SenseHealth: A Platform to Enable Personalized Healthcare through Context-aware Sensing and Predictive Modeling Using Sensor Streams and Electronic Medical Record Data

Project ID
Project Categories
Computer Science
NSF Grant Number
Current healthcare diagnostics and assessment systems are limited by health data, which is sporadic, periodic, and incomplete. Owing to paucity of data collection during periods of wellness or good health, periods of disease remission are not well characterized. Clinicians therefore commonly do not have all the data necessary to recognize disease or illness patterns in order to anticipate patient needs to enable prospective, preventive care and early intervention. Wireless devices and health sensor technologies are increasing in use for continuous monitoring and assessment of key physiologic, psychologic, and environmental variables and reduce the current gaps in health data. However, uptake of such data by current health systems has been slow because of the reliance upon the already overburdened physician/healthcare team to interpret and manage incoming data. Nevertheless, the large streams of data generated by these devices in conjunction with traditional clinical data (Electronic Medical Records) have the potential provide real and important insights into patient health and behavior.
To address this gap, we propose to develop a novel cyberinfrastructure (CI) to automatically process and incorporate volumes of real-time data from sensors tailored to the individual in the context of personal EMRs and available environmental data.  Such data will be integrated into the clinical care workflow to enable system usability, feasibility, and ultimately utility. A core component of the CI is a collection of quantitative, predictive models that are sensitive to concerns across age, diseases, and health and variety of patient situations (ranging from low priority with no consequence on patient management to high priority requiring emergency evaluation), and sensor failures.  The models will be integrated with a distributed real-time stream data processing system and a complex event stream processing engine to process sensor data in a scalable and fault-tolerant manner. In order to develop the models we will leverage the research at Rady Children’s Hospital of San Diego, an affiliate of UCSD. In each of these studies, we will identify clinically relevant events (i.e. events that require clinical intervention) and then develop disease specific models that will predict clinical relevance or the need for intervention. We will evaluate incoming data and resulting clinical management activity from studies evaluating various types of health sensors in two different patient populations: (1) MyGlucoHealth application for evaluating the use of a Bluetooth-enabled glucometer (for blood sugar measurements) in 40 youths with Type 1 diabetes, and (2) Asthma Tracking application for evaluating the ability of a metered dose inhaler (MDI) tracking device to track asthma medication use in 50 mild-to-moderate asthma subjects over a period of 6 months. We will then evaluate the models using multiple sensor streams in youth with diabetes and in a prospective study in youth with asthma to determine their validity, efficacy, and utility in identifying patient scenarios of concern. SenseHealth will also incorporate a novel context-aware dynamic power management framework that adapts hardware and application behavior. Finally, it will fuse sensor and clinical data in a visual format that will increase interpretability and comprehension independent of literacy levels and will provide timely and relevant feedback to the user (patient and clinician).

We will use FutureGrid resources only for software and infrastructure development and performance benchmarking and we will not store and analyze any patient/medical data on FutureGrid machines. Medical data will be analyzed only on HIPPA compliant infrastructure.

Use of FutureSystems
For performance benchmarking, software development, and scalability studies.
Scale of Use
scalability study will involve simulating workloads consisting of hundreds of users and tens of thousands of sensor streams.