panel discussion: big data; holly jimison, phd
DESCRIPTION
Tuesday, October 23, 2012 Panel Discussion: Big Data Moderator: Roozbeh Jafari, PhD – Electrical Engineering, UT Dallas Panelists: Holly Jimison, PhD – Medical Informatics & Clinical Epidemiology, OHSU James McClain, PhD – Physical Activity Epidemiologist , Risk Factor Monitoring & Methods Branch, National Cancer Institute (NCI) Lucila Ohno-Machado, MD, PhD – Associate Dean for Informatics & Technology, School of Medicine; Founding Chief, Division of Biomedical Informatics; Professor of Medicine, UC San DiegoTRANSCRIPT
Holly Jimison, PhD, FACMIMedical Informatics, Oregon Health & Science University
Technology Advisor IPA, Office of Behavioral and Social Science Research, NIH
Opportunities and Challenges in Monitoring Health Behaviors in the Home and Environment
Big Data
Behavioral Markers = Continuous Monitoring + Computational Models
Home health based on unobtrusive, continuous monitoring
Hayes, ORCATECH 2007
Bedroom
Bathroom
Living Rm
Front Door
Kitchen
Sensor EventsPrivate Home
Activity Monitoring in the Home
Hayes, ORCATECH 2007
Sensor EventsResidential Facility
Bedroom
Bathroom
Living Rm
Front Door
Kitchen
Activity Monitoring in the Home
Measuring Gait in the Home
5
• Unobtrusive gait measurement in-home with passive infrared (PIR) sensors - Hagler, et al., IEEE Trans Biomed Eng, 2010
– Four restricted view PIR sensors– Measure gait velocity whenever a
subjects passes through the “sensor-line”
– Deployed for the Intelligent Systems for Assessing Aging Changes (ISAAC) study– 200+ subjects monitored for up
to 4 years and counting
Subject 1
12/07 08/08 11/09 12/10
30
40
50
60
70
80
90
Time
Ve
loci
ty (
cm/s
)
0.005
0.01
0.015
0.02
0.025
0.03
0.035
Stroke
Austin et al, Sept 2011 - EMBC (Gait)6
Subject 2
07/07 02/09 09/10
50
60
70
80
90
Time
Ve
loci
ty (
cm/s
)
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05CDR=0.5and MCIdiagnosis
7Austin et al, Sept 2011 - EMBC (Gait)
Monitoring->Care
8
ECG
EEG
Pulmonary Function
Gait
Balance
Step Size
BloodPressure
SpO2
Posture
Step Height
GPS
Performance
Early Detection
Prediction
Inference
Datamining
Training
Health Information
Coaching
Chronic Care
Social Networks
Decision Support
Population Statistics
EpidemiologyEvidence
M Pavel, H Watclar, Ref
ChallengesBig Data Challenges with Behavior Monitoring• Need low cost sensors / intelligent algorithms• Frequent data, but noisy and context dependent• Models of sensors, noise, context • Data harmonization• New modeling techniques –
• Robust estimation and classification framework• Need advances in machine learning, data mining,
fusion algorithms, modeling and visualization• Information fusion from multiple sources• Need dynamic user models, just-in-time feedback• Privacy / security advances• Address alert fatigue - containment of false alarms
Big Data Skill Sets• Sensor characterization (accuracy, bias, drift
sampling rate, setting, etc.)• Intelligent data sampling• Data cleaning / missing data / understanding • Data visualization techniques, data representation• Data storage / transfer• Privacy / security of data• Modeling techniques• Analysis methods, sensor fusion
Big Data Skill Sets• Sensor characterization (accuracy, bias, drift sampling
rate, setting, etc.)• Intelligent data sampling• Data cleaning / missing data / understanding • Data visualization techniques, data representation• Data storage / transfer• Privacy / security of data• Modeling techniques• Analysis methods, sensor fusion• Clinical or health relevance• Managing multidisciplinary teams, IRB, etc.
NIH OBSSR Big Data Training: [email protected]