Dylan J Jester1, Ross Andel1,2,3, Katerina Cechová2,3, Jan Laczó2,3, Ondrej Lerch2,3, Hana Marková2,3, Tomás Nikolai2,3, Martin Vyhnálek2,3, Jakub Hort2,3. 1. School of Aging Studies, University of South Florida, Tampa, FL33612, USA. 2. Department of Neurology, Memory Clinic, 2nd Faculty of Medicine, Charles University and Motol University Hospital, Prague, 150 06, Czech Republic. 3. International Clinical Research Center, St. Anne's University Hospital Brno, Brno, 656 91, Czech Republic.
Abstract
OBJECTIVE: To compare cognitive phenotypes of participants with subjective cognitive decline (SCD) and amnestic mild cognitive impairment (aMCI), estimate progression to MCI/dementia by phenotype and assess classification error with machine learning. METHOD: Dataset consisted of 163 participants with SCD and 282 participants with aMCI from the Czech Brain Aging Study. Cognitive assessment included the Uniform Data Set battery and additional tests to ascertain executive function, language, immediate and delayed memory, visuospatial skills, and processing speed. Latent profile analyses were used to develop cognitive profiles, and Cox proportional hazards models were used to estimate risk of progression. Random forest machine learning algorithms reported cognitive phenotype classification error. RESULTS: Latent profile analysis identified three phenotypes for SCD, with one phenotype performing worse across all domains but not progressing more quickly to MCI/dementia after controlling for age, sex, and education. Three aMCI phenotypes were characterized by mild deficits, memory and language impairment (dysnomic aMCI), and severe multi-domain aMCI (i.e., deficits across all domains). A dose-response relationship between baseline level of impairment and subsequent risk of progression to dementia was evident for aMCI profiles after controlling for age, sex, and education. Machine learning more easily classified participants with aMCI in comparison to SCD (8% vs. 21% misclassified). CONCLUSIONS: Cognitive performance follows distinct patterns, especially within aMCI. The patterns map onto risk of progression to dementia.
OBJECTIVE: To compare cognitive phenotypes of participants with subjective cognitive decline (SCD) and amnestic mild cognitive impairment (aMCI), estimate progression to MCI/dementia by phenotype and assess classification error with machine learning. METHOD: Dataset consisted of 163 participants with SCD and 282 participants with aMCI from the Czech Brain Aging Study. Cognitive assessment included the Uniform Data Set battery and additional tests to ascertain executive function, language, immediate and delayed memory, visuospatial skills, and processing speed. Latent profile analyses were used to develop cognitive profiles, and Cox proportional hazards models were used to estimate risk of progression. Random forest machine learning algorithms reported cognitive phenotype classification error. RESULTS: Latent profile analysis identified three phenotypes for SCD, with one phenotype performing worse across all domains but not progressing more quickly to MCI/dementia after controlling for age, sex, and education. Three aMCI phenotypes were characterized by mild deficits, memory and language impairment (dysnomic aMCI), and severe multi-domain aMCI (i.e., deficits across all domains). A dose-response relationship between baseline level of impairment and subsequent risk of progression to dementia was evident for aMCI profiles after controlling for age, sex, and education. Machine learning more easily classified participants with aMCI in comparison to SCD (8% vs. 21% misclassified). CONCLUSIONS: Cognitive performance follows distinct patterns, especially within aMCI. The patterns map onto risk of progression to dementia.