain the impact of this approach for improving understanding and treatment of psychiatric disorders. This is an ambitious and long-term project, and we don’t yet know whether it is achievable. As a tool for classification, it will have to be judged on the same criteria as existing diagnostic systems, namely, reliability, validity, and utility.
Another example is the investigation of transdiagnostic cognitive and behavioural processes (Harvey et al., 2004; Nolen-Hoeksema & Watkins, 2011). The transdiagnostic approach suggests that psychiatric classification could be based around the presence, absence, or relative strength of transdiagnostic processes. The presence or absence of particular mechanisms would be tested as a predictor of prognosis and as an indicator of appropriate treatment, for example, the selection of a rumination module for a patient strongly displaying this characteristic. In this approach, patients with different traditional diagnoses may receive overlapping treatments because they present with common cognitive-behavioural mechanisms, whilst patients with the same traditional diagnosis (e.g., major depression) may receive different interventions because they display different mechanisms.
Mansell, Harvey, Watkins, and Shafran (2008) argued that adopting this transdiagnostic mechanistic approach may improve treatment outcomes. First, it enables goodness-of-fit matching of interventions to those specific vulnerabilities and processes relevant to each individual. Second, it directly targets fundamental active mechanisms, rather than symptom clusters, with the hypothesis that this should enhance outcomes. Third, it provides a flexible treatment approach that can be applied across a range of presentations, including comorbidity. For example, because avoidance is a causal factor common to both depression and anxiety disorders, then interventions targeting avoidance are hypothesized to simultaneously treat comorbid depression and anxiety.
For this article, from the potential range of transdiagnostic processes (e.g., avoidance, metacognitive beliefs, attentional bias, poor attentional control), I focus on the specific example of repetitive negative thought (RNT) to illustrate this mechanism-focused approach. I focus on RNT because (a) the evidence for RNT as a transdiagnostic process is at least as robust as any other process; (b) it is implicated in depression and grief; (c) there are emerging accounts of what determines whether repetitive thought contributes to normal versus abnormal emotion (Ehring & Watkins, 2008; Nolen-Hoeksema & Watkins, 2011; Watkins, 2008), all relevant to the focus of this special section.
Segerstrom, Stanton, Alden, and Shortridge (2003, p. 3) initially defined “repetitive thought” (RT) as the “process of thinking attentively, repetitively or frequently about one’s self and one’s world.” RT is a common process with both constructive and unconstructive consequences for cognition and emotion, with repetitive thought focused on negative content implicated in exacerbating negative affect and the onset and maintenance of psychiatric disorders, but with RT also implicated in problem-solving and recovery from upsetting events (see further discussion in Watkins, 2008).
The most studied forms of RT clinically are depressive rumination and anxious worry. Because measures of worry and rumination are highly correlated and load on a common factor in structural equation modelling, and both influence anxiety and depression (e.g., Segerstrom et al., 2003; Watkins, 2008), Ehring and Watkins (2008) hypothesized an underlying common process of repetitive thinking about one or more negative topics that is experienced as difficult to control (i.e., repetitive negative thought, RNT), shared across both rumination and worry (although worry and rumination may differ in specific thought content, e.g., worry future-focused and rumination past-focused). RNT therefore discriminates pathological forms within the wider construct of RT, and subsumes worry and rumination.
There is robust evidence that RNT is a transdiagnostic process contributing to multiple psychopathologies, with elevated RNT found in depression, generalized anxiety disorder, social anxiety, posttraumatic stress disorder, eating disorders, alcohol abuse and dependency, and psychosis (see detailed reviews in Ehring & Watkins, 2008; Harvey et al., 2004; Nolen-Hoeksema & Watkins, 2011; Nolen-Hoeksema, Wisco, & Lyubomirsky, 2008). Critically, in large-scale longitudinal studies, RNT prospectively predicts the onset of multiple psychiatric disorders and symptoms, after controlling for baseline symptoms. For example, in 1,109 community adults followed over 1 year, RNT predicted onset of major depression, and depressive and anxious symptoms (Nolen-Hoeksema, 2000). In 496 female adolescents followed over 4 years, RNT predicted increases in bulimic and substance abuse symptoms, and onset of major depression, binge eating, and substance abuse (Nolen-Hoeksema, Stice, Wade, & Bohon, 2007). RNT predicted the onset and severity of posttraumatic stress symptoms and diagnosis of PTSD up to 3 years after traumatic events (Ehlers, Mayou, & Bryant, 1998). RNT is therefore not simply a consequence or associate of psychopathology. Experimental evidence further indicates that RT causally exacerbates clinical symptoms, with inductions of worry and rumination increasing anxiety, depression, and negative cognition, relative to control conditions (see Nolen-Hoeksema et al., 2008; Watkins, 2008).
Consistent with the earlier argument, RT is a normal and universal process. Nearly everyone has had the experience of dwelling repeatedly on disappointments and upsets, such as the end of a relationship, ongoing difficulties, and major losses. There is evidence consistent with the hypothesis from control theory that RT is triggered and maintained by discrepancies between an individual’s current state and their desired or expected state, such as when experiencing a loss or an unresolved goal (Martin & Tesser, 1996; Watkins, 2008). Therefore, RT is a normal cognitive operation common to all. Moreover, RT can be adaptive and instrumental when this focus helps to resolve the goal discrepancy. Consistent with this, Watkins (2008) reviewed evidence that RT can be helpful in coming to terms with an upset and in problem-solving.
What then determines whether repeated dwelling on a problem or upset leads to useful working through and problem-solving, versus unhelpful brooding that gets stuck and exacerbates negative mood, negative cognitions, and psychopathology? One answer is that the way an individual thinks determines whether RNT is helpful or unhelpful (Watkins, 2008). RNT characterized by an abstract processing mode, characterized by general and decontextualized representations focused on causes, meanings, and implications (“why”), has more unconstructive consequences than RNT characterized by a concrete processing mode, characterized by direct, specific, and contextualized representations focused on details and means (“how”; at least when focused on negative content—in other contexts, abstract processing can be more adaptive than concrete processing; Watkins, 2008). For example, in depressed patients, compared to abstract rumination where participants thought about the causes, meanings, and consequences of feelings and symptoms, concrete rumination in which participants were instructed to “focus attention on the experience of” feelings, mood, and symptoms reduced negative global self-judgments, improved social problem-solving, and increased specificity of autobiographical memory recall (see Watkins, 2008). Prompting abstract RT (e.g., “Why did this problem happen?”) impaired social problem-solving in a recovered depressed group, who performed as well as never-depressed participants in a no-prompt control condition, whereas prompting concrete RT (e.g., “How are you deciding what to do next?”) ameliorated the problem-solving deficit found in currently depressed patients (Watkins & Baracaia, 2002). Relative to abstract RT, concrete RT produced faster recovery from negative affect and reduced intrusions after a previous negative induction (Watkins, 2004). Individuals trained to imagine the concrete details of what is happening in emotional scenarios had reduced emotional reactivity to a subsequent experimental stressor relative to those who repeatedly practiced evaluating the causes, meanings, and implications of the same emotional scenarios (Watkins, Moberly, & Moulds, 2008).
Thus, psychological research into RT indicates that (a) degree of RNT about loss or difficulties, and (b) the style of such thinking (how abstract or concrete) both impact on symptom development in affective disorders. The study of RNT illustrates how a transdiagnostic mechanism can reflect a normal cognitive operation found in everyone, and also how it can be psychopathological versus adaptive under particular conditions. To date, such understanding of contextual and functional effects on adaptive versus maladaptive functioning has had limited impact on diagnostic classification, indicating the value of incorporating understanding from basic emotion research.
Classifying by mechanisms may bring more rigor, utility, and empirical justification to the key theoretical and clinical question of how to distinguish normality from abnormality. A pertinent example is the debate about whether to distinguish bereavement from major depression, highlighted by the removal in DSM-V of recent bereavement as a diagnostic exclusion for major depression. It has been argued that the emotional and behavioural responses following bereavement (and other major losses) are a normal response, which typically naturally resolves, rather than a psychiatric disorder like major depression that requires treatment, despite the overlap in symptoms (Horwitz & Wakefield, 2007). An approach based solely on symptom clusters is at risk of making an arbitrary distinction between normal bereavement and major depression. In contrast, a focus on mechanistic differences underpinning how individuals respond to loss has potential to clarify our understanding, better predict those at risk for worse prognosis, and develop targeted therapy.
RNT provides a good example of how this mechanistic approach could work. In the initial months after the death of a loved one, nearly all bereaved individuals report RNT (e.g., daily thoughts about the deceased; Bonanno & Kaltman, 2001; Bonanno, Wortman, & Neese, 2004). Nonetheless, the majority of bereaved individuals adapt in the initial months and are no longer distressed about the loss (Bonanno & Kaltman, 2001; Bonanno et al., 2004), indicating that early RNT is not necessarily maladaptive. Indeed, there is evidence that RNT has constructive and unconstructive consequences following bereavement. Increased levels of RNT about loss are found in the 10–15% of individuals who develop chronic grief symptoms (Bonanno et al., 2004). Rumination postbereavement predicts subsequent depression (Nolen-Hoeksema, Parker, & Larson, 1994). Yet, in contrast, consistent with adaptive consequences of RNT, RT postbereavement predicted greater subsequent sense of meaning and better health outcomes (Bower, Kemeny, Taylor, & Fahey, 1998).
Based on the evidence for distinct consequences of abstract versus concrete processing modes, a plausible hypothesis is that RNT about the loss of a loved one is adaptive when concrete, contributing to normal recovery, but maladaptive when abstract, contributing to complicated grief and depression. Consistent with this hypothesis, healthy adjustment over the first 2 years after a spouse’s death is associated with the bereaved individual having more concrete, behaviour-focused self-evaluations, relative to abstract trait-focused evaluations (Bauer & Bonanno, 2001). Furthermore, concrete goals and concrete plan narratives at the first month of bereavement predicted healthier outcomes 1 year later (Stein, Folkman, Trabasso, & Richards, 1997). Bereaved individuals who report not searching for meaning are more resilient and have better outcomes than those who engage in abstract evaluations (Bonanno et al., 2004; Davis, Wortman, Lehman, & Silver, 2000). These results suggest a potential application of a mechanism-focused framework in which individuals are classified on the extent, valence, and processing mode of their RT, rather than their symptom presentation, postbereavement. The assumption is that mechanism-focused approach will better predict prognosis and need for treatment. The next step is to test such predictions empirically.
The mechanism-focused approach hypothesizes that treatments specifically designed to target identified dysregulated mechanisms improve treatment outcome. This requires the development of interventions targeting specific mechanisms. With respect to RNT, rumination-focused cognitive-behavioural therapy (RFCBT) has been designed to coach patients to shift from unconstructive RT to constructive RT, through the use of functional analysis, experiential/imagery exercises, and behavioural experiments (see Watkins et al., 2007). In a Phase II randomized trial of 42 treatment-resistant patients with residual depression RFCBT added to treatment-as-usual (continuation antidepressants) reduced self-reported and observer-rated depression significantly more than treatment-as-usual, and reduced self-reported rumination to normative never-depressed levels (Watkins et al., 2011). Although assessment of rumination necessarily involves subjective experiential report, and is potentially prone to response and demand biases, we do not believe that these account for the observed effect of RFCBT because the measure is of trait rumination and is relatively unresponsive to other interventions that explicitly emphasized the targeting of negative thinking and rumination in their rationales (e.g., Watkins et al., 2012). Future research needs to test whether targeting rumination has impact across a range of comorbidity disorders, as predicted by the transdiagnostic account.
How might a transdiagnostic mechanistic approach to classification work? Brown and Barlow (2009) proposed that individuals could be classified across dimensions of neuroticism, positive affectivity, and avoidance. Each individual is rated on the severity of the dimension for each mechanism, with the overall profile being examined to provide a summary of their symptomatic and mechanistic presentation. I propose that these dimensions could be extended to include identified transdiagnostic mechanisms, such as tendency towards RNT, abstract versus concrete processing; selective attentional bias, selective memory, etcetera. Within this classification, as well as specifying mechanisms, the foci and current concerns and goals (including potential goal conflict) of the patient are referenced (Harvey et al., 2004; Nolen-Hoeksema & Watkins, 2011) as these are hypothesized to interact with cognitive-behavioural processes to impact symptom presentation (e.g., focus of anxiety, foci of attention). Brown and Barlow (2009) further propose that this framework can be used to derive categorical diagnostic labels based on empirically derived cut-points from diagnostic profile across the dimensions.
This dimensional-mechanisms approach may better capture salient clinical features, whilst retaining the benefits sought in traditional diagnoses (demarcation, tractability, homogeneity). However, ratification of this approach requires new methods and new approaches to ascertain samples, relying on hypothesized psychopathological mechanisms to define experimental groups under study (as proposed within RDoC), instead of traditional diagnostic categories. At this point, any potential benefit of this approach is speculative and without empirical foundation. It will be complex and challenging to develop, evaluate, and implement this approach, and to determine which (if any) dimensions are valid, reliable, and useful for predicting prognosis and response to treatment. Until attempts are made to conceptualize clinical presentations this way, we won’t know if this approach is discriminative and reliable. Ultimately, the key consideration will be its clinical utility, relative to the traditional diagnostic approach.
In conclusion, I have reviewed limitations to current categoric diagnostic classification schemes, and argued that classification of psychiatric disorders would benefit from incorporating knowledge from clinical psychology research, including the recognition of similarities across disorders, transdiagnostic mechanisms, and the dimensional nature of cognitive, behavioural, and emotional processes. One possibility is the shift to classification based on underlying mechanisms rather than observed symptom clusters, as proposed within the RDoC initiative. Although it remains unresolved whether such an approach will improve on current classification systems, and considerable work and resource is required to investigate these approaches, this seems a worthwhile endeavour given the importance of enhancing our understanding and treatment of affective disorders.
Footnotes
Declaration of Conflicting Interests
None declared.
