Vous êtes sur la page 1sur 25

HealthMap org

HealthMap.org

John Brownstein, PhD

Children’ss Hospital Boston, Harvard Medical School  
Children Hospital Boston Harvard Medical School

WHO investigated outbreaks (1996‐2009)
Lab 
Confirmation

Detection

Start

Time from outbreak start to detection
i f b k d i

Time from outbreak start to detection
Improves 2 days per year – 20 days in 2008
Source of outbreak news verified by WHO
Source of outbreak news verified by WHO

Adapted from Heymann 2001


www.healthmap.org
h lth

HealthMap Article Processing


HealthMap Article Processing

Acquisition Extraction Categorization Aggregation


>20,000 sites 1800 disease patterns 6 million phrases Text Matching
Every hour; 24/7 5000 location patterns 91% accuracy Similarity Score
lh
HealthMap Reports

>> 600 alerts per day from 22 sources 
600 alerts per day from 22 sources
from over 20,000 websites

Alerts in 201 countries & territories  
Alerts in 201 countries & territories
and 175 disease categories

Seven languages –
g g English, French, 
g , ,
Spanish, Russian, Chinese, 
Portuguese, Arabic
Focus on Haiti

Enhanced surveillance for measles, malaria, meningitis,  
gastrointestinal diseases, repiratory illnesses. 

Sick Individuals

Informal Report
Informal Report

Automated collection

Feed to public health


Feed to public health
Sick Individuals

Informal Report
Informal Report

Automated collection

Feed to public health


Feed to public health

A Community of Users
f

Goal: develop a
community collaboration
that incorporates the
strengths of automation
and human input
p
¾ Improves timeliness
¾ Improves global reach
¾ Improves curation
Public Health 2.0 and surveillane
bl lh d ll

Feed to local, state,


national public health
Geo‐alerting Geo‐reporting
>100k downloads in first month
>100k downloads in first month 

1k submissions first week
k b f k
Limitations

Validation
Lack of details
Issues of self‐diagnosis
Intentional false alarms

What about aggregation?

iPhone and H1N1
h d
iPhone and possible H1N1
h d bl

iPhone
h and not H1N1
d
iPh
iPhone Submissions vs  CDC sentinel surveillance
S b i i CDC ti l ill

R2=0 74
R2=0.74

crowd‐sourced surveillance
d d ll
Closures “Outbreak of laboratory‐confirmed H1N1 in schools in the
provinces of Lucca and Pisa, Tuscany, Central Italy” “Canterbury
elementary school closed until 10/23 due to 30% percent of students
out with flu.
flu.”

Clinical “10 year old boy confirmed test positive. Innitial onset came
on fast with extreme headache, fatigue, and low/med grade fever.
He began tamiflu same day. Is asthmatic. So far is recovering well.”

Clusters “First my 5 year old son got it then my 18 month old


daughter got it. Now my wife and I both have it.”
Android
Potential for increased global coverage

HealthMap Hotline

919‐MAP‐1‐BUG (627‐1284)
Leave a voicemail or Send SMS
Combining informal data with traditional surveillance
Combining informal data with traditional surveillance
Conclusions

Value in the fusion and visualization of distributed electronic 
resources

Complements traditional public health surveillance systems


Complements traditional public health surveillance systems

Importance of multi‐lingual, collaborative approach that 
minimizes information overload and engages users
i i i i f ti l d d

HealthMap’s mission is to provide customized real‐time 
intelligence for the broad scope of public health activities

HealthMap Team Acknowledgments


Clark Freifeld
Susan Aman
Mikaela Keller, PhD Funding
Amy Sonricker, MPH
Emily Chan, MSc
Sumiko Mekaru,
Mekaru DVM
Annie Gatwood, PhD
Leila Amerling, MBA
Children’ss Hospital Informatics Program
Children
Ken Mandl, MD MPH
Ben Reis, PhD
Isaac Kohane, MD PhD
ProMED
Larry Madoff, MD
Marjorie Pollack , MD
Tim Brewer, MD
Slope: ‐0.007 days/day (95% CI: ‐0.011, ‐0.003)
Or ‐2.5days/year (95% CI: ‐3.9, ‐1.0)
Æ SIGNIFICANT
Calculations excluded outlier at around 2003‐01‐01
Lightweight client side HealthMap
g g p

Adobe Air Application
h l h ll
Technical Challenges
Formatting, quality
Georeferencing while preserving privacy
Potential for abuse
Scalability
Rules and features for automated filtering
¾ Distance from phone to event
¾ Photo
¾ Reputation

ld l
Validation Process: Example
Galvin middle school teacher; teacher diagnosed
with bacterial meningitis.

Additional info
Coordinates of iphone
Email for follow‐up
Associated news media
ld l
Validation Process: Example
Galvin middle school teacher; teacher diagnosed
with bacterial meningitis.
Submitted from wakefiled,, ma

Also have email address for follow


follow‐up
up

ld l
Validation Process: Example
Galvin middle school teacher; teacher diagnosed
with bacterial meningitis.
Submitted from wakefiled,, ma

Also have email address for follow


follow‐up
up
New collaborations/Data sources

Clinical data (CDC/GeoSentinel)

Wildlife data (WCS/GAINS)

Baseline epi data (WHO)

Global Spread of H1N1
l b l d f

Surveillance of
Surveillance of
Clusters
Closures
Community transmission
y
Changes in transmission
Changes in age distribution
f lh h
Limitations of current HeatlhMap approach
Can t mine all possible web sources
Can’t
Can’t mine in all languages
Can’tt mine all media types
Can
Delay required for searching, curating and processing
Resource limited process (both machine and human)

Traditional public health reporting Labs

Practitioners Public

Local officials

Ministry of Health Practitioners Public

Practitioners Public

Local officials

Practitioners Public
World Bodies 
(UN, WHO, FAO, OIE)
Practitioners Public

Local officials

Practitioners Public

Ministry of Health Practitioners Public

Local officials

Practitioners Public

Labs
Informal reporting
Informal reporting
World Bodies 
(UN, WHO, 
FAO, OIE)

Local Health  Ministries of 
Officials Health

Informal 
Surveillance

General Public Laboratories

Healthcare 
Public health 
workers, 
practitioners
Clinicians

Surveillance and IHR 2005


Surveillance and IHR 2005

Outbreak detection
Pre: 43.1 (37.0,49.2)
Post: 23.2 (14.2,32.1)

Outbreak report
Pre: 49 6 (43 3 55 9)
Pre: 49.6 (43.3, 55.9)
Post: 37.3 (25.2, 49.3)

N=244 N=30
l h l
Preliminary iphone results
Improved international coverage
¾ Reporting in place and languages not covered by 
HealthMap (articles in Swedish)
¾ Increased reporting of smaller stories (school 
absences)

Improve timeliness
¾ 148 submissions with supporting validated reporting
148 b i i ith ti lid t d ti
¾ 63% of submissions collected before HealthMap

HealthMap Users
HealthMap Users

10,000 unique visitors/day

HHS Command Center >1,000,000 since launch in 9/06

Top visitors:
¾ CDC.gov
¾ WHO.int
¾ DHS.gov
DHS gov
¾ National, state, local public health depts
ECDC Command Center ¾ NGOs
¾ National Conventions 

Collaborations: CDC, WHO, ProMED,


GeoSentinel, Argus, GPHIN, HHS, local/state
public health

Liberty Science Museum, NYC


Community Submissions
b

• Many ways users can submit reports to HealthMap

• User submissions appear on map after being reviewed

Engaging ProMED Users

Human network has proven 
value (ProMED)

Automated Internet‐based 
surveillance offers adjunct to 
human effort

Value in customized delivery of 
information that provides early 
warning and situation context 
g

Vous aimerez peut-être aussi