Prepared by Michael B. First, M.D., DSM Consultant to the American Psychiatric Institute for Research and Education (APIRE), a subsidiary of the American Psychiatric Association
A diagnosis-related research planning conference focusing on dimensional approaches in diagnostic classification was held at the Natcher Conference Center on the NIH campus in Bethesda, Maryland on July 27th and 28th, 2006. The conference was the seventh in a series of 12 conferences on "The Future of Psychiatric Diagnosis: Refining the Research Agenda" convened by the American Psychiatric Institute for Research and Education (APIRE) in collaboration with the WHO and NIH, with funding from NIH. Dimensional conference co-chairs were John Helzer, MD, from the University of Vermont in Burlington, Vermont, Helena Kraemer, PhD, from Stanford University in Palo Alto, California, Robert Krueger, PhD, from the University of Minnesota in Minneapolis, Minnesota, and Hans Ulrich Wittchen, PhD, from the Institute of Clinical Psychology and Psychotherapy in Dresden, Germany. Twenty-eight invited scientists from around the world participated.
John Helzer, MD (Burlington, VT) opened the conference with a description of its rationale and goals These were 1) to review the advantages and drawbacks of offering a quantitative "dimensional" component in DSM-V and ICD-11; 2) to discuss various alternative dimensional approaches; and 3) to discuss a research agenda relevant to designing and testing dimensional alternatives. He stressed that the six diagnostic areas scheduled for formal presentation - i.e., substance use disorders, mood disorders, psychotic disorders, anxiety disorders, childhood disorders, and personality disorders - were illustrative examples and that it was not the intention of the conference to confine discussion to these areas. Dr. Helzer noted that although DSM-III was a revolutionary paradigmatic shift, the diagnostic progress it fostered revealed the inadequacies of a strictly categorical diagnostic model. He then offered four recommendations for DSM-V: 1) that the DSM-V criteria should include options for dimensional approaches; 2) that the categorical approach of DSM should be retained given the ongoing need for diagnostic categories for clinical work and research; 3) that the content of DSM-V dimensional components be determined by categorical definitions given the need to be able to relate the dimensional scales back to the categorical definitions; and 4) that DSM-V should be structured to ensure maximum utility for future taxonomic needs.
Helena Kraemer, PhD (Palo Alto, CA) presented on the clinical and research contexts of categories and dimensions. She first distinguished between the term "disorder," which she defined as "something wrong in the patient that is of clinical relevance," and "diagnosis of a disorder," which is an expert opinion that a disorder is present. The decision to use a dimensional versus categorical diagnosis has nothing to do with the nature of the disorder itself but everything to do with the quality of the diagnosis for a disorder. A dimensional diagnosis has three or more ordinal values, which can range from a three point scale (at a minimum) up to continuum (e.g., BMI for eating disorders). A categorical diagnosis has only two possible values: present and absent. Further elaborating the purpose of the conference, she said the key issue is to consider whether to add a dimensional component to the DSM-V categorical diagnoses rather than to try to replace categorical diagnoses with dimensional constructs. Every dimensional diagnosis can be made categorical by setting a cut-point; conversely, every categorical diagnosis can be made dimensional a cross any of a number of possible dimensions, including - but not limited to - symptom count, symptom duration, symptom severity, degree of impairment, and certainty of diagnosis. Dr. Kraemer illustrated the advantages of dimensional over categorical diagnoses in research settings by describing a randomized controlled trial of cognitive-behavioral therapy vs. self-help for the treatment of eating disorders. The actual trial, which used the presence or absence of a categorical diagnosis of bulimia nervosa as the outcome variable, showed no statistically significant differences in the two groups. However, if a dimensional measure of frequency of binges and purges were to have been used instead, a statistically significant moderator effect would have been detected, indicating that the treatment worked in low-risk but not high-risk groups. From a statistical perspective, advantages of using dimensional vs. categorical outcomes include greater power to detect treatment effects, less attenuation and greater precision in estimates of effect sizes, and better ability to detect signals Dr. Kraemer recommended that any proposed DSM-V dimensional diagnosis correspond well with the respective categorical diagnosis, be transparent to clinicians, and have clinical validity and test-retest reliability. She concluded that the time to add dimensional diagnosis to DSM has come and that a dimensional approach is needed in order to prepare for the future inclusion of genetic, imaging, biochemical elements to psychiatric diagnoses. A dimensional approach can be as simple or as complicated as is appropriate for each disorder, but most importantly, clinicians must be able to use it.
Steven Haffner, MD (San Antonio, TX), an epidemiologist working in cardiovascular medicine, provided a perspective on the use of dimensional approaches in that field. He described the rise and fall of the "metabolic syndrome," conceptualized as a combination of risk factors for predicting cardiovascular disease (CVD). Although the the metabolic syndrome concept is only eight years old it has already gone through three iterations. The original definition of metabolic syndrome required three out of five factors (abdominal obesity, elevated triglycerides, low HDL-C, high blood pressure, and elevated fasting glucose). However, because having fewer than three factors also is associated with an elevated risk of disease, while having more than three factors is indicative of a further increased risk, defining risk in terms of this categorical syndrome mischaracterizes people and loses important information. Problems with the existing categorical definition of the metabolic syndrome construct include: 1) loss of data due to the use of dichotomous variables to define each component of the syndrome; 2) questions about the appropriateness of using these cut-offs in other countries with different lifestyles; 3) the differing ability of the components to predict diabetes and CVD; and 4) questions about whether the syndrome predicts CVD independently of its components. Potential advantages of the categorical syndromal definition include 1) its utility as an operational definition for "cardio-metabolic" risk; 2) its ease of use as compared to multi-variate predicting equations; 3) its capacity to encourage providers to look for multiple risk factors in their patients; and 4) its tendency to encourage behavioral interventions (e.g., weight loss and increased activity) rather than treatment of individual risk factors.
At the core of the conference were a series of presentations centered on the clinical and scientific feasibility of dimensional models for specific diagnostic areas. The first, by Kathleen Bucholz, MD (St. Louis, MO), reviewed research evidence for a dimensional approach to substance use disorders. Dr. Bucholz noted that the change from a monothetic criteria set for substance dependence in DSM-III to a polythetic criteria set in DSM-III-R foreshadowed a dimensional conceptualization of dependence. Summarizing 26 studies of alcohol use disorders, Dr. Bucholz reported that some of the early studies rejected the unidimensionality of alcohol symptoms, finding instead two highly correlated dimensions that reflected mild and severe symptomatology. However, neither of the factors corresponded to the DSM construct of abuse. More recent studies have provided strong evidence of the unidimensionality of alcohol items, with the consistent implication that alcohol problems lie on a continuum. Furthermore, converging evidence indicates that some DSM abuse items are severe and some dependence items are mild. Studies of other drugs (18 studies) support unidimensionality for some but not all drugs, with the findings varying by type of population studied (e.g., unidimensionality was not supported in problem users). Two studies of nicotine suggest, however, that nicotine dependence does not appear to be unidimensional but instead yields two factors: one factor for failed cessation and the other factor for general dependence symptoms. Dr. Bucholz suggested that future research investigate gradation within symptoms, examine weighting of symptoms, consider why findings might differ in clinical and population samples, reconsider clustering, investigate symptoms outside the DSM, and reconsider other aspects of dimensionality in addition to severity.
John Helzer, MD (Burlington, VT) then offered an example of how DSM-V might accommodate a dimensional approach to substance use disorders. He proposed a four-step process for DSM-V, beginning with the status quo (i.e., diagnostic workgroups creating categorical definitions) in recognition of the ongoing need for diagnostic categories for clinical work and for research. The second step would dimensionalize each of the individual criterion items, ranging from simple 3-point scales that could apply across diagnoses (e.g., a severity scale of absent, mild, and severe) or a complex diagnostic-specific scale that could include laboratory values. The third step would entail creating a diagnostic scale from dimensional items using statistical techniques such as factor analysis, latent trait modeling, item response theory, etc. The final step would relate the dimensional scale back to the categorical definition in step one, using methods such as ROC analyses or logistical regression. Dr. Helzer then proposed a more ambitious alternative which would entail developing the dimensions empirically from the "bottom up," using a mix of current criterion items and associated features which are not currently part of the criteria sets, including behaviors, family history, biological test results, and other features of putative diagnostic value.
Michael Gossop, PhD (London, UK), reviewed several conceptual, technical, and practical problems raised by a dimensional approach to substance use disorders. He noted that substance dependence is overwhelmingly the reason why people come to clinical services and he asked whether the current criteria for dependence are in fact correct and, more pointedly, whether there are specific core criteria which should be weighted. He suggested that the criteria for dependence be geared to what clinicians need to know, namely whether there is a need to manage withdrawal symptoms and whether long-term treatment is needed. A diagnosis of substance abuse, on the other hand is based only on harmful consequences, i.e., it is basically a problem-based conceptualization of a disorder. Dr. Gossop argued that dependence on the one hand, and problems due to substance use on the other are fundamentally different concepts, each with its own management implications and thus should be kept apart. Furthermore, the majority of people who seek treatment are multiple substance users, although the literature primarily focuses only on single substance dependence. Is dependence on multiple substances simply a sum of two or more disorders? Bridget Grant, PhD (Bethesda, MD), representing the National Institute on Alcohol Abuse and Alcoholism, led the general discussion by noting that one cannot consider dimensionality apart from other nosological issues that have proven recalcitrant. A researcher can take any disorders and throw the symptoms into an item response theory analysis and find dimensionality. It is important instead to consider the nature of the disorder being measured and determine whether one is starting with the right criteria.
Terry Brugha MD (Leicester, UK) presented on dimensional approaches and mood disorders. Key opening points were that a categorically-based classification does not reflect reality as evidenced in community epidemiological data; that having a single threshold for all clinical and research purposes is problematic; and that categories have a tendency to be reified naïvely. Although clinical decision making is inherently binary (i.e., to treat or not), both clinical and epidemiological studies demonstrate that dimensional approaches to mood disorders may have advantages over diagnostic categories; for example, most subthreshold depressions predict later depression. Dr. Brugha offered the NICE (National Institute for Clinical Excellence) guidelines for the treatment of depression in primary and secondary care settings as an example of a marriage of categorical and dimensional approaches. These guidelines use a stepped care approach that recognizes the different treatment requirements for different levels of severity of depression as defined in ICD-10. In concluding, Dr. Brugha cautioned against making rules needed for levels of severity too complicated for clinical use. Epidemiologists should use dimensional measures that can accommodate a range of disorder thresholds and categories.
Michael Thase MD (Pittsburgh, PA) followed with a presentation in which he contended that symptom severity of depression is the critical dimension for predicting treatment response. Although depressive disorders are heterogeneous, a unitary dimension of symptom severity conveys important descriptive and prognostic information. As depression severity increases, the probability of clinical and biological correlates of dysphoric activation increases. Clinical correlates of high pre-treatment severity include melancholic features, psychotic features, comorbid anxiety and neuroticism, borderline personality disorder, and suicidality. Neurobiological correlates include hypercortisolism, changes in regional cerebral metabolism (increased activation of amygdala, decreased activation of prefrontal cortical structures) and increased peripheral levels of norepinephrine metabolites. Increased symptom severity has important treatment implications including longer time to remission and recovery, a lower absolute likelihood of remission or recovery within 6-8 weeks, relatively lower likelihood of placebo response compared to antidepressant response, and a greater likelihood of response to combined psychotherapy and pharmacotherapy compared to therapy with either alone.
Gavin Andrews, MD (Sydney, Australia) considered various reasons that might explain why categories continue to dominate in the classification of mental disorders given the clear advantages of dimensional approaches. One problem with categorical classification is that small differences in the wording of diagnostic criteria can lead to big differences in diagnostic concordance between different classification systems. For example, according to the Australian CIDI data, which collected both DSM-IV and ICD-10 diagnoses in the community, only two-thirds of individuals who met criteria for a disorder in one classification met criteria for the corresponding disorder in the other classification; factors primarily responsible for the discordance were differences in exclusion criteria and diagnostic thresholds. Early hopes that external markers (e.g., laboratory tests) would validate categories has not born out. With respect to the anxiety and mood disorders, Dr. Andrews said, genetic data is more compatible with a dimensional than a categorical approach, and it appears likely that ongoing research into fear circuitry and HPA axis pathophysiology will also support a dimensional approach. Much of the current high rate of diagnostic comorbidity between depression and anxiety may be explained by the existence of higher order dimensions (i.e., both disorders appear to be part of a single superordinate dimension). Moreover, epidemiological data from community studies suggests that depression is best conceptualized, measured and classified as a continuously distributed syndrome rather than as a discrete diagnostic entity. Given the numerous advantages of a dimensional approach, , encouraging clinicians to use a dimensional approach is a pressing question. Dr. Andrews proposed that patients could complete a symptom screener on computer before seeing the doctor, which would indicate those diagnoses that are highly probable. The clinician could use this information to guide the clinical interview and/or to confirm his or her findings and to provide routine outcome measurement. Noting that the rules for developing a taxonomy that permits dimensions will not be simple or easy, he called for a systematic research effort using epidemiological survey data which contain both categorical diagnosis and dimensional measures on over 100,000 subjects.
William T. Carpenter, Jr., MD (Catonsville, MD) presented on dimensional approaches and psychotic disorders. He first summarized the presentation of Jim van Os, MD (Maastricht, the Netherlands) who was unable to attend the meeting at the last minute because of travel logistics. Dr. Van Os noted that by focusing on the severely mentally ill, clinicians have a very biased view of psychiatric symptoms as they might occur in the general population. Because phenotypes depend on multiple genes and their interaction with multiple environmental factors, behavioral traits tend to be widely distributed in populations without any points of rarity, which is ill-suited to a categorical conceptualization of mental illness. Research suggests, for example, that major depression may be an arbitrary diagnostic convention imposed on a continuum of depressive symptoms, with subclinical manifestation of depression highly related to the incidence of major depressive disorder. Dimensional approaches to psychotic disorders may be particularly useful as an indication of the need for treatment-the more dimensions that are present, the greater the unmet needs. Dr. Carpenter then continued with his own presentation, noting that the heterogeneity among patients with schizophrenia reflects different interactions between genes and the environment and that the schizophrenia symptom areas can be reduced to three domains: psychotic symptoms, negative symptoms, and interpersonal problems. He then discussed in detail a "deficit" subgroup of individuals with schizophrenia. This subgroup has less depression, less suicidal ideation, less insight, more social and physical anhedonia, a different course of illness (e.g., poorer social function as compared to a non-deficit group), a different pattern of information processing problems, and a different neuroanatomy. Furthermore, relatives of probands with the deficit syndrome tend to have more schizophrenia, more non-affective psychosis, and are more likely to themselves manifest the deficit subgroup of schizophrenia as compared to the relatives of non-deficit who have more affective psychosis. Commenting on this presentation, Trisha Suppes, MD, PhD (Dallas, TX), noted that it is important to consider the relationship between dimensions and categories insofar as they would result in a change in treatment. In this regard, the current categorical nature of psychosis is DSM-IV is potentially problematic because psychotic symptoms occur on a continuum. At the most severe end are symptoms such as command hallucinations and delusions that drive one's behavior; in the middle are symptoms such as hearing whispers, grandiosity, and paranoia, and on the least severe end would be the phenomenon of hearing one's name called and fearfulness without frank paranoia. Perhaps a dimensional approach to psychosis would encourage better recognition of a schizophrenia, psychotic depression, or bipolar psychosis prodrome which might lead to earlier therapeutic intervention.
Katherine Shear, MD (New York, NY) presented a novel dimensional approach for anxiety disorders in the form of a panic-agoraphobic spectrum. Because most patients present with an array of co-occurring symptoms and associated clinical features, the DSM diagnosis alone results in an oversimplification of clinical reality. The spectrum model provides an operational system for describing a range of phenomena that are the presumed expression of psychopathological processes related to given DSM syndromal categories. The panic-agoraphobic spectrum is a group of 114 typical and atypical panic-agoraphobic symptoms and related behavioral tendencies and temperament traits, grouped as separation sensitivity, stress sensitivity, medication/substance sensitivity, anxious expectation, illness phobia/hypochondriasis and reassurance orientation. The spectrum approach enhances the precision of categorical diagnoses by employing a wider band-width assessment of the array of clinical features associated with DSM. This approach defines a gradient of psychopathology within each category; quantifies features of a given disorder to generate a novel dimensional measure; includes a broad array of features that provides a more sensitive threshold for clinically significant symptoms; and offers the possibility of better identifying boundaries between disorders by evaluating multiple dimensions in a range of DSM disorders. Furthermore, it is feasible to assess the panic-agoraphobia spectrum in clinical settings using a self-report instrument or a fifteen-minute interview. Dr. Shear recommended that future research focus on analyses of existing spectrum data to further tease apart the boundaries of different spectrum features, and to identify features that can be used to guide management (e.g. treatment selection, targets, dose and termination decisions).
Ingvar Bjelland, MD, PhD (Bergen, Norway) compared dimensional and categorical approaches to co-occurring anxiety and depressive symptoms by examining results from the HUNT study, a population-based general health survey conducted in Norway in the mid 1990's. The survey included two 7-item subscales for anxiety and depression, as well as measures of subjective impairment. Dr. Bjelland concluded that given the loss of information caused by the categorization of continuously distributed phenomena, a dimensional approach was superior to the categorical approach in describing co-occurring symptoms of anxiety and depression and in predicting the impact of such co-occurrence. He noted that a future challenge is to determine the most characteristic symptom dimensions of anxiety and depression and how to measure them.
Hans-Ulrich Wittchen, PhD (Dresden, Germany), discussing the anxiety disorder presentations, suggested that dimensional approaches offer several advantages over categorical approaches. These include incorporating a greater amount of potentially relevant information (eg., severity), having better predictive and discriminating power, being more stable over time, being less affected by minor shifts in psychopathology, and having higher levels of reliability. Moreover, dimensional assessment is likely to be acceptable to both clinicians and researchers given that dimensional instruments (self-rated and clinician-rated) are available for various domains (e.g., somatic and cognitive symptoms of anxiety, panic, avoidance) and that dimensional approaches are the standard in clinical research (e.g., number of panic attacks). He cautioned, however, that dimensional self-report scales have some significant limitations including difficulties in determining duration and persistence; and their almost exclusive focus on current state. Furthermore, the lumping together of the anxiety disorders based simply on factor analytic or taxonomic approaches of categorical diagnostic data seem to be premature and clinically of restricted value. Unlike the fairly coherent unified domain of externalizing disorders (conduct disorder, antisocial PD, substance use disorders), there is little evidence for a clear structure of anxiety disorders. Dr. Wittchen concluded that fuller understanding of the structure of mental disorders might require taking into account the developmental pathways and the mechanisms involved in symptom progression.
James Hudziak, MD (Burlington, VT) reviewed research evidence for a dimensional approach in developmental psychopathology. Thousands of studies on child and adolescent psychopathology have used dimensional approaches in epidemiologic and clinical populations; many of these have combined and compared dimensional and categorical approaches. The fact that categorical approaches fail to manage sources of variance that are important in child psychiatry further buttresses the case for use of dimensional approaches. These sources of variance include developmental stage, gender, informant (teacher vs. parent vs. child), ethnicity, and comorbidity. A variety of standardized and epidemiologically normed dimensional/quantitative approaches have been developed for the study of developmental psychopathology. For example, a study using the quantitative Child Behavior Check List (CBCL) scale demonstrated that teachers and parents had significant disagreements on whether and to what degree girls are aggressive. Developmental stages are also important in understanding sources of variance, for example, in early childhood, 50% of the variance is due to genetic factors whereas in older age groups there is a remarkable reduction in genetic contributions to the development of aggression. In summary, Dr. Hudziak said that quantitative approaches, which already exist in child and adolescent psychiatry, are needed in order to incorporate potentially useful endophenotypic and genetic discoveries into psychiatric assessment approaches and that the taxonomy should not impede research discoveries.
Thomas M. Achenbach, Ph.D (Burlington, VT) reviewed the use and usefulness of dimensional vs. categorical approaches to the diagnosis of externalizing disorders across the transition from childhood to adulthood. Both "overt" and "covert" factors have been found in analyses of symptoms of conduct disorder (CD) in children and adolescents. "Overt" symptoms include bullying, physical fighting, and cruelty to people and animals, conning, and destroying other's property. "Covert" symptoms include stealing, staying out at night, running away from home and truancy. These two patterns are not mutually exclusive and may be found simultaneously or at different stages of development. Overt CD symptoms are strongly associated with the aggressive syndrome scale derived from the CBCL whereas covert symptoms are associated with the rule-breaking syndrome. Longitudinal data reveal, however, that children who were aggressive at ages 13-16 evolved either into an aggressive adult behavior pattern or an intrusive adult syndrome (bragging, showing off, demanding attention) in which the more overtly aggressive behaviors disappear but socially obnoxious behaviors are retained. Over a 10 year period, aggressive scores in female adolescents predicted substance abuse and disruptive behavior in adulthood. However, rule breaking in adolescent males predicted disruptive behavior in adult males. Furthermore, the aggressive syndrome has higher heritability than the rule-breaking syndrome. Dr. Achenbach concluded from this data that although conduct disorder criteria can be quantified by computing symptom scores, they should be separated into at least two patterns because of differences in features, predictive value, and heritability.
Commting on these presentations, Daniel Pine, MD (Bethesda, MD), identified numerous advantages to integrating continuous approaches into a categorical system. These include: 1) the continuous nature of external validators (e.g., genetics); 2) the availability of rich data and the opportunity to integrate with work in typical development; 3) the prospect of solving the problems with categorical cutoffs (e.g., below threshold cases can also be impaired); 4) their application to other developmental disorders (e.g., depression); and 5) the availability of perspectives from diseases in other branches of medicine (e.g., hypertension, obesity). Perhaps the only disadvantage of continuous approaches (as opposed to categorical approaches) is that with limited resources, classification is needed to facilitate allocation of treatment. Key questions that need to be addressed regarding the adoption of continuous approaches in the diagnosis of childhood disorders include: 1) how to integrate information across informants; 2) how to objectify age-related variations in scales; 3) how to accommodate multiple domains within each construct (e.g., for ADHD, there are inattention, hyperactivity, and impulsivity domains; for conduct disorder there are rule-violation, aggression, and emotional [i.e., lack of remorse] domains); and 4) how to integrate these approaches with neuroscience, where sometimes continua are evident (e.g., all children will have some level of improved attention after taking stimulants) and sometimes categories (e.g., improvement in mood after taking antidepressants occurs only in those children with a mood disorder). While continuous approaches need to be part of a DSM-V, Dr. Pine concluded that it is hard to objectify how to do this and come up with really good rules that maximize the advantages of the continuous approach the nosology.
Robert Krueger, PhD (Minneapolis, MN) presented on the research base for a dimensional approach to personality diagnosis in the DSM, noting that the first of the research planning conferences in the "Future of Psychiatric Diagnosis" series (which took place in December 2004) focused on dimensional models of personality disorders. Eighteen alternative dimensional models for personality were summarized at the meeting, raising the question about which dimensional system would be best for diagnosing personality disorders. Although the various models appear to be superficially diverse, extensive evidence documents systematic links between the models, leading many investigators to believe that a hierarchical model can be developed that that integrates these models. The top level of the hierarchy would have two dimensions, internalizing and externalizing, each of which would have two lower dimensions, emotional dysregulation and introversion vs. extraversion under internalizing, and impulsivity vs. constraint and antagonism vs. compliance under externalizing. Beneath the four dimension level would be lower-level traits akin to the current personality disorder items (e.g., difficulty with intimacy may lie under the introversion/extraversion factor) which would facilitate clinically optimal conceptualization of specific patients. This hierarchical structure reflects not only observable variation but also the underlying genetic risk factors, is also applicable to children and adolescents, and is applicable cross-culturally. Dr. Krueger recommended that an integrative dimensional structure should be a starting point for the classification of personality disorders in DSM-V. He proposed, however, that ultimately it will be necessary to reconsider the deeper structural aspects of the DSM given that the Axis I-II distinction is not highly compatible with the empirical organization of mental disorders. Also, he said, augmenting DSM-V with dimensions would help to generate the data needed to formulate a "bottom-up" structural organization for DSM-VI.
John Livesley, MD, PhD (Vancouver, Canada) presented on the clinical relevance of a proposed dimensional classification of personality disorder (PD). He argued against just taking the DSM categories and converting them to a set of dimensions because they are not "natural kinds," not clinically useful, and provide poor coverage. He proposed a two component structure for classifying PD: 1) developing diagnostic criteria for general PD; and 2) developing a system for representing individual differences in PD (different forms of disorder). The DSM-IV definition of PD as maladaptive traits would be extended by defining the individual disorders as categories of primary traits. For example, borderline personality disorder could be defined in terms of the primary traits of affective lability, impulsivity, cognitive dysregulation, insecure attachment, and self-harm. Dr. Livesley then presented an example of this approach using the Dimensional Assessment of Personality Pathology (DAPP) Model, which contains 28 primary traits and four higher-order factors (emotional dysregulation, dissocial behavior/psychopathy, inhibitedness, and compulsivity). He argued for the clinical utility of this dimensional approach on several grounds: 1) that it corrects coverage problems in the DSM (e.g., it includes sadism and pessimistic anhedonia), 2) it allows for flexibility in assessment on either the higher-order factor level or the primary trait level, 3) most interventions for PD focus on specific symptom or behavior clusters (i.e., the primary traits) rather than global diagnoses; and 4) psychotherapeutic interventions are typically organized around incidents, scenarios, recurrent themes, and maladaptive cognitions which may represent the behavioral expressions of primary traits.
Andrew Skodol, MD (New York, NY), in his discussion, raised several issues and questions that will need to be addressed in the future: 1) although the dimensional models for PD converge on broad domains, they differ at trait/lower item levels, so ascertaining which of the many existing models to choose will be challenging; 2) how should comorbidity that occurs both within and across broad spectra be interpreted (e.g., comorbidity between mood disorders in the internalizing dimension and substance use in the externalizing dimension); 3) although traits long have been presumed to be stable over time, there is some evidence of change over time, raising the question of how to incorporate such change into conceptualizations of traits; 4) identifying and achieving consensus on the core features of any new global definition of PD; numerous features of such a definition that could be dimensionalized - e.g., duration of traits, pervasiveness, inflexibility, distress, and functional impairment; and 5) given that personality disorders vary in dimensional complexity, it will be necessary to determine which traits are the most important and most useful. . Dr. Skodol concluded by suggesting that a personality disorder might possibly be reconceptualized as a combination of both enduring predispositions (static phenotypes) and characteristic behaviors (dynamic processes) and that it will be important to incorporate both static and dynamic elements when formulating a hybrid model for DSM-V.
Maritza Rubio-Stipec, ScD (Arlington, VA) discussed how existing data can inform the revision of the nosology to include both categorical and dimensional diagnostic measures. In discussing the availability of epidemiological data sets to inform the dimensionality of disorders she highlighted the array of considerations that must be taken when using these data. Although most existing information about the epidemiology of psychiatric disorders comes from structured diagnostic instruments that collect data on signs, symptoms and duration, they were designed to operationalize the DSM and ICD diagnostic systems by producing dichotomous diagnoses. Dr. Rubio-Stipec advised that when developing dimensional measures, the diagnostic algorithms can be skipped in favor of using the raw data. Caution should be exercised as the presence of skip patterns may affect the clustering of symptoms and thus the type of dimensional measure that can be generated with the existing data. She also noted that certain aspects of the data sets might lend themselves to different types of analyses; for example, data sets that capture multiple diagnoses may be useful for studying diagnostic boundaries whereas data sets with only one or two disorders may allow for an analysis that goes deeper into the disorder. She described a directory of data sets from different countries now being compiled by APIRE, that is intended to inform the DSM-V workgroups about available data sources that could be used to develop dimensional approaches. For each available data set, the directory describes the aim of the survey, the disorders included, the diagnostic instruments used, the target population, and the sampling strategies.
Helena Kraemer, PhD (Palo Alto, CA) discussed statistical "tactics" and questions to be considered when doing research on dimensional approaches. Among these: 1) Once items are selected, do they tap one dimension, two dimensions, or more?; 2) In doing a factor analysis, if only one factor emerges then go on. If more than one factor emerges, split the list of items and deal with each separately. 3) Are the items in the dimension reliable? If not, either reframe the questions assessing the item or else discard the item. 4) If items are redundant to each other, the redundant item should be either reframed or discarded. 5) When considering how to scale individual items, item response theory can be helpful, with more difficult or discriminative items scored more heavily. It is important, however, to insure that the scaling is as simple as possible to make it clinically useful. 6) Receiver Operating Curve (ROC) analysis can be used to establish optimal cut points by showing the correspondence between the binary DSM-V diagnosis and a proposed DSM-V ordinal diagnosis, with the higher the ROC curve the better the correspondence between the binary and ordinal diagnosis. For optimal effect, ROC analysis should be done for various populations (men/women, ethnic groups, different countries, socio-economic status, community, high risk groups, and clinical samples) and for various usages: screening (sensitivity), discrimination, definitive diagnosis (specificity). Each of these combinations may or may not identify a different cut point. 7) When looking at a dimensional measure to determine whether there is one disorder or two, there is one disorder if there are shared risk factors, if one leads to the other (i.e., multiple stages of one disorder) or if the same treatment principles apply.
Following these presentations and commentaries, the participants reconvened in six breakout groups corresponding to the six diagnostic areas; each group was asked to select one diagnosis in DSM and to develop a concrete proposal for a dimensional approach that might serve to help future DSM workgroups visualize the feasibility of introducing dimensional components to the diagnostic system. (The proposals are summarized in the table below).
The externalizing disorders of childhood breakout group used ADHD as an exemplar. A number of quantitative scales are available to measure ADHD symptoms and functioning. When setting categorical thresholds to determine caseness and treatment, it may be that thresholds should differ by age, gender, informants, and culture/ethnicity. Furthermore, severity can be considered at the symptom level (which is often presented in terms of frequency) or at the syndrome level, in which the overall impact on functioning is considered. Regarding the practical issue of how to implement a dimensional approach in clinical settings, the group noted that a dimensional approach could actually reduce the workload for clinicians especially in settings where data collection is automated, given that parent and teacher reports have been shown to be highly predictive of clinical diagnoses of ADHD.
The personality disorders breakout group proposed a dimensional model that used roughly 28 personality descriptive facets based on the empirical literature for fine-grained personality description. For the clinician using the DSM, each facet could be rated on a 3 point scale with scales varying by facet (i.e., some bipolar, some unipolar). A profile of scores across the 28 facets could then provide clinically useful dimensional data on personality description which can be summarized into the four broad domains described previously. The group also recommended the development of general criteria for personality disorder that go beyond personality description and that could include elements such as requiring a disturbance of self and interpersonal relations, having a minimum level of impairment (perhaps rated on a scale such as the SOFAS), requiring that it be present more often than not for at least a certain number of years, and requiring that it not be due to some other reason (e.g., chronic substance use).
The substance disorders breakout group selected alcohol dependence as its focus and suggested a bottom-up, data-based development of criteria. The process could begin by using the existing DSM-IV criteria as a starting point and then creating possible new criteria (e.g., measure of substance consumption, craving, rapid reinstatement), eliminating candidate criteria that are uninformative or questionably relevant to the diagnosis, rare, highly unreliable, or redundant with other criteria. Criteria could then be dimensionalized using a 3 point scale (i.e., no/mild/severe, or never/sometimes/frequent). Then, instead of a committee setting the diagnostic thresholds in advance, thresholds could be determined in a more empirical fashion, with different thresholds depending on the population of interest (e.g., community, clinical, high-risk), the purpose (severity, impairment, diagnostic certainty), developmental level (e.g., adolescents vs. adults), demography (e.g., gender, race) and the development of new findings (e.g., candidate genes).
The anxiety disorders breakout group noted that shared features of anxiety disorders could easily be dimensionalized across these disorders. Among the shared elements are panic, anticipatory anxiety, avoidance behavior, and other dysphoric affects.. Although it is neither feasible nor useful at this time to attempt to reconfigure anxiety disorders based on dimensional ratings of core symptoms, the use of such an approach by researchers could be advantageous. Instruments already exist to assess each of these elements, although they vary in degree of practicality. Panic spectrum assessment, a dimensional measure of panic-related symptoms, has been shown by researchers to be an important predictor of illness severity and treatment outcomes. Although requiring dimensional measures of all features of all anxiety disorders is not practical for routine clinical purposes at this time, where simple instruments are available, these could be recommended as an enhancement to categorical diagnosis. Using panic disorder as an example, the group identified the Panic Disorder Severity Scale (PDSS) because it is sensitive to treatment response, is a simple way of operationalizing severity, has a 1-month time frame, and includes defined anchor points (0-4). A further way that dimensional assessment could be useful in anxiety disorders would be to adopt a staging approach to illness, similar to that used in other areas of medicine to inform treatment decisions and predict course. Again using panic disorder as an example, dimensional assessment of panic, anxiety, phobia and other psychopathology, could lead to the ability to stage the evolution of the disorder, with stage 1 being isolated panic attacks, stage 2 panic disorder, uncomplicated, and stage 3 panic disorder complicated by other problems such as agoraphobia, other anxiety or depressive disorders, etc.
The mood disorders breakout group proposed adding dimensionality to major depressive disorder by exploring the implications of adopting the PHQ-9, a nine-item self-report scale with a 0-27 range of scores. Used as a severity measure, suggested cutpoints include 5-9 for mild, 10-14 for moderate, 15-19 for moderately severe, and 20 or greater for severe, with "significantly improved" defined as a 5-point drop after 6 weeks, "responded to treatment" as a 50% drop in score, and "remission of depression" as a score of less than 5. Such an approach has clinical utility in that it can identify probable cases, it identifies cases at risk who need referral, it confirmed current treatment in 60% of cases, it informed a change of treatment in 40% of cases, and in >90% of visits, clinicians rated it as useful. It has research utility in that it sets benchmarks for improvement, recovery, and remission and permits clinical audits of practice. The scale still needs validation in different settings and with different types of patients (e.g., high risk patients with comorbid conditions).
The psychotic disorders breakout group recommended that DSM should assess a variety of domains relevant for care in a quantitative fashion, using a simple scale. The final quantitative scores for DSM purposes should be made by the clinician and may be based on a variety of information ranging from clinical observation to sophisticated, computer-based measurements. Diagnostic-specific domains could include deficiency in reality testing, disorganization of thought, cognitive dysfunction, negative symptoms (avolition, anhedonia), and social impairment. Treatment-relevant domains include affective disturbances, activities of daily life, social support networks/assets, participation in productive activities, and substance use. For example, for the domain of reality testing there are numerous validated research-level measures available (e.g., BPRS, PANSS). Ancillary information may include a full, research-level assessment if available and if ustified by the purpose and resources, but the final decision about the domain score will be made by the clinician.
||Summary of Proposed Dimensional Examples:|
||ADHD using dimensional scales and parent-teacher reports|
||Dimensional approach with 28 facets and 4 overarching factors plus definition for general PD|
|Substance Use Disorders
||Bottom-up data-based development of new criteria to be dimensionalized using 3-point scale|
||Dimensional scale for panic attack based on Panic Disorder Severity Scale (PDSS) with staged approach|
||Dimensional approach for MDD based on PHQ-9|
||Dimensional approach to assess a variety of diagnostic-specific and treatment-relevant domains|
The full presentations were published in a monograph by American Psychiatric Publishing, Inc., which can be purchased here.