Clinical Manifestations

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Remote neurocognitive assessment tools like the Visual Short-Term Memory Binding Task (VSTMBT) show promise for Alzheimer's Disease (AD) risk assessment.
• Machine learning (ML) can enhance these tools by providing insights into AD risk mechanisms.
• This study investigates the influence of self-reported variables on VSTMBT performance using ML and regression.

Purpose of the Study:

• To evaluate the impact of self-reported variables on Visual Short-Term Memory Binding Task (VSTMBT) performance.
• To identify predictors of weak-binding effects, indicative of potential pathology.
• To compare the efficacy of regression and machine learning models in AD risk assessment.

Main Methods:

• Participants were categorized as strong-binders (SB) or weak-binders (WB) based on binding-cost.
• Bivariate and multivariate regression analyses identified significant variables (p < 0.05).
• Machine learning models (SVM, RF, DT) were developed and validated using an 80/20% train-test split.

Main Results:

• Higher PDQ5 scores, multi-morbidity, and MHC were significantly associated with increased odds of WB effects.
• Consistent sleep quality showed a marginal association with lower odds of WB.
• Random Forest (RF) model achieved 81.0% accuracy, identifying overlapping risk factors with regression models.

Conclusions:

• Integrating ML with regression effectively identifies VSTMBT predictors for AD risk stratification.
• ML complements traditional regression methods in population-based AD assessments.
• Future research should focus on applying these models for individual prediction and early AD detection.