Abstract
PURPOSE: The Geriatric Depression Scale (GDS) is widely used to screen for depression in clinical practice and to assess symptoms of depression in research about older adults. To determine whether the 15-item GDS can be used in adults with dementia, this study investigated whether item- or test-level bias in the GDS-15 is associated with the respondent's level of cognitive impairment. METHODS: Using a large U.S. sample of 24,674 adults, we first conducted sample matching procedures between the five groups defined by CDR® Dementia Staging Instrument scores to control for potential confounding effects of common demographic variables. Then, we employed confirmatory factor analysis (for single-group and configural-invariance models only) and item response theory models to test potential differential item/test functioning effects associated with the GDS-15 across the five CDR groups. Practical consequences of the identified biases were quantified using sample-based Cohen's d effect sizes and misclassification rates. RESULTS: In general, people with higher CDR scores were older and had fewer years of education. In comparison to the normal cognition group (CDR-0), negligible biases in GDS-15 scores were found for the groups with questionable, mild, or moderate cognitive impairment (CDR-0.5/1/2). For individuals with severe cognitive impairment (CDR-3), their responses were inconsistent with the normal cognition group and their depression scores were significantly biased with a small-to-medium effect size. CONCLUSIONS: The GDS-15 can be used to assess depression in individuals with mild or moderate cognitive impairment, but not in individuals with severe cognitive impairment.