| Many scientists have expressed concerns about the validity of psychiatric diagnoses. Because psychiatric illnesses are not directly observable and because latent class analysis has a number of beneficial properties (e.g., does not require symptoms to be continuously distributed or additive in the way they combine, aims for homogeneity within the classes), several authors have advocated the use of latent class analysis for the evaluation of psychiatric diagnostic criteria. The overall goal of the research was to further develop statistical methodology for latent class analysis in order to evaluate the value for improved diagnostic accuracy/precision of adding symptoms to diagnostic criteria. The research examined a Standard Procedure, which tested the null hypothesis that the candidate symptom had equal prevalence in all classes, and new statistics, which were designed to characterize and test value added for diagnosis. One new statistic focused on posterior probabilities and predictive precision/discrimination and the second focused on sensitivity/specificity and diagnostic accuracy. The statistics were investigated using simulations with differing levels of symptom and diagnostic prevalence and sample sizes. The results showed that the tests developed for the new statistics afforded considerably lesser power than the Standard Procedure to reject the hypothesis of equal symptom prevalence across classes, but the interpretation information that was obtained from the new statistics could aid clinical research because the statistics were grounded in interpretable standards of diagnostic accuracy, involving sensitivity, specificity and Area Under the Curve. In addition, the Standard Procedure was found to be sensitive to data sparseness so might not be appropriate to use to examine rare symptoms and diagnoses. The methods were also applied to data collected in a substantive study of child anxiety disorders. Three candidate symptoms were examined for inclusion in the diagnosis of social phobia: (1) inhibited behavior, (2) discomfort with disapproval/praise, and (3) fear of scolding/reprimand. Based upon the research, none were recommended for inclusion in the diagnosis of social phobia. |
