Concordance between Composite International Diagnostic Interview and self-reports of depressive symptoms: a re-analysis.

Concordance between sum scores of self-reported depressive symptoms and structured interview diagnoses has been studied extensively, but are these the best attainable self-report-based predictions for interview diagnoses? We maximized the cross-validated concordance between World Health Organization's Composite International Diagnostic Interview (CIDI) diagnosis and Beck's Depression Inventory (BDI), and General Health Questionnaire (GHQ), from the viewpoint of exploratory statistics, re-analysing Health 2000 general-population sample of adults over 30 years in mainland Finland (N = 5200-5435). BDI sum-score prediction of CIDI diagnosis could be superseded by using (1) weighted sums of items, (2) classification trees constructed from items, or (3) a single item. Best solution (2) yielded cross-validated Youden's Index 0.757 [standard error (SE) = 0.001, sensitivity = 0.907, specificity = 0.851], improving the concordance to 1.07-fold (1.18-fold for 12-month diagnosis). A single-item solution was best for the GHQ. All positive predictive values remained low (0.09-0.31). Thus, CIDI-to-questionnaire concordance can be improved by using all information in the questionnaires instead of just sum scores, but latent-trait theory for questionnaires is incompatible with interview diagnoses (single item achieved better concordance than summing all). Self-reports have low predictive value for CIDI diagnoses in the general population, but better in settings with higher major depressive disorder (MDD) base rates.