Imported and autochthonous cases in the dynamics of dengue epidemics in Brazil

OBJECTIVE: To estimate the basic reproduction number (R0) of dengue fever including both imported and autochthonous cases. METHODS: The study was conducted based on epidemiological data of the 2003 dengue epidemic in Brasília, Brazil. The basic reproduction number is estimated from the epidemic curve, fi tting linearly the increase of initial cases. Aiming at simulating an epidemic with both autochthonous and imported cases, a “susceptible-infectious-resistant” compartmental model was designed, in which the imported cases were considered as an external forcing. The ratio between R0 of imported versus autochthonous cases was used as an estimator of real R0. RESULTS: The comparison of both reproduction numbers (only autochthonous versus all cases) showed that considering all cases as autochthonous yielded a R0 above one, although the real R0 was below one. The same results were seen when the method was applied on simulated epidemics with fi xed R0. This method was also compared to some previous proposed methods by other authors and showed that the latter underestimated R0 values. CONCLUSIONS: It was shown that the inclusion of both imported and autochthonous cases is crucial for the modeling of the epidemic dynamics, and thus provides critical information for decision makers in charge of prevention and control of this disease. DESCRIPTORS: Dengue, epidemiology. Flavivirus Infections, transmission. Risk. Communicable Period. Disease Outbreaks. Epidemiologic Surveillance. Artigos Originais | Original Articles 2 Dynamics of dengue cases in Brazil Degallier N et al Dengue fever has been endemic in Brazil since 1986.5 Its transmission involves only urban mosquitoes3 and humans, and many fatal cases occur every year due to epidemics where different serotypes are transmitted.8 The quick expansion of vectors and viruses throughout nearly all Brazilian states is facilitated by ever increasing moving and traveling of people in the country. Until recent years in most localities the origin of dengue cases was neither reported nor investigated. To the authors’ best knowledge, no study has been published investigating the impact of “imported” cases on the dynamics of the epidemics. The origin of each case in the 20022003 epidemics in Brasília was inferred after a detailed survey of the patient’s background, taking into account the duration of the incubation phase of the disease. As these data about the most probable origin of these cases were available, we investigated its possible impact on the dynamics of dengue epidemics. We compared its dynamics with the outputs of a simple “susceptibleinfectious-resistant” (SIR) model modifi ed in order to include the input of daily numbers of imported cases when estimating the basic reproduction number (R0). Mathematical modeling is a powerful tool to simulate real past situations, to input data from the present and hopefully to forecast future outcomes.6 It has been widely used to understand the dynamics of infectious RESUMO OBJETIVO: Estimar o número básico de reprodução da dengue (R0), com base nos casos importados, além dos casos autóctones. MÉTODOS: O estudo foi feito sobre dados epidemiológicos da epidemia de dengue em Brasília, 2003. O número básico de reprodução é determinado a partir da curva epidêmica, ajustando uma reta ao crescimento inicial do número de casos. Para simular uma epidemia com casos autóctones e importados, foi criado um modelo compartimentado do tipo “suscetíveis-infectadosresistentes”. O R0 real foi estimado pela fração entre R0 dos casos autóctones e dos importados. RESULTADOS: A comparação de ambos valores de reprodução (apenas autóctones versus todos os casos) mostrou que considerando todos casos como autóctones, o valor de R0 foi superior a um, enquanto o R0 real era inferior a um. O mesmo resultado foi obtido com o conjunto de dados simulando uma epidemia com R0 fi xo. O método foi também comparado a outros, observandose que estes últimos subestimaram os valores do R0. CONCLUSÕES: A inclusão de tanto casos autóctones como os importados é essencial para modelar a dinâmica da epidemia, possibilitando informação crítica aos tomadores de decisão, responsáveis pelo controle da doença. DESCRITORES: Dengue, epidemiologia. Infecções por Flavivirus, transmissão. Risco. Período de Transmissibilidade. Surtos de Doenças. Vigilância Epidemológica.

Dengue fever has been endemic in Brazil since 1986. 5ts transmission involves only urban mosquitoes 3 and humans, and many fatal cases occur every year due to epidemics where different serotypes are transmitted. 8e quick expansion of vectors and viruses throughout nearly all Brazilian states is facilitated by ever increasing moving and traveling of people in the country.Until recent years in most localities the origin of dengue cases was neither reported nor investigated.To the authors' best knowledge, no study has been published investigating the impact of "imported" cases on the dynamics of the epidemics.The origin of each case in the 2002-2003 epidemics in Brasília was inferred after a detailed survey of the patient's background, taking into account the duration of the incubation phase of the disease.As these data about the most probable origin of these cases were available, we investigated its possible impact on the dynamics of dengue epidemics.We compared its dynamics with the outputs of a simple "susceptibleinfectious-resistant" (SIR) model modifi ed in order to include the input of daily numbers of imported cases when estimating the basic reproduction number (R 0 ).Mathematical modeling is a powerful tool to simulate real past situations, to input data from the present and hopefully to forecast future outcomes. 6It has been widely used to understand the dynamics of infectious RESUMO OBJETIVO: Estimar o número básico de reprodução da dengue (R 0 ), com base nos casos importados, além dos casos autóctones.

INTRODUCTION
diseases. 1 A commonly used family of epidemiological models are the compartmental or "box ones": individuals are grouped in classes according to their immunological status (e.g.SIR models consider the progress in the numbers of susceptible, infectious, and removed immune people). 1 The basic reproduction number (R 0 ) is an essential parameter to describe the dynamics of an outbreak.It is the number of secondary infections generated by one case when the virus is introduced into a wholly susceptible population and when the probability of contact with the pathogen is homogeneous. 7e objective of the present study was to estimate the basic reproduction number (R 0 ) of dengue fever including both imported and autochthonous cases.

METHODS
Data analyzed refer to the dengue epidemics occurred during 2002-2003 in Brasília, Brazilian Federal District (Figure 1).Brasília is a city situated in Central-West Brazil and has a rather recent history of contact with dengue. 4en the number of imported cases is disregarded, the basic reproduction number is generally estimated from the epidemiological curve.At least four different methods for this estimation have been proposed.According to the classical formula, 1,13,15 where m is the relative density of mosquitoes to humans, a is the daily biting rate of the mosquitoes, b and c are the probabilities of viral transmission from an infected mosquito to a susceptible human, and from an infected human to a mosquito, respectively, μ is the daily mortality rate of the vector,τ e the extrinsic incubation period, and γ the inverse of the duration of viremia.In the present study, we modifi ed the method used by Massad et al 13 (2001) to estimate the basic reproduction number. 2 Considering that the initial growth of the number of cases is exponential with parameter λ, with τ i the intrinsic incubation period and the other parameters as above.Then, λ is estimated by a linear least squares fi tting of the initial part of the curve of the cumulated number of cases against the daily number of new cases.The number of cases, which was taken into account for the regression of the increase at the beginning of each outbreak (slope λ in Fig. 2 A-D), was chosen to ensure a statistical signifi cance of p<10 -4 .
A simple SIR model is used to characterize the progress of local (autochthonous) cases with continuous infl ux of exogenous cases (imported).We set S A , I A and N A as susceptible, infectious and cumulative number of autochthonous cases, N as the total susceptible population and I I and N I infectious and cumulative numbers of imported cases.The progress of N I is thus considered as an external forcing.
The parameter η is introduced to represent the possible differences between the durations of exposure times of autochthonous and imported hosts, the latter arriving at various times of viremia.The basic reproductive rate of autochthonous cases is while for imported cases is is thus the ratio between the basic reproductive rate of imported and autochthonous cases.Since the fi nal number of autochthonous cases is far lesser than the whole population, the total number of people can be assumed very large (here N = 100,000) and, as we only consider the initial model behavior, results are independent of N provided it is large enough.The other two parameters are computed by a fi t of the ratio between the cumulative autochthonous cases over the cumulative imported cases in the Nelder-Mead simplex algorithm pre-implemented using MATLAB.Using this index (the ratio between autochthonous and imported cumulative cases) for the fi t has yielded results that were more robust to the initial guess than the cumulative number of autochthonous cases.Bootstrap methods were used to estimate confi dence intervals and R 0 s values were compared to R 0 estimated as if all cases were autochthonous.

RESULTS
In the case of the 2003 dengue epidemic in Brasília, the imported cases were included, and the modifi ed model fi tted when only autochthonous cases were considered (Figure 3B), as well when autochthonous cases were divided by the daily numbers of imported cases (Figure 3A).The comparison of R 0 estimated using the method above (when the number of imported cases is unknown or disregarded) with R 0 estimated when Table 1.Estimates of the initial slope of the fi tted curve (λ) and R 0 with standard error, according to various methods, for the epidemics in Brasília (Federal District), Belém (Northern Brazil) and Fortaleza (Northeastern Brazil).

Methods for estimating R 0 λ ± 1 standard deviation
First doubling time of the number of cases 12 Initial exponential phase of the number of cases 13 Fitting of the SIR model on the initial slope of accumulated cases plotted against the daily number of cases exogenous cases are explicitly included is striking.
In the fi rst case, R 0 is far greater than one, indicating eco-epidemiological conditions favorable to local transmission of the virus.In the second case, the local R 0 (estimated for autochthonous transmission) is clearly lower than one.The results of the comparison between the three methods are shown in Table 1, for different epidemics and for a SIR-simulated one with pre-defi ned R 0 = 8.0.Besides great variability between the epidemics, the R 0 values estimated by the method here described are always greater than those obtained using other methods.Furthermore, this method is the only one that provides a correct estimation of the R 0 of the SIR-modeled epidemic.Figure 4 shows an almost linear relationship between R 0aut and R 0imp , when the former is set under one.Thus, under such assumptions, it becomes evident that in case of an epidemic, the reproduction number due to imported cases cannot be above 0.45 and that if the reproduction number due to autochthonous cases is near one, the latter becomes negligible.

DISCUSSION
The methods used by Marques et al 12 (1994) and Massad et al 13 (2001) systematically underestimated R 0 , probably because they failed to consider the whole delay between two successive cases (viremia, and intrinsic and extrinsic incubation periods).The differences between the R 0 values obtained for different epidemics (Table 1) refl ect various eco-epidemiological conditions infl uencing hidden parameters (such as mosquito density, effective exposure rates between hosts and vectors 10 ).
An unknown proportion of cases remain undetected as they are asymptomatic or are diagnosed as other fevercausing diseases.This fact makes the determination of the number of cases uncertain but not the estimation of R 0 if we postulate that it is independent of the intensity of symptoms.A similar, or even more serious diffi culty is the number of uncharacterized imported cases, which may negatively affect the estimation of R 0 and thus invalidate any decision about prevention and control.This clearly implies that dengue epidemics may occur even when local ecological conditions are unfavorable, as was probably the case in Brasília in 2003.This is the fi rst report emphasizing this aspect.Taking into account the imported vs. autochthonous nature of each diagnosed case will allow a better evaluation of the risk of dengue transmission or epidemic.Health authorities involved with dengue monitoring and prevention should develop and strengthen their methods to determine the imported or autochthonous origin of the cases.Future studies should incorporate these data in models, along with the yet missing entomological fi eld data.As the populations of arthropod vectors are much infl uenced by climatic factors, it may thus be possible to link the dynamics of outbreaks to climatic factors through the underlying R 0 and vectorial capacity.

Figure 2 .
Figure 2. Estimation of R 0 including only the autochthonous cases or the total number of cases when the number of imported ones was not known.(A) Simulated epidemic, obtained by the SIR model where R 0 was set at 8.4 (see methods), (B) Belém (Northern Brazil) 1996 epidemic, (C) Brasília (FD), 2002 epidemic, and (D) Fortaleza (Northeastern Brazil), 2003 epidemic.The number of cumulative cases used to estimate R 0 was chosen for p<10-4; λ is the slope of the regression curve.

Figure 3 .
Figure 3. Fitting of the modifi ed SIR model (with imported cases included) with the initial cumulative curve of daily numbers of autochthonous cases (B), and daily ratio of autochthonous / imported cases (A) in Brasília, DF, 2003.Best-fi t model was obtained with γ = 0.04 and β = 0.015 for η = 1 and with γ = 0.07 and β = 0.048 for η = 0.3 (η is the ratio between the basic reproductive rate of imported cases and the rate of autochthonous cases).

Figure 4 .
Figure 4. Relationships between imported and autochthonous cases during the 2002 epidemic in Brasília, FD, Brazil.R 0I and R 0A were both set below one, with R 0I ≤ R 0A , and the R 0A /R 0I ratio was set at various values from 1 to 0.2.Error bars represents the 0.95 confi dence interval.