Explainable Machine Learning Reveals Time-Dependent Cognitive Risk in Minor Neurocognitive Disorder: Implications for...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Gerontology

Background:

• Minor neurocognitive disorder (NCD) is a critical, modifiable stage before dementia.
• Early identification of individuals at risk of progression is crucial for intervention.
• Preventive strategies can delay functional decline in aging populations.

Purpose of the Study:

• Develop a transparent machine learning (ML) framework to predict progression from minor to major NCD.
• Utilize baseline demographic, clinical, and neuropsychological data for prediction.
• Enhance early detection for timely health promotion interventions.

Main Methods:

• Retrospective analysis of 162 memory clinic patients.
• Employed nested stratified cross-validation and SMOTE for data balancing.
• Evaluated logistic regression, SVM, and XGBoost, using SHAP for interpretability.

Main Results:

• Support Vector Machines (SVM) demonstrated strong predictive performance (12 months: accuracy=0.90; 24 months: accuracy=0.81).
• Short-term progression linked to cognitive inefficiencies; longer-term risk associated with cognitive vulnerability and metabolic factors like diabetes.
• Identified distinct cognitive risk signatures for short- and long-term progression.

Conclusions:

• Explainable ML serves as a valuable health promotion tool for early NCD detection.
• ML frameworks can uncover clinically meaningful cognitive risk signatures.
• Supports precision prevention and proactive monitoring for individuals with minor NCD.