Relative accuracy of three automated systems for neuropsychological interpretation.

Abstract

Three computer programs for automated interpretation of variants of the Halstead-Reitan Neuropsychological Test Battery were tested on cases in two data sets. The Key approach of Russell, Neuringer, and Goldstein (1970), Brain I (Finkelstein, 1977), and Adams' (1975) ability-based algorithm were employed in the study. The first data set included 63 well-documented cases with precise criterion data and multiple sources of direct verification. The second data set consisted of 30 equally well-studied cerebrovascular disease patients whose cerebral circulation disorders resulted in clinical manifestations encompassing the entire range of stroke. Results suggested that, while none of the programs do poorly at identifying the presence of brain damage, lateralization and possibly other localization/process predictions are not done well by these programs. The failure described in this particular study does not imply that automated methods are potentially less effective than true actuarial or clinical ones. Rather, we suggest that the translation process from clinical interpretation to the mechanical combinatory logic of the digital computer is at an early stage.