Early Prediction of Alzheimer’s Disease via Longitudinal Clinical Informatics

Alzheimer's disease (AD) is a progressive neurodegenerative disease in which cognitive and functional decline may begin years before the onset of dementia. Therefore, early identification of high-risk individuals is essential for prompt monitoring and intervention. However, advanced diagnostic approaches based on MRI, PET, and cerebrospinal fluid biomarkers are expensive, infrastructure-intensive, and difficult to implement for large-scale screening. This paper proposes ClinicoFusion-Net, a non-invasive longitudinal machine-learning framework for early AD risk prediction using routinely available clinical and cognitive data. The framework integrates cognitive assessments, functional measures, demographic characteristics, medical history, and longitudinal changes across clinical visits. A multi-domain feature representation is processed through a Bidirectional Gated Recurrent Unit (Bi-GRU) with temporal attention to capture non-linear patterns of cognitive and functional decline. SHAP-based explainability is incorporated to identify the clinical factors contributing to individual risk predictions. The proposed framework aims to provide an accessible and interpretable approach for risk stratification across Cognitively Normal (CN), Early Mild Cognitive Impairment (EMCI), Late Mild Cognitive Impairment (LMCI), and AD, while supporting future progression-risk estimation