Rethinking the residual approach: leveraging statistical learning to operationalize cognitive resilience in Alzheimer’s disease

Rethinking the residual approach: leveraging statistical learning to operationalize cognitive resilience in Alzheimer’s disease

Abstract Cognitive resilience (CR) describes the phenomenon of individuals evading cognitive decline despite prominent Alzheimer’s disease neuropathology. Operationalization and measurement of this latent construct is non-trivial as it cannot be directly observed. The residual approach has been wide...

Full description

Saved in:
Bibliographic Details
Main Authors: Colin Birkenbihl, Madison Cuppels, Rory T. Boyle, Hannah M. Klinger, Oliver Langford, Gillian T. Coughlan, Michael J. Properzi, Jasmeer Chhatwal, Julie C. Price, Aaron P. Schultz, Dorene M. Rentz, Rebecca E. Amariglio, Keith A. Johnson, Rebecca F. Gottesman, Shubhabrata Mukherjee, Paul Maruff, Yen Ying Lim, Colin L. Masters, Alexa Beiser, Susan M. Resnick, Timothy M. Hughes, Samantha Burnham, Ilke Tunali, Susan Landau, Ann D. Cohen, Sterling C. Johnson, Tobey J. Betthauser, Sudha Seshadri, Samuel N. Lockhart, Sid E. O’Bryant, Prashanthi Vemuri, Reisa A. Sperling, Timothy J. Hohman, Michael C. Donohue, Rachel F. Buckley
Format: Article
Language:English
Published: SpringerOpen 2025-01-01
Series:Brain Informatics
Subjects:
Cognitive resilience
Alzheimer’s disease
Dementia
Machine learning
Artificial intelligence
Cognitive reserve
Online Access:https://doi.org/10.1186/s40708-024-00249-4
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Cognitive resilience (CR) describes the phenomenon of individuals evading cognitive decline despite prominent Alzheimer’s disease neuropathology. Operationalization and measurement of this latent construct is non-trivial as it cannot be directly observed. The residual approach has been widely applied to estimate CR, where the degree of resilience is estimated through a linear model’s residuals. We demonstrate that this approach makes specific, uncontrollable assumptions and likely leads to biased and erroneous resilience estimates. This is especially true when information about CR is contained in the data the linear model was fitted to, either through inclusion of CR-associated variables or due to correlation. We propose an alternative strategy which overcomes the standard approach’s limitations using machine learning principles. Our proposed approach makes fewer assumptions about the data and CR and achieves better estimation accuracy on simulated ground-truth data.
ISSN:2198-4018
2198-4026

Similar Items