Biomarkers.

Area of Science:

• Artificial Intelligence
• Neuroscience
• Biostatistics

Background:

• Predicting Alzheimer's disease (AD) risk is vital for precision medicine and timely interventions.
• Early identification of individuals progressing from Mild Cognitive Impairment (MCI) to AD is a significant clinical challenge.

Purpose of the Study:

• To develop and evaluate a novel AI-based joint prediction model for longitudinal and survival data in MCI patients.
• To dynamically update AD risk predictions using evolving patient data.

Main Methods:

• A Kolmogorov-Arnold Networks (KANs)-based joint prediction model (JM-KAN) was developed.
• Real-world data from the National Alzheimer's Coordinating Center (NACC) dataset of MCI patients was used.
• Joint models were constructed with landmark times up to 4 years, predicting survival outcomes one year post-landmark. Performance was assessed using integrated AUC (iAUC) and integrated Brier score (iBS).

Main Results:

• The study analyzed 2711 MCI patients, with 821 progressing to AD.
• The JM-KAN model achieved an iAUC of 0.789 and an iBS of 0.118, outperforming existing joint models in discrimination and overall performance.
• The model demonstrated success in individualized predictions of both longitudinal risk factors and survival probability as new data emerged annually.

Conclusions:

• The novel AI architecture, JM-KAN, significantly enhances the dynamic prediction of AD risk.
• This approach improves upon existing methods by leveraging new longitudinal data to continuously update risk assessments.
• The findings support the use of advanced AI models for dynamic risk stratification in neurodegenerative diseases.