Revisiting the Geriatric Depression Scale: An IRT-Based 10-Item Screen Outperforms the GDS-15 in Diagnostic Accuracy and Efficiency

Abstract

Background/Objective: Existing abbreviated Geriatric Depression Scales (GDSs), derived via Classical Test Theory (CTT), often sacrifice accuracy for brevity and retain non-specific items. We aimed to develop a minimum-item GDS maintaining diagnostic performance equivalent to the full 30-item scale (GDS30) using Item Response Theory (IRT). Methods: This cross-sectional study employed rigorous 5:5 split-sample cross-validation. Participants included 6525 older adults (aged >= 60 years) from community-based (Korean Longitudinal Study on Cognitive Aging and Dementia) and clinical settings (geropsychiatry clinic). Depression was diagnosed through standardized clinical interviews based on DSM-IV criteria. Two-parameter logistic IRT models estimated item discrimination and difficulty parameters. Sequential item reduction with DeLong tests identified the minimum number of items required to maintain GDS30-equivalent area under the curve (AUC). Results: The 10-item IRT-optimized scale (GDS10-IRT) achieved an AUC of 0.856 (95% CI: 0.809-0.895) in the validation set, showing no significant difference from GDS30 (AUC = 0.883; p = 0.396). Conversely, the 15-item GDS (GDS15) demonstrated significantly lower AUC than GDS30 (p < 0.001) despite having more items. GDS10-IRT achieved a 234% improvement in efficiency ratio (AUC/items) over GDS30. Notably, Item 16 ("feeling downhearted and blue"), identified as the most discriminating symptom (a = 2.53), is absent from the GDS15 but included in GDS10-IRT. Conclusions: IRT-based item selection achieves GDS30-equivalent diagnostic accuracy with only 10 items, outperforming the widely used GDS15. By recovering high-discrimination items excluded by CTT, the GDS10-IRT offers a more efficient, specific screening tool for late-life depression.