It is a melancholy truth that even great men have their poor relations.

Nature-morte aux trois crânes, Paul Cézanne, 1900.

Diagnostic consensus should reflect empirically derived data as well as experience. In the former effort, syndromes are identified as they seem to occur naturally among large groups of patients or in community samples. Rather than assigning each patient to a category predetermined by expert consensus, the sample’s inherent groupings are determined by statistical modeling and then given descriptive names.

Psychopathology and other clinical features expressed by each subject are systematically recorded and that database subjected to statistical analyses designed to reveal ‘‘patterns’’ of signs and symptoms. A well-defined pattern is assumed to represent a coherent condition when it is supported by mathematical and logical assumptions.

The validity of the pattern, however, must be tested against external variables not used to generate the patterns. Different statistical techniques have been used. Factor, discriminant function, cluster, and latent class analyses are among the strategies employed. The identified
groups are observed for treatment response, laboratory findings, family history, and other independent criteria to determine if the groups remain distinct.

Such an empirical process begins, however, with preconceptions. The patient groups are typically hospitalized psychiatric patients who were first clinically diagnosed by the expert consensus system. The starting point is almost always the ‘‘psychiatric patient,’’ and the search for patterns of depressive illness typically begins with patients labeled ‘‘depressed’’ by other systems. Examining non-patient community samples partially overcomes this bias, but these studies are few and have their own limitations. Thus, the empirical strategy is a refinement of the expert consensus, not a distinctly diVerent approach.

The empirical technique also has methodological limitations. The quality of psychopathology rating scales is uneven. The items to be measured and their ranges are arbitrary. The degree of interrater reliability and the raters’ clinical skills influence results. In community sample studies, the training of the assessors is often poor compared to the quality and training of raters of hospital-based patients. The instruments used in community studies, while psychometrically sophisticated, are phenomenologically simplistic. They ignore the duration, sequence of appearance, and severity relationships among symptoms.

Ascertainment bias influences results. Psychotic features are more commonly recognized in hospitalized depressed patients than in outpatients. A study of outpatients might not identify psychotic depression as readily as an analysis of inpatients. Differences in samples and sampling, clinical instruments used, the skills of the examiners, and the statistical methods employed are better explanations for differences in patterns identified across studies than any inherent differences in psychopathology in the samples.