Abstract
BACKGROUND: Prevalence estimates for depression vary considerably by the type of assessment instrument, and there is limited information on their overlap in population-based samples. Our aim was to compare the Patient Health Questionnaire-9 (PHQ-9) with the Composite International Diagnostic Interview (CIDI) as measures for current major depressive syndrome (MDS) in a large population-based sample. METHODS: Data derived from the mental health module of the nationwide cross-sectional German Health Interview and Examination Survey for Adults (DEGS1-MH) (n = 4483; age 18-79 years). MDS in the past two weeks was assessed (a) using the PHQ-9 diagnostic algorithm (PHQ-MDS) and (b) based on CIDI information about the latest symptom occurrence (recency) (CIDI-MDS). Prevalences, overall concordance and percentages of overlap of both MDS measures were determined. Prevalences of affirmed PHQ-9 depression symptoms and the mean and median PHQ-9 sum scores were analyzed per measure. RESULTS: Prevalence of current MDS was 2.7% (95% CI: 2.0-3.6) for PHQ-MDS and 3.9% (95% CI: 3.1-5.0) for CIDI-MDS. The overall agreement between both measures was moderate (kappa: 0.43). Of all the participants, 1.5% (95% CI: 1.0-2.2) were classified as MDS cases by both measures, with 54.5% (95% CI: 42.7-65.9) of PHQ-MDS cases and 37.9% (95% CI: 27.8-49.1) of CIDI-MDS cases also being classified as MDS by the respective other MDS measure. However, 94.8% (95% CI: 93.6-95.8) of the participants were classified as non-MDS by both measures, with 97.5% (95% CI: 96.6-98.1) of non-PHQ-MDS and 98.7% (95% CI: 98.2-99.1) of non-CIDI-MDS being classified as non-MDS by the respective other MDS measure. The mean and median PHQ-9 sum score was higher in those with PHQ-MDS than in those with CIDI-MDS. CONCLUSIONS: Both measures have a high level of agreement for ruling out current MDS, but the overlap in their classification of cases is moderate. Our results indicate that they cannot be interpreted as equal measures of the same construct, suggesting limited comparability of their prevalence estimates. However, further exploration of algorithms and correlates and a proper labeling of measures in epidemiological studies are required.