Introducing a gatekeeping system for amyloid status assessment in mild cognitive impairment

Area of Science:

• Neuroscience
• Biomarker Discovery
• Machine Learning in Medicine

Background:

• Cerebral amyloid-β plaque burden is a high-risk factor for Alzheimer's disease (AD) in mild cognitive impairment (MCI) patients.
• Direct assessment of amyloid status (A-status) is not always accessible.
• Identifying pre-diagnostic biomarkers can optimize patient selection for further A-status workup.

Purpose of the Study:

• To develop and validate a machine learning-based gatekeeping system for predicting A-status in MCI patients.
• To utilize pre-existing information including APOE-genotype, 18F-FDG PET, age, and sex for A-status prediction.
• To assess the utility of predicted A-status in estimating dementia progression risk.

Main Methods:

• Trained machine learning classifiers on 342 MCI patients using 18F-FDG-PET, age, and sex to predict A-status.
• Classified patients into APOE-ε4 non-carriers (APOE4-nc) and carriers (APOE4-c) with majority classes of amyloid-negative (Aβ-) and amyloid-positive (Aβ+) respectively.
• Tested classifier performance on two independent datasets and compared dementia progression rates between gold standard and predicted A-status.

Main Results:

• Achieved 87% precision for predicting Aβ- in APOE4-nc and 51% recall for predicting Aβ+ in APOE4-c.
• The predicted A-status demonstrated at least equal indication of dementia progression risk compared to the gold standard A-status.
• The developed algorithm showed good reliability in approximating A-status using readily available patient information.

Conclusions:

• An algorithm was developed to reliably approximate A-status in MCI patients using APOE-genotype, 18F-FDG PET, age, and sex.
• This tool can enhance individual risk estimation for AD development based on existing biomarkers.
• The algorithm supports efficient selection of patients for etiological clarification and may have utility in clinical routine and trials.