PROJECT SUMMARY In both the United States and globally, we are witnessing an unprecedented increase in the elderly population (> 65 years) and the very elderly population (> 85 years). This demographic shift has led to a significant rise in the number of individuals affected by age-related diseases and conditions, with a particularly profound impact on cognitive decline, Alzheimer's disease, and related dementias (ADRD). There is an urgent imperative to create more accurate and practical predictive models for these conditions, usable across a wide range of clinical settings in the United States. Much of ADRD related advances (e.g., risk factors such as lifestyles, cognitive signs and symptoms, latest treatments) is often embedded in Pubmed literatures, which are not directly available for computational analysis and manual extraction is very time consuming and costly. To the best of our knowledge, there are no investigations on predictive models for dynamic ADRD risk assessment leveraging knowledge from literatures. Our current parent grant is focusing on the development of a computable rare disease knowledge hub to accelerate rare disease knowledge discovery. To respond to NOSI-AG-23-032, this administrative supplement application will utilize the increasing availability of EHR and the graph embedding methodology created in the parent proposal, alongside our proficiency in real-time and real-world risk modeling, to evaluate the effectiveness, implementability, and clinical utility of dynamic ADRD risk modeling in elderly population. In our preliminary work, we have investigated methods to extract entities and relationships from unstructured text data and showcased GRU-D based architectures in real-time risk assessment of post surgical complications (PSC) 1,2 and chronic kidney disease (CKD) 3,4 with a reasonable model explainability at population level. We will further construct an ADRD knowledge hub (ADKH), and integrate the ADKH with GRU-D based dynamic ADRD risk assessment to overcome both data quality issues and difficulties in feature selection and representation of EHR data. Our specific aims include (1) construction of a computable ADKH to accelerate ADRD knowledge discovery, and (2) performance evaluation, feasibility, and model explainability research of gated GRU-D based models for ADRD risk assessment in the elderly population, leveraging ADKH and EHR.