Download Sensor Systems for Monitoring Congestive Heart Failure: Location

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Electronic prescribing wikipedia , lookup

Transcript
Autumn 2011 Conference
November 2-3, 2011
Washington, DC
AGENDA
Sensor Systems for Monitoring
Congestive Heart Failure: Locationbased Privacy Encodings
WEDNESDAY, NOVEMBER 2, 2011
TIME
TOPIC
0730 – 0845
Breakfast
0845 – 0900
Conference Welcome
0900 – 1000
Keynote Address – Virtualization of R&D Driving New Health IT Requirements
Vijay Pillai (Oracle Corporation)
1000 – 1030
Break
Session1: Trustworthy Health Information Systems
1030 – 1050 Regret Minimizing Audits: A Learning-Theoretic Basis for Privacy Protection
1050 – 1110
Edmund Seto, Posu Yan, Ruzena Bajcsy
Declarative Privacy Policy: Finite Models and Attribute-Based Encryption
Ellick Chan
(Stanford University)
University
of California, Berkeley
1110 – 1130
Context-Aware Anomaly Detection for Electronic Medical Record Systems
Arunesh Sinha (Carnegie Mellon University)
Yuan Xue (Vanderbilt University)
1130 – 1150
Sensor System for Monitoring Congestive Heart Failure Patients
TRUST Autumn 2011 Conference
Uncovering Anomalous Usage of Medical Records via Social Network Analysis
November
2, 2011
Bradley Malin (Vanderbilt
University)
Edmund Seto (University of California, Berkeley)
1150 – 1210
1210 – 1340
Lunch
Congestive Heart Failure





Inability for the heart to pump enough blood to the rest of the body.
Cardiovascular disease is the #1 killer in the U.S.
Approximately 5.7 million Americans have Congestive Heart Failure.
Each year 670,000 Americans will be newly diagnosed with CHF.
The estimated direct and indirect cost of CHF in the U.S. for 2009 is $37.2
billion.
Congestive Heart Failure
 CHF is a chronic disease
 Treatable




Medications
Lifestyle changes (diet, smoking, physical activity, weight, etc.)
Frequent monitoring (every 3-6 months w/doctor)
Attention to symptoms (cough, fatigue, weight gain, swollen feet)
 Telemonitoring
 Systematic review by Louis, et al., 2003
 18 observational studies and 6 randomised controlled trials
 Findings suggest telemonitoring benefits:





Early detection of deterioration
Reduce readmission rates
Reduce length of hospital stay
Reduce readmissions
Reduced mortality
Case Study: Congestive Heart Failure
Mobile device
GPS
Accelerometer
BT digital scale
BT blood pressure
Data sent to server
at Vanderbilt
Patient receives
regular feedback
messages
Privacy Considerations
• Device security (authentication, device loss, etc.)
• Wireless security (eavesdropping, DoS, Phishing, etc.)
• Data security (encryption, access rights, audit trails, etc.)
• Privacy policies
– Patients control their data
– Some potential benefits to sharing their data
– But, also some potential risks to sharing their data
Recipe dictates permitted behavior for a client application, specifically:
What sensor/video/audio data the application can access, and at what
Secureresolution
Communication Framework for Networked Tele-Health Applications
Aaron What
Bestick,output
Posu Yan,
Ruzena
Bajcsy
data
streams
the application is allowed to create
Figure 1: A sample smartphone client collecting accelerometer data, showing the sensor
interface, network interface, and recipe components of the client library.
Defining Contextual Exposure
For example, doctor may be interested in:
• Where is patient getting physical activity?
• Where is patient having high blood pressure?
• Where is patient having lunch?
Elaboration on contextual exposure
Problem: Where is patient getting physical activity?
• “Physical activity” defined by p(t)
(e.g., physical activity obtained from accelerometry)
• “Where” defined by x(t)
(e.g., location obtained from GPS)
• Hence: x(t) for all t when p(t)>threshold intensity of activity
• Furthermore: g(x(t)) = places (e.g., parks, schools, home, etc.)
• and… Σ g(x(t)) / T
(i.e., proportion of monitoring period that exposure occurred)
Privacy of Inferred Context
• Location of home, work, etc.
Introduce random error
Aggregation (1 km)
Aggregation (2 km)
Model of the patient
What might influence a patient’s encoding decisions?
• Risk adversity (cost)
– Less data shared, the lower the privacy risk
– Factors in various aspects of “trust” (of their physician, the
network, data security, laws, etc.)
• Possible reward
– Sharing more data, might lead to better care
• … and obviously, these vary between individuals
Model of the doctor
What might influence a doctor’s perspective on
encoded data?
• Generally more detailed data is better than less
• Up to a point (saturation)
• … and presumably, less variation between doctors
(e.g., standard treatment protocols)
Privacy in the Federal Health IT Plan: a Game Theoretic Approach
Daniel Aranki, Ruzena Bajcsy
What is the optimal “move” of the device?
Future work
• Finish implementation of the recipe architecture, including
the collaboration server
• User studies to define useful encodings
• User studies to define utility functions
• Analyze (and optimize) the patients’ decisions by extending
this framework to consider various privacy and security
threats.
THANKS!
Autumn 2011 Conference
November 2-3, 2011
Washington, DC