Rethinking the residual approach: leveraging statistical learning to operationalize cognitive resilience in...

Area of Science:

• Neuroscience
• Biostatistics
• Cognitive Science

Background:

• Cognitive resilience (CR) is the ability to resist cognitive decline despite Alzheimer's disease (AD) neuropathology.
• Measuring CR is challenging as it's an unobservable (latent) construct.
• The residual approach, using linear model residuals, is a common but potentially flawed method for estimating CR.

Purpose of the Study:

• To identify limitations of the standard residual approach for estimating cognitive resilience.
• To propose and validate a novel machine learning-based strategy for more accurate CR estimation.

Main Methods:

• Analysis of the assumptions and limitations of the linear residual approach for CR estimation.
• Development of a machine learning-based alternative strategy for CR measurement.
• Validation of the proposed method using simulated data with known ground-truth CR.

Main Results:

• The standard residual approach makes strong, often unmet, assumptions leading to biased CR estimates.
• The proposed machine learning approach demonstrates superior estimation accuracy on simulated data.
• The new method is less sensitive to confounding variables and data correlations.

Conclusions:

• The conventional residual method for cognitive resilience estimation is unreliable.
• Machine learning offers a more robust and accurate framework for quantifying cognitive resilience.
• This advancement has significant implications for understanding and potentially intervening in cognitive aging and AD.