Advanced Modeling in Alzheimer's Cohorts and Trials

We aim to develop robust, calibrated, and reliable predictive models by utilizing machine learning techniques on a rich dataset of imaging, omics, and phenotypic data. Our approach is twofold:

1. We will employ supervised machine learning methods to construct predictive models for defined outcomes such as clinical disease stages, meaningful cognitive decline, and imaging and neuropathologic markers (e.g., amyloid or tau positivity). We acknowledge that a major challenge in implementing predictive models is their integration into real-world settings. Therefore, once validated, we will convert these models into web applications or simplify them into risk scores to improve their usability and practical application.
2. Given the significant heterogeneity in Alzheimer's disease and existing gaps in our mechanistic understanding, we will utilize semi-supervised machine learning to subtype the disease. This method will incorporate genetics, biofluid biomarkers, imaging, and neuropsychological tests. Our methodology offers a significant advantage over the unsupervised machine and deep learning techniques commonly used in the field. By incorporating "partial labeling" and incorporating knowledge of disease states, our semi-supervised approach enables more clinically relevant disease subtyping and results.

For this project, we are collaborating with experts from other institutions, including Dr. Christos Davatzikos from the University of Pennsylvania, who is an expert in neuroimaging and machine learning, and Dr. Richard Lipton from Albert Einstein College of Medicine, known for his expertise in epidemiology and the design of trials and cohorts.

Despite the extensive research on Alzheimer's and other dementias, the success rate of randomized clinical trials has been disappointingly low. This is primarily due to the significant clinical variations among patients, which limits our ability to predict their outcomes accurately. To address this challenge, we aim to develop a predictive framework specifically focusing on Alzheimer's and related dementias. We will use several cohorts and will employ machine learning and advanced statistical modeling on various types of data, including demographics, neuropsychological measures, biofluid and imaging biomarkers from existing international cohorts and concluded clinical trials. The framework will enable us to predict disease progression which will facilitate early intervention and personalized care for individuals at risk of developing dementia. Our project seeks to harness the power of machine learning and advanced analytics to improve our understanding of Alzheimer's Disease, enhance prediction accuracy, and ultimately contribute to more effective treatments and interventions.