Brief presentation of selection of social media research conducted at Robert H. Smith School of Business; used for seeding discussion of how to use innovative social network analytical techniques applied to FDA priorities.
Social media research at CHIDS for FDA patient prefs wg 05 20 2013
1. Social Media Research
Brief Presentation for FDA
May 2013
Slides available for download at:
http://bit.ly/CHIDSSMforFDAWG
2. Presenters
Ritu Agarwal, PhD
Professor and Dean’s Chair of Information Systems
Co-Director of CHIDS
Kenyon Crowley, MBA, MSIS, CPHIMS
Deputy Director of CHIDS
Director of Health Innovation, COEHITR
Gordon Gao, PhD
Associate Professor of Information Systems
Co-Director of CHIDS
Bill Rand, PhD
Director, Center for Complexity in Business
Assistant Professor of Marketing
Il-Horn Hann, PhD
Associate Professor if Information Systems
Co-Director, Center for Digital Innovation, Thought and Strategy (DIGITS)
2 May 2013
3. Topic agenda
Introduction of research team and CHIDS
Gao
Online ratings of healthcare services
Information diffusion in online communities
Pain management preferences concept
Rand
Trust and Influence in social media
Urgent diffusion
Preference mapping
Social media recruiting
Hann
Social influence and biases in online rating communities
Discussion, concluding remarks and questions
3 May 2013
4. Strategy
TechnologyPolicy
CHIDS
Mission
Designed to research, analyze, and
recommend solutions to challenges
surrounding the introduction and
integration of information and decision
technologies into the health care system
Improve the practice and delivery of health
care by offering researched solutions that
impact safety, quality, access, efficiency, and
return on investment
Works with many partners across
academia, government, clinical orgs and
industry
4 May 2013
5. Research focus areas
Impact and
Comparative
Effectiveness of
Health
Information
Systems
New Models of
Care (ACO, HIE,
PCMH, Care
Transitions)
Healthcare
Analytics (Data-
driven Health
Services Insights,
Modeling,
Operations)
Consumers,
Quality &
Transparency,
and Social Media
May 20135
6. Lack of information in the information age
Jessie Gruman
Three-time Cancer Survivor
Hodgkin’s disease, cervical
cancer, colon cancer
Then a fourth one hit her
“I searched online but found that
comparative quality information
on surgeons specializing in
stomach cancer was virtually
nonexistent.” (Health Affairs,
2013)
6 May 2013
7. Web 2.0 comes to doctors
53 + and growing!
0
100000200000300000400000
2005 2006 2007 2008 2009 2010
ratings physicians
Sources: Gao, McCullough, Agarwal, and Jha 2012
0
2
4
6
8
10
12
14
2005 2006 2007 2008 2009 2010
7 May 2013
8. Do online ratings reflect doctor quality?
Sources: Gao, Greenwood, McCullough, and Agarwal 2012
8 May 2013
9. Knowledge flow in online patient community
Sources: Goh, Gao and Agarwal 2013
9 May 2013
10. Stopping the biggest man-made epidemic
in the United States
Prescription drug overdose
Kills 40,000 people last year
More than deaths from traffic accidents
Distribution of morphine increased 600% from 1997-2007
Use social media as a surveillance tool
Text mining analysis on conversations in pain mgmt forums
Optimize strategies to use social media to curb the growth of drug
overdose?
Identify factors that affect a patient's risk attitude toward drug
overdose;
Leverage the social network structure to reduce drug overdose
(information hub, influentials, followers, etc.)
10 May 2013
11. Trust and Influence in Social Media
Who are the most influential in social
media?
Not just the people who have the
most fans, but the boundary spanners
Who are the most trusted?
Can be measured by observing
behavior over time
Trust is the best predictor of “virality”
and information diffusion
11
12. Observed data and model predictions for Hurricane Sandy data
(first tweets by hour)
0
5
10
15
20
25
30
35
40
45
50
55
Thur23
Fri7
Fri15
Fri23
Sat7
Sat15
Sat23
Sun7
Sun15
Sun23
Mon7
Mon15
Mon23
Tue7
Tue15
Tue23
Wed7
Wed15
Wed23
Thur7
Thur15
Thur23
Fri7
Fri15
Fri23
Sat7
Sat15
Sat23
Sun7
Hour
NumberofTweetsperHour
Observed data
Geometric decay NHPP model
Constant intensity NHPP model
Sum of geometric decay NHPP model
Urgent Diffusion
Constructing a model of the
diffusion of information when the
time scale of external events is as
fast or faster than the diffusion
process
Social media can be used to
identify key individuals who get
information out soon
Sometimes the social network is
irrelevant
The process really depends on the
presence of new information
12 May 2013
13. Preference Mapping from Social Media
Rather than doing focus
groups and surveys, we
can monitor social
media to understand
consumer preferences
Enables a quicker and
faster way to estimate
consumer preferences
13 May 2013
14. Social Media Recruiting
Social Media Monitoring
can be used to identify
individuals with an interest
or concern
Piloting with flu
identification
Can target based on
location
14 May 2013
17. There Is Social Influence in Online Ratings
May 20, 2013
What Implications does this have?
What is the value of online rating systems?
How credible is this information for consumers?
Risk reduction still possible?
In general: Can we separate ‘true’ evaluations from
social influence in social networks?
17 May 2013
18. Bias in Online Ratings
May 20, 2013
Hu, Pavlou and Zhang (2009)
Acquisition bias
Under-reporting bias
18 May 2013
19. Bias in Online Ratings
May 20, 2013
Li and Hitt (2008)
19 May 2013
20. Bias in Online Ratings
May 20, 2013
Wu and Huberman(2008)
Perceived uniqueness in identity
20 May 2013
21. The Social Bias in Online Ratings
May 20, 2013
Social Context Influences Ratings
Social Bias in Social Networks21 May 2013
22. Demo: Negative Rating Condition
May 20, 2013
Friend:1.00
AFTER Case
My rating: 2.30
BEFORE case
My rating: 2.80
Average Rating: 3.30
Homophily
Social Bias
Social Bias in Social Networks22 May 2013
23. Insights from the Model
May 20, 2013
Social Bias
Focal rating is unbiased ONLY when there is no social
factor
Path Dependence
Early ratings affect later ratings, therefore changing overall
rating trajectory
Un-Correctability
Without knowing each user's social preference, it is
impossible to correct this bias
23 May 2013
24. Discussion
How can these findings, techniques and future
extensions of this research be applied to FDA
patient preferences goals and objectives?
Questions
24 May 2013
25. Follow up
Thank You!
Kenyon Crowley
kcrowley@rhsmith.umd.edu
Mobile: (919) 649-2279
Slides available for download at:
http://bit.ly/CHIDSSMforFDAWG
25 May 2013