Evaluating hierarchical items of the geriatric depression scale through factor analysis and item response theory

• Main Authors: Nahathai Wongpakaran, Tinakon Wongpakaran, Pimolpun Kuntawong
• Format: Journal
• Published: 2019
• Subjects: Multidisciplinary
• Online Access: https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=85070525764&origin=inward; http://cmuir.cmu.ac.th/jspui/handle/6653943832/66750
• Institution: Chiang Mai University

© 2019 Background: Geriatric depression scale (GDS) is a common screening tool for measuring depression among older adults. It employs a multi-factor structure and some differential item functioning (DIF) allowing different versions of GDS across cultures. The present study aimed to identify the short version of the hierarchical scale of GDS in which all items comprised the invariant item ordering, and items without DIF. Methods: Participants and Measurement: A total of 803 participants, 70% female, with a mean age of 69.24 years (SD = 6.88) were enrolled from three geriatric units of tertiary care hospitals. All completed the 15-item GDS. Three methods of confirmatory factor analysis (CFA) with multiple indicators, multiple cause model, Mokken analysis and Rasch analysis were applied. Results: Item 9 (prefer to stay at home) showed poor discriminatory power among all three methods. After removing DIF items due to sex and age, nine items remained suitable for the shortened version by CFA. When Mokken and Rasch analysis were applied, only six items remained for the hierarchical scale. Compared with other related shortened version of GDS, the new GDS-6 proved to have a comparable ability to detect depression as did the original 15-item GDS. Limitation: The new GDS-6 needs to be investigated for test-retest reliability to ensure temporal stability of the scale. This cross-sectional analysis needs replication. Conclusion: The GDS-6 derived from IRT had measurement properties and met criteria related to unidimensionality and ability to separate levels of depression. It was shown to be equal to or better in predicting performance compared with the original 15-item GDS.