UCLA

Alzheimer's disease (AD) represents one of the most pressing challenges in modern medicine, affecting over 6.7 million individuals in the United States with projections reaching 13 million by 2050. Despite significant advances in understanding individual risk factors, current approaches to AD research and clinical care face three fundamental limitations: the lack of robust methods for quantifying disease relationships, insufficient understanding of temporal disease progression patterns, and persistent health disparities in genetic risk assessment that exclude underrepresented populations from precision medicine benefits. This dissertation presents an integrated framework for precision medicine in Alzheimer's disease that addresses these limitations through three interconnected studies spanning computational methodology, temporal analysis, and health equity. The overarching goal is to develop predictive, preventive, personalized, and participatory approaches to AD research that can transform how we understand, predict, and prevent this devastating condition. The first study established a foundational framework for quantifying disease relationships using SNOMED CT embeddings. By integrating semantic, structural, comorbidity-based, and genetic correlation-based distance metrics, this work demonstrated that embedding-based approaches significantly outperform traditional methods in capturing clinically meaningful disease relationships. This semantic framework provides the essential foundation for meaningful trajectory analysis and clinical decision support systems. Building upon this foundation, the second study identified distinct temporal pathways leading to Alzheimer's disease. Four trajectory clusters were discovered: mental health, encephalopathy, mild cognitive impairment/neurodegenerative, and vascular cluster. These clusters exhibited significant differences in demographic characteristics, comorbidity patterns, and progression rates. Furthermore, the confirmation that multi-step trajectories confer higher risk than individual diagnoses underscores the importance of considering sequential disease progression in AD risk assessment. The third study addressed critical health disparities in genetic research by developing machine learning approaches for dementia risk modeling in underrepresented populations. Traditional polygenic risk scores showed poor performance in non-European populations, particularly African Americans where they performed worse than simple APOE-ε4 counting. The novel Elastic Net SNP models incorporating functional genomic information achieved substantial improvements while identifying both shared and ancestry-specific risk factors. Validation in the All of Us Research Program confirmed the generalizability of this approach across diverse healthcare systems. The integration of these three studies creates a comprehensive framework that spans from semantic disease understanding through temporal progression modeling to equitable genetic risk assessment. This framework enables trajectory-based risk stratification years before clinical symptoms appear, provides specific targets for personalized prevention strategies, and ensures that precision medicine benefits are accessible to all populations regardless of ancestry. The clinical implications are immediate and transformative. The trajectory patterns enable clinicians to identify high-risk patients and implement targeted interventions based on specific progression pathways. The genetic risk models provide accurate assessment in diverse populations, supporting personalized prevention strategies. The semantic framework supports clinical decision-making through quantified disease relationships and automated risk assessment systems. Beyond Alzheimer's disease, this work contributes methodological innovations applicable to other complex diseases with multi-step progression patterns. The semantic embedding approach provides a generalizable framework for disease relationship modeling, the trajectory analysis methodology can be applied to other chronic conditions, and the equity-focused genetic modeling addresses fundamental limitations in precision medicine more broadly. This dissertation establishes foundations for a future where Alzheimer's disease prevention is predictive (identifying risk years before symptoms), preventive (targeting specific progression pathways), personalized (accounting for individual genetic and clinical profiles), and participatory (including all populations in research and clinical benefits). The substantial lead times identified for intervention, combined with specific biological targets and equitable risk assessment approaches, provide unprecedented opportunities to transform the clinical course of this devastating disease while ensuring that precision medicine advances benefit all populations equitably.