3.3 Flu Prediction Better Than Google
39
them, while two nearby cities without any direct flights between them might be
largely separated.
16
Dirk Brockmann and I started to collaborate in 2011, when Germany was
witnessing the spread of the deadly, food-borne EHEC epidemic. I got in touch
with Dirk and suggested that we could combine a model of the spread of epidemics
with a model of food supply chains. In this way, we wanted to identify the location
where the disease originated, which was unknown at that time. Unfortunately, we
could not obtain proper supply chain data then. But our discussion triggered a number
of important ideas. In particular, the research activities shifted from predicting the
spread of diseases toward detecting the locations where they originate.
In fact, when analyzing empirical data of infections as a function of effective
distance from the perspective of all airports worldwide, we found that the most
circular spreading pattern identifies the most likely origin of the disease. More importantly, however, once the location of origin of a disease is known, one can use the
circular spreading dynamics as a function of effective distance to predict the order
in which cities will be hit by a pandemic.
17 This helps to put medical drugs (such as
immunization shots) and doctors in place where they are most effective in countering
the impact and spread of the disease.
When Ebola broke out, Dirk furthermore used the method discussed above to
make early predictions about possible cases in other countries. This helped to inform
international preparations to contain the virus.
18 However, I would also like to highlight here the fantastic research teams of Alessandro Vespignani and Vittoria Colizza,
both partners of the FuturICT initiative. To predict the spread of diseases, they have
built a very detailed and sophisticated simulator. Whenever a disease breaks out,
this simulator can be used to test the effectiveness of countermeasures and inform
policy-makers around the world.
19 It was found, for example, that closing down
some airline connections can only delay the spread of the disease, while the best way
for industrialized countries to protect themselves from diseases such as Ebola is to
16 Independently of Dirk Brockmann’s activitiess, I became interested in the modeling of epidemic
spread back in 2002. In the wake of the September 11 attacks the year before, there were fears that
terrorists could use anthrax or other deadly germs to threaten the USA and the rest of the world.
At this time, I proposed to Otto Schily, the then German Minister of Internal Affairs, to build a
self-calibrating epidemic simulator to predict the spread of pandemics. Directly after the outbreak
of a disease, accurate data about infection and recovery rates is often not available. Thus, the idea
was that a self-adaptive calibration model could produce increasingly accurate predictions, as more
data became available. At that time, I received a letter stating that such an approach was not feasible.
But of course, it was!
17 See the movie http://www.youtube.com/watch?v=ECJ2DdPhMxI. It turns out that this technique
can be successfully applied even in cases where certain key information (such as the infectiousness
of the disease and recovery rate) is not well-known, which is typical after the outbreak of a new
disease. The only data besides the outbreak location which is important for our analysis is the
volume of passenger air traffic between all airports. This is needed to specify the effective distance.
18 See http://rocs.hu-berlin.de/projects/ebola/.
19 See https://www.youtube.com/watch?v=YstB9VWDUqE and http://www.gleamviz.org/.
39
them, while two nearby cities without any direct flights between them might be
largely separated.
16
Dirk Brockmann and I started to collaborate in 2011, when Germany was
witnessing the spread of the deadly, food-borne EHEC epidemic. I got in touch
with Dirk and suggested that we could combine a model of the spread of epidemics
with a model of food supply chains. In this way, we wanted to identify the location
where the disease originated, which was unknown at that time. Unfortunately, we
could not obtain proper supply chain data then. But our discussion triggered a number
of important ideas. In particular, the research activities shifted from predicting the
spread of diseases toward detecting the locations where they originate.
In fact, when analyzing empirical data of infections as a function of effective
distance from the perspective of all airports worldwide, we found that the most
circular spreading pattern identifies the most likely origin of the disease. More importantly, however, once the location of origin of a disease is known, one can use the
circular spreading dynamics as a function of effective distance to predict the order
in which cities will be hit by a pandemic.
17 This helps to put medical drugs (such as
immunization shots) and doctors in place where they are most effective in countering
the impact and spread of the disease.
When Ebola broke out, Dirk furthermore used the method discussed above to
make early predictions about possible cases in other countries. This helped to inform
international preparations to contain the virus.
18 However, I would also like to highlight here the fantastic research teams of Alessandro Vespignani and Vittoria Colizza,
both partners of the FuturICT initiative. To predict the spread of diseases, they have
built a very detailed and sophisticated simulator. Whenever a disease breaks out,
this simulator can be used to test the effectiveness of countermeasures and inform
policy-makers around the world.
19 It was found, for example, that closing down
some airline connections can only delay the spread of the disease, while the best way
for industrialized countries to protect themselves from diseases such as Ebola is to
16 Independently of Dirk Brockmann’s activitiess, I became interested in the modeling of epidemic
spread back in 2002. In the wake of the September 11 attacks the year before, there were fears that
terrorists could use anthrax or other deadly germs to threaten the USA and the rest of the world.
At this time, I proposed to Otto Schily, the then German Minister of Internal Affairs, to build a
self-calibrating epidemic simulator to predict the spread of pandemics. Directly after the outbreak
of a disease, accurate data about infection and recovery rates is often not available. Thus, the idea
was that a self-adaptive calibration model could produce increasingly accurate predictions, as more
data became available. At that time, I received a letter stating that such an approach was not feasible.
But of course, it was!
17 See the movie http://www.youtube.com/watch?v=ECJ2DdPhMxI. It turns out that this technique
can be successfully applied even in cases where certain key information (such as the infectiousness
of the disease and recovery rate) is not well-known, which is typical after the outbreak of a new
disease. The only data besides the outbreak location which is important for our analysis is the
volume of passenger air traffic between all airports. This is needed to specify the effective distance.
18 See http://rocs.hu-berlin.de/projects/ebola/.
19 See https://www.youtube.com/watch?v=YstB9VWDUqE and http://www.gleamviz.org/.