Characterizing the reproduction number of epidemics with early sub-exponential
growth dynamics
Gerardo Chowell1,2 , Cécile Viboud2, Lone Simonsen3,4, Seyed Moghadas5
1
School of Public Health, Georgia State University, Atlanta, GA, USA
2
Division of International Epidemiology and Population Studies, Fogarty International
Center, National Institutes of Health, Bethesda, MD, USA
3
Department of Public health, University of Copenhagen, Copenhagen, Denmark
4
Department of Global Health, School of Public Health and Health Services, George
Washington University, Washington DC, USA
5
Agent Based Modelling Laboratory, York University, Toronto, Canada.
Corresponding author:
Gerardo Chowell, PhD
School of Public Health, Georgia State University, Atlanta, GA, USA
Division of International Epidemiology and Population Studies, Fogarty International
Center, National Institutes of Health, Bethesda, MD, USA
Email: [email protected]
Body word count: 3205
Abstract word count: 192
1
Abstract
The reproduction number is a central parameter in epidemiology used to quantify the
average number of secondary cases generated by a primary infectious individual during the
early epidemic growth phase. Existing methods to estimate the reproduction number
generally assume early exponential epidemic growth, but do not account for the possibility
of early sub-exponential (i.e., polynomial) growth. Here, we introduce a novel method for
estimating the reproduction number without making explicit assumptions about the early
epidemic growth profile. We demonstrate our methods using both synthetic and real
epidemic datasets. Our results indicate that the effective reproduction number for
epidemics characterized by early sub-exponential growth exhibits a natural downward
trend over time approaching unity, even in the absence of control interventions, or
depletion of susceptibles. This pattern is in stark contrast with the invariant reproduction
number predicted for epidemics with an initial exponential growth phase. Our findings
provide a compelling argument for understanding the early extinction of some emerging
disease outbreaks during the early ascending phase of sub-exponential growth. A reliable
data-driven characterization of the early epidemic phase is crucial for estimating the
reproduction number, forecasting disease dynamics, and guiding public health intervention
strategies.
Keywords: Reproduction number, exponential growth; sub-exponential growth,
polynomial growth; epidemics; infectious disease.
2
Introduction
There is a long and successful history of using compartmental transmission models to
study epidemic dynamics, often calibrated using time series data describing the
progression of the epidemic [1-6]. A fundamental tenet of classic epidemic theory is that
the initial growth phase should be exponential in the absence of susceptible depletion,
control interventions and behavior changes. Structured compartmental models integrating
age, high-risk groups, geography, or changes in behavior or interventions, can alleviate up
to some extent departures from exponential growth theory when disease transmission is
heterogeneous (see e.g., [7-11]). In some cases however, the underlying mechanisms are
difficult to disentangle and hence to model, or the models may become too complex and
intractable. Indeed, early sub-exponential (e.g., polynomial) growth patterns have been
observed during the early phase of outbreaks of HIV/AIDS [12-14] Ebola [15], and footand-mouth disease [16], for which the biological mechanisms remain debated.
Accordingly, a generalized-growth extension to standard compartmental models has been
proposed to accommodate more diverse epidemic profiles, where a tuning parameter
(deceleration of growth, p) can mirror a range of epidemic dynamics from constant
incidence (p=0) to exponential growth (p=1) [16].
Application of generalized-growth models to empirical data support a range of epidemic
behaviors across human pathogens, with notably slow spread (p <0.5) for district-level
Ebola outbreaks in parts of West Africa, intermediate spread profiles for historical plague
and smallpox outbreak (p=0.8), and near exponential dynamics for pandemic influenza
(p~1) [17]. Hence, departure from standard epidemic theory may be more common than
previously thought since transmission heterogeneities are the rule rather than the exception
[16]. The generalized growth model is a convenient approach to faithfully capture the
early dynamics of a range of infectious disease outbreaks without the need to understand
these heterogeneities.
The reproduction number R0 is a key parameter that characterizes the early epidemic
spread, and can be used to inform public health authorities on the level of risk posed by an
infectious disease and the potential effects of control interventions [18]. According to the
3
classical epidemic theory of epidemics, largely based on compartmental modeling (e.g., [1,
2, 19, 20]), R0 is expected to remain invariant during the early phase of an epidemic that
follows exponential growth and before susceptible depletion sets in [2]. Here, we expand
on the generalized-growth method [16] to estimate the reproduction number of outbreaks
in the context of early exponential and sub-exponential growth dynamics. We illustrate our
approach using case incidence data for a diverse set of historic and contemporary
infectious disease outbreaks and show that consideration of sub-exponential growth
dynamics is important for accurate assessment of the reproduction number.
Materials and Methods
Methods
The reproduction number for exponential and sub-exponential growth epidemics
We extend a previously described generalized-growth model to estimate the effective
reproduction number Rg over disease generations g [16]. Briefly, the generalized growth
model departs from the standard compartmental modelling theory by relaxing the
assumption of exponential growth dynamics in the early ascending phase of an outbreak,
following:
C '(t) = rC p
(1)
where C '(t) describes the incidence curve over time t, the solution C(t) describes the
cumulative number of cases at time t , r is a positive parameter denoting the growth rate
(1/time), and p∈ [0,1] is a ‘deceleration of growth’ parameter. If p=0, this equation
describes constant incidence over time and the cumulative number of cases grows linearly,
whereas p=1 describes exponential growth dynamics (i.e., Malthus equation) and the
solution is given by: C(t) = C0 ert . For early exponential growth, the average number of
4
secondary cases generated by initial cases during the first generation interval, Tg (assumed
to be fixed) is estimated by [21, 22]:
exp
0
R
C '(Tg ) rC0 erTg
rT
=
=
=e g
C '(0)
rC0
(2)
The expression for R0exp only depends on r and Tg . Moreover, in the absence of control
interventions or behavior changes R0exp remains invariant at e
rTg
. This can be shown by
analyzing, Rgexp , the ratio of case incidences over consecutive generation intervals, which is
given by:
Rgexp =
C '[(g + 1)Tg ] rC0 er (g+1)Tg
r (g+1)Tg −rgTg
rT
=
=e
=e g
rgTg
C '[gTg ]
rC0 e
(3)
In the case of sub-exponential growth, i.e. when p < 1 , no expression for the reproduction
number has been derived. For such polynomial epidemics, equation (1) exhibits an explicit
solution that describes the cumulative number of cases over time, C sub exp (t) , in the form of
[23]:
C sub exp (t) = (r(1− p)t + A)1/(1− p)
(4)
where A = C01− p is a constant that depends on the initial number of cases, C0 . Hence, the
corresponding incidence equation is given by:
C sub exp' (t) = r[r(1− p)t + A] p/(1− p)
5
(5)
These formulae can be used to derive an expression for the reproduction number in the first
generation interval, denoted by R0sub exp , which quantifies the average number of cases
generated by initial cases during the first generation interval and is given by:
sub exp
0
R
C sub exp' (T ) ⎛ r(1− p)Tg + A ⎞
= sub exp' g = ⎜
⎟⎠
C
(0) ⎝
A
p/(1− p)
r(1− p)Tg ⎞
⎛
= ⎜ 1+
⎟⎠
⎝
A
p/(1− p)
(6)
In general, the expression for R0sub exp depends on r, p,Tg and the initial number of cases, C0 .
When C0 = 1 , this ratio depends only on r, p and Tg and is given by:
(
R0sub exp = 1+ r(1− p)Tg
)
p/(1− p)
(7)
The expression can be generalized to later disease generations, g , by analyzing, Rgsub exp , the
ratio of case incidences over consecutive disease generations, which is given by:
Rgsub exp =
C sub exp' [(g + 1)Tg+1 ] r[r(1− p)(g + 1)Tg + A] p/(1− p)
=
=
C sub exp' [gTg ]
r[r(1− p)gTg + A] p/(1− p)
⎛
r(1− p)Tg ⎞
= ⎜ 1+
r(1− p)gTg + A ⎟⎠
⎝
p/(1− p)
(8)
In contrast to the exponential growth model where R0exp was independent of disease
generation during the early growth phase, we observe that Rgsub exp varies over time as a
function of g . Since A in Equation (8) is small, the ratio
g increases, and thus Rgsub exp approaches 1.0.
6
r(1− p)Tg
declines to zero as
r(1− p)gTg + A
Numerical estimation of the reproduction number
The reproduction number can be estimated from incidence data comprising early epidemic
growth phase and information about the distribution of the generation time of the disease
(Table 1) [24]. Specifically, based on the early growth phase in case incidence at calendar
time ti denoted by I i , and the discretized probability distribution of the generation interval
denoted by ρ i , the reproduction number is estimated by [24, 25]:
R(ti ) =
Ii
i
∑I
i− j
(9)
ρj
j=0
where the denominator represents the disease prevalence at time ti [24].
Simulations of the reproduction number
We simulated the temporal variation in the effective reproduction number for outbreaks
that are characterized by initial sub-exponential and exponential growth dynamics. We
analyzed these temporal profiles in the reproduction number in the first 5 disease
generations using Equation (9). For this purpose, we simulated early growth outbreak data
using the generalized growth model (Equation 1) and assumed different distributions of the
disease generation interval (e.g., exponential, gamma, uniform, delta). For this purpose, we
fixed the growth rate parameter r and varied the ‘deceleration of growth’ parameter p
between 0 and 1 [16]. Furthermore, we also compared our numerical simulations of the
reproduction number assuming a fixed generation interval using Equation (9) with
analytical results obtained using Equations 3 and 7-8.
7
Application to real outbreak data
We obtained case incidence data for various infectious disease epidemics including
pandemic influenza, measles, smallpox, bubonic plague, cholera, foot-and-mouth disease
(FMD), HIV/AIDS, and Ebola. The temporal resolution of the datasets varied from daily,
weekly, to annual. We employed these datasets to illustrate the estimation of the
reproduction number and represent a convenience sample encompassing a range of
pathogens, geographic contexts, and time periods (Table 1). For each outbreak, the onset
week corresponds to the first observation associated with a monotonic increase in incident
cases, up to the peak incidence. We estimated the reproduction number based on Equation
(9), assuming that the generation interval of the disease follows a gamma distribution with
mean and variance given in Table 1 [11, 26-31].
We estimated parameters r and p of the generalized growth model based on Equation 1, in
line with past work [16] . Specifically, a nonlinear least-squares fitting procedure was
applied to case incidence curves modeled by equation C’( t ) during the initial epidemic
phase comprising approximately 3 to 5 disease generations, when the proportion of
susceptible individuals in the population approximates its initial value [16]. We used the
Levenberg-Marquardt algorithm implemented in MATLAB (The Mathworks, Inc.) as in
prior studies (e.g, [17]). The initial number of cases C(0) was fixed according to the first
observation. Confidence intervals for the model parameter estimates were constructed by
simulating 200 realizations of the best-fit curve C’( t ) using parametric bootstrapping with
a Poisson error structure, as in prior studies [17, 32]. Parameters r and p were then
estimated from each of 200 simulated epidemic curves to derive nominal 95% confidence
intervals.
8
Results
We first analyzed simulations of early epidemic growth for different values of the growth
rate r and the “the deceleration of growth” parameter p, and assuming a fixed generation
interval (Figure 1 and Figure S1). Our simulations assuming a fixed generation interval
confirm the analytical results described in Equations 3 and 7-8 in relation to the behavior
of the reproduction number. In particular, in the case of sub-exponential growth (p<1), and
for a given growth rate, simulations show that the effective reproduction number
approaches 1.0 as disease generations progress. In contrast, for exponential growth (p=1),
the effective reproduction number settles at a constant value following the first disease
generation, in the absence of interventions or susceptible depletion.
We also ran simulations under different assumptions regarding the distribution of the
generation interval as p varied in the range 0 < p ≤ 1 (Figure 2). The declining trend in
the effective reproduction number persists independently of the form of the generation
interval distribution for the sub-exponential growth regime ( p < 1 ). Moreover, as departure
from exponential growth increased (i.e., as p decreased towards 0), reproduction number
estimates became less dependent on the generation time distribution (Figure 2).
Importantly, this indicates that for a sufficiently small value of p < 1 , the mean of the
generation interval distribution provides sufficient information to estimate the reproduction
number, without the need for information on the shape of the generation interval
distribution.
Next, we analyzed a variety of empirical outbreak data to test the importance of subexponential growth behavior in real disease dynamics and the resulting impact on
reproduction number estimates. We found some variability in estimates of the
reproduction number across 21 outbreaks representing 8 different pathogens (median=1.4,
IQR: 1.7, 2.6; Figure 3) and the deceleration of growth parameter (median p=0.47, IQR:
0.64, 0.82; Figure 4). Not surprisingly, parameter uncertainty declined with increasing
length of the early epidemic phase used for estimation (Figures 3-4). However, the mean
estimates of the effective reproduction number showed a decreasing trend as more data of
9
the early epidemic phase comprising 3-5 generation intervals were used in estimation
(Figure 3). On the other hand, mean estimates of the deceleration of growth parameter p
(Table 1) were stable when using early growth outbreak data comprising 3-5 generation
intervals (ANOVA, P=0.9). We found that a significant variation in the mean estimates of
the reproduction number could occur as a result of small changes in the deceleration of
growth parameter across outbreaks (Spearman rho>0.62, P<0.002; Supplementary Figure
S2).
A subset of model fits to empirical data is displayed in Figures 5-9, illustrating a variety of
exponential and sub-exponential growth profiles. For instance, the 1918 influenza
pandemic in San Francisco was characterized by near-exponential growth dynamics, with a
high value of parameter p~ 0.8-0.9 and relatively stable R ~1.7- 1.8 (Figure 5). In contrast,
the foot-and-mouth disease outbreak in Uruguay at the farm level displayed a profile of
slower growth (Figure 5) with mean p~0.4-0.5 and more variable mean R~1.6-2.8 (Figure
6). For the HIV/AIDS epidemic in Japan (1985-2012), a disease that is largely spread
through sexual contact via bodily fluids, we estimated mean R~ 1.3-1.6 with p~0.5
assuming a mean generation interval of 4 years, consistent with sub-exponential growth
(Figure 7).
The wealth of Ebola data available at the district level for the 2014 epidemic in West
Africa provides a good opportunity to gauge geographic variations in growth profiles, and
the resulting reproduction numbers. We found significant variability across locations in
estimates of R (median=1.27, IQR: 1.5-1.86) and p (median=0.44, IQR: 0.60-0.72). The
estimates of R showed a significant correlation with the corresponding estimates of the
deceleration of growth parameter p (Spearman rho=0.78, P<0.001). For the Ebola outbreak
in the district of Gueckedou, Guinea where the West Africa Ebola outbreak is assumed to
have originated, we estimated R at 1.2 (95%CI: 1.1, 1.3) and p=0.44 (95%CI: 0.2, 0.64)
(Figure 8). We obtained similar estimates for the Ebola outbreak in Western Area Urban in
Sierra Leone (R~1.2-1.3 and p~0.5; Figure 9).
10
Discussion
We have introduced a quantitative framework to characterize the transmission potential of
epidemics that exhibit an initial sub-exponential (e.g., polynomial) growth phase. This
framework is based on the generalized-growth model [16], and requires information about
the disease generation interval. In this context, using disease-specific outbreak data, we
demonstrated that the effective reproduction number displays a downward trend that
approaches 1.0 over time (typically, as early as 3-5 disease generations), in the absence of
control interventions, depletion of susceptibles, or population behavioral changes. This is
in stark contrast to the standard theory of compartmental models, which predicts that the
reproduction number remains invariant during the initial exponential growth phase. We
illustrated our methodology using mathematical analysis, simulations and applications to
various empirical infectious disease datasets representing directly and sexually transmitted
viral pathogens, including Ebola, as well as historic outbreaks of pandemic influenza,
smallpox, plague, cholera, measles, foot-and-mouth disease, and HIV/AIDS. The
discrepancy in estimates of the reproduction number observed here arises when the model
is allowed to naturally capture the profile of initial growth and its uncertainty using the
deceleration of growth parameter, rather than making the assumption of exponential
growth. Our results indicate that the concept of sub-exponential growth is important to
consider for accurate assessment of the reproduction number, which is a key epidemiologic
parameter to guide interventions.
As shown here, for epidemics that truly depart from exponential growth theory, use of
traditional estimation methods relying on exponential-growth assumptions are expected to
inflate estimates of the reproduction number. The bias between theoretical values and
estimates increases as departure from exponential theory becomes more pronounced, i.e.,
when p decreases towards 0, representing slower epidemic spread compared to the
exponential case where p=1. For instance, our estimate of the reproduction number for the
1972 smallpox epidemic in Khulna, Bangladesh at ~2 (95% CI: 1.6, 2.6) is significantly
lower than earlier historic estimates of smallpox based on exponential growth assumptions
(range 3.5 – 6.0) [33]. Not surprisingly, when the deceleration of growth parameter p is
11
near 1.0, suggesting initial exponential growth, our estimate of the reproduction number
remains consistent with those of compartment models, as is the case for the 1905 bubonic
plague epidemic in Bombay, India [34]. Similarly, our mean R estimate at 1.67 (95% CI:
1.43, 2.08) for the 1918 influenza pandemic in San Francisco with near exponential-growth
dynamics is consistent with that derived using the SEIR model assuming a 3 day
generation interval. Overall, our estimates for Ebola outbreaks tend to be slightly lower
than those reported in prior studies, possibly because our data are consistent with subexponential growth at the district levels (e.g., [28, 35-42]). While our goal was not to carry
out extensive comparisons of reproduction number estimates across studies, similar
conclusions can be drawn for the other epidemics analyzed here (Table 1).
While providing a quantitative framework for estimation of the reproduction number, this
study adds to our understanding of infectious disease dynamics and control. Classical
theory of epidemics predicts that the reproduction number remains unchanged over time
unless other factors that blunt the transmission rate take effect in the population, such as
control interventions or behavior changes [1, 2]. For instance, in the simple susceptibleinfectious-removed (SIR) model, the critical fraction of the population needed to be
effectively vaccinated to prevent an epidemic is given by: 1-1/R0, which is in the range 5090% of the population for most epidemic diseases [1, 41]. However, this fraction may be
potentially considerably lower for epidemics rendering sub-exponential growth, where the
reproduction number naturally declines towards unity irrespective of other intervention
measures and before susceptible depletion sets in. For example, the 2014 West-African
Ebola outbreak ended with less than 1% of the population registered as cases, which defies
expectations from SIR models, and the contribution of large-scale interventions on these
low attack rates remains debated [43]. These data-driven observations suggest that more
attention should be paid to the shape of the early ascending phase of emerging infectious
diseases outbreaks, via the generalized-growth model, and the associated uncertainty in the
reproduction number estimates should be considered. In the context of novel infectious
disease threats for which little or no knowledge of the generation interval exists, the
deceleration of growth parameter alone could provide useful information about the early
12
epidemic growth pattern, which could then be improved in real-time as more data become
available.
Given our observations, it is natural to expect that the probability that an emerging disease
outbreak goes extinct due to stochastic effects is higher for epidemics with sub-exponential
growth, exhibiting a declining trend in the effective reproduction number. This may
partially explain the small magnitude and a short duration of most Ebola outbreaks since
1976 [44-46]. In fact, simulations using an individual-level stochastic model calibrated to
Ebola transmission with an early sub-exponential growth phase yielded a probability of
~0.4 that an outbreak would die out spontaneously during the first month of transmission
[42].
In a public health context, sub-exponential epidemic growth should not be interpreted as an
indicator of the level of difficulty we may face to control an epidemic. However, it
suggests a greater window of opportunity for the implementation of control interventions
compared to an exponential-growth epidemic for a given growth rate, r. In the context of
geographically widespread epidemics such as the West African 2014 Ebola outbreak, an
Ebola transmission model that incorporates a household and community contact network
structure [42] has displayed an effective reproduction number that asymptotically declines
toward unity as the virus spreads through the population. Yet, the size and speed of this
epidemic wave was shown to depend on the contact network properties and the
transmissibility at the household and community levels [42]. Thus, regardless of whether
or not epidemic growth is sub-exponential, public health interventions are critical to
decrease the reproduction number below the threshold of 1 for halting disease spread.
In this article we introduced for the first time a practical quantitative framework to
characterize the reproduction number of epidemics with early sub-exponential growth
dynamics. Our results underscore the need to carefully characterize the shape of the
epidemic growth phase via the generalized-growth model in order to accurately capture
changes in the effective reproduction number. Consideration of the sub-exponential growth
13
phenomenon will improve our ability to appropriately model transmission scenarios, assess
the potential effects of control interventions, and provide accurate forecasts of epidemic
impact. Due to the importance of early quantification of disease characteristics for rapid
and informed public health responses, the development of new mechanistic transmission
models is needed to provide a better understanding of the factors driving sub-exponential
growth dynamics. Such models, as has historically been shown, would allow for a
systematic evaluation of epidemic outcomes and disease control policies. A recent review
of forecasting models for the West African Ebola epidemic highlighted a range of
approaches to investigating disease spread from simple phenomenological models, to
compartmental epidemic models, to intricate contact networks [47]. The vast majority of
these approaches considered models with early exponential growth dynamics, an
assumption that had led to substantial overestimates in projecting the 2014 Ebola epidemic
size and peak time. Chretien et al. [47] underscore the need to design new mechanistic
models that incorporate “dampening approaches” to improve the characterization of the
force of infection and provide uniform forecasting approaches and evaluation metrics. We
believe the present study represents a significant step in this direction.
14
Table 1. Summary of epidemic datasets and the corresponding estimates of the
reproduction number derived using the methodology described in the main text.
Disease
Outbreak
Disease
generatio
n interval
mean
(S.D) &
Time
series
temporal
resolution
Epidemic
peak
timing (no.
data
points)
Early
epidemic
phase (no.
data points)
Reproduction
number (95%
CI)
Data
Source
Pandemic
influenza
Ebola
San Francisco
(1918)
Khulna,
Bangladesh
(1972)
Bombay
(1905-06)
London
(1948)
Aalborg
(1853)
Japan (19852012)
NYC (19822002)
Uruguay
(2001)
Uganda
(2000)
Congo
(1976)
Gueckedou,
Guinea
(2014)
Montserrado,
Liberia
(2014)
Margibi,
Liberia
(2014)
3 (1)
Days
33
14
1.67 (1.43,2.08)
[17]
14 (2)
Weeks
10
7
1.97 (1.61,2.62)
[48]
7 (2)
Weeks
19
7
1.37 (1.19,1.61)
[49]
14 (2)
Weeks
16
8
1.31 (1.26,1.36)
[50]
5 (1)
Days
39
20
1.87 (1.64,2.23)
[51]
4 (1.4)
Years
24
16
1.40 (1.37,1.43)
[52]
4 (1.4)
Years
12
8
2.39 (2.37,2.44)
[53]
5(1)
Days
33
10
1.77 (1.33,2.44)
[54, 55]
13.1 (6.6)
Weeks
8
7
1.65 (1.42,2.08)
[35, 56]
13.1 (6.6)
Days
22
16
3.79 (2.31,6.39)
[57, 58]
19 (11)
Weeks
22
12
1.46 (1.24,1.90)
[59]
13.1 (6.6)
Weeks
16
8
1.46 (1.24,1.93)
[59]
13.1 (6.6)
Weeks
14
8
1.99 (1.52,2.86)
[59]
Ebola
Bomi, Liberia
(2014)
13.1 (6.6)
Weeks
10
8
1.05 (1.00,1.20)
[59]
Ebola
Grand Bassa,
Liberia
(2014)
13.1 (6.6)
Weeks
7
8
1.25 (1.03,1.64)
[59]
Ebola
Western Area
Urban, Sierra
Leone (2014)
11.6 (5.6)
Weeks
18
8
1.26 (1.16,1.41)
[59]
Ebola
Western Area
Rural, Sierra
Leone (2014)
11.6 (5.6)
Weeks
19
8
1.79 (1.50,2.26)
[59]
Smallpox
Plague
Measles
Cholera
HIV/AIDS
AIDS
FMD
Ebola
Ebola
Ebola
Ebola
15
Bo, Sierra
Leone (2014) 11.6 (5.6)
Weeks
15
8
1.50 (1.21,1.99)
Bombali,
Sierra Leone
Ebola
(2014)
11.6 (5.6)
Weeks
12
7
1.94 (1.53,2.14)
Kenema,
Sierra Leone
Ebola
(2014)
11.6 (5.6)
Weeks
12
7
1.24 (1.11,1.41)
Port Loko,
Sierra Leone
Ebola
(2014)
11.6 (5.6)
Weeks
16
7
1.32 (1.18,1.50)
FMD=foot-and-mouth disease
&
Mean and SD for generation interval is given in days except for HIV/AIDS datasets given in years.
Ebola
16
[59]
[59]
[59]
[59]
Figure captions
Figure 1. Simulated profiles of epidemic growth and the corresponding curves of the
effective reproduction number supported by the generalized growth model in the absence
of control interventions and depletion of susceptibles during the first 5 generation intervals
of disease transmission. The deceleration of growth parameter p is varied from 0.2 to 1.0
while the growth rate parameter r is fixed at 0.65 per day and C(0)=1. The generation
interval distribution is assumed to be fixed at Tg = 3 days. When p=1 (exponential growth),
the reproduction number is stationary at R0exp = erT = 7.03 . When p<1 (sub-exponential
growth) the reproduction number gradually declines over time and approaches 1.0 (See
Methods). The black circles correspond to Rgsub exp which are explicitly calculated using
Equation (8) while the solid lines correspond to the numerical solutions computed using
Equation (9).
Figure 2. Simulated profiles of the effective reproduction number during the first 5
generation intervals estimated from case incidence curves derived using the generalizedgrowth model with different values of the deceleration of growth parameter, p. The growth
rate parameter r is fixed at 0.65 per day and C(0)=1. Estimates of the effective
reproduction number are numerically obtained using Equation (9) assuming four different
distributions for the generation interval distribution: a) uniform distribution in the range 24 days, b) exponential distribution with mean of 3 days, c) gamma distribution with mean
of 3 days and variance of 1 day, and d) fixed generation interval at 3 days. When p=1
(exponential growth dynamics) the reproduction number is invariant irrespective of the
shape of the generation interval distribution and is exactly given by 1+ rTg = 2.95 when
g
the generation interval is exponentially distributed and e g = 7.03 when the generation
interval is fixed (see Methods). For p<1 (sub-exponential growth) the reproduction number
approaches 1.0 over disease generations. The reproduction number Rgsub exp at fixed
generation intervals can be explicitly estimated using Equation (8) (black circles).
Figure 3. Estimates of the reproduction number and corresponding 95% confidence
intervals derived from various infectious disease outbreak datasets of case incidence series
by fitting the generalized-growth model to the initial phase of the epidemics comprising
approximately the first 3 (green), 4 (blue), and 5 (red) generation intervals of disease
transmission. The generation interval is assumed to follow a gamma distribution with the
corresponding mean and variance provided in Table 1.
rT
Figure 4. Estimates of the deceleration of growth parameter and corresponding 95%
confidence intervals derived from various infectious disease outbreak datasets of case
incidence series by fitting the generalized-growth model to the initial phase of epidemics
comprising approximately the first 3 (green), 4 (blue), and 5 (red) generation intervals of
disease transmission.
Figure 5. The 1918 influenza pandemic in San Francisco. Estimates and 95%
confidence intervals of the effective reproduction number derived from fitting the
17
generalized growth model to an increasing length of the early phase of epidemic growth
comprising approximately 3-5 generation intervals of the disease. The generation interval
is assumed to be gamma distributed with mean of 3 days and standard deviation of 1 day.
Estimates and 95% confidence intervals for the growth rate parameter r and the
deceleration of growth parameter p are also shown.
Figure 6. The 2001 foot-and-mouth disease epidemic in Uruguay.
Estimates and 95% confidence intervals of the effective reproduction number derived from
fitting the generalized growth model to an increasing length of the early phase of epidemic
growth comprising approximately 3-5 generation intervals of the disease. The generation
interval is assumed to be gamma distributed with mean of 5 days and standard deviation of
1 day. Estimates and 95% confidence intervals for the growth rate parameter r and the
deceleration of growth parameter p are also shown.
Figure 7. The HIV/AIDS epidemic in Japan (1985-2012).
Estimates and 95% confidence intervals of the effective reproduction number derived from
fitting the generalized growth model to an increasing length of the early phase of epidemic
growth comprising approximately 3-5 generation intervals of the disease. The generation
interval is assumed to be gamma distributed with mean of 4 years and standard deviation
of 1.4 years. Estimates and 95% confidence intervals for the growth rate parameter r and
the deceleration of growth parameter p are also shown.
Figure 8. The 2014-15 Ebola epidemic in Gueckedou, Guinea.
Estimates and 95% confidence intervals of the effective reproduction number derived from
fitting the generalized growth model to an increasing length of the early phase of epidemic
growth comprising approximately 3-5 generation intervals of the disease. The generation
interval is assumed to be gamma distributed with mean of 19 days and standard deviation
of 11 days. Estimates and 95% confidence intervals for the growth rate parameter r and the
deceleration of growth parameter p are also shown.
Figure 9. The 2014-15 Ebola epidemic in Western Area Urban, Sierra Leone.
Estimates and 95% confidence intervals of the effective reproduction number derived from
fitting the generalized growth model to an increasing length of the early phase of epidemic
growth comprising approximately 3-5 generation intervals of the disease. The generation
interval is assumed to be gamma distributed with mean of 11.6 days and standard deviation
of 5.6 days. Estimates and 95% confidence intervals for the growth rate parameter r and
the deceleration of growth parameter p are also shown.
18
References
[1]Anderson,R.M.&May,R.M.1991Infectiousdiseasesofhumans.Oxford,Oxford
UniversityPress.
[2]Diekmann,O.&Heesterbeek,J.2000Mathematicalepidemiologyofinfectious
diseases:modelbuilding,analysisandinterpretation,Wiley.
[3]Kermack,W.O.&McKendrick,A.G.1937Contributionstothemathematicaltheory
ofepidemics:IV.Analysisofexperimentalepidemicsofthevirusdiseasemouse
ectromelia.JHyg(Lond)37,172-187.
[4]Ross,R.1911ThePreventionofMalaria.London,JohnMurray.
[5]Mollison,D.1995Epidemicmodels:theirstructureandrelationtodata,Cambridge
UniversityPress.
[6]Bailey,N.T.J.1975Themathematicaltheoryofinfectiousdiseaseandits
applications.NewYork,Hafner.
[7]Ferguson,N.M.,Donnelly,C.A.&Anderson,R.M.2001Thefoot-and-mouth
epidemicinGreatBritain:patternofspreadandimpactofinterventions.Science292,
1155-1160.(doi:10.1126/science.1061020
1061020[pii]).
[8]Keeling,M.J.,Woolhouse,M.E.,Shaw,D.J.,Matthews,L.,Chase-Topping,M.,
Haydon,D.T.,Cornell,S.J.,Kappey,J.,Wilesmith,J.&Grenfell,B.T.2001Dynamicsof
the2001UKfootandmouthepidemic:stochasticdispersalinaheterogeneous
landscape.Science294,813-817.(doi:10.1126/science.1065973
1065973[pii]).
[9]Merler,S.,Ajelli,M.,Fumanelli,L.,Gomes,M.F.,Piontti,A.P.,Rossi,L.,Chao,D.L.,
Longini,I.M.,Jr.,Halloran,M.E.&Vespignani,A.2015Spatiotemporalspreadofthe
2014outbreakofEbolavirusdiseaseinLiberiaandtheeffectivenessofnonpharmaceuticalinterventions:acomputationalmodellinganalysis.TheLancet.
Infectiousdiseases15,204-211.(doi:10.1016/S1473-3099(14)71074-6).
[10]Riley,S.&Ferguson,N.M.2006Smallpoxtransmissionandcontrol:spatial
dynamicsinGreatBritain.ProceedingsoftheNationalAcademyofSciencesofthe
UnitedStatesofAmerica103,12637-12642.(doi:0510873103[pii]
10.1073/pnas.0510873103).
[11]Halloran,M.E.,Longini,I.M.,Jr.,Nizam,A.&Yang,Y.2002Containingbioterrorist
smallpox.Science298,1428-1432.(doi:10.1126/science.1074674
298/5597/1428[pii]).
[12]Colgate,S.A.,Stanley,E.A.,Hyman,J.M.,Layne,S.P.&Qualls,C.1989Risk
behavior-basedmodelofthecubicgrowthofacquiredimmunodeficiencysyndrome
intheUnitedStates.ProceedingsoftheNationalAcademyofSciencesoftheUnited
StatesofAmerica86,4793-4797.
[13]Szendroi,B.&Csányi,G.2004Polynomialepidemicsandclusteringincontact
networks.ProceedingsoftheRoyalSocietyofLondon.SeriesB:BiologicalSciences271,
S364-S366.
[14]May,R.M.&Anderson,R.M.1987TransmissiondynamicsofHIVinfection.
Nature326,137-142.(doi:10.1038/326137a0).
19
[15]Chowell,G.,Viboud,C.,Hyman,J.M.&Simonsen,L.2015TheWesternAfrica
ebolavirusdiseaseepidemicexhibitsbothglobalexponentialandlocalpolynomial
growthrates.PLoScurrents7.
(doi:10.1371/currents.outbreaks.8b55f4bad99ac5c5db3663e916803261).
[16]Viboud,C.,Simonsen,L.&Chowell,G.2016Ageneralized-growthmodelto
characterizetheearlyascendingphaseofinfectiousdiseaseoutbreaksEpidemics15,
27–37.
[17]Chowell,G.,Nishiura,H.&Bettencourt,L.M.2007Comparativeestimationofthe
reproductionnumberforpandemicinfluenzafromdailycasenotificationdata.
JournaloftheRoyalSociety,Interface/theRoyalSociety4,155-166.
(doi:R1512W8743042865[pii]
10.1098/rsif.2006.0161).
[18]Anderson,R.M.&May,R.M.1982Directlytransmittedinfectionsdiseases:
controlbyvaccination.Science215,1053-1060.
[19]vandenDriessche,P.&Watmough,J.2002Reproductionnumbersandsubthresholdendemicequilibriaforcompartmentalmodelsofdiseasetransmission.
Mathematicalbiosciences180,29-48.(doi:S0025556402001086[pii]).
[20]Diekmann,O.,Heesterbeek,J.A.&Roberts,M.G.2010Theconstructionofnextgenerationmatricesforcompartmentalepidemicmodels.JournaloftheRoyalSociety,
Interface/theRoyalSociety7,873-885.(doi:rsif.2009.0386[pii]
10.1098/rsif.2009.0386).
[21]Wallinga,J.&Lipsitch,M.2007Howgenerationintervalsshapetherelationship
betweengrowthratesandreproductivenumbers.Proceedings.Biologicalsciences/
TheRoyalSociety274,599-604.
[22]Roberts,M.G.&Heesterbeek,J.A.2007Model-consistentestimationofthebasic
reproductionnumberfromtheincidenceofanemerginginfection.Journalof
mathematicalbiology55,803-816.(doi:10.1007/s00285-007-0112-8).
[23]Tolle,J.2003CanGrowthBeFasterthanExponential,andJustHowSlowIsthe
Logarithm?TheMathematicalGazette87,522-525.
[24]Nishiura,H.&Chowell,G.2009Theeffectivereproductionnumberasaprelude
tostatisticalestimationoftime-dependentepidemictrends.InMathematicaland
statisticalestimationapproachesinepidemiology(ed.G.e.a.Chowell),pp.103-121.,
Springer(TheNetherlands).
[25]Fraser,C.2007Estimatingindividualandhouseholdreproductionnumbersinan
emergingepidemic.PloSone2,e758.(doi:10.1371/journal.pone.0000758).
[26]Carrat,F.,Vergu,E.,Ferguson,N.M.,Lemaitre,M.,Cauchemez,S.,Leach,S.&
Valleron,A.J.2008Timelinesofinfectionanddiseaseinhumaninfluenza:areviewof
volunteerchallengestudies.Americanjournalofepidemiology167,775-785.
(doi:10.1093/aje/kwm375).
[27]Fine,P.E.2003Theintervalbetweensuccessivecasesofaninfectiousdisease.
Americanjournalofepidemiology158,1039-1047.
[28]Team,W.H.O.E.R.2014EbolaVirusDiseaseinWestAfrica-TheFirst9Monthsof
theEpidemicandForwardProjections.TheNewEnglandjournalofmedicine371,
1481-1495.(doi:10.1056/NEJMoa1411100).
20
[29]Gani,R.&Leach,S.2004Epidemiologicdeterminantsformodelingpneumonic
plagueoutbreaks.Emerginginfectiousdiseases10,608-614.
(doi:10.3201/eid1004.030509).
[30]Burrows,R.1968Excretionoffoot-and-mouthdiseaseviruspriorto
developmentoflesions.VeterinaryRecord82,387.
[31]Nishiura,H.2010Correctingtheactualreproductionnumber:asimplemethod
toestimateR(0)fromearlyepidemicgrowthdata.Internationaljournalof
environmentalresearchandpublichealth7,291-302.(doi:10.3390/ijerph7010291).
[32]Chowell,G.,Ammon,C.E.,Hengartner,N.W.&Hyman,J.M.2006Transmission
dynamicsofthegreatinfluenzapandemicof1918inGeneva,Switzerland:Assessing
theeffectsofhypotheticalinterventions.JTheorBiol241,193-204.(doi:S00225193(05)00509-6[pii]
10.1016/j.jtbi.2005.11.026).
[33]Gani,R.&Leach,S.2001Transmissionpotentialofsmallpoxincontemporary
populations.Nature414,748-751.(doi:10.1038/414748a).
[34]Bacaer,N.2012ThemodelofKermackandMcKendrickfortheplagueepidemic
inBombayandthetypereproductionnumberwithseasonality.Journalof
mathematicalbiology64,403-422.(doi:10.1007/s00285-011-0417-5).
[35]Chowell,G.,Hengartner,N.W.,Castillo-Chavez,C.,Fenimore,P.W.&Hyman,J.M.
2004ThebasicreproductivenumberofEbolaandtheeffectsofpublichealth
measures:thecasesofCongoandUganda.Journaloftheoreticalbiology229,119-126.
(doi:10.1016/j.jtbi.2004.03.006
S0022519304001092[pii]).
[36]Althaus,C.L.2014EstimatingthereproductionnumberofZaireebolavirus
(EBOV)duringthe2014outbreakinWestAfrica.PLOSCurrentsOutbreaks.Edition1.
doi:10.1371/currents.outbreaks.91afb5e0f279e7f29e7056095255b288.
[37]Nishiura,H.&Chowell,G.2014EarlytransmissiondynamicsofEbolavirus
disease(EVD),WestAfrica,MarchtoAugust2014.Eurosurveillance:bulletin
Europeensurlesmaladiestransmissibles=Europeancommunicablediseasebulletin19.
[38]Towers,S.,Patterson-Lomba,O.&Castillo-Chavez,C.2014TemporalVariations
intheEffectiveReproductionNumberofthe2014WestAfricaEbolaOutbreak.PLOS
CurrentsOutbreaks.
[39]Lewnard,J.A.,NdeffoMbah,M.L.,Alfaro-Murillo,J.A.,Altice,F.L.,Bawo,L.,
Nyenswah,T.G.&Galvani,A.P.2014DynamicsandcontrolofEbolavirus
transmissioninMontserrado,Liberia:amathematicalmodellinganalysis.TheLancet.
Infectiousdiseases14,1189-1195.(doi:10.1016/S1473-3099(14)70995-8).
[40]Yamin,D.,Gertler,S.,Ndeffo-Mbah,M.L.,Skrip,L.A.,Fallah,M.,Nyenswah,T.G.,
Altice,F.L.&Galvani,A.P.2014EffectofEbolaProgressiononTransmissionand
ControlinLiberia.Annalsofinternalmedicine.(doi:10.7326/M14-2255).
[41]Pandey,A.,Atkins,K.E.,Medlock,J.,Wenzel,N.,Townsend,J.P.,Childs,J.E.,
Nyenswah,T.G.,Ndeffo-Mbah,M.L.&Galvani,A.P.2014Strategiesforcontaining
EbolainWestAfrica.Science346,991-995.(doi:10.1126/science.1260612).
[42]Kiskowski,M.&Chowell,G.2015Modelinghouseholdandcommunity
transmissionofEbolavirusdisease:epidemicgrowth,spatialdynamicsandinsights
forepidemiccontrol.Virulence,1-11.(doi:10.1080/21505594.2015.1076613).
21
[43]Meltzer,M.I.,Atkins,C.Y.,Santibanez,S.,Knust,B.,Petersen,B.W.,Ervin,E.D.,
Nichol,S.T.,Damon,I.K.&Washington,M.L.2014Estimatingthefuturenumberof
casesintheebolaepidemic---liberiaandsierraleone,2014--2015.Morbidityand
mortalityweeklyreport.Surveillancesummaries63,1-14.
[44]House,T.2014EpidemiologicalDynamicsofEbolaOutbreaks.eLife3,e03908.
(doi:10.7554/eLife.03908).
[45]Althaus,C.L.2015RapiddropinthereproductionnumberduringtheEbola
outbreakintheDemocraticRepublicofCongo.PeerJ3,e1418.
(doi:10.7717/peerj.1418).
[46]Ajelli,M.,Parlamento,S.,Bome,D.,Kebbi,A.,Atzori,A.,Frasson,C.,Putoto,G.,
Carraro,D.&Merler,S.2015The2014EbolavirusdiseaseoutbreakinPujehun,
SierraLeone:epidemiologyandimpactofinterventions.BMCmedicine13,281.
(doi:10.1186/s12916-015-0524-z).
[47]Chretien,J.P.,Riley,S.&George,D.B.2015MathematicalmodelingoftheWest
AfricaEbolaepidemic.eLife4.(doi:10.7554/eLife.09186).
[48]Sommer,A.1974The1972smallpoxoutbreakinKhulnaMunicipality,
Bangladesh.II.Effectivenessofsurveillanceandcontainmentinurbanepidemic
control.Americanjournalofepidemiology99,303-313.
[49]1907XXII.EpidemiologicalobservationsinBombayCity.JHyg(Lond)7,724–
798.
[50]Measlestime-seriesdata.ProfessorBenBolker'sPersonaldatarepositoryat
McMasterUniversity.Availablefrom:
https://ms.mcmaster.ca/~bolker/measdata.html(
[51]DetKongeligeSundhedskollegiumsAarsberetningfor1853.UddragfraAalborg
Physikat.(
[52]HIV/AIDSinJapan,2013.InfectiousAgentsSurveillanceReport(IASR).Month
Vol.35No.9(No.415).(
[53]CDCWonder-AIDSPublicInformationDatasetU.S.Surveillance.Availablefrom:
http://wonder.cdc.gov/aidsPublic.html(
[54]Chowell,G.,Rivas,A.L.,Smith,S.D.&Hyman,J.M.2006Identificationofcase
clustersandcountieswithhighinfectiveconnectivityinthe2001epidemicoffootand-mouthdiseaseinUruguay.AmJVetRes67,102-113.
(doi:10.2460/ajvr.67.1.102).
[55]Chowell,G.,Rivas,A.L.,Hengartner,N.W.,Hyman,J.M.&Castillo-Chavez,C.2006
Theroleofspatialmixinginthespreadoffoot-and-mouthdisease.PrevVetMed73,
297-314.(doi:S0167-5877(05)00275-8[pii]
10.1016/j.prevetmed.2005.10.002).
[56]WHO.2001WorldHealthOrganization(WHO),2001.OutbreakofEbola
hemorrhagicfever,Uganda,August2000–January2001.Epidemiol.Record76,41-48.
[57]BremanJG,PiotP,JohnsonKM,WhiteMK,MbuyiM,SureauP,HeymannDL,Van
NieuwenhoveS,McCormickJB,RuppolJP,etal.1978TheepidemiologyofEbola
hemorrhagicfeverinZaire,1976.Ebolavirushaemorrhagicfever,103-124.
[58]Camacho,A.,Kucharski,A.J.,Funk,S.,Breman,J.,Piot,P.&Edmunds,W.J.2014
PotentialforlargeoutbreaksofEbolavirusdisease.Epidemics9,70-78.
(doi:10.1016/j.epidem.2014.09.003).
22
[59]2015Ebolaresponseroadmap-Situationreport-14October2015.Available
from:http://apps.who.int/ebola/current-situation/ebola-situation-report-14october-2015(accessedon17October2015).(
Figure 1
23
Figure 2
24
Figure 3
25
Figure 4
26
Figure 5
27
Figure 6
28
Figure 7
29
Figure 8
30
Figure 9
31
Supplementaryinformation
FigureS1.
Simulated profiles of the effective reproduction number during the first 5 generation
intervals derived from case incidence curves of the generalized-growth model with
different values of the growth rate (r) and the deceleration of growth parameter (p). The
initial number of cases is set to C(0)=1. Estimates of the effective reproduction number are
generated assuming a fixed generation interval at 3 days.
32
Figure S2
Mean estimates of the reproduction number and the deceleration of growth parameter p
derived from our sample of infectious disease datasets (Table 1) were significantly
correlated for three estimation periods with an initial phase length comprising 3 to 5
generation intervals.
33