IEEE Spectrum June, 2015 - 65

Data sources: worlD health orGanization; virGinia bioinforMatics institute, virGinia tech

Special Report

World Health Organization

Liberia

29 Apr 2015 12K

12 K
Cumulative cases

Cumulative cases

8K
6K

World Health Organization

29 Apr 2015 11 K

10 K

10 K
8K

Sierra Leone

6K
4K
2K
0

0
Mar
2014

May

Jul

Sep

Nov

Jan
2015

Mar
2014

Mar

Virginia Tech
200K

May

Jul

Sep

Nov

Jan
2015

Mar

Jan
2015

Mar

Virginia Tech
60 K

31 Dec 2014 175K
Cumulative cases

Cumulative cases

WHO declared a "public health emergency of international concern." The
world was playing catch-up from the
start, and so were the modelers.
Virginia Tech graduate student
Caitlin Rivers, who was finishing her
doctorate in computational epidemiology, was one of those running
the race. When her advisor, Bryan
Lewis, asked her to turn her attention to Ebola in July, she started with
a model built from the data of a 1995
outbreak in Central Africa and published by a French scientist in 2007.
That model partitioned the population into categories: susceptible,
exposed, infectious, hospitalized,
dead but not yet buried, and those
no longer capable of virus transmission because of burial or recovery. It
considered everyone, from infants
to the elderly, to be at equal risk of
infection and assumed the disease
would move through a population
in a homogeneous fashion. In other
words, it was a gross oversimplification, especially for a complex pathogen like the Ebola virus. But it was
better than nothing.
To sharpen her model, Rivers knew
she needed this outbreak's case counts
per day or week. These are essential
for building an epidemic curve, which
visually depicts the growth of an outbreak over time. Rivers searched the
Web and found that the Liberia and
Sierra Leone governments published
reports (sporadically, but sometimes
twice a day) listing the numbers of new
cases, cases in treatment, contacts
being sought, and contacts found. The
data was in PDF form, so she did the
"very unglamorous work" of transcribing it onto a machine-readable spreadsheet. She knew other researchers
would be interested, so she posted it
on GitHub, a website for open-source
data. By October, her database was
getting 2,000 visits a day.

Hacking the Human OS

150K
Liberia
100K
50K

31 Dec 2014 46K

50 K
40 K

Sierra Leone

30 K
20 k
10 K

0
Mar
2014

May

Jul

Sep

Nov

Jan
2015

Mar

Mar
2014

May

Jul

Sep

Nov

The Problem With Prophecy
The Virginia Tech modelers produced their forecasts [bottom row] for the Ebola epidemic's
course in Liberia and Sierra Leone in October 2014 based on data from preceding months,
when the number of people infected by the virus was growing exponentially. Their forecasts,
and those of many other modelers, vastly overpredicted the actual toll of the disease. As of
May 2015, the World Health Organization's record [top row] put the cumulative total of Ebola
cases at 10,564 for Liberia and 12,440 for Sierra Leone.

Once Rivers and Lewis had four
weeks of data in hand, they cranked
the first curve out of the model. "It
showed exponential growth," Rivers
recalls. "Which is really the last thing
you want to see in an outbreak."
The data Rivers compiled didn't
include some information essential
for forecasting, such as the length of
time from exposure to infectiousness
and the duration of infectiousness.
Some modelers used values measured in previous Ebola outbreaks,
but Rivers doubted that historical data
would apply to the current scenario.
So she had her computer program estimate those values based on what the
epidemic curve was showing. It was
a good strategy; her values turned

out to be close to the actual ones the
WHO eventually published. Her model
also did well in calculating a key variable called the "basic reproduction
number," which is the number of secondary infections caused by each primary infection in the early phase of
an epidemic. Rivers's model came up
with the number 2.2, not far from the
WHO's final figure of 1.8.
What wasn't close was Virginia
Tech's crucial forecast of how big the
epidemic might get. On 16 October
2014, the team published a paper saying there could be 175,000 cumulative cases in Liberia by the end of the
year. It was a startling figure, especially in light of the fact that at the
time of publication, there had been

SPectrum.ieee.orG

north american

jun 2015

http://SPectrum.ieee.orG

Table of Contents for the Digital Edition of IEEE Spectrum June, 2015