Background: Two-phase diagnostic surveys are popular in psychiatric epidemiology. The Geriatric Depression Scale (GDS) and the General Health Questionnaire (GHQ) are commonly used to screen in older and younger populations, respectively.

Method: In Phase I, in this Brazilian population-based study, we screened 392 participants aged ≥ 75 years. In Phase II, half of those scoring ≥ 11 in the GDS and ≥ 4 in the GHQ and 20% of others were selected for detailed evaluation with ICD-10 diagnoses assessed by the Schedules for Clinical Assessment in Neuropsychiatry (SCAN) (n = 126).

Results: Internal consistencies were good for all scales. At the optimal cut-off points GDS-30 (14/15) yielded 73% sensitivity and 65% specificity, while classical GHQ (4/5), Likert (15/16) and c-GHQ (5/6) yielded 66–75% sensitivity and 57–62% specificity. Receiver operating characteristic (ROC) curve analysis against SCAN indicated only moderate screening potential [area under the ROC curve (AUROC): GDS = 0.76; classical GHQ = 0.74; Likert = 0.76; c-GHQ = 0.73], with no statistically significant differences. All measures were biased by disability and self-reported health.

Conclusions: In this study, neither screen was sufficiently predictive of SCAN ICD-10 diagnosis to recommend their use in two-phase surveys. Despite its theoretical advantages, the GDS-30 performed no better than the GHQ-12, and was biased in similar ways.