AI-Enabled Modeling for Alzheimer's Disease Risk Prediction and Validation

Area of Science:

• Neurology
• Biostatistics
• Machine Learning in Healthcare

Background:

• Alzheimer's disease (AD) poses a significant challenge, necessitating early detection and intervention strategies.
• Identifying multimodal clinical factors influencing AD onset is crucial for risk stratification.

Purpose of the Study:

• To investigate multimodal clinical factors associated with Alzheimer's disease (AD) onset.
• To develop and validate a risk prediction tool for early AD intervention and classification in high-risk individuals.

Main Methods:

• A retrospective cohort of 502 high-risk individuals was analyzed.
• Machine learning models (Random Forest, XGBoost, Deep Learning) were trained and validated.
• Multivariate logistic regression identified independent risk and protective factors.

Main Results:

• APOE ε4 genotype, CSF p-tau181/Aβ42 ratio, and diabetes were identified as independent risk factors for AD.
• Serum folate, MMSE, and MoCA scores were independent protective factors.
• The Random Forest model achieved the highest predictive performance (AUC=0.879) in the validation set.

Conclusions:

• The Random Forest model shows promise for AD risk stratification in high-risk populations.
• Integrating multimodal clinical data enhances predictive accuracy for Alzheimer's disease.
• This approach facilitates early intervention and risk classification for individuals at high risk for AD.