Profiling and forecasting individual unique progression in ADRD using machine learning

Massachusetts General Hospital, Boston MA

Investigators

Abstract

PROJECT SUMMARY Alzheimer's disease and related dementias (ADRD) are progressive disorders that can begin over 20 years before symptoms appear. However, less than 50% of patients are diagnosed accurately in the early stage. Early and accurate detection of ADRD is critical for timely intervention but is extremely difficult due to the large heterogeneity among patients. No single diagnostic or prognostic tool is sufficient to fully explain the heterogeneity of ADRD. Relying solely on a single modality restricts our grasp of the complex dynamics at play within ADRD, significantly reducing diagnostic and prognostic capabilities in our current care settings. The overall goal of this project is to address this critical unmet need through the proposed research and mentored training of the applicant. The scientific goals of this proposal are to 1) deploy a multi-modal information framework that leverages decades-long in vivo and digital biomarkers, 2) profile the inter-individual heterogeneity of ADRD, and 3) build a risk prediction model of time-to-diagnosis from machine learning and time series analysis. The applicant will generate digital biomarkers collected through passive infrared motion sensors, wearables, sleep mats, electronic scales, computer usage software, and weekly online questionnaires. The applicant will establish acoustic and linguistic markers from semi-interview conversation recordings. In addition, the applicant will process in vivo biomarker data such as plasma, cerebrospinal fluid, and neuroimaging data. Such multi-site, multi-factorial data will be used to classify subtypes of ADRD and predict time-to-diagnosis. In order to conduct the proposed study and prepare for an independent research career, the applicant will be trained through taking courses and attending conferences in the following areas: 1) the underlying neurodegenerative processes related to ADRD, with a specific focus on the utilization of in vivo biomarkers in research; 2) methods of using technology and digital biomarkers in dementia research; 3) methods of deep learning and time series analysis for building risk prediction models of time-to-diagnosis; and 4) development of professional skills for conducting successful and ethically responsible clinical research. The proposed team of mentors each provide expertise in one or more of these areas and are together committed to collaboratively facilitating the applicant’s training. The applicant will apply these new skills to the proposed research project and obtain R01 or equivalent support to use the methods for improving the diagnostic capabilities in current care settings. Such findings are likely to make significant advances in reshaping personalized dementia care for millions of people.