Applied Bayesian Modelling, 2nd Edition

Latent categories may be useful in medical diagnosis where the observed indicators are various diagnostic tests or criteria of illness (or different clinicians), with none being certain or ‘gold standard’ indicators of the presence of a disease. One may assess whether a single latent categorisation (e.g. if the latent categories might be ill vs. not ill) underlies several observed binary diagnostic items, none of which provide a ‘gold standard’ test (Rindskopf, 2002).

Example 9.7 Psychiatric caseness

As an example of latent class analysis applied to clinical diagnosis, consider responses to three binary diagnostic items for patients (Dunn, 1999). The binary responses are taken by dichotomising more extensive scales, and are denoted CIS (Clinical Interview Schedule), GHQ (General Health Questionnaire) and HADS (Hospital Anxiety and Depression Scale). Categories 1 of CIS, GHQ and HADS define less ill patients and categories 2 of CIS, GHQ and HADS define more ill patients. Let denote the observed counts of patients in category of CIS, of GHQ, and of HADS.

A latent class model is ...