CONDITIONAL STANDARDS: IDENTIFYING CUTOFFS FOR PREDICTING AD SURROGATES USING TRADITIONAL AND MACHINE LEARNING METHODS

Abstract

divided based on the traditional classification of MCI (single domain amnestic MCI, multidomain amnestic MCI and non-amnestic MCI). Eight tests, assessing executive functions were administered at baseline and at 1-year-follow-up. Screening tests of cognitive and functional abilities were also used. A new dysexecutive MCI classification was developed relying on k-means cluster analysis through which three clusters were identified. Baseline data entered simple lineal regression models to examine whether such a classification based on executive profiles could significantly predict progression to AD dementia a year later. Results: The dysexecutive classification accounted for 63% of the variance linked to MCI to AD conversion even when controlling for the severity of disease at baseline (F(1, 68) 1⁄4 116.25, p1⁄40.000, R1⁄40.63). Such a prediction power was not observed when the classical MCI classification based on memory profiles alone entered the model as a predictor (F(1, 68) 1⁄4 5.09, p1⁄40.955, R1⁄40.07). Conclusions: Considering dysexecutive profiles of MCI patients may increase the accuracy of prediction models aimed at detecting risk of progressing to AD dementia. MCI patients with worse performance on executive tests seem to hold a higher risk of conversion and such a risk seems to be accounted for neither by memory impairments nor by the severity of the disease at baseline.