1: \documentclass[12pt]{article}
2: \usepackage{amsfonts,amsmath,amsthm}
3: \usepackage{graphicx}
4: \newtheorem{thm}{Theorem}
5: \def\be{\begin{equation}}
6: \def\ee{\end{equation}}
7: \def\ba{\begin{array}}
8: \def\ea{\end{array}}
9:
10: \def\bee{\begin{eqnarray}}
11: \def\eee{\end{eqnarray}}
12:
13: \def\dis{\displaystyle}
14: \def\cR{{\cal R}}
15: \def\cK{{\cal K}}
16:
17: \linespread{1.9}
18: \setlength{\textwidth}{6.7in}
19: \setlength{\textheight}{8.52in}
20: \setlength{\oddsidemargin}{0.7in}
21: \setlength{\evensidemargin}{0.7in}
22:
23:
24: \addtolength {\topmargin}{-.5in}
25: \addtolength {\oddsidemargin}{-.65in}
26: \addtolength {\evensidemargin}{-.65in}
27:
28: \begin{document}
29: \title {The Basic Reproductive Number of Ebola and the Effects
30: of Public Health Measures: The Cases of Congo and Uganda}
31:
32: \author{G.\ Chowell$^{1,4}$, N. \ W. Hengartner$^{2}$,
33: C. Castillo-Chavez$^{1,4}$, \\ P. \ W. Fenimore$^{1}$, J. \ M. Hyman$^{3}$ \\
34: \footnotesize $^{1}$ Center for Nonlinear Studies (MS B258) \\
35: \footnotesize $^{2}$ Statistical Science (MS F600) \\
36: \footnotesize $^{3}$ Mathematical Modeling and Analysis (MS B284) \\
37: \footnotesize Los Alamos National Laboratory \\
38: \footnotesize Los Alamos, NM 87545 \\
39: \footnotesize $^{4}$ Department of Biological Statistics and
40: Computational Biology\\
41: \footnotesize Cornell University \\
42: \footnotesize Warren Hall, Ithaca, NY 14853\\
43: \footnotesize LA-UR-03-8189 \\
44: \footnotesize (Abstract: 174 words, Text: 3,295 words)
45: }
46:
47: \date{}
48:
49: \maketitle
50:
51: \begin{abstract}
52:
53: Despite improved control measures, Ebola remains a serious
54: public health risk in African regions where recurrent outbreaks have been observed
55: since the initial epidemic in $1976$. Using epidemic modeling and data from two
56: well-documented Ebola outbreaks (Congo $1995$ and Uganda $2000$), we
57: estimate the number of secondary cases generated by an index case
58: in the absence of control interventions ($R_0$). Our estimate of
59: $R_0$ is $1.83$ (SD $0.06$) for Congo (1995) and
60: $1.34$ (SD $0.03$) for Uganda (2000). We model the course of the outbreaks via
61: an SEIR (susceptible-exposed-infectious-removed) epidemic model that
62: includes a smooth transition in the transmission rate after control interventions are put in
63: place. We perform an uncertainty analysis of the basic reproductive number $R_0$ to quantify its sensitivity to other disease-related parameters. We also analyze the sensitivity of the final epidemic size to the time interventions begin and provide a distribution for the final epidemic size. The control measures implemented during these two outbreaks (including education and contact tracing followed by quarantine) reduce the final epidemic
64: size by a factor of $2$ relative the final size with a two-week delay in their implementation.
65: \end{abstract}
66:
67: \section{Introduction}
68:
69: \indent Ebola hemorrhagic fever is a highly infectious and lethal disease named after a
70: river in the Democratic Republic of the Congo (formerly Zaire) where it
71: was first identified in 1976 \cite{CDC1}. Twelve outbreaks of Ebola
72: have been reported in Congo, Sudan, Gabon, and Uganda as of September 14, 2003
73: \cite{CDC2,WHO0}. Two different strains of the Ebola virus (Ebola-Zaire and the Ebola-Sudan)
74: have been reported in those regions. Despite extensive search,
75: the reservoir of the Ebola virus has not yet been identified
76: \cite{Breman2, Leirs1}. Ebola is transmitted by physical contact with
77: body fluids, secretions, tissues or semen from infected persons \cite{CDC1, WHO1}. Nosocomial
78: transmission (transmission from patients within hospital settings) has
79: been typical as patients are often treated by
80: unprepared hospital personnel (barrier nursing techniques need to be observed).
81: Individuals exposed to the virus who become infectious do so after a mean incubation
82: period of $6.3$ days ($1-21$ days) \cite{Breman1}. Ebola is
83: characterized by initial flu-like symptoms
84: which rapidly progress to vomiting, diarrhea, rash, and internal and external bleeding.
85: Infected individuals receive limited care as no specific treatment
86: or vaccine exists. Most infected persons die within $10$ days of
87: their initial infection \cite{Nature1} ($50\%-90\%$ mortality \cite{WHO1}).\\
88:
89: \noindent Using a simple SEIR (susceptible-exposed-infectious-removed)
90: epidemic model (Figure \ref{myfig0}) and data from two
91: well-documented Ebola outbreaks (Congo $1995$ and Uganda $2000$), we
92: estimate the number of secondary cases generated by an index case
93: in the absence of control interventions ($R_0$). Our estimates of
94: $R_0$ are $1.83$ (SD $0.06$) for Congo (1995) and
95: $1.34$ (SD $0.03$) for Uganda (2000). We model the course of the outbreaks via
96: an SEIR epidemic model that includes a smooth transition in the transmission rate after
97: control interventions are put in place. We also perform an uncertainty
98: analysis on the basic reproductive number $R_0$ to
99: account for its sensitivity to disease-related parameters and analyze the
100: model sensitivity of the final epidemic size to the time at which
101: interventions begin. We provide a distribution for the final epidemic size. A
102: two-week delay in implementing public health measures results in an
103: approximated doubling of the final epidemic size.
104: \section{Methods}
105:
106: We fit data from Ebola hemorrhagic fever outbreaks in Congo (1995) and Uganda (2000)
107: to a simple deterministic (continuous
108: time) SEIR epidemic model (Figure \ref{myfig0}). The least-squares fit
109: of the model provides estimates for the epidemic parameters. The fitted model can then be used to estimate
110: the basic reproductive number $R_0$ and quantify the impact of intervention measures on the
111: transmission rate of the disease. Interpreting the fitted model as an
112: expected value of a Markov process, we use multiple stochastic
113: realizations of the epidemic to estimate a distribution for the final
114: epidemic size. We also study the sensitivity
115: of the final epidemic size to the timing of interventions
116: and perform an uncertainty analysis on $R_0$ to account for
117: the high variability in disease-related parameters in our model.
118:
119: \subsection{Epidemic Models}
120: Individuals are assumed to be in one of the following epidemiological
121: states (Figure \ref{myfig0}): susceptibles (at risk of contracting the disease),
122: exposed (infected but not yet infectious), infectives (capable of
123: transmitting the disease), and removed (those who recover or die from the
124: disease).
125:
126: \subsubsection{Differential Equation Model}
127: Susceptible individuals in class $S$ in contact with the
128: virus enter the exposed class $E$ at the per-capita rate $\beta
129: I/N$, where $\beta$ is transmission rate per
130: person per day, $N$ is the total effective population size,
131: and $I/N$ is the probability that a contact is made with a infectious
132: individual (i.e. uniform mixing is assumed).
133: Exposed individuals undergo an average incubation period (assumed
134: asymptomatic and uninfectious) of $1/k$ days before progressing to the
135: infectious class $I$. Infectious individuals move to the $R$-class
136: (death or recovered) at the per-capita rate $\gamma$ (see Figure
137: \ref{myfig0}). The above transmission process is modeled by the following system of
138: nonlinear ordinary differential equations \cite{AM, BC}:\\
139: \begin{equation}
140: \label{eqn1}
141: \begin{array}{rcl}
142: {\dis \dot{S}(t)}&=& -\beta S(t) I(t)/N\\
143: {\dis \dot{E}(t)}&=& \beta S(t) I(t)/N -k E(t)\\
144: {\dis \dot{I}(t)}&=& k E(t) - \gamma I(t)\\
145: {\dis \dot{R}(t)}&=& \gamma I(t)\\
146: {\dis \dot{C}(t)}&=& k E(t),\\
147: \end{array}
148: \end{equation}
149:
150: \noindent where $S(t)$, $E(t)$, $I(t)$, and $R(t)$ denote the number of
151: susceptible, exposed, infectious, and removed individuals at
152: time $t$ (the dot denotes time derivatives). $C(t)$ is not an epidemiological
153: state but serves to keep track of
154: the cumulative number of Ebola cases from the time of onset of symptoms.\\
155:
156: \subsubsection{Markov Chain Model}
157: \noindent The analogous stochastic model (continuous time Markov
158: chain) is constructed by considering three events: \textit{exposure}, \textit{infection} and
159: \textit{removal}. The transition rates are defined as:\\
160:
161: \begin {tabular}{l c l}
162: \hline
163: Event & Effect & Transition rate\\
164: \hline
165: Exposure & (S, E, I, R) \ $\rightarrow$ \ (S-1, E+1, I, R) & $\beta(t) S I/N$ \\
166: Infection & (S, E, I, R) \ $\rightarrow$ \ (S, E-1, I+1, R) & $k E$ \\
167: Removal & (S, E, I, R) \ $\rightarrow$ \ (S, E, I-1, R+1) & $\gamma I$ \\
168: \hline
169: \\
170: \\
171: \end{tabular}
172:
173:
174: \noindent The event times $0 < T_1 < T_2 < ...$ at which an
175: individual moves from one state to another are modeled as a renewal
176: process with increments distributed exponentially, \\
177:
178: $$P(T_k - T_{k-1} > t | T_j, j \le k-1) = e^{-t \mu(T_{k-1})}$$
179:
180: \noindent where $\mu(T_{k-1}) = (\beta(T_{k-1}) S(T_{k-1})I(T_{k-1})/N
181: + kE(T_{k-1}) +\gamma I(T_{k-1}))^{-1}$.\\
182:
183: \noindent The final epidemic size is $Z=C(T)$ where
184: $T= min\{t>0, E(t)+I(t)=0 \}$, and its empirical distribution can be
185: computed via Monte Carlo simulations \cite{Renshaw1}.\\
186:
187:
188: \subsection{The Transmission Rate and the Impact of Interventions}
189:
190: \noindent The intervention strategies to control the spread of Ebola
191: include surveillance, placement of suspected cases in quarantine
192: for three weeks (the maximum estimated length of the incubation
193: period), education of hospital personnel and community members on the
194: use of strict barrier nursing techniques (i.e protective clothing and
195: equipment, patient management), and the rapid burial or cremation of patients
196: who die from the disease \cite{WHO1}. Their net effect, in our model, is
197: to reduce the transmission rate $\beta$ from
198: $\beta_0$ to $\beta_1 < \beta_0$. In practice, the impact of the
199: intervention is not instantaneous. Between the time of the onset of
200: the intervention to the time of full compliance, the transmission rate
201: is assumed to decrease gradually from $\beta_0$ to $\beta_1$ according
202: to\\
203: $$
204: \beta(t) = \left \{ \begin{array}{ll}
205:
206: \beta_0 & t<\tau \\
207: \beta_1+ (\beta_0 - \beta_1)e^{-q (t-\tau)} & t \ge \tau
208: \end{array} \right.
209: $$
210:
211: \noindent where $\tau$ is the time at which interventions start and
212: $q$ controls the rate of the transition from $\beta_0$ to
213: $\beta_1$. Another interpretation of
214: the parameter $q$ can be given in terms of $t_h = \frac{ln(2)}{q}$,
215: the time to achieve $\beta(t) = \frac{\beta_0 + \beta_1}{2}$.
216:
217: \subsection{Epidemiological data}
218:
219: The data for the Congo (1995) and Uganda (2000) Ebola hemorrhagic
220: fever outbreaks include the identification dates of
221: the causative agent and data sources. The reported data are ($t_i$,
222: $y_i$), $i=1,...,n$ where $t_i$ denotes the $i^{th}$ reporting time
223: and $y_i$ the cumulative number of infectious cases from the beginning of the
224: outbreak to time $t_i$.\\
225:
226: \noindent \textbf{Congo 1995.} This outbreak began in the Bandundu region, primarily
227: in Kikwit, located on the banks of the Kwilu River. The first case
228: (January 6) involved a 42-year old male charcoal worker and farmer
229: who died on January 13. The Ebola virus was not
230: identified as the causative agent until May $9$. At that time, an international team
231: implemented a control plan that involved active
232: surveillance (identification of cases) and education
233: programs for infected people and their family members. Family members
234: were visited for up to three weeks (maximum incubation period) after
235: their last identified contact
236: with a probable case. Nosocomial transmission occurred in Kikwit General Hospital but it
237: was halted through the institution of strict barrier nursing
238: techniques that included the use of protective equipment and
239: special isolation wards. A total of $315$ cases of Ebola
240: were identified ($81\%$ case fatality). Daily Ebola cases by date of
241: symptom onset from March $1$ through
242: July $12$ are available (Figure \ref{figDailycases}) \cite{Khan1}.\\
243:
244: \noindent \textbf{Uganda 2000.} A total of $425$ cases ($53\%$ case
245: fatality) of Ebola were identified in three
246: districts of Uganda: Gulu, Masindi and Mbara. The onset of symptoms for
247: the first reported case was on August $30$, but the cause was
248: not identified as Ebola until October $15$ by the National Institute of
249: Virology in Johannesburg (South Africa). Active
250: surveillance started during the third week of October. A plan that
251: included the voluntary hospitalization of probable cases
252: was then put in place. Suspected cases were closely followed
253: for up to three weeks. Other control measures included community
254: education (avoiding crowd gatherings
255: during burials) and the systematic implementation of
256: protective measures by health care personnel and
257: the use of special isolation wards in hospitals. Weekly Ebola cases
258: by date of symptom onset are available from the WHO (World Health
259: Organization) \cite{WHO2} (from August $20$, $2000$ through January $7$, $2001$)
260: (Figure \ref{figDailycases}).\\
261:
262: \subsection{Parameter Estimation}
263:
264: Empirical studies in Congo suggest that the
265: incubation period is less than $21$ days with a mean of $6.3$ days
266: \cite{Breman1} and the infectious period is between $3.5$ and $10.7$
267: days. The model parameters $\Theta =(\beta_0$, $\beta_1$, $k$, $q$, $\gamma$) are
268: fitted to the Congo (1995) and Uganda (2000) Ebola outbreak data by
269: \textit{least squares} fit to the cumulative number of cases $C(t,\Theta)$ in
270: eqn. (\ref{eqn1}). We used a computer program
271: (Berkeley Madonna, Berkeley, CA) and appropriate initial
272: conditions for the parameters ($0<\beta<1$, $0<q<100$, $1<1/k<21$
273: \cite{Breman1}, $3.5<1/\gamma<10.7$ \cite{Piot1}). The optimization
274: process was repeated $10$ times (each time the program is fed with two
275: different initial conditions for each parameter) before the ``best fit'' was
276: chosen. The asymptotic variance-covariance $AV(\hat{\theta})$ of
277: the least-squares estimate is\\
278: $$AV(\hat{\theta}) = \sigma^2 (\sum^n_{i=1} \nabla C(t_i, \Theta_0) \nabla C(t_i,\Theta_0)^{T})^{-1}$$ \\
279: \noindent which we estimate by \\
280: $$\hat{\sigma}^2 (\sum_{i=1}^n \hat{\nabla C}(t_i,\hat{\Theta}) \hat{\nabla
281: C}(t_i,\hat{\Theta})^{T})^{-1}$$\\
282: \noindent where $n$ is the total number of observations,
283: $\hat{\sigma}^2 = \frac{1}{n-5} \sum(y_i - C(t_i, \hat{\Theta}))^2$
284: and $\hat{\nabla C}$ are numerical derivatives of $C$.\\
285:
286:
287: \noindent For small samples, the confidence intervals based on these
288: variance estimates may not have the nominal coverage probability. For
289: example, for the case of Zaire $1995$, the $95 \%$ confidence interval for $q$ based on
290: asymptomatic normality is ($-0.26, 2.22$). It should be obvious that this interval
291: is not ``sharp'' as it covers negative values whereas we know $q \ge
292: 0$. The likelihood ratio provides an attractive alternative to build
293: confidence sets (Figure \ref{figq}). Formally, these sets are of the form\\
294: $$\left \{ \Theta: \frac{\sum(y_i - C(t_i,\Theta))^2}{\sum(y_i - C(t_i,\hat{\Theta}))^2} \le A_{\alpha} \right \}$$ \\
295: \noindent where $A_{\alpha}$ is the $1-\alpha$ quantile of an $F$ distribution with
296: appropriate degrees of freedom. Parameter estimates are given in
297: Table \ref{TableParameters}.
298:
299: \subsection{The Reproductive Number}
300:
301: \noindent The basic reproductive number $R_0$ measures the average number
302: of secondary cases generated by a primary case in a pool of mostly
303: susceptible individuals \cite{AM,BC} and is an estimate of the
304: epidemic growth at the start of an outbreak if everyone is susceptible. That is, a primary case
305: generates $R_0 = \frac{\beta_0}{\gamma}$ new cases on the average where $\beta_0$ is the
306: pre-interventions transmission rate and $1/\gamma$ is the mean
307: infectious period. The effective reproductive number
308: at time $t$, $R_{eff}(t) = \frac{\beta(t)}{\gamma} x(t)$, measures the average number of
309: secondary cases per infectious
310: case $t$ time units after the introduction of the initial
311: infections and $x(t) = \frac{S(t)}{N} \approx 1$ as the population size is much larger than the resulting size of the outbreak (Table \ref{TableOutbreaks}). Hence, $R_{eff}(0)=R_0$. In a closed population,
312: the effective reproductive number $R_{eff}(t)$ is
313: non-increasing as the size of the susceptible population
314: decreases. The case $R_{eff}(t) \le 1$ is of special interest as it highlights
315: the crossing of the threshold to eventual control of the outbreak.
316: An intervention is judged successful if it reduces the effective
317: reproductive number to a value less than one. In our model, the
318: post-intevention reproductive number $R_p = \frac{\beta_1}{\gamma}$ where $\beta_1$ denotes
319: the post-intervention transmission rate. In general, the smaller $\beta_1$, the faster
320: an outbreak is extinguished. By the delta method \cite{Bickel1}, the variance of the estimated
321: basic reproductive number $\hat{R_0}$ is approximately\\
322:
323: $$ V(\hat{R_0}) \approx \hat{R_0}^2 \ \{ \frac{V(\hat{\beta_0})}{\hat{\beta_0}^2} +
324: \frac{V(\hat{\gamma})}{\hat{\gamma}^2} - \frac{2
325: Cov(\hat{\beta_0}, \hat{\gamma})}{\hat{\beta_0}
326: \hat{\gamma}} \}. $$
327:
328: \subsection {The Effective Population Size}
329:
330: \noindent A rough estimate of the population size in the Bandundu
331: region of Congo (where the epidemic developed) in 1995 is computed
332: from the population size of the Bandundu region in $1984$
333: \cite{WGazzetter1} and annual population growth
334: rates \cite{unhabitat1} (Table \ref{TableOutbreaks}). For the case
335: of Uganda (2000), we adjusted the population sizes of the
336: districts of Gulu, Masindi and Mbara in $1991$ and
337: annual population growth rates \cite{UBOS1} (Table
338: \ref{TableOutbreaks}). These estimates are an upper bound of the
339: effective population size (those at risk of becoming infected) for
340: each region. Estimates of the effective population
341: size are essential when the incidence is modeled with the pseudo mass-action
342: assumption ($\beta(t) S I$) which implies that transmission grows linearly with the
343: population size and hence the basic reproductive number $R_0 (N) = \beta_0 N
344: /\gamma$. In our model, we use the true mass-action assumption
345: ($\beta(t) S I/N$) which makes the model parameters (homogeneous
346: system of order $1$) independent of $N$ and hence the basic reproductive number can be
347: estimated by $R_0 = \beta_0 / \gamma$ \cite{CVF}. In fact, comparisons between
348: the pseudo mass-action and the true mass-action assumptions with
349: experimental data have concluded in favor of the later
350: \cite{JDH}. The model assumption that $N$ is constant is not critical as the
351: outbreaks resulted in a small number of cases compared to the size of the
352: population.
353:
354: \subsection{Uncertainty Analysis on $R_0$}
355: Log-normal distributions seem to model well the incubation period distributions for
356: a large number of diseases \cite{Sartwell1}. Here, a log-normal
357: distribution is assumed for the incubation period of Ebola in our
358: uncertainty analysis. Log-normal distribution parameters are set from empirical
359: observations (mean incubation period is $6.3$ and the $95\%$ quantile
360: is $21$ days \cite{Breman1}). The infectious period is assumed to be uniformly
361: distributed in the range ($3.5-10.7$) days \cite{Piot1}. \\
362:
363: \noindent A formula for the basic reproductive number $R_0$ that depends on the initial per-capita rate of
364: growth $r$ in the number of cases (Figure \ref{figR0uncertainty}), the incubation period
365: ($1/k$) and the infectious period ($1/\gamma$) can be obtained by linearizing
366: equations $\dot{E}$ and $\dot{I}$ of system (\ref{eqn1}) around the disease-free equilibrium
367: with $S=N$. The corresponding Jacobian matrix is given by:\\
368: \[
369: J=\left(\begin{array}{cc}
370: -k & \beta \\
371: k & -\gamma \\
372: \end{array}\right),
373: \]
374:
375: \noindent and the characteristic equation is given by:
376: \[
377: r^2 + (k+\gamma) r + (\gamma - \beta) k = 0
378: \]
379:
380: \noindent where the early-time and per-capita free growth $r$ is essentially the dominant eigenvalue. By solving for $\beta$ in terms of $r$, $k$ and $\gamma$, one can
381: obtain the following expression for $R_0$ using the fact that $R_0 = \beta/\gamma$:\\
382: $$ {R_0} = 1 + \frac{r^2 + (k+\gamma) r}{k \gamma}.$$ \\
383: \noindent Our estimate of the initial rate of growth $r$ for the Congo 1995 epidemic
384: is $r=0.07$ day$^{-1}$, obtained from the time series $y(t)$, $t<\tau$ of the
385: cumulative number of cases and assuming exponential growth ($y(t)
386: \propto e^{rt}$). The distribution of $R_0$ (Figure \ref{figR0uncertainty})
387: lies in the interquartile range (IQR) ($1.66-2.28$) with a median of $1.89$, generated from
388: Monte Carlo sampling of size $10^5$ from the distributed epidemic
389: parameters ($1/k$ and $1/\gamma$) for fixed $r$ \cite{Blower1}. We give the median of $R_0$ (not the mean) as the resulting distribution of $R_0$ from our uncertainty analysis is skewed to the right.
390:
391: \section{Results}
392:
393: \indent Using our parameter estimates (Table \ref{TableParameters}), we
394: estimate an $R_0$ of $1.83$ (SD $0.06$) for Congo (1995) and
395: $1.34$ (SD $0.03$) for Uganda (2000).
396: \noindent The effectiveness of interventions is often quantified in terms of the
397: reproductive number $R_p$ after interventions are put in place. For
398: the case of Congo $R_p = 0.51$ (SD $0.04$) and $R_p = 0.66$ (SD
399: $0.02$) for Uganda allowing us to conclude that in both cases, the intervention
400: was successful in controlling the epidemic. Furthermore, the time to
401: achieve a transmission rate of $\frac{\beta_0 + \beta_1}{2}$ ($t_h$)
402: is $0.71$ ($95\%$ CI ($0.02, 1.39$)) days and $0.11$ ($95 \%$ CI ($0, 0.87$)) days
403: for the cases of Congo and Uganda respectively after the time at which interventions begin.\\
404: \noindent We use the estimated parameters to simulate the Ebola outbreaks in Congo (1995)
405: and Uganda (2000) via Monte Carlo simulations of the stochastic model of Section $2.1$ \cite{Renshaw1}.
406: There is very good agreement between the mean of the stochastic
407: simulations and the reported cases despite the
408: ``wiggle'' captured in the residuals around the time $\tau$ of the
409: start of interventions (Figure \ref{figmodel2}). The
410: empirical distribution of the final epidemic sizes for the cases of Congo
411: $1995$ and Uganda $2000$ are given in Figure \ref{figOutbreaksizedistr}.\\
412: \noindent The final epidemic size is sensitive to
413: the start time of interventions $\tau$. Numerical solutions
414: (deterministic model) show that the final epidemic
415: size grows exponentially fast with the initial time of
416: interventions (not surprising as the intial epidemic growth is driven by
417: exponential dynamics). For instance, for the case of
418: Congo, our model predicts that there would have been $20$ more cases
419: if interventions had started one day later (Figure \ref{figSensInterv}).
420:
421: \section{Discussion}
422:
423: \indent Using epidemic-curve data from two major Ebola hemorrhagic fever
424: outbreaks \cite{Khan1, WHO2}, we have estimated the basic reproductive
425: number ($R_0$) (Table \ref{TableOutbreaks}). Our estimate of $R_0$ (median is $1.89$)
426: obtained from an uncertainty analysis \cite{Blower1} by simple random sampling (Figure
427: \ref{figR0uncertainty}) of the parameters $k$ and $\gamma$ distributed
428: according to empirical data from the Zaire (now the Democratic Republic of Congo) $1976$ Ebola outbreak \cite{Breman1, Piot1} is in agreement with our estimate of $R_0 = 1.83$ from the outbreak in Congo $1995$ (obtained from least squares fitting
429: of our model (\ref{eqn1}) to epidemic curve data). \\
430: The difference in the basic reproductive
431: numbers $R_0$ between Congo and Uganda is due to our different estimates for the
432: infectious period ($1/\gamma$) observed in these two places. Their
433: transmission rates $\beta_0$ are quite similar (Table \ref{TableParameters}). Our
434: estimate for the infectious period for the case of Congo ($5.61$ days)
435: is slightly larger than that of Uganda ($3.50$ days). Clearly, a larger infectious
436: period increases the likelihood of infecting a susceptible
437: individual and hence increases the basic reproductive number.
438: The difference in the infectious periods might be due to differences
439: in virus subtypes \cite{Niikura1}. The Congo outbreak was caused by the Ebola-Zaire
440: virus subtype \cite{Khan1} while the Uganda outbreak was caused by
441: the Ebola-Sudan virus subtype \cite{WHO2}.\\
442: \noindent The significant reduction from the basic reproductive number ($R_0$) to the post-intervention reproductive number ($R_p$) in our estimates for Congo and Uganda shows that the implementation of control measures such as education, contact tracing and quarantine will have a significant effect on lowering the effective reproductive rate of Ebola.
443: Furthermore, estimates for the time to achieve
444: $\frac{\beta_0 + \beta_1}{2}$ have been provided (Table \ref{TableParameters}).\\
445: \noindent We have explored the sensitivity of the final epidemic size to the
446: starting time of interventions. The exponential increase of the final
447: epidemic size with the time of start of interventions (Figure
448: \ref{figSensInterv}) supports the idea that the rapid
449: implementation of control measures should be considered as a critical
450: component in any contingency plan against disease outbreaks specially
451: for those like Ebola and SARS for which no specific treatment or
452: vaccine exists. A two-week delay in implementing public health
453: measures results in an approximated doubling of the final outbreak
454: size. Because the existing control measures cut the transmission rate to
455: less than half, we should seek and support further improvement in the effectiveness of
456: interventions for Ebola. A mathematical model that considers basic public health interventions for SARS control in Toronto supports this conclusion \cite{Chowell1, Chowell2}. Moreover, computer simulations show that small perturbations to the rate $q$ at which interventions are put fully in place do not have a significant effect on the final epidemic size. The rapid identification of an outbreak, of course, remains the strongest determinant of the final outbreak size.\\
457: \noindent Field studies of Ebola virus are difficult to conduct due to
458: the high risk imposed on the scientific and medical personnel
459: \cite{Nature2}. Recently, a new vaccine that makes use of an
460: \textit{adenovirus technology} has been shown to give cynomolgus macaques
461: protection within $4$ weeks of a single jab \cite{Nature3,
462: Nature4}. If the vaccine turns out to be effective in humans, then
463: its value should be tested. A key question would be ``What are the
464: conditions for a successful target vaccination campaign during an Ebola outbreak?''
465: To address questions of this type elaborate models need to be developed.
466:
467: \begin{thebibliography} {99}
468:
469: \bibitem{CDC1} Centers for Disease Control (CDC). Ebola Hemorrhagic Fever.
470: \texttt{(http://www.cdc.gov/ncidod/dvrd/spb/mnpages/dispages/ebola.htm)},
471: accessed on August 24, 2003.
472:
473: \bibitem{CDC2} Centers for Disease Control (CDC). Ebola Hemorrhagic Fever: Table
474: Showing Known Cases and Outbreaks, in Chronological Order.
475: \texttt{(http://www.cdc.gov/ncidod/dvrd/spb/mnpages/dispages/ebotabl.htm)},
476: accessed on August 24, 2003.
477:
478: \bibitem{WHO0} World Health Organization (WHO). Ebola Hemorrhagic
479: Fever: Disease Outbreaks.
480: \texttt{(http://www.who.int/disease-outbreak-news/disease/A98.4.htm)},
481: accessed on October 17, 2003.
482:
483: \bibitem{Breman2} Breman, JG, Johnson, KM, van der Groen, G, et al.
484: A search for for Ebola virus in
485: animals in the Democratic Republic of the Congo and Cameroon:
486: ecologic, virologic, and serologic surveys
487: 1979-1980. J Inf Dis 1999;179:S139-47.
488:
489: \bibitem{Leirs1} Leirs, H, Mills, JN, Krebs, JW, et al.
490: Search for the Ebola Virus Reservoir in
491: Kikwit, Democratic Republic of the Congo: Reflections on a
492: Vertebrate Collection. J Inf Dis 1999;179:S155-63.
493:
494: \bibitem{WHO1} World Health Organization (WHO). Ebola Hemorrhagic Fever.
495: \texttt{(http://www.who.int/inf-fs/en/fact103.html)},
496: accessed on August 24, 2003.
497:
498: \bibitem{Breman1} Breman, JG, Piot, P, Johnson, KM, et al.
499: The Epidemiology of Ebola Hemorrhagic Fever in Zaire, 1976.
500: Proc Int Colloquium on Ebola Virus Inf held in Antwerp, Belgium,
501: $6-8$ December 1977.
502:
503: \bibitem{Nature1} Birmingham, K and Cooney, S. Ebola:
504: small, but real progress (news feature). Nature Med 2002;8:313.
505:
506: \bibitem{AM} Anderson, RM and May, RM. Infectious Diseases
507: of Humans. Oxford University Press, Oxford, 1991.
508:
509: \bibitem{BC} Brauer, F, Castillo-Chavez, C. Mathematical Models
510: in Population Biology and Epidemiology, Springer-Verlag, New York, 2000.
511:
512: \bibitem{Renshaw1} Renshaw, E. Modelling Biological Populations
513: in Space and Time. Cambridge University Press, Cambridge, 1991.
514:
515: \bibitem{Khan1} Khan, AS, Tshioko, FK, Heymann, DL, et al. The
516: Reemergence of Ebola Hemorrhagic Fever, Democratic Republic of the Congo, 1995.
517: J Inf Dis 1999;179:S76-86.
518:
519: \bibitem{WHO2} World Health Organization (WHO).
520: Outbreak of Ebola hemorrhagic fever, Uganda, August $2000$ - January $2001$.
521: Weekly epidemiological record 2001;76:41-48.
522:
523: \bibitem{Piot1} Piot, P, Bureau, P, Breman, G. D. Heymann, et al.
524: Clinical Aspects of Ebola Virus Infection in Yambuku Area, Zaire,
525: 1976. Proc. Int. Colloquium on Ebola Virus Inf. held in Antwerp, Belgium,
526: $6-8$ December 1977.
527:
528: \bibitem{Bickel1} Bickel, P, Doksum, KA. Mathematical
529: Statistics. Holden-Day, Oakland, California, 1977.
530:
531: \bibitem{WGazzetter1} The World Gazetteer. Democratic Republic of the
532: Congo. \texttt{(http://www.world-gazetteer.com/fr/fr\_cd.htm)},
533: accessed on August $24$, 2003.
534:
535: \bibitem{unhabitat1} UN-HABITAT (United Nations Human Settlelments
536: Programme). Republic of Congo. website:
537: \texttt{(http://www.unhabitat.org/habrdd/conditions/midafrica/zaire.html)},
538: accessed on August 24, 2003.
539:
540: \bibitem{UBOS1} Uganda Bureau of Statistics (UBOS). Website:
541: \texttt{(http://www.ubos.org/)}, accessed on August 24, 2003.
542:
543: \bibitem{CVF} Castillo-Chavez, C, Velasco-Hernandez, JX and
544: Fridman, S. Modeling Contact Structures in Biology. In: S.A. Levin
545: ed., Frontiers of Theoretical Biology, Lecture Notes in
546: Biomathematics 100, Springer-Veralg,Berlin-Heidelberg-New York
547: 1994:454-91.
548:
549: \bibitem{JDH} De Jong, MCM, Diekmann, O and Heesterbeek, H.
550: How does transmission of infection depend on population size?. In:
551: Mollison, D, ed. Epidemic models: Their structure and relation to
552: data, Cambridge University Press, 1995:84-94.
553:
554: \bibitem{Sartwell1} Sartwell, PE. The distribution of incubation
555: periods and the dynamics of infectious disease. Am J Epidem 1966;83:204-216.
556:
557: \bibitem{Blower1} Blower, SM, Dowlatabadi, H. Sensitivity and
558: uncertainty analysis of complex models of disease transmission: an
559: HIV model, as an example. Int Stat Rev 1994;2:229-243.
560:
561: \bibitem{Niikura1} Niikura, M, Ikegami, T, Saijo, M, et al. Analysis
562: of Linear B-Cell Epitopes of the
563: Nucleoprotein of Ebola Virus That distinguish Ebola Virus Subtypes,
564: Clin Diagn Lab Immunol 2003;10:83-87.
565:
566: \bibitem{Chowell1} Chowell, G, Fenimore, PW, Castillo-Garsow, MA, et
567: al. SARS Outbreaks in Ontario, Hong Kong and
568: Singapore: the role of diagnosis and isolation as a control
569: mechanism. J \ Theor \ Biol 2003;24:1-8.
570:
571: \bibitem{Chowell2} Chowell, G, Castillo-Chavez, C, Fenimore, PW, et
572: al. Implications of an Uncertainty and Sensitivity analysis for SARS's Basic Reproductive
573: Number for General Public Health Measures (submitted).
574:
575: \bibitem{Nature2} Nabel, GJ. Surviving Ebola virus infection
576: (news feature). Nature 1999;4:373-374.
577:
578: \bibitem{Nature3} Sullivan, NJ, Geisbert, TW, Geibsert, JB, et
579: al. Accelerated vaccination for Ebola virus
580: haemorrhagic fever in non-human primates. Nature 2003;424:681-684.
581:
582: \bibitem{Nature4} Clarke T and Knight, J. Fast vaccine offers
583: hope in battle with Ebola. Nature 2003;424:602.
584:
585: \end{thebibliography}
586:
587: \newpage
588:
589: \section*{Tables \& Figures}
590:
591: % sets the footnote to symbols instead of numbers
592: \renewcommand{\thefootnote}{\fnsymbol{footnote}}
593:
594: \begin{table}[h*]
595: \begin {tabular}{||l|l|c|c|c|c||}
596: \hline
597: & & \multicolumn{2}{c|}{Congo 1995}
598: & \multicolumn{2}{c||}{Uganda 2000} \\
599: \hline
600: Parameter & Definition & Estim. & S.\ D. & Estim. & S. \ D. \\
601: \hline
602: $\beta_0$ & Pre-interventions transmission rate (days$^{-1}$) & $0.33$ & $0.06$ &
603: $0.38$ & $0.24$\\
604: $\beta_1$ & Post-interventions transmission rate (days$^{-1}$) & $0.09$ & $0.01$ &
605: $0.19$ & $0.13$\\
606: $t_h$ & Time to achieve $\frac{\beta_0 + \beta_1}{2}$ (days) & $0.71$ & $(0.02, 1.39)$\footnotemark[2] &
607: $0.11$ & $(0, 0.87)$\footnotemark[2]\\
608: $1/k$ & Mean incubation period (days) & $5.30$ & $0.23$ & $3.35$ & $0.49$ \\
609: $1/\gamma$ & Mean infectious period (days) & $5.61$ & $0.19$ & $3.50$ & $0.67$\\
610: \hline
611: \end{tabular}
612: \caption{Parameter definitions and baseline estimates obtained
613: from the best fit of the model equations (\ref{eqn1}) to the
614: epidemic-curve data of the Congo $1995$ and Uganda $2000$
615: outbreaks (Figure \ref{figmodel2}). The parameters were optimized
616: by a computer program (Berkeley Madonna, Berkeley, CA) using a \textit{least
617: squares} fitting technique and appropriate initial
618: conditions for the parameters ($0<\beta<1$, $0<q<100$, $1<1/k<21$
619: \cite{Breman1}, $3.5<1/\gamma<10.7$ \cite{Piot1}). The optimization
620: process was repeated $10$ times (each time the program is fed with two
621: different initial conditions for each parameter) before the ``best fit'' was
622: chosen.}
623: \label{TableParameters}
624: \end{table}
625: \vfill
626:
627: \footnotetext[2]{95 $\%$ CI (Figure \ref{figq}).}
628:
629: \begin{table}[h*]
630: \begin {tabular}{||c|c|l|c|c|c|c|c||}
631: \hline
632: Outbreak & Eff. Pop. Size (N) & Start of interv. & Fatality rate
633: ($\%$) & Estim. $R_0$ & S.D. $R_0$ \\
634: \hline
635: Congo 1995 & $5,364,500$\footnotemark[1] & May $9$, $1995$ \cite{Khan1} & $81\%$
636: \cite{Khan1} & $1.83$ & $0.06$ \\
637: Uganda 2000 & $1,867,200$ \footnotemark[5] & Oct $22$, $2000$ \cite{WHO2} &
638: $53\%$ \cite{WHO2} & $1.34$ & $0.03$ \\
639: \hline
640: \end{tabular}
641: \caption{Population parameters and estimated $R_0$ for the Congo $1995$
642: and the Uganda $2000$ Ebola outbreaks. Notice that even though our expression for $R_0$ is independent of $N$, our model is not independent of $N$ and hence the corresponding population sizes for Congo and Uganda are used in the least-squares estimation of the parameters.}
643: \label{TableOutbreaks}
644: \end{table}
645:
646: \footnotetext[1]{Adjusted from population size of the Bandundu region
647: in $1984$ \cite{WGazzetter1} using the annual population growth
648: rates \cite{unhabitat1}.}
649:
650: \footnotetext[5]{Adjusted from the population sizes of the districts of Gulu,
651: Masindi and Mbara (where the outbreak developed) in $1991$ using the
652: annual population growth rates \cite{UBOS1}.}
653:
654: \vfill
655:
656: %%%%%%%%%% SEIJR model %%%%%%%%%%%%%%
657: \begin{figure}[h*]
658: \begin{center}
659: \scalebox{0.5}{\includegraphics{img_SEIRmodel.eps}}
660: \end{center}
661: \caption{A schematic representation of the flow of individuals
662: between epidemiological classes. $\beta \frac{I}{N}$ is the
663: transmission rate to susceptibles $S$ from $I$; $E$ is the class
664: of infected (not yet infectious) individuals; $k$ is the rate at
665: which $E$-individuals move to the symptomatic and infectious class
666: $I$; Infectious individuals ($I$) either die or recover at rate
667: $\gamma$. $C$ is not an epidemiological state but keeps track of
668: the cumulative number of cases after the time of onset of symptoms.}
669: \label{myfig0}
670: \end {figure}
671: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
672: \vfill
673:
674: %%%%%%%%%% Daily number of cases %%%%%%%%%%%%%%
675: \begin{figure}[h*]
676: \begin{center}
677: \scalebox{0.5}{\includegraphics{img_dailycases.eps}}
678: \end{center}
679: \caption{On the left, we have the \textit{daily} number of cases by date of symptom onset
680: during the Ebola outbreak in Congo 1995 (Mar $6$-Jul $12$). On the right,
681: we have the \textit{weekly} number of cases by date of symptom
682: onset during the Ebola outbreak in Uganda $2000$ (Aug $20$-Jan
683: $07$). Data has been taken from refs. \cite{Khan1, WHO2}.}
684: \label{figDailycases}
685: \end {figure}
686: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
687:
688: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
689: \begin{figure}[h*]
690: \begin{center}
691: \scalebox{0.45}{\includegraphics{img_q.eps}}
692: \end{center}
693: \caption{$95 \%$ confidence intervals for $t_h$ ($t_h =
694: \frac{log(2)}{q}$), the time to achieve a transmission rate of
695: $\frac{\beta_0 + \beta_1}{2}$, obtained
696: from the likelihood ratio as described in the text.}
697: \label{figq}
698: \end {figure}
699: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
700:
701:
702: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
703: \begin{figure}[h*]
704: \begin{center}
705: \scalebox{0.45}{\includegraphics{img_R0uncertainty.eps}}
706: \end{center}
707: \caption{(Top) cumulative number of cases (log-lin scale) during the
708: exponential growth phase of the Congo $1995$ epidemic as
709: identified by the date of start of interventions ($09$ May
710: $1995$ \cite{Khan1}). The model-free initial growth
711: rate of the number of cases for Congo 1995 is $0.07$ (linear
712: regression); (bottom)
713: estimated distribution of $R_0$ from our uncertainty
714: analysis (see text). $R_0$ lies in the
715: interquartile range (IQR) ($1.66-2.28$) with a median of
716: $1.89$. Notice that $100\%$ of the weight lies above $R_0=1$.}
717: \label{figR0uncertainty}
718: \end {figure}
719: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
720:
721:
722: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
723: \begin{figure}[h*]
724: \begin{center}
725: \scalebox{0.45}{\includegraphics{img_model2.eps}}
726: \end{center}
727: \caption{(Top) Comparison of the cumulative number of Ebola cases
728: during the Congo 1995 and Uganda 2000 Ebola outbreaks, as a function of the
729: time of onset of symptoms. Circles are the data. The solid line
730: is the average of $250$ Monte Carlo
731: replicates and the error bars represent the standard
732: error around the mean from the simulation
733: replicates using our parameter estimates (Table
734: \ref{TableParameters}). For the case of Congo $1995$, simulations were begun on $13$
735: Mar $1995$. A reduction in the transmission rate $\beta$ due to
736: the implementation of interventions occurs on $09$ May 1995 (day $56$)
737: \cite{Khan1}. For the case of Uganda $2000$, simulations start on
738: $27$ August $2000$ and interventions take place on $22$
739: October $2000$ (day $56$) \cite{WHO2}; (bottom) comparison of the residuals
740: (difference between the data and the model best fit) scaled by
741: the standard deviation for the cases of Congo and Uganda.}
742: \label{figmodel2}
743: \end {figure}
744: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
745:
746: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
747: \begin{figure}[h*]
748: \begin{center}
749: \scalebox{0.45}{\includegraphics{img_finalsize.eps}}
750: \end{center}
751: \caption{The final epidemic size distributions for
752: the cases of Congo $1995$ and Uganda $2000$ obtained from $250$
753: Monte Carlo replicas. Crosses (X) represent the final
754: epidemic size from data.}
755: \label{figOutbreaksizedistr}
756: \end {figure}
757: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
758:
759: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
760: \begin{figure}[h*]
761: \begin{center}
762: \scalebox{0.45}{\includegraphics{img_sensInterv.eps}}
763: \end{center}
764: \caption{Sensitivity of the final epidemic size to the time of
765: start of interventions. Here negative numbers represent
766: number of days before the actual reported intervention date (Table
767: \ref{TableOutbreaks}) and positive numbers represent a delay
768: after the actual reported intervention date ($\tau=0$). All other
769: parameters have been fixed to their baseline values (Table
770: \ref{TableOutbreaks}). The final epidemic size grows
771: exponentially as expected with the time of interventions
772: with a rate of $0.06$ for the case of Congo and $0.05$ for the
773: case of Uganda.}
774: \label{figSensInterv}
775: \end {figure}
776: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
777:
778: \end{document}
779: