Development of Simple Risk Scores for Prediction of Brain β-Amyloid and Tau Status in Older Adults With Mild...

Area of Science:

• Neurology
• Biomarkers
• Machine Learning

Background:

• Alzheimer's disease (AD) diagnosis in mild cognitive impairment (MCI) is challenging.
• Biomarkers like beta-amyloid (Aβ) and tau are crucial for AD progression.
• Predictive tools can improve early detection and clinical trial enrollment.

Purpose of the Study:

• To develop simple risk scores for predicting Aβ and tau positivity in MCI patients.
• To utilize a machine learning framework for risk score development.
• To assess the utility of these scores in improving clinical trial screening.

Main Methods:

• Employed the AutoScore machine learning tool on 657 MCI individuals from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Developed risk scores using demographics, neuropsychological tests, APOE4 status, and imaging biomarkers.
• Evaluated model performance using area under the receiver operating characteristic curve (AUC).

Main Results:

• A model predicting Aβ positivity achieved an AUC of 0.79 using age, ADAS-cog, APOE4 status, and either TMT-B or white matter hyperintensity.
• A model for tau positivity achieved an AUC of 0.91, incorporating age, ADAS-13, TMT-B, APOE4 status, hippocampal volume, and amyloid status.
• The developed risk scores are simple and integer-based.

Conclusions:

• Simple integer-based risk scores can effectively predict Aβ and tau positivity in individuals with MCI.
• These predictive models show potential for enhancing participant selection in AD clinical trials.
• The findings support the use of accessible data for early AD biomarker prediction.