Development of a short form of the Geriatric Depression Scale-30 based on item response theory and the RiskSLIM algorithm.

Abstract

Recently, methods of quickly and accurately screening for geriatric depression have attracted substantial attention. Short forms of the 30-item Geriatric Depression Scale have been developed based on classical test theory, such as the GDS-4, GDS-5, and GDS-15, but they have shown low diagnostic accuracy. Therefore, in this study, we developed a new short form of the GDS-30 based on item response theory and the RiskSLIM, a machine learning method, and validated it based on gray matter volume. We found that the short form based on IRT (GDS-9) and the short form based on the RiskSLIM (GDS-14) had higher diagnostic accuracy than other short forms of the scale. In addition, in the Region of Interest based brain analysis, we found that the GDS-9 was significantly negatively correlated with the gray matter volumes of the right hippocampus, the right parahippocampal gyrus, and the right superior temporal gyrus, whereas the other short forms were not significantly associated with the gray matter volumes of any regions. This implies that the GDS-9 has higher empirical validity than other short forms and corresponds with brain structure. Therefore, the GDS-9 can be used to screen for geriatric depression and may improve the efficiency and accuracy of screening.