Accumulating evidence reveals a strong link between human
mobility and the spread of epidemics. In order to control the spread of an
epidemic, governments can implement mobility restrictions to its citizens. The
effect of such restrictions on the mobility of the population has not been
adequately studied at a large scale mainly due to the lack of relevant data.
Nevertheless, the recent adoption of ubiquitous computing technologies enables
the design of such studies. In this paper we measure the impact that the alerts
issued by the Mexican government had on the mobility of the Mexican
population during the H1N1 flu outbreak in April and May of 2009. The
mobility of individuals was characterized using anonymized Call Detail
Records (CDRs) traces. The results indicate a statistically significant reduction,
of up to 80% in some cases, in the diameter of mobility of individuals.
Top Health Trends: An information visualization tool for awareness of local h...
Measuring the impact of epidemic alerts on human mobility using cell-phone network data
1. Measuring the impact of
epidemic alerts on human
mobility using cell-phone
network data
Enrique Frias-Martinez
Telefonica Research, Madrid, Spain
efm@tid.es
2. Human mobility plays a central role in the spatial
spreading of infectious diseases.
WHO recommends suspension of activities as a
measure to reduce the transmission of a disease.
Actions implemented by the Mexican government
to control the H1N1 flu outbreak of April 2009
constitute an illustrative example.
3. In this paper we measure the impact that the
actions taken by the Mexican government during
April and May of 2009 had on human mobility
using phone Call Detail Records (CDRs).
We evaluate the impact of the alerts using two
approaches:
Population Mobility Analysis
Geographic Mobility Analysis
4. Closed Reopen
Preflu 27th April 6th May
Alert Shutdown
17th April 1st May
Medical Alert (Stage 1)-> Closing of Schools (Stage 2)->
Suspension (Stage 3)
5. Digital Footprints
For the first time in human history, we have
access to large-scale human behavioral data
at varying levels of spatial and temporal
granularities
8. CDRs for 1,000,000 anonymized customers from
one of the most affected Mexican cities were
obtained for a period of 5 months from January
2009 to May 2009.
Only considered users with an average of two
daily calls or more (20% population).
10. Definition of baselines:
baseline 1 - quantifies changes in mobility during Medical Alert
(from Thursday 16th to Wednesday 22nd). Computed using
mobility data from four 7-day time periods – from Thursday to
Wednesday – prior to April 16th.
baseline 2 - defined to evaluate changes in mobility during stage 2
(from Monday 27th to Thursday 30th). Computed using four 4-
day time periods – from Monday to Thursday –prior to April 16th.
baseline 3- defined to identify mobility changes during the stage 3
alert (Friday May 1st to Tuesday May 5th), was computed using
data from Easter holidays.
11. Distribution of the Diameter of Influence for each
baseline and its corresponding alert period.
Lilliefors test indicates normal distribution.
In order to quantify the impact of the
government calls on mobility, a one-sided t-test
(p<0.01) was used to compare the distribution of
diameters between each stage and its
corresponding baseline.
12. Stage 1 alert period: no significant change in
human mobility.
Stage 2 and Stage 3: t-test revealed statistically
significant differences with p<0.01:
the distribution of the diameter of mobility was
reduced during these two stages compared to their
baselines.
13. Quantify the changes in mobility during alert
periods 2 and 3.
CDF of diameter change for April 27th relative to its
baseline - 80% of the population reduced its diameter
- around 50% of them by 20km or more.
14. Evaluates impact that alerts had on critical
infrastructures.
Understand whether the number of individuals
that visited these infrastructures changed.
University Campus
Hospital
Airport
BTS Coverage Area / Correction with Home
algorithm
18. Lilliefors indicates normal distribution of hourly
representation – Not-normal for hourly representation.
Daily representation: t-test
Hospital: No change in any period.
University Campus: Reduction number of visitors in Stage 2 and
Stage 3.
Airport: statistically significant increase in the number of visitors
to the airport during the same stage 2 alert period. No change
period 1 or 3.
Hourly representation: Kolgomorov-Smirnof test and a
Wilcoxon rank-sum test
No statistical differences at hourly level
19. Use of CDR for qualitative & quantitative evaluating the
impact of government mandates.
Reduction in mobility is higher when schools are closed
during regular days (stage 2) than when all non-essential
activities are closed (stage 3) during a period that already
was a holiday.
Total closing of infrastructures (stage 3 alert) might
provoke an increase in the number of individuals that
visit transport hubs before its enactment.
No statistical significant change in the number of visitors
to the Hospital
20. Thanks Mexico!
"An Agent-Based Model of Epidemic Spread using Human Mobility and
Social Network Information", The 3rd IEEE Int. Conf. on Social Computing
(SocialCom 2011), Boston, MA, USA
efm@tid.es
www.enriquefrias-martinez.info