While lifestyle accounts for up to 3/4 of healthcare costs, most people do not find exercise fun. Some have tried numerous diet and exercise programs with the primary goal of weight loss, only to fail and become discouraged, resulting in worse health outcomes over time. In this paper, we explore crucial elements in consumer engagement that can lead to improved health outcomes. We deploy data mining algorithms to understand the impacts of various interventions, such as social network, visual dashboard, event processing, games and challenges, on health outcomes using data obtained from a clinical trial. We explain what worked, what did not work, and why. More importantly, we describe salient attributes of social health games that are crucial in both consumer engagement and health outcomes.
David Kil is the founder and CEO of HealthMantic, focusing on lifestyle-medical informatics and sensor-based health gaming. Prior to founding HealthMantic, he was Chief Scientist at SKT Americas and Chief Science Officer at Humana, responsible for the development and deployment of healthcare informatics applications. At SKTA, he founded the iWell project and built an integrated wellness platform consisting of sensors, platform, and informatics. The system underwent a successful clinical trial at PeaceHealth with promising results. At Humana, he led the design and development efforts in enterprise knowledge engine, predictive modeling, and outcomes analytics while working with Samsung on U-Health initiatives. The enterprise insight engine won the best-of-breed technology award from the ComputerWorld magazine. He co-authored a book entitled “Pattern Recognition and Prediction with Applications to Signal Characterization” by Springer-Verlag, published over 30 papers, and holds 8 US/European patents. He graduated from the University of Illinois at Urbana-Champaign with BSEE/Chemistry (Highest Honor and Bronze Tablet), the Polytechnic University of New York (MSEE), and Arizona State University (MBA).
8. Web service features
Customizable health
summary page
Zite-like
personalized
reading section Goals
Virtual
Virtual
CoachCoach
Social/coach
nudging based
on CEP
Challenges, gam
Coach es, friends,
& role models
12. Taking a closer look
PeaceHealth results
BMI
Social nudging
Biomarkers
Activities
Tracking systems
Real time feedback
13. BMI changes vs. steps per day
Relationship is not as strong as expected
BMI % Change
3.2%
weight
loss
8600 steps a day
Steps per day (1000 steps)
14. Improving SN improves health
Social nudging plays an important role
Emerging influencers
87%role model request accepted
78%friend request accepted
15. Passing the smell test
Relationships among activities, biometrics and biomarkers
Steps per day: Time (y) vs. users
Steps per day: 10/12/03
Time (y) vs users
11/03/13
11/06/21
SPD in K steps
5 BMI change 15
0 10
-5 5
Correlation between
SPD & HDL = 0.25 -10 0
0 50 100 150 200 250
LDL: = -0.059 HDL: = 0.251 TC: = -0.001 TG: = -0.085
SPD, BMI
= -0.24 40 50
20 10 100
0 0 0
-20 0
Correlation between 0 -40 -10 -50
-100
-60
SPD & BMI = –0.24
-20
-5 -80 -100 -200
-30
5 10 15 5 10 15 5 10 15 5 10 15
BMI
-10 FPG: = -0.023 HbA1c: = -0.062 TG/HDL: = -0.143 TG/HDL vs. BMI : = 0.197
4 4
-15 50 0 2 2
0 0
0 -1
Correlation between BMI -20 -2
-4
-2
-4
5 10 15 -2
& TG/HDL = –0.20
-50 -6 -6
SPD (K steps) 5 10 15 5 10 15 5 10 15 -20 -10 0
SPD (K steps) BMI % change
16. Disease risk vs. health outcomes
Better health outcomes for high risk participants
Lowering FPG
64.0% of high-risk people lowered FPG while
47.5% of the rest moved in the right direction
Weight Loss
Greater weight loss for the higher-risk
subgroup 6.4 lbs vs. the rest 3.7 lbs. Seniors lost
7.8% weight
Cholesterol
223 to 214 mg/dL little change for the rest
Triglycerides
183 to 157 mg/dL little change for the rest
17. Good predictors for weight loss
Social nudging & key service features
SPD FDR= 0.54
20
Weight loss
18 No weight loss
Feature Fisher’s ratio 16
Average steps per day 0.54 14
PDF comparison of
steps per day
# of users
12
# of goals 0.52 between those
10
# of competitions 0.36 withlargeweight loss
8
# of wellness games 0.35 and small weight
6
loss
# of students 0.31 4
2
# of UGC nudging
0.22
received from friends 0
0 0.5 1 1.5 2 2.5
Average steps per day x 10
4
# of UGC nudging Ngoals FDR= 0.52
0.20 25
written Weight loss
No weight loss
# of family members 0.14
20
PDF comparison of the
UGC = User generated content number of goals set
# of users
15
between those with
large weight loss and
10
small weight loss
5
0
0 5 10 15 20 25 30
# of goals
18. BRN is better at predicting weight loss
As we add more features, performance improves
Prediction accuracy
improves as we add
good features
19. Best predictor
Success begets more success Lab data
Steps
Wellness
score
PA
trends
Social
reputation
Competition
& games
Weight Goal difficulty
SN Goals
activities
20. Why not better results?
Lessons learned from listening to participants and data
• Pedometer = blunt instrument
• Real-time feedback vs. near real-time
feedback
• Full credit, not partial credit
• 30% hunger factor
21. The right approach
Gentler, more effective, shorter, personalized workout
• When is enough enough?
• Sufficient statistics concept
• Motion sensing and classification
• Exercise to improve flexibility, balance, strength, and cardio
Full credit & real-time feedback
• Games during rest for fun & relaxation
• Where is the “fitness science” for the masses?
23. My 2 year health data
I ran a total of 2 marathons & 8 half marathons
Min-by-min Activity Pattern: Co pa raw 20
½ marathon 10 mph
3:20 No
6:40
device
10:00
Seoul
13:20
16:40 Madrid
20:00
23:20 0 mph
160 Steps per day (K) 40
155 Weight (lbs)
20
150
145 0
40
# friends
20
0
100
50
Wellness meter
Social reputation score
0
10/05/13 10/07/02 10/08/21 10/10/10 10/11/29 11/01/18 11/03/09 11/04/28 11/06/17
24. My last half marathon run
Are treadmills designed to torture prisoners?
Short walk Long walk
Run, run, and run some
more
25. Can we differentiate?
Cycling, elliptical and walk/run for full credit
Dave tread ellipt cycle walk 04 Dec 2011.csv
3000 Parameters for real-time
Magnitude
2000
feedback
Treadmill Elliptical Cycling Walk • Duration
1000
• VO2max
0 1 2 3 4 5 6 • Speed/distance
Sample index (30 Hz) 4
SPM, strides per min, and RPM over time
x 10 • Calories
300 • HIIT parameters
Activity intensity
200
• RPM parameters
• …
100
0
0 1 2 3 4 5 6
Sample index (0.1172 Hz) 4
x 10
Activity Classification Decision
stairs 800
cycling 600
elliptical 400
walk/run 200
no activity
0 1 2 3 4 5 6 0
0 20 40 60 80 100 120 140 160 180 200
4
x 10
26. Adding HIIT for greater effectiveness
Helps build strength and endurance fast
800
600
Outside walk
400
High-intensity interval training
200
Elliptical
0
0 50 100 150 200
27. Recognition of motion building blocks
Personalized fitness programs
A library of 300 motions consisting of The Happy Body, Tai Chi, Qi-Gong, Resistance, Tibetan
Rites, Martial Arts, Dynamic/Static Stretching, Cardio, Tabata, Yoga, Pilates, etc.
28. Importance of resistance training
Improving body composition – fallacy of BMI
Raw gyroscope sensor data after signal processing
5000
0
Roll
Pitch
Yaw
-5000
0 2000 4000 6000 8000 10000
Combined accelerometer and gyroscope outputs
1200
Chin-ups Bicep curl Pull
1000
One-arm side pull downs
800 Accelerometer
Bench press Gyroscope
600
400
200
0
0 500 1000 1500 2000 2500 3000 3500
30. Losing the last 10 – 15lbs
The right way
1. Easy, short effective Weight & body fat/muscle %
• No punishing workout
• Fun motion interval training
including free weights
2. Positive Health outcomes
• Instant performance feedback
• Sensible nutrition plan – no
diet, more awareness
• Healthy micro-habit formation
3. Financial ROI
34. Lessons learned
Easy on ramp, effective, science, enjoyable
• Right way to improve health
• Personalized and effective workout & relaxation plans
• Body fat and muscle science (no BMI obsession)
• Enough workout, enough food, & self- awareness
Lifestyle habit formation with real-time learning Enjoy
Life & Share with Others!
35. Future directions
Take the solution to the masses
• Larger pilot with an academic medical center
• Personalized lifestyle-medical informatics
• Holistic mind-body health
• Motion interval training
• Games for relaxation, brain fitness, and social nudging
• More meaningful biometric data & feedback
Case study: Start with a powerful message and demo. 2010 start study We got these results. Background, demo, slides
Mixed biomarker results highlighting distinct physiological events …6 months is too short to sort these out effectively…but iWell shows that it is possible to document clinically relevant co-occurrences that will eventually allow us to separate out weight loss that is health promoting from weight loss that is not.
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
FPG: 64.0% of high-risk people lowered FPG while 47.5% of the rest moved in the right directionWeight Loss: Greater weight loss for the higher-risk subgroup vs. the rest 6.4 vs. 3.7 lbs Seniors lost 7.8% weightCholesterol: Reduction from 223 to 214 mg/dL vs. little change for the rest Triglycerides (more meaningful measure for insulin resistance): Reduction from 183 to 157 mg/dL vs. little change for the rest
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)
Statistically significant weight loss 3.2% for overall7.8% for those ≥ 65Predictive model showed the importance of social components and key service features in weight loss Social network perturbation and social nudging effective in improving activities The more friends, the more steps Role model and social nudgingCompetitions, wellness games, and goal-related tasks Emerging influencers Peer-to-peer Significant increase in physical activities 2-4K to 8.5K steps per dayGood engagement in service usage stats 20% drop out50% still using after the pilotThe higher the disease burden, the better outcomesGood biomarker results for those at high risk, but mixed results for the rest highlighting distinct physiological events (6 months too short)