The latent variable "δ", can accurately diagnose dementia. Its generalizability across populations is unknown. We constructed a δ homolog ("dT2J") in data collected by the Texas Alzheimer's Research and Care Consortium (TARCC). From this, we calculated a composite d-score "d". We then tested d's generalizability across random subsets of TARCC participants and to a convenience sample of elderly Japanese persons with normal cognition (NC), mild cognitive impairment (MCI), and dementia (AD) (n = 176). dT2J was indicated by Instrumental Activities of Daily Living and psychometric measures. Embedded in this battery were the Mini-Mental Status Examination (MMSE) and an executive clock-drawing task (CLOX). Only MMSE and CLOX were available in both TARCC and the Japanese cohort. Therefore, a second composite variable, "T2J", was constructed solely from the factor loadings of CLOX and MMSE on d. The diagnostic accuracy of T2J was estimated in the validation sample, the remainder of the TARCC cohort, and in the Japanese sample. The areas under the receiver operating curve (AUC; ROC) for T2J were compared in each sample, and against d in TARCC. The AUCs for T2J were statistically indiscriminable within TARCC, and in Japanese persons. In Japanese persons, AUCs for T2J were 0.97 for the discrimination between AD versus NC, 0.86 for AD versus MCI, and 0.79 for NC versus MCI. The AUCs for T2J in Japanese persons were higher than any individual psychometric measure in that sample. Valid d-score composites can be abstracted from a subset of δ's indicators. Moreover, those composites are exportable across cultural and linguistic boundaries.