Related Experiment Video
Updated: Mar 9, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Determining optimal impairment rating methodology for a new HIV-associated neurocognitive disorder screening
Jody Kamminga1, Mark Bloch2, Trina Vincent2
1a Neuroscience Research Australia , Randwick , NSW , Australia.
Introduction:
The current study sought to determine the optimal impairment rating definition for a new HIV-associated neurocognitive disorder (HAND) screening procedure as compared to standard neuropsychological testing.
Method:
A total of 55 HIV-infected (HIV+; 19% AIDS; 87% on combination antiretroviral therapy, cART; 80% plasma undetectable) and 22 demographically comparable HIV-uninfected (HIV-) control adults (all male) enrolled in an urban Australian primary care cohort study completed the CogState computerized cognitive screen, a standard independence in activities of daily living (ADL) questionnaire, and a standard neuropsychological test battery. Local or American demographically adjusted norms were applied to the neuropsychological data, taking into account premorbid reading level. CogState norms that corrected for age and sex were applied to raw CogState data. The HAND Frascati classification criteria were implemented to determine "impairment" on both the standard neuropsychological test and the CogState-based screen using two established methods: a battery-wide summary score (global deficit score; GDS), and cognitive domain rating, both combined with ADL independence data. Criterion validity was operationalized by comparing rate of impairment derived from the CogState-based screen to that obtained from the standard neuropsychological test battery first in the combined HIV- and HIV+ sample, and then in the HIV+ sample only.
Results:
In the combined sample, CogState-based screen criterion validity was higher using the GDS (79% correct classification, 73% sensitivity, 82% specificity) over the cognitive domain rating (correct classification, sensitivity, specificity all 69%) method. A similar pattern was found for the HIV+ group separately [GDS (74% correct classification, 76% sensitivity, 71% specificity) versus cognitive domain rating (64% correct classification, 72% sensitivity, 57% specificity)]. The cases that resulted in disagreement across the two methods were those with borderline impaired/unimpaired cognitive performance.
Conclusions:
The GDS is a relatively easy statistical method for computing impairment rate when using the CogState-based screen that yields adequate criterion validity compared to standard neuropsychological testing. Feasibility of standardized test administration and appropriate interpretation of results for this CogState-based screen in primary care was enhanced by the training and consultation provided by study neuropsychologists.
