A latent transition analysis for the assessment of structured diagnostic interviews

Structured diagnostic interviews administered by lay people are commonly used to assess psychiatric disorders, including depression, in large epidemiologic studies. Many interviews utilize "gate" questions, such as screening questions, that allow interviewers to skip entire survey sections for a particular respondent, saving time and reducing respondent fatigue. However, most depression estimates based on these response data are predicated on the assumption that the gate questions function without measurement error or bias. The tenability of this assumption is questionable, and its violation could compromise the reliability and validity of those estimates of depression. In this study, we used a novel application of latent transition analysis to cross-sectional data, accounting for measurement error in different response pathways through the depression module in the World Mental Health Composite International Diagnostic Interview. The analysis included data from 19,734 participants ≥18 years of age in the Comprehensive Psychiatric Epidemiologic Surveys. The latent transition analysis, allowing for measurement error in screening questions and exclusion criteria, produced a higher estimate of the lifetime probability of experiencing depression than did the algorithm based on the Diagnostic and Statistical Manual for Mental Disorders, 4th Edition, Text Revision. This illustration of latent transition analysis applied to item-level data from a complex structured diagnostic tool with gate questions demonstrates the potential utility of an analytic approach that does not automatically assume gate questions function without measurement error. This model could also be used to probe for evidence of measurement bias in the form of differential item function when using structured diagnostic tools in different cultures and languages.