38
3 Social Forces
3.2 Monitoring the Flu and Other Diseases
Pandemics are a major threat to humanity. Some of them have killed millions of
people. The Spanish flu in 1918 was a shocking example of this. In fact, such
pandemics are expected to happen time and again because viruses keep mutating,
such that our immune systems might be unprepared. For instance, the world was
caught by surprise by the Ebola outbreak, and recently by COVID-19.
To contain the spread of epidemics, the World Health Organization (WHO) is
continuously monitoring emerging diseases. It takes about two weeks to collect
the data from all the hospitals in the world, meaning that each overview of the
current situation is two weeks out of date. However, Google Flu Trends pioneered an
approach called “nowcasting”, which was celebrated as major success of Big Data
analytics at that time. It was claimed that it is possible to estimate the number of
infections in real-time, based on the search queries of Google users. The underlying
idea was that queries such as “I have a headache” or “I don’t feel well” or “I have a
fever”, and so on, might indicate that the user has the flu. While this makes a lot of
sense, the Google Flu approach was recently found to be unreliable, partly because
Google constantly changes its search algorithms and also because advertisements
bias people’s behavior.
13
3.3 Flu Prediction Better Than Google
Fortunately, a model using much less data than Google Flu can be applied to analyze
how a disease spreads, namely by augmenting data of infections with a model based
on air travel data. Dirk Brockmann and I found this approach in 2012/13. About ten
years back, Dirk started to investigate the spread of diseases by analyzing the time and
geographic location of infections using computer simulations. He also analyzed the
paths of dollar bills in his famous “Where is George?” study.
14 But when visualizing
the spatio-temporal spread of epidemics, the patterns looked frustratingly chaotic
and unpredictable. The relationship between the arrival time of a new disease as
a function of the distance from the place where it originated was so scattered that
it was hard to make much sense of the data. Eventually, however, it became clear
that this problem resulted from the high volume of passenger air travel. Thus, Dirk
had the idea to define an “effective distance”, based on the volume of travel between
airports in the world, and to study the spread of disease as a function of this alternative
measure of distance.
15 In effective distance, two airports such as New York City and
Frankfurt are close to each other because of the large passenger flows connecting
13 Lazer et al. [5].
14 Brockmann et al. [6], see also http://www.youtube.com/watch?v=kn32vavZqvg.
15 See http://rocs.hu-berlin.de/ and http://rocs.hu-berlin.de/corona/docs/model/visual_analytics/,
https://www.youtube.com/watch?v=zEO8yZoNBsk.
3 Social Forces
3.2 Monitoring the Flu and Other Diseases
Pandemics are a major threat to humanity. Some of them have killed millions of
people. The Spanish flu in 1918 was a shocking example of this. In fact, such
pandemics are expected to happen time and again because viruses keep mutating,
such that our immune systems might be unprepared. For instance, the world was
caught by surprise by the Ebola outbreak, and recently by COVID-19.
To contain the spread of epidemics, the World Health Organization (WHO) is
continuously monitoring emerging diseases. It takes about two weeks to collect
the data from all the hospitals in the world, meaning that each overview of the
current situation is two weeks out of date. However, Google Flu Trends pioneered an
approach called “nowcasting”, which was celebrated as major success of Big Data
analytics at that time. It was claimed that it is possible to estimate the number of
infections in real-time, based on the search queries of Google users. The underlying
idea was that queries such as “I have a headache” or “I don’t feel well” or “I have a
fever”, and so on, might indicate that the user has the flu. While this makes a lot of
sense, the Google Flu approach was recently found to be unreliable, partly because
Google constantly changes its search algorithms and also because advertisements
bias people’s behavior.
13
3.3 Flu Prediction Better Than Google
Fortunately, a model using much less data than Google Flu can be applied to analyze
how a disease spreads, namely by augmenting data of infections with a model based
on air travel data. Dirk Brockmann and I found this approach in 2012/13. About ten
years back, Dirk started to investigate the spread of diseases by analyzing the time and
geographic location of infections using computer simulations. He also analyzed the
paths of dollar bills in his famous “Where is George?” study.
14 But when visualizing
the spatio-temporal spread of epidemics, the patterns looked frustratingly chaotic
and unpredictable. The relationship between the arrival time of a new disease as
a function of the distance from the place where it originated was so scattered that
it was hard to make much sense of the data. Eventually, however, it became clear
that this problem resulted from the high volume of passenger air travel. Thus, Dirk
had the idea to define an “effective distance”, based on the volume of travel between
airports in the world, and to study the spread of disease as a function of this alternative
measure of distance.
15 In effective distance, two airports such as New York City and
Frankfurt are close to each other because of the large passenger flows connecting
13 Lazer et al. [5].
14 Brockmann et al. [6], see also http://www.youtube.com/watch?v=kn32vavZqvg.
15 See http://rocs.hu-berlin.de/ and http://rocs.hu-berlin.de/corona/docs/model/visual_analytics/,
https://www.youtube.com/watch?v=zEO8yZoNBsk.