Large-Scale Personalized Health Surveillance by Collaborative Learning and Selective Sensing
Recent advance in sensing and information technology provides abundance of risk predictive data, leading to the development of personalized health surveillance. However, effective use of the sensing technology is prohibited by the complexity of disease progression, heterogeneity in a large population, and the lack of cost-effective monitoring strategy. To scale up personalized health surveillance for a large population, we developed innovative methodologies for individual prognostic and personalized monitoring strategy design in this thesis. We proposed a statistical framework, collaborative learning, for characterizing the individual disease progression trajectories from sparse and irregular data. We then developed a decision support algorithm, selective sensing, to adaptively allocate limited monitoring resources to high-risk individuals. We further proposed a rule-based method and a prognostic-based monitoring framework to translate the sensing data into risk-predictive patterns for individual prognostic and identify the cost-effective monitoring strategies for disease prevention. We applied the proposed methods to real-world applications, including cognitive monitoring in Alzheimer’s Disease (AD), and follow-up monitoring in depression treatment.