Greater than the sum of its parts: δ improves upon a battery's diagnostic performance

Objective: "δ", a latent variable constructed from batteries that contain both cognitive and functional status measures, can accurately diagnose dementia relative to expert clinicians. The minimal assessment needed is unknown. Methods: We validated 2 δ homologs in a convenience sample of elderly Japanese persons with normal cognition (NC), mild cognitive impairment (MCI), and dementia (n = 176). The latent δ homolog "d" (for dementia) was constructed from Instrumental Activities of Daily Living (IADL) and Japanese translationsof the Executive Clock-Drawing Task (CLOX), Frontal Assessment Battery (FAB), and Executive Interview (EXIT25). The latent delta homolog "d3" was constructed from a restricted set of d's factor loadings. Results: d and d3 were highly intercorrelated (r = .97) and strongly related to both IADL and dementia severity, as rated blindly by the Clinical Dementia Rating Scale (CDR). d was more strongly related to IADL and CDR than any of its indicators. In multivariate regression, d explained more variance in CDR scores than all of its indicators combined. d's areas under the receiver operating characteristic curve (AUC) were 0.95 for the discrimination between Alzheimer's disease (AD) vs. NC, 0.84 for AD vs. MCI and 0.81 for NC vs. MCI. d3's AUC's were statistically indiscriminable. These AUC's are higher than any of d's indicators, as reported recently by Matsuoka et al. (2014), as well as the Mini-Mental State Examination (MMSE), which had been made available by Matsuoka et al. to the CDR raters. Conclusions: Latent variables can improve upon a battery's diagnostic performance and offer the potential for accurate dementia case-finding after a minimal bedside assessment.