Measurement bias in depression assessment according to brain amyloid burden, cognitive function, and sex in the A4 study

Abstract

AbstractBackgroundWe sought to identify measurement bias (differential item functioning, DIF) in the Geriatric Depression Scale (15 item version) attributable to cerebral amyloid level (Aβ) among cognitively normal older adults. Specifically, we were interested in determining if endorsement of self‐reported memory problems on the GDS‐15 differed by Aβ level. We hypothesized that persons with high Aβ would have a greater probability of endorsing memory problems than those with low Aβ, consistent with the literature showing that subjective cognitive decline in older adults is associated with increased AD risk. We also examined DIF attributable to level of cognitive functioning and sex as positive controls that may account for differences in self‐reported depression.MethodAnalyses were conducted using the pre‐randomization database of the Anti‐Amyloid Treatment in Asymptomatic Alzheimer's Disease (A4)/Longitudinal Evaluation of Amyloid Risk and Neurodegeneration (LEARN) Study (N=3,961). We used the MIMIC (multiple indicators, multiple causes) approach to detect DIF. The MIMIC approach is consistent with item response theory, and allows for the detection of uniform DIF for predictors that are either discretely observed (sex) or continuously observed (Aβ level, cognitive function).ResultWe found DIF with respect to Aβ level, sex, and cognitive functioning. Most DIF detected was of a trivial effect size with some showing small‐to‐moderate effect sizes. Mild DIF, attributable to Aβ level, was shown for the GDS memory item in the hypothesized direction: after adjustment for depressive symptom severity (overall GDS symptoms), persons with high Aβ had a greater probability of endorsing memory problems than those with low Aβ (effect size = 0.22, p <.001). Interestingly, DIF for 2 items assessing hopelessness were also observed according to high Aβ versus low Aβ. Despite these findings we determined that there was negligible impact of DIF attributable to Aβ level on overall depressive symptom level.ConclusionParticipant responses on items of the GDS‐15 are sensitive to Aβ burden, sex, and level of cognitive functioning over and above what would be expected given the effect of those factors on depressive symptom severity overall. However, these direct effects are of small magnitude and do not appreciably impact the validity of inferences based on the GDS‐15.