Biomarkers.

Area of Science:

• Neurology
• Biomedical Data Science
• Gerontology

Background:

• Alzheimer's disease (AD) dementia risk prediction is crucial for early intervention, potentially preventing 30% of cases through vascular risk factor management.
• Emerging disease-modifying drugs for AD are most effective in the earliest clinical stages, necessitating tools for early detection in cognitively healthy individuals.
• A cost-effective, scalable screening tool is needed to identify high-risk cognitively healthy (CH) subjects for improved access to early treatments.

Purpose of the Study:

• To develop a predictive tool for assessing Alzheimer's disease dementia risk using routine blood test data.
• To evaluate the feasibility of utilizing routine blood counts and chemistry panels for early AD risk screening in a large population.

Main Methods:

• A community cohort of 381,754 cognitively healthy subjects (aged 45+) from Clalit health care services was analyzed.
• Machine learning (ML) models were trained using historical blood exam data (1-10 years) and prediction horizons (1-10 years).
• Model performance was evaluated using accuracy, AUC, precision, recall, F1 score, and other statistical metrics.

Main Results:

• Models demonstrated consistent accuracy around 0.73 across various historical periods and prediction horizons.
• Area Under the Curve (AUC) reached up to 0.84 for a 2-year history and 3-year prediction horizon.
• Precision varied (0.19-0.28), while recall remained high (0.73-0.82), indicating potential for identifying at-risk individuals.

Conclusions:

• Routine blood counts and chemistry contain valuable information for predicting future Alzheimer's disease dementia risk.
• Machine learning and AI can transform routine blood exams into effective screening tools for AD dementia risk assessment.
• This approach offers a scalable and low-cost method for identifying individuals who may benefit from early AD interventions.