ARTICLE IN PRESS

Journal of Theoretical Biology 245 (2007) 125–129

www.elsevier.com/locate/yjtbi

Epidemic outbreaks on structured populations

Alexei Vazquez
The Simons Center for Systems Biology, Institute for Advanced Study, Einstein Dr, Princeton, NJ 08540, USA
Received 24 May 2006; received in revised form 18 September 2006; accepted 18 September 2006
Available online 23 September 2006

Abstract

Our chances to halt epidemic outbreaks rely on how accurately we represent the population structure underlying the disease spread.
When analysing global epidemics this force us to consider metapopulation models taking into account intra- and inter-community
interactions. Here I introduce and analyze a metapopulation model which accounts for several features observed in real outbreaks. First,
I demonstrate that depending on the intra-community expected outbreak size and the fraction of social bridges the epidemic outbreaks
die out or there is a finite probability to observe a global epidemics. Second, I show that the global scenario is characterized by resurgent
epidemics, their number increasing with increasing the intra-community average distance between individuals. Finally, I present
empirical data for the AIDS epidemics supporting the model predictions.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Epidemic outbreaks; Community structure; Metapopulation; Infectious diseases

1. Introduction numerical results indicate the existence of a transition from

local to global epidemics when the expected number of
Human populations are structured in communities infected individuals changing community reaches one
representing geographical locations and other factors (Watts et al., 2005). Here I introduce a related but
leading to partial segregation. This population structure simplified model which can be solved analytically. This
has a strong impact on the spreading patterns of infectious analytical solution allow us to obtain a much deeper insight
diseases among humans, forcing us to consider metapopu- into the main features of global epidemic outbreaks.
lation models making an explicit distinction between the
intra- and inter-community interactions (Rvachev and
2. Model
Longini, 1985; Sattenspiel and Dietz, 1995). The increase
in model realism is paid, however, by an increase in model
Fig. 1 illustrates the general features of an epidemic
complexity. Detailed metapopulation models are difficult
outbreak on a population structured in different commu-
to build and as a consequence they are available for a few
nities. Starting from an index case a disease spreads widely
locations in the world (Rvachev and Longini, 1985;
inside a community, thanks to the frequent intra-commu-
Flahault et al., 1988; Eubank et al., 2004; Germann et
nity interactions. In addition the disease is transmitted to
al., 2006) or they cover a single route of global transmis-
other communities via individuals belonging to different
sion (Hufnagel et al., 2004; Colizza et al., 2006).
communities. While the inter-community interactions may
Recently Watts et al. (2005) introduced a simple
be rare they are determinant to understand the overall
metapopulation model making an explicit distinction
outbreak progression. Based on this picture I divide the
between the intra- and inter-community interactions. In
population in two types or classes. The locals belonging to
spite of the model simplicity it accounts for several features
a single community and the social bridges belonging to
observed in real epidemic outbreaks. In particular, the
different communities. In a first approximation I assume
that (i) all communities are statistically equivalent, (ii) the
Tel.: +1 609 734 8025. mixing between the local and bridges is homogeneous, and
E-mail address: [email protected]. (iii) social bridges belong to two populations. While these

0022-5193/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jtbi.2006.09.018
 ARTICLE IN PRESS
126 A. Vazquez / Journal of Theoretical Biology 245 (2007) 125–129

distance to the index case. Furthermore, the tree can have

at most D generations, where D is the average distance
between individuals inside a community.

3. Spreading regimes

Let us focus on a primary case at generation d and its

secondary cases at the following generation (see Fig. 2). Let
N d ðtÞ denote the expected number of descendants of the
primary case at generation d. In particular N 0 ðtÞ gives the
expected number of descendants from the index case, i.e.
the expected outbreak size. In turn, N dþ1 ðtÞ is the expected
number of descendants generated by a local secondary case
at generation d þ 1. Otherwise, if the secondary case is a
bridge, it starts a new outbreak in a different community
with expected outbreak size N 0 ðtÞ. Putting together the
contribution of locals and bridges we obtain the recursive
equation
8  Rt 
>
< ð1  bÞ1 þ R R0 dGðtÞN dþ1 ðt  tÞ;
> d ¼ 0;
N d ðtÞ ¼ ð1  bÞ 1 þ R~ 0t dGðtÞN dþ1 ðt  tÞ þ bN 0 ðtÞ; 0odoD;
>
>
: 1  b þ bN 0 ðtÞ; d ¼ D:
Fig. 1. Epidemic outbreak on a structured population: schematic
representation of a population structured in communities (big circles) (1)
and the spread of an infectious disease inside and between communities. Iterating this equation from d ¼ D to d ¼ 0 we obtain
The individuals inside a community are divided between locals (open Z t
circles) and bridges (filled circles). The locals transmit the disease to other
individuals inside the community (solid arcs) while the bridges transmit N 0 ðtÞ ¼ 1 þ ð1  bÞF ðtÞ þ b dF ðtÞN 0 ðt  tÞ, (2)
0
the disease to individuals in other communities (dashed arcs). For
simplicity individuals that are not affected by the outbreak are not shown. where
X
D
assumptions are off course approximations they allow us to F ðtÞ ¼ R ~ d1 G %d ðtÞ
½ð1  bÞR (3)
gain insight into the problem. They could be relaxed in d¼1
future works to include other factors such as degree and G %d ðtÞ denotes the d-order convolution of GðtÞ, i.e.
correlations among interacting individuals (Vazquez, Rt
G%0 ðtÞ ¼ 1 and G %dþ1 ðtÞ ¼ 0 dGðtÞG%d ðt  tÞ. F ðtÞ repre-
2006c) and more realistic mixing patterns (Vazquez, sents the expected outbreak size inside a community at time
2006d). t and
An epidemic outbreak taking place inside a community
is then modeled by a multi-type branching process (Mode, N C ¼ lim F ðtÞ, (4)
t!1
1971) starting from an index case (see Fig. 1). The key
intra-community magnitudes are the reproductive number
and the generation times (Anderson and May, 1991;
Vazquez, 2006b). The reproductive number is the average
number of secondary cases generated by a primary case.
The disease transmission introduces some biases towards
individuals that interact more often. Therefore, I make an
explicit distinction between the index case and other
primary cases and denote their expected reproductive
numbers by R and R, ~ respectively. The generation time t
is the time elapse from the infection of a primary case and
the infection of a secondary case. It is a random variable
characterized by the generation time distribution function Fig. 2. Local disease transmission: diagram representing the disease
GðtÞ. These magnitudes can be calculated for different transmission from a primary case at generation d to secondary cases in the
models such as the susceptible infected recovered (SIR) following generation. The secondary cases are locals with probability
model and they can be estimated from empirical data as 1  b, potentially leading to subsequent infections inside their community,
and bridges with probability b, transmitting the disease to other
well. Finally, a community outbreak is represented by a communities. Note that the expected number of descendants generated
causal true rooted at the index case (Vazquez, 2006a, b). In by a secondary case is evaluated at a delayed time t  t, where t is the
this tree the generation of an infected case is given by the generation time.
 ARTICLE IN PRESS
A. Vazquez / Journal of Theoretical Biology 245 (2007) 125–129 127

is the final expected outbreak size inside a community. Calculating the inverse Laplace transform of (6) I finally
When b ¼ 0 it coincides with the expected outbreak size obtain
inside a community (Vazquez, 2006b). When b40 (2)
provides a self-consistent equation to determine the overall X
1
lðltÞDðmþ1Þ1 elt
expected outbreak size after taking into account the inter- IðtÞ ¼ N C ðbN 1 Þm , (14)
m¼0
G½Dðm þ 1Þ
community transmissions.
To calculate N 0 ðtÞ I use the Laplace transform method. where GðxÞ is the gamma function. Fig. 3 shows the
Consider the incidence progression of the incidence as obtained from (14). As
IðtÞ ¼ N_ 0 ðtÞ (5) predicted above, the outbreak dies out when RC o1 while
when RC 41 it grows exponentially. More important, the
and its Laplace transform incidence exhibits oscillations at the early stages, their
Z 1 number increasing with increasing D. For example, we
^
IðoÞ ¼ dt eot IðtÞ. (6) distinguish about two oscillations for D ¼ 10 while for D ¼
0
30 several oscillations are observed. These oscillations
Substituting the recursive equation (2) in (6) I obtain represent resurgent epidemics, which are often observed in
f^ðoÞ real outbreaks (Riley et al., 2003; Anderson et al., 2004)
^
IðoÞ ¼ , (7) and simulations (Sattenspiel and Dietz, 1995; Watts et al.,
1  bf^ðoÞ 2005).
where
Z 1
f^ðoÞ ¼ dt eot F_ ðtÞ. (8) 0.25
0
0.2 D=10
The validity of (6) is restricted to o values satisfying
1  bf^ðoÞ40, resulting in different scenarios depending on I(t)/NC 0.15
the value of the parameter
0.1
RC ¼ bN C . (9)
^ 0.05
Local outbreaks: When RC o1 then IðoÞ is defined for all
oX0 and IðtÞ is obtained inverting the Laplace transform 0
^ is defined from (7) it follows
in (6). Furthermore, since Ið0Þ 0 20 40 60 80 100
that IðtÞ decreases to zero when t ! 1, i.e. the epidemic
0.1
outbreak dies out.
Global outbreaks: When RC 41 the incidence grows D=20
0.08
exponentially IðtÞeoc t , where oc is the positive root of the
equation 0.06
I(t)/NC

bf^ðoÞ ¼ 1. (10) 0.04

These two scenarios are equivalent to those obtained for a 0.02

single community (Anderson and May, 1991). RC repre-
sents the effective community’s reproductive number and 0
0 50 100 150
the threshold condition
0.08
RC ¼ 1 (11)
D=30
delimits the local and global scenarios. 0.06
To go beyond the final outbreak I analyse the progres-
I(t)/NC

sion of the inter-communities outbreak. I assume that the 0.04

disease is transmitted at a constant rate l from a primary
case to a secondary case independently of their type. In this 0.02
case the intra-community incidence is given by (Vazquez,
2006b) 0
D1 lt
0 50 100 150 200
lðltÞ e λt
F_ ðtÞN C , (12)
ðD  1Þ!
Fig. 3. Epidemic outbreak progression: the incidence IðtÞ as a function of
for tbt0 , where time in units of the local disease transmission rate l, for RC ¼ 0:1 (dash-
dotted), 0.5 (dotted), 0.9 (dashed), 1.0 (solid) and 1.1 (dash-dash-dotted).
D  11 The panels from top to bottom corresponds to different average distances
t0 ¼ . (13)
R~ l D between individuals inside a community.
 ARTICLE IN PRESS
128 A. Vazquez / Journal of Theoretical Biology 245 (2007) 125–129

4. Case study: AIDS epidemics 5. Discussion and conclusions

To understand the relevance of these results in a real RC in (11) represents the expected number of infected
world scenario I analyse data reported for the AIDS individuals leaving their community. The numerical
epidemics. First, I estimate the parameter RC determining simulations reported in Watts et al. (2005) indicated the
the spreading regime, local or global. Fig. 4a shows the existence of a transition at RC ¼ 1, from local outbreaks
value of RC across the USA by state. For most states when RC o1 to global epidemics when RC 41. I have
RC 41, reaching significantly large values for several states. demonstrated that there is indeed a phase transition at
For example, RC exceeds 1000 for California and New RC ¼ 1. Furthermore, the analytical solution provides an
York. These numbers indicate that the USA AIDS expression of RC as a function of the bridge’s fraction and
epidemics is in the global spread scenario (RC 41), in the intra-community expected outbreak size (9). RC
agreement with the general belief. represents a measure of the reproductive number at the
Second, I analyse the temporal evolution of the AIDS inter-community level. Its value can be estimated from the
incidence. Fig. 3b and c show the AIDS incidence in USA expected outbreak size inside a community and the bridge’s
and UK by year, indicating a similar temporal pattern. The fraction. Based on the resulting estimate we can determine
epidemics started with an increasing tendency of the if an epidemics is in the local or global epidemics scenario
incidence which, after reaching a maximum, switched to and react accordingly.
a decreasing trend. After some years, however, the The inter-community disease transmission is character-
epidemics resurges with a new incidence increase. This ized by oscillations at the early stages which represents
picture coincides with the model predictions in Fig. 4. resurgent epidemics, the number of these resurgencies
Therefore, a possible explanation of the observed being determined by the characteristic distance between
multiple peaks is the existence of a community structure, individuals within a community. In essence, when D is
which can be attributed to geographical location and other small the time scale characterizing the outbreak progres-
factors. sion within a community is very small (Barthélemy et al.,

6000

4000
RC

2000

0
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
D. Columbia
Florida
Georgia
Hawai
Idaho
Illinois
Indiana
Iowa
Kansas
Kentuchy
Luisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
Soth Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming

(a)

80000 2000

60000 1500
I(t) (AIDS)

40000 1000

20000 500
USA UK
0 0
1980 1985 1990 1995 2000 2005 1990 1995 2000 2005
(b) t (years) (c) t (years)

Fig. 4. USA AIDS epidemics: estimated RC ¼ bN C for the different USA states. (a) b was computed as the ratio between the number of state out-
immigrants and the total state population according to the 1995–2000 USA census (http://www.census.gov). N C was computed as the number of habitants
living with AIDS according to the 2005 statistics published by the US Department of Health (http://www.hhs.gov). AIDS incidence in the USA (b) and
UK (c) by year, as reported by the US Department of Health and the UK Health Protection Agency (http://www.hpa.org.uk), respectively.
 ARTICLE IN PRESS
A. Vazquez / Journal of Theoretical Biology 245 (2007) 125–129 129

2004, 2005; Vazquez, 2006b). Therefore, the time it takes to Colizza, V., Barrat, A., Barthelemy, M., Vespignani, A., 2006. The role of
observe the infection of a social bridge is very small as well, the airline network in the prediction and predictability of global
epidemics. Proc. Natl Acad. Sci. USA 103, 2015–2020.
resulting in the mixing between the intra- and inter-
Eubank, S., Guclu, H., Kumar, V.S.A., Marathe, M., Srinivasan, A.,
community transmissions. In contrast, when D is large it Toroczcai, Z., Wang, N., 2004. Modelling disease outbreaks in realistic
takes a longer time to observe the infection of a social urban social networks. Nature 429, 180–184.
bridge and by that time the intra-community outbreak has Flahault, A., Letrait, S., Blin, P., Hazout, S., Menares, J., Valleron, A.J.,
significantly developed. Therefore, in this last case the 1988. Modelling the 1985 influenza epidemic in France. Stat. Med. 7,
outbreak within communities is partially segregated in 1147–1155.
Germann, T.C., Kadau, K., Longini, I.M., Macken, C.A., 2006.
time. Mitigation strategies for pandemic influenza in the united states. Proc.
When multi-agent models are not available these results Natl Acad. Sci. USA 103, 5935–5940.
allow us to evaluate the potential progression of an Hufnagel, L., Brockmann, D., Geisel, T., 2004. Forecasts and control of
epidemic outbreak and consequently determine the magni- epidemics in a globalized world. Proc. Natl Acad. Sci. USA 101,
tude of our response to halt it. They are also valuable when 15124–15129.
Mode, C.J., 1971. Multitype Branching Processes. Elsevier, New York.
a detailed metapopulation model is available, funneling the Riley, S., et al., 2003. Transmission dynamics of the etiological agent of
search for key quantities among the several model sars in Hong Kong: impact of public health interventions. Science 300,
parameters. More important, this work open avenues for 1961–1966.
future analytical works that side by side with multi-agent Rvachev, L.A., Longini, I.M., 1985. A mathematical model for the global
spread of influenza. Math. Biosci. 75, 3–22.
models will increase our chances to control global
Sattenspiel, L., Dietz, K., 1995. A structured epidemic model incorporat-
epidemics. ing geographic mobility among agents. Math. Biosci. 128, 71–91.
Vazquez, A., 2006a. Causal tree of disease transmission and the spreading
of infectious diseases. In: Discrete Methods in Epidemiology,
DIMACS Series in Discrete Mathematics and Theoretical Computer
References Science, vol. 70. AMS, Providence, RI, pp. 163–179.
Vazquez, A., 2006b. Polynomial growth in age-dependent branching
Anderson, R.M., May, R.M., 1991. Infectious Diseases of Humans. processes with diverging reproductive number. Phys. Rev. Lett. 96,
Oxford University Press, New York. 038702.
Anderson, R.M., et al., 2004. Epidemiology, transmission dynamics and Vazquez, A., 2006c. Spreading dynamics on small-world networks with
control of sars: the 2002–2003 epidemic. Philos. Trans. R. Soc. connectivity fluctuations and correlations. Phys. Rev. E (In press).
London, Ser. B 359, 1091–1105. hhttp://arxiv.org/q-bio.PE/0603010i.
Barthélemy, M., Barrat, A., Pastor-Satorras, R., Vespignani, A., 2004. Vazquez, A., 2006d. Spreading of infectious diseases on heterogeneous
Velocity and hierarchical spread of epidemic outbreaks in scale-free populations: multi-type network approach. hhttp://arxiv.org/q-bio.PE/
networks. Phys. Rev. Lett. 92, 178701–178704. 0605001i.
Barthélemy, M., Barrat, A., Pastor-Satorras, R., Vespignani, A., 2005. Watts, D., Muhamad, R., Medina, D.C., Dodds, P.S., 2005. Multiscale,
Dynamical patterns of epidemic outbreaks in complex heterogeneous resurgent epidemics in a hierarchical metapopulation model. Proc.
networks. J. Theor. Biol. 235, 275–278. Natl Acad. Sci. USA 102, 11157–11162.