Our paper on ``Federated learning of predictive models from federated Electronic Health Records''...
SHB:Large:Collaborative Research: Algorithmic Approaches to Personalized Health Care
Funding Agency: National Science Foundation (NSF), Biology.
Award Number: IIS-1237022.
Principal Investigators: Yannis Paschalidis and W. Adams at Boston Univ., collaborative with D. Bertsimas at MIT.
Intellectual Merit. The US health care system is considered costly and highly inefficient. It spends huge resources on the treatment of acute conditions in a hospital setting rather than focusing on prevention and keeping patients out of the hospital. While there is no broad agreement on the potential solutions, structural reform is on the horizon. To that end, the meaningful use of Electronic Health Records (EHRs) is seen as a key to improving efficiency. At the same time, medical devices are undergoing a revolution as they become smaller, often implantable, with embedded intelligence, and with the ability to (wirelessly) transmit information on a daily basis to the clinic. The health care system, however, is not well equipped to deal with the impending deluge of personalized health-related data from EHRs and medical devices. This proposal puts forth a comprehensive and systematic approach to intelligently process such data and help direct physician attention to the prevention of serious conditions while relieving them of rudimentary tasks.
In the proposed framework, early risk assessment starts with algorithms for mining individual insurance claims and EHR data to classify patients in terms of the risk they have for developing a specific disease. Anomaly detection techniques monitor individual patients to detect changes in their status and guide their re-classification. Risk clusters produced by our approach have an associated set of actions, including tests, additional measurements (including the possibility to instrument the patient with medical devices – sensors), and physician involvement. The sensors we plan to leverage include body sensors, from implantable to simpler measurement devices, that can wirelessly send information to the clinic in near real-time. Using EHR and sensor data, we further propose to develop algorithms that help manage chronic conditions, such as diabetes and heart disease, and develop a distributed epidemiology approach suitable for the emerging landscape where lots of data about each patient will be distributed among many different locations.
The research will leverage the PIs expertise in data models, optimization, and decision theory. The team includes primary care physicians, the chief medical information officer at the Boston Medical Center (BMC) and the BMC director of enterprise analytics responsible for an existing data warehouse we will leverage for our work.
Broader Impact. If the proposed vision becomes reality, the potential for revolutionary improvements in the quality of health care is significant. Risk assessment combined with continuous health monitoring of high risk individuals and intelligent management of chronic conditions, offers the potential of preventing acute health episodes. The research will certainly impact health care economics — a timely contribution in an era of ever increasing health care costs fueled by a rapidly aging population. In general, the potential societal benefits of medical informatics can only be understated. Further, the paradigm-shift in epidemiology we advocate can greatly advance our understanding of disease, its progression, and its management. An important feature of the proposed work is that the BMC data we plan to utilize concern a significant percentage of underrepresented groups, hence, the research can benefit this often overlooked segment of the US population. On the educational front, plans include new courses, training a diverse set graduate students, involving undergraduate students, collaborating with medical doctors, and reaching out to high school students through existing programs embraced by the PIs. Dissemination plans include capitalizing on the BU Sensor Network Consortium and organizing a major medical informatics workshop which can lead to the creation of an influential interest group in this area.