Mental health articles

Classification research for the development of classification schemes requires a framework for describing the approach to and evaluating classif ication schemes. In evaluation, there are four major aspects that are very important in producing a reliable and valid classification system:

1.The classification system should be reliable and consistent from user to user (interrater agreement) and within the same user over different time periods (stability).

2.The system should describe the domains of behaviors it encompasses. This aspect of a classification system always involves a trade-off between the extent of coverage and the precision of measurement. A good example of broad coverage is the series of APA Diagnostic and Statistic Manuals that include behavioral disorders and organic conditions, as well as numerous mental disorders.

3.The system should have descriptive validity, which is the degree to which the categories within it are homogenous with respect to their relative attributes (e.g., symptoms, personality constructs, biographical data, and other defining attributes of the categories).

4.The system should have predictive and clinical validity, which addresses the question of whether classification predicts other variables and whether the predicted variables are clinically significant, respectively. As a framework for clinical research, the threestates paradigm described by Skinner, evolving from the work of Loevinger, is a useful model and is shown in Figure

1. The theoretical formulation should lead to hypotheses that are open to empirical evaluation so that the classif ication system can be shown capable of scientific evaluation.

As Popper has noted, a scientific theory is only useful if it suggests repudiation rather than truth. Thus, a classification system should include a precise definition of each type of functional relationship among the various subtypes (i.e., a nomological network), a method of explicating the development and etiology of the disorders, a description of the prognosis, the appropriate treatment interventions, and a discussion of the population for which the classification occurs. Irrefutable theories and classification systems that cannot be evaluated are of little value to the scientific clinician. The content domain should describe clearly the variables on which the classification is based. In other words, is the individual or the behavior placed in a category on the basis of clinical symptoms (aberrant behavior), the course of the disorder, its etiology, or some combination of these variables? The two major competing classification variables are behavior and etiology.

However, any system that uses both variables to classify a disorder hopelessly confounds the issue of whether a certain etiological factor causes a specific type of deviant behavior. There is considerable argument that the use of etiology as a method of classifying behavior is premature and that the description of behavior is most appropriate for the current state of the science. The validity of this argument has been recognized in DSM-III, III-R, and IV, which reversed the trend toward classification by etiology in favor of more descriptive, observational notions. Klerman, Vaillant, Spitzer, and Michels  clearly indicated the reasons for the reversal, stating that the reliance on descriptive rather than etiological criteria ‘‘represents a strategic mode of dealing with the frustrating reality that, for most of the disorders we treat, there is only limited evidence for their etiologies.’’

The obvious solution would be to construct one classification system for abnormal behavior and a second for etiology, in accordance with Hempel’s  view of logical empiricism, which holds that we can claim to understand concepts scientifically only if we are able (1) to describe them and (2) to explain or predict them through general laws and theories. An example of etiological classification is Zubin and Spring’s description of six approaches to the classification of schizophrenia: the ecological, developmental, learning, genetic, internal, and neurophysiological models.When using two classification systems, it is possible to demonstrate by research that a specific behavior type is associated with specific kinds of etiology. This type of research leads to refinements in classification systems. As a matter of fact, a multiaxial system—similar to that introduced in DSM-III, in which the clinical syndrome is diagnosed separately from its etiology—would be one possible step toward that end. Nevertheless, it must be emphasized that scientific documentation rather than theoretical speculation is the only method by which a classif ication scheme can use both etiology and behavior (or symptoms) in a classification system. There are basically three types of structural models that can be used to develop a classification system. The first is the hierarchical model that classifies individuals into subsets that are themselves successively classified into groups at a higher level in the hierarchy. One example is schizophrenia, which in the figure is subdivided into acute or chronic types and then further subdivided into various subtypes, like hebephrenic (disorganized), catatonic, and paranoid, where each category is distinct from other subcategories but all share characteristics at higher levels. For instance, all subsets of schizophrenia would include the defining attributes of schizophrenia, such as disorders of cognition. Another example is the four-tiered system of Foulds and Bedford, which begins with neurosis and personality disorders and progresses up the hierarchy to depression, mania, and paranoia; schizophrenia; and organic brain syndromes. These classes are ordered in terms of increasing severity and assume that a person with symptoms of a given class necessarily has the symptoms of all of the lower classes in the hierarchy. Another approach is to use categorical systems, which assume that individuals can be classified into discrete classes that are seen as internally coherent but distinct from each other. This was the primary approach adopted in DSM-III, DSM-IIIR, and DSM-IV. An example of this type is the cluster analysis depicted in Figure

3. The third method is the dimensional system, which orders individuals along axes in a multidimensional scale. This has been the most common method employed in personality research, perhaps because the literature in this area has failed to establish any sharp boundaries between personality styles or between adaptive and maladaptive traits. An example of this approach is the threedimensional model of DSM-III personality disorders developed by Widiger et al., which identified these disorders along the dimensions of social involvement, assertiveness, and acting-out behavior. A more recent alternative is the translation of the DSM-IV personality disorders along the dimensions of the five-factor model of personality. Dimensional models provide more flexible, specific, and comprehensive information, whereas categorical models tend to be Procrustean, to lose information, and to result in many classificatory dilemmas for borderline cases. If one accepts these advantages, however, one is still confronted with the difficult question of which dimensions to choose in defining these disorders. Figure 4 demonstrates one option. Still another dimensional approach is the circumplex, a circular model in which disorders that appear across from each other are considered opposites, whereas disorders that are next to each other are considered similar. They further discussed how the classical view assumes a particular relationship between categories. For example, if two categories have a subset relationship, the defining features of the more abstract category are included or nested in those of the subset. To illustrate, the defining features of schizophrenia are also included in the subset of paranoid schizophrenia. Thus, in a categorization system, according to the classic view, the features of any higher level category should occur in every subset of that category. Cantor and her colleagues pointed out that there are a number of difficulties with this approach. They claimed that the classical view suffers from a failure to specify defining features for most categories, the existence of borderline cases, the heterogeneity of the members of a category with respect to their features, the variation among category members with respect to their typicality and ease of categorization, and less than perfect nesting of a general category’s features among its subsets. What Cantor et al. suggested as a solution to these problems is prototype categorization, where an ideal type is specified in terms of its features, but it is assumed that the features in most cases need only be similar, not necessarily identical. Attributes need only be correlated with category membership and need not be necessary and sufficient. In other words, if there are four features of a category such as histrionic personality disorder, an individual could have two, three, or four of the features and still be labeled histrionic. Thus, the features are not defining but correlational. This type of approach, as Cantor and her colleagues mentioned, eliminates the existence of borderline cases and causes heterogeneity of categories. It does not require perfect nesting of a general category’s features in the subset, and, they claimed, prototypes make sense of variations in typicality. This point was expanded upon by Schwartz and Wiggins, who stated that by depicting the perfect case, the ‘‘ideal types’’ permit us to pinpoint in specific terms the areas that require further inquiry. For any given disorder, deviations from the pure case—the ideal case—should spur the clinician to focus on these omissions (or additions) as indicative of the uniqueness of the individual within the framework of the general disorder. In short, framed in terms of ideal types, the clinician’s inquiry is given shape and direction in a heuristic manner, without regard to a truth value, in a practical effort to understand a person’s problems. This lack of truth or falsity, however, makes prototypes (or ideal types) problematic from a scientific point of view, because it may violate Popper’s doctrine of the ability to falsify a hypothesis. Furthermore, the fact that the criticisms by Cantor et al. of the traditional view (existence of borderline cases, inability to specify defining features, the heterogeneity of the members of a category, and less than perfect nesting of a general category’s features) are in themselves the criteria used to evaluate classification systems that further violate Popper’s doctrine. Schwartz and Wiggins disputed this criticism, stating that the ideal type guides the clinician to particular beliefs about an individual that are themselves falsifiable with respect to that particular patient. In essence, however, advocates of this approach are abandoning the logical empiricism that has guided science for more than 100 years in favor of practical considerations. This approach aggravates the situation described by Birley, who maintained that psychiatry is already littered with a mixture of irrefutable theories, including classification systems, that explain a great deal and refutable theories that explain very little. More complex models of classification may solve the problem that some behavioral disorders differ in kind and others in degree from normal behavior. The multiaxial diagnostic system, for example, is a hybrid model that allows evaluating an individual in terms of several different domains. Each of the domains is assessed quasiindependently of the others, but together they represent a more comprehensive evaluation than when an evaluation is limited to one mental disorder. The use of various domains permits a class-quantitative structure in which quantitative dimensions are superimposed on qualitative categories. For example, Axes I and II of the DSM-IV allow differentiations on qualitative dimensions, whereas Axes IV and V (which indicate severity of psychosocial stressors and global assessment of functioning) are quantitative. Another interesting approach is the Radex that integrates dimensions and spheroids. An example is the vector model of disease described by Sneath: Disease states are represented by a swarm of points around the origin (which represents health) in dimensional space. As people become ill, they move away from the origin; the length of the vector (or the distance from the origin) indicates a symptom’s severity, and the direction of the vector indicates the type of disease. In terms of the internal validation) of a classification scheme, some decision should be made about the appropriate statistical techniques, usually such multivariate statistical methods as factor or cluster analysis. If the disorder is assumed to be categorical, then Meehl’s taxonomic methods should be used to establish this fact empirically. There are a number of other methods that can be used, but as Skinner  indicated, there is no real way of evaluating which statistical procedure is more appropriate. Indeed, few would disagree that thus far factor analysis and discriminant functions have yielded few new insights. However, Zubin reminded us that statistical treatment lies in the realm of verification rather than discovery and that therefore statistics are still important to scientists.

In any case, internal validation would also involve replicating the types in other samples of the population and evaluating the reliability, the homogeneity, and the coverage of the system. External validation in this framework would involve studies of the generalizability of types to the overall population, the predictive validity of the categories, the descriptive or content validity of the disorders, and the clinical usefulness of the system.

Tags: abnormal, abnormal behavior, behavior, classification