Abstract
The authors have developed a neuro-psychological test for dementia based on working memory theory and item response theory, and the test, named TKW test, proved useful for diagnosis of severity and etiology of dementia. In the present study, the authors tried to construct short and concise versions of the test to increase its practical utility in a clinical setting. A new sample of patients was added to the last sample, and the same results were confirmed using the same analytical procedures as in the last study : high agreement between item difficulty estimated from obtained scores and working memory demand, test homogeneity as a prerequisite for application of item response theory, high discriminant power between dementia of the Alzheimer type and vascular dementia, and high concurrent validity with several tests widely used today. Based on these results, six concise versions were constructed in line with the following five selection criteria : item information function, first factor loadings, correlation coefficients of structural matrices in discriminant analysis, item discriminating power, and item difficulty. All concise versions were effective for measurement of severity, and one version, named B2, was the most effective for discrimination between etiological categories. In addition, single versions of the test using only one item were also examined in terms of merit of item response theory. The above five selection criteria plus the maximum working memory demand were used to select items for this version. One item, named hierarchical classification 1, was useful for measurement of severity, and another item, named pointing out the odd kanji of the group, was accurate for discriminating between etiological categories, although these items were necessarily less effective than the whole test because of loss of the total amount of information.
