Skip to main content
An official website of the United States government
Grant Details

Grant Number: 1R01CA297869-01 Interpret this number
Primary Investigator: Qu, Annie
Organization: University Of California-Irvine
Project Title: SCH: Individualized Learning and Prediction for Heterogeneous Multimodal Data From Wearable Devices
Fiscal Year: 2024


Abstract

Recent technological advances in mobile health have catalyzed the rapid growth of large volumes of digital health data due to its potential for scientific impact and health relevance for general populations. This growth imposes unique challenges for machine learning, computation, and information science due to the sheer volumes and high frequencies of longitudinal measurements over time. While machine learning, particularly deep learning, has led to major advances across various data types, its application in modeling time series health data to develop smarter health care systems remains challenging, with relatively low adoption in real-world healthcare settings. Primary reasons for these challenges lie in the irregular, multi-resolution health data collected from diverse subjects who display significant variations in their behaviors and responses to different treatments. Neglecting population diversity contributes significantly to health disparities, as policies and decisions often disregard marginalized minority groups. This proposal provides fundamental and rigorous solutions to address these challenges for complex heterogenous data with a particular focus on mobile health data. More specifically, our research is inspired by social determinants of health (SDoH) that impact health disparities among women in the U.S using two ethnically diverse women’s health datasets, where the data present high heterogeneity and multi-resolution features over time. We aim to flexibly extract the essential information from heterogeneous signals across high-dimensional omni-channels over time and integrate information from multi-resolution variables, to enhance prediction precision and interpretability. We further target online sequential policy optimization that provides non-invasive intervention treatments over time to maximize individuals’ health outcomes according to their heterogeneity. In summary, our research aims to: (1) Develop deep neural models for understanding and predicting individual health over time, while learning shared patterns across multiple individuals to enhance interpretability of the results; (2) Create novel data integration techniques for multi-resolution mobile health data, accommodating irregular time intervals, diverse sampling rates, and measurement variations among subjects; (3) Implement a reinforcement learning platform to find subject-specific optimal policies under populational heterogeneity; (4) Apply the proposed methods to identify biopsychosocial factors affecting women’s health, enabling the development of early prevention and intervention strategies to enhance overall well-being in women.



Publications


None


Back to Top