Evaluating diagnostic criteria: A latent class paradigm

https://doi.org/10.1016/0022-3956(82)90007-3Get rights and content

Abstract

A latent class analysis paradigm for evaluating and developing diagnostic criteria was presented. How latent class analysis statistically models the conceptual structure of diagnostic systems was demonstrated. These models were then used to study the Taylor and Abrams diagnostic criteria for schizophrenia. The findings of this analysis were used in reanalyzing a previous study, yielding improved results. Finally, the nature of internal and external construct validity, and the particular appropriateness for diagnostic research of models with discrete variables was discussed.

References (21)

There are more references available in the full text version of this article.

Cited by (79)

  • The Vascular Depression Subtype: Evidence of Internal Validity

    2008, Biological Psychiatry
    Citation Excerpt :

    The method used in this study also has important implications for the process of establishing valid diagnoses. Sneed et al. (16) articulated the psychometric approach to diagnostic validity (39–41), which is represented in Figure 1. According to this approach, the first step in establishing diagnostic validity is the specification of a theory.

  • Vascular Depression: A Distinct Diagnostic Subtype?

    2006, Biological Psychiatry
    Citation Excerpt :

    Of course, the researcher is more interested in finding the probabilities of latent class membership given a particular response set on the manifest indicators, and these reverse conditional probabilities can be easily obtained with Bayes’ theorem and are standard in most statistical packages with a LCCA module (e.g., Mplus, Latent Gold, LEM). Importantly, LCCA allows the researcher to evaluate the relative accuracy of competing diagnostic criteria in the absence of a “gold standard” (Faraone and Tsuang 1994; Young 1983). In other words, LCCA can be used to determine whether multiple definitions of a construct (e.g., vascular MDD) reflect a single underlying latent class or separate diagnostic entities.

  • Hui and Walter's latent-class reference-free approach may be more useful in assessing agreement than diagnostic performance

    2005, Journal of Clinical Epidemiology
    Citation Excerpt :

    Observer responses were unconditionally dependent, which can be regarded as a prerequisite for a meaningful latent class analysis. Indeed, the absence of unconditional dependence would suggest that observers did not identify the same latent diagnostic classification, thus precluding any useful interpretation of results from latent class analysis, as argued by Rindskopf and Rindskopf [13] and by Young [37]. The issue of conditional dependence in the framework of latent class models has been considered by several authors who proposed introducing either combinations of variables or covariance terms among observers in their models [7,14,25,34,38].

View all citing articles on Scopus

The author wishes to thank David Clark, Jan Fawcett, Martin Harrow, William Scheftner and Martin Tanner for their valuable comments.

View full text