ACSUIUC Partnership: Measuring Population Health with Internet Health Monitors

1
ACS-UIUC PartnershipMeasuring Population
Healthwith Internet Health Monitors
Bruce R. SchatzDepartment of Medical Information
ScienceCollege of Medicine, Institute for
Genomic Biology Department of Computer Science,
Illinois Informatics Institute University of
Illinois at Urbana-Champaign schatz_at_uiuc.edu ,
www.canis.uiuc.edu
American Cancer Society, Illinois
Division Chicago, Illinois, November 30, 2007
2
Outline of Talk

• The Promise (What)
• Population Monitoring of Everyday Health
• The Technology (How)
• Full-Spectrum Quality-of-Life Indicators
• The Plan (from Here to There)
• Pilot Projects for Population Management

3
The Promise

• Population Monitoring
• of
• Everyday Health

4
Measuring Population Health

• CPS Cancer Prevention Study (ACS)
• 500K persons measured every other year
• 100 questions paper survey, self-administer
• BRFSS Behavioral Risk Factor Survey (CDC)
• 350K persons measured per year (sampled)
• 100 questions paper survey, phone interview

5
Beyond Screening

• Why are Some People Healthy? (R. Evans)
• Major categories are disease, health care,
  health function, genetic endowment, physical
  environment, social environment, individual
  response, behavior, well-being, prosperity.
• Healthy People 2010
• 467 objectives in 28 focus areas
• www.health.gov/healthypeople
• Measure Full-Spectrum Health Status
• Detailed QoL in each detailed category

6
Measuring Population Health

• Internet Health Monitors (UIUC)
• 10M persons measured every day
• Times 1000 in frequency
• Times 10 in quantity
• 10K questions total sampled over time
• Times 100 in quality
• So 1M times finer resolution!

7
Population Health Categories

• Demographics (Cancer Registry)
• Disease Survey and Screening (Diagnosis )
• Medications and Vitamins (Treatment)
• Exercise and Diet (Behavioral Risks)
• Alcohol and Smoking (Behavioral Risks)
• Family History (Genetics)
• Pain and Physical Disability (Lifestyle)
• Social and Physical Environment (Lifestyle)
• Healthcare Availability (e.g. Insurance)

8
The Ms from Here to There

• Measuring 100 to 1000 to 10,000 questions
• Monitoring these questions every day
• Mining patterns across population database
• Managing individuals across cohorts

9
The Technology

• Full-Spectrum
• Quality-of-Life
• Indicators

10
Prototype with 100 Questions

• Questionnaire from Merged QoL
• 120 questions from 20 questionnaires
• General plus some Specific questions
• Simple Clusters do Coarse Prediction
• Students simulate sick or well patients
• K-means with random seeds does correct clustering
  from actual health monitor sessions with 100
  answers

11
Sample General Health Questions
12
Sample Specific Health Questions

• Heart Condition

• Arthritis Condition

13
Health Monitor Session
14
Prototype with 1000 Questions

• Disease-oriented Brunner model
• Mental
• Physical
• Social
• Recognizes role of diet, exercise, and stress
• 1200 questions evenly distributed

15
Health Monitor System

• Prototype Demonstration
• www.canis.uiuc.edu/healthmonitor
• Lifestyle Tracker and Teachable Moments
• Comparative Status and Customized Brochures
• Questions and Categories at
• www.canis.uiuc.edu/schatz/monitors/health.questio
  ns.pdf
• www.canis.uiuc.edu/schatz/monitors/health.measure
  .pdf
• Future Discussion Forums and Social Networks

16
Informatics Technologies

• Measure Population Health
• Adaptive Question Asking of Quality of Life
  Questionnaires
• Answers for Individuals creates Database for the
  Population
• Manage Population Health
• Statistical Information Retrieval cluster
  patients into care cohorts
• Data Mining of Feature Vectors discovers common
  causal patterns

17
Statistical Technologies

• Calibrated Averages (Jeff Douglas)
• Fill in missing values so complete vectors
• Requires calibration runs from user classes
• Correct normalization of user answers
• True adaptive sampling with even distribution
• Weighted Clusters (ChengXiang Zhai)
• Compute distances for natural clusters
• Requires semantic weights for healthcare
• Dynamic clustering for cohort assignment
• True orientation towards treatment classes

18
Small World Social Networks

• Community Structure enables Dynamic Clustering
  with High Coherence

19
Where to cut?
20
Small Worlds Clustering

• Mutual Information

• Initially all nodes are in their own community.
• Look for the join that maximizes ?Q.
• Modularity is determined by two terms

21
BeeSpace FIBR Project

• BeeSpace project is NSF FIBR flagship
• Frontiers Integrative Biological Research,
• 5M for 5 years at University of Illinois
• Genomic technologies in wet lab and dry lab
• Bee Biology gene expressions
• Space Informatics concept navigations
• 10 PhD Students 8 Computer Science, 4
  Informatics
• Zhai, Han, Roth, Schatz, Sinha, Zhong

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
The Plan

• Pilot Projects
• for
• Population Monitoring

27
Getting from Here to There

• Develop Full-spectrum Questionnaire
• Merge existing Quality of Life instruments
• Encode knowledge from Medical Professionals
• Develop Dynamic Adaptive Administration
• Software to handle Interactive Sessions
• Software to build Individual History
• Software to build Population Database
• Develop Cohort Similarity Clustering
• Algorithms for Statistical Feature Matching
• Lifestyle Coaching via Cohort Switching
• Deploy Test to Trial Population

28
Partner Strengths

• ACS American Cancer Society
• Volunteer Network
• Health Management (Intervention Strategy)
• UIUC University Illinois Urbana Champaign
• Supercomputer Network
• Data Management (Discovery Strategy)

29
Partnership Projects

• 1K Persons (Monitor Development)
• Regional Volunteers through MCCD
• Compare Clusters to Nurse Assessments
• Develop 10K questions and Evolve softwares
• 100K Persons (Mining Deployment)
• Statewide Volunteers through Illinois ACS
• 10K questions and hundreds of cohorts
• Data Mining across thousands of factors
• 10M Persons (Manage Universally)
• SOS (Save Our State) or Full-Scale CPS

30
Further Information

• See R. Berlin, MD, and B. Schatz, PhD, under
  Publications under Papers at www.canis.uiuc.edu
• Population Monitoring of Quality of Life,
  Congestive Heart Failure, invited review, 7(1)
  13-22, Jan 2001.
• www.canis.uiuc.edu/schatz/monitors/chf-heart.doc
• The Inevitable Evolution of Healthcare
  Infrastructure, Medscape General Medicine, May
  2004.
• www.canis.uiuc.edu/schatz/monitors/healthcare.inf
  rastructure.doc
• CS598HI/LIS590HI Healthcare Infrastructure
• http//www.cs.uiuc.edu/class/fa07/cs598hi/