Evaluating the Beck Depression Inventory-II Fast Screen in middle-aged and older adults using random forests and factor analytic methods.

The Beck Depression Inventory-Fast Screen (BDI-FS) consists of seven items from the Beck Depression Inventory-II and is used to screen for depression in individuals with medical illness. The BDI-FS omits somatic items common in physical health conditions. However, these items may more accurately predict depression for some individuals, and their utility has not been empirically evaluated. This study aimed to (a) evaluate the convergent validity for the Beck Depression Inventory-II and BDI-FS within an ethnoculturally heterogeneous sample of middle-aged and older adults, including individuals with chronic medical conditions and (b) identify the subset of items from the Beck Depression Inventory-II that best predict current major depressive episode. We hypothesized that an alternate, empirically derived model would outperform the existing BDI-FS. Participants included 186 adults (age M = 65.66 years, SD = 6.36) enrolled in an observational research study. Two random forests (RF)-derived scales were compared to the existing BDI-FS in predicting current major depressive episode. Scales were compared using confirmatory factor analysis, measurement invariance, and receiver operating characteristic analysis. Overall, results were comparable across scales. Models based on RF-selected items did not demonstrate better model fit than those based on the BDI-FS; however, scores derived from the RF-selected items exhibited higher reliability. In addition, models based on the RF-derived scales and the BDI-FS showed invariance across demographic and clinical variables. Scores derived from all scales, including RF-derived scales and BDI-FS, showed excellent classification performance in receiver operating characteristic analysis. Future research should further investigate these RF-derived scales, with the BDI-FS, to evaluate their reliability and validity across other diverse populations. (PsycInfo Database Record (c) 2026 APA, all rights reserved).