Dynamics of dengue virus circulation: a silent epidemic in a complex urban area


M. G. Teixeira, Instituto de Saúde Coletiva, Universidade Federal da Bahia, Rua Padre Feijó, 29 Canela, 40.110-170 Salvador-Bahia, Brazil. Tel.: +21 245 0544; Fax: +71 237 5856; E-mail: magloria@ufba.br


Serotypes of dengue DEN-1 and DEN-2 have been reported in much of Brazil over the last 15 years, and DEN-3 serotype was only recently detected. This prospective study was conducted in Salvador, a large city in north-east Brazil, where two epidemics were previously recorded (DEN-1 and DEN-2). We obtained the seroprevalence and 1-year incidence of dengue infections in the population of 30 sampling areas of Salvador and analysed the relationship between intensity of viral circulation, standard of living and vector density. High seroprevalence (68.7%) and annual incidence (70.6%) of infection for one or both circulating serotypes (DEN-1 and DEN-2) were found. High rates of transmission were observed in all studied areas, from the highest to the lowest socio-economic status. The mean PI (Premise Index) for Aedes aegypti was 7.4% (range 0.27–25.6%). Even in the areas with the lowest PI (≤3%) the observed seroincidence was 54.6%. These findings highlighted the existence of a silent epidemic during a period perceived by the Health Services as of low endemicity, indicating the strength and speed of dengue transmission in the city of Salvador.


The occurrence of epidemics of classical dengue, haemorrhagic dengue and dengue shock syndrome in the America make this vector-borne disease an important public health concern throughout the New World. The serotypes DEN-1 and DEN-2 have been reported in Brazil over the last 15 years, and the DEN-3 serotype was detected only recently. Few cases of dengue haemorrhagic fever have been registered. Since 1995 the virus of dengue (DEN-1 and DEN-2) is circulating in Salvador, a large city in the north-east region of the country, and two previous epidemics have been recorded (Teixeira et al. 2001).

A number of factors have been pointed out as contributing to the resurgence of dengue. Demographic and societal changes, increased air travel, decay of the public health infrastructure (Gubler 1998) and the co-circulation of multiple serotypes increase the potential for the emergence of dengue haemorrhagic fever (Halstead 1981). In the absence of a vaccine, dengue prevention currently depends on effective vector control, but a new research initiative is required because this strategy is not enough to prevent dengue virus transmission (Reiter 1998).

One of the controversial aspects about dengue is the relationship between environmental conditions and the risk of infection, which has been related to both good and bad standards of living (Medronho 1995; Vasconcelos et al. 1998). The dynamics of dengue transmission in urban settings are still poorly understood. Better knowledge about the epidemiology of this disease requires epidemiological studies, especially those based on incidence (Kuno 1995). For instance, it would be useful to know the threshold of transmission in terms of vector density, so as to develop new strategies for control of this disease (WHO 1999). Our study of the seroprevalence and incidence of dengue infection in a large Brazilian city examines the relationship between intensity of viral circulation and standard of living, environmental quality and vector density.

Materials and methods

This prospective cohort study was analysed both as an ecological study and in terms of risk to individuals. In 1998 the city of Salvador had around 2.3 million inhabitants, with wide differences between specific areas in the socio-economic status of the population and the quality of their environment.

The sampling scheme is described in detail elsewhere (Teixeira et al. 2002). The study population was drawn from a spatial sample of 30 neighbourhoods throughout the city, purposively selected from the census enumeration areas which had been stratified on the basis of sanitation coverage and income level, so as to represent a wide range of living conditions. These included areas where living standards were

  • High: > 80% of households with sanitation and >50% of households earning more than five times the national minimum wage (US$ 80.00, six areas);
  • Medium: 50–80% of households with sanitation and over 50% of households earning one to four times the national minimum wage (19 areas);
  • Low: less than half of households with sanitation and > 50% of households earning less than the national minimum wage (five areas).

The data about level of income and population density were supplied by the Brazilian Statistics Institute (Brazil 1996). The association between sanitation coverage and income was such that all census tracts in the city fell into one of the three categories above.

For the sample size calculation, a seroprevalence of dengue of 50% was assumed, based on previous surveys in other Brazilian cities (Cunha et al. 1995; Vasconcelos et al. 1998). Assuming a precision of ±3%, a cluster effect of 1.5, aconfidence level of 95%, and an inflation of 30% to compensate for those individuals lost gave a final figure of 2149 individuals. Using a previous census as a sampling frame, 2149 residents of the 30 sample areas were selected from the population of 68 749 by random sampling without replacement and with post-stratification (Cochran 1977).

A questionnaire which included name, address, sex, age, educational level, income and yellow fever (YF) vaccination status was administered to all participating individuals between May and July 1998. After informed consent was obtained, the first sample of blood was also collected at thistime. Three individuals who were vaccinated were excluded to avoid false positives because of cross-reaction. One year later, a second sample of blood was collected from all individuals except those who had tested positive for both serotypes in the first collection.

Blood samples were collected in sterilized vacuum tubes of 10 ml, the serum was separated by centrifugation and stored at −20 °C. These samples were sent in thermal boxes containing ice to the Arbovirus Laboratory of the Evandro Chagas Institute, where the Haemagglutination Inhibition/HI test (Clark & Casals 1958), modified by Shope (1963), was carried out using antigens for the four serotypes of the dengue virus and four other flaviviruses: YF, Rocio, Ilhéus and St Louis Encephalitis, although these do not circulate in Salvador. Serological response to flaviviruses is of controversial interpretation and differs between the first (primary) infection and any subsequent (secondary) infection with another flavivirus or serotype, as flaviviruses show an increasingly strong response on subsequent infection and serological cross-reaction is frequently observed. Thus, the criterion adopted was that defined by World Health Organization (1997): titres by HI of 1:20 or higher, exclusively to a specific dengue serotype, or titres four times higher for one serotype than for another (DEN-1 or DEN-2) were considered positive and specific for that serotype (primary response). Titres indicative of secondary response were also as defined by WHO (1997). These were confirmed by IgG–enzyme-linked immunosorbent assay (Chungue et al. 1989), and considered positive to both serotypes meaning that infection with both DEN-1 and DEN-2 had occurred.

Crude and age-standardized seroprevalence and incidence of infection for dengue were calculated by the direct method (Rothman 1986) for each sampling area and all the categories of living standards, using the overall studied population as the reference population. Since the survey were 1 year apart, seroincidences were all expressed as yearly rates in percentage. The prevalence ratios (PR) and relative risks (RR) of infection with dengue virus were calculated, as well as 95% confidence intervals. The chi-square test for trend was used. Exploratory analysis was performed to characterize the study population and to verify the association among the variables using the Pearson correlation coefficient. An exploratory analysis was also performed at the individual level. Risk factors for occurrence of infection by the dengue virus were analysed by logistic regression, thereby obtaining the measures of association (PR and RR) by the Delta method (Armitage & Berry 1987), setting the thresholds so that the ‘exposed’ individuals were those ≥ 15 years, with less than 8 years of schooling and with an income of less than twice the minimum wage.

The Premise Index (PI) was obtained for each sampling area as the percentage of all inspected premises which were found with at least one positive breeding site for Aedes aegypti larvae. Using ancova (Montgomery 1991) the incidence of infection for the neighbourhood in each range of PI (≤3%; 3.1% to 5%; 5.1% to 10%; and > 10%) was calculated and adjusted for age and mean seroprevalence. The preventable fraction (PF) was also calculated by treating ‘unexposed’ individuals as if they lived in areas where the PI was 3% or less.

Data were entered in Epi-Info 6.0 (Atlanta, Center for Disease Control, 1994) and further analysis was performed using SAS and STATA. The study protocol was approved by the ethics commission for scientific research of the Oswaldo Cruz Institute (FIOCRUZ).


Of the 2149 selected individuals, 1515 (70%) participated in this study with complete data. However, this did not prejudice the power of the study because the sample size had been inflated by 30% and there were no major demographic differences between the studied and the missing cases. The age distribution of the studied cases showed that 5.1% were aged < 5 years; 20.9% were 5–14 years old, and 74% were ≥ 15 years old. A total of 57.9% were females. Sixty-eight per cent of the adults (age > 15) had less than 8 years of schooling. Around 25% of the population reported a household income lower (< 160 US$ per month) and 50% had an income between two and five times the minimum wage.

The overall seroprevalence was 68.7%, varying from 16.2% to 97.6% between the 30 areas. The overall seroprevalence of infection with two serotypes of dengue was 43.2%, varying from 0% to 87.9% across the 30 study areas. In two areas, no individuals with double infections were found.

Seroprevalence values were lower in younger age groups: 39.0% among 0–4 year olds and reaching a maximum of 76.4% in the 30–39-year-old age group. The incidence of infection was lower in the group aged 0–4 (46.2%), but showed a sharp increase in the next age group (5–9 years old) (78.3%) and maintained levels between 62.7% and 82.8% in the older age groups. Between the age groups of ≥ 15 and < 15-year-old there was a statistically significant difference both in seroprevalence (57.4%vs. 76.1%; P < 0.001) and in seroincidence (73.3%vs. 81.0%; P = 0.03). Neither the seroprevalence nor the incidence of dengue infection was associated with education, income or sex.

Ecological analysis

No significant correlation was found between seroprevalence and the proportion of individuals with lower levels of education or the mean income in the 30 sampling areas (Table 1). On the other hand, seroprevalence showed a strong positive correlation with population density (r = +0.49; P < 0.01).

Table 1.  Correlation coefficients for age-standardized seroprevalence and seroincidence for one or two dengue serotypes with several selected variables related to the 30 sampling areas, Salvador, Brazil 1998–99
 Seroprevalence (1998)Seroincidence (1998–99)
  • *

    Per cent of individuals

  • 15 years old with elementary school not completed.

Low level of schooling*−0.28 0.14−0.50<0.01
Mean income−0.06 0.76−0.12 0.52
Population density (hab/km2)+0.49<0.01−0.15 0.43

Table 2 shows that the overall seroprevalence increased with worsening living standards (inline image = 8.386; P < 0.01).

Table 2.  Age-standardized seroprevalence and seroincidence of dengue according to living standards in the 30 sampling areas, Salvador, Brazil 1998–99
 Seroprevalence (%) (1998)Seroincidence (%) (1998–99)
Living standards of areaNumber examinedAge-standardizedPR* 95% CINumber examinedAge-standardizedRR 95% CI
  • *

     Prevalence ratio;

  •  Risk ratio;

  • ‡ 

    ‡ In parenthesis number of sampling areas.

High (6) 29362.81.0 – 7976.61.0 –
Medium (19)101469.21.06 (0.96–1.16)44570.30.94 (0.81–1.09)
Low (5) 20878.41.19 (1.07–1.33) 7166.70.88 (0.71–1.09)


Of the 860 individuals who tested negative or positive for only one serotype in the first survey, 595 participated in the seroincidence survey, meaning a loss of 30.8%, primarily because of address changes of the individuals. However, the social and demographic characteristics of the sample remained similar to those in the first survey.

The overall annual incidence rate of dengue infection was 70.6%. Specifically, the incidence rate in individuals who had previously tested positive for one serotype was 83.0%, significantly higher than among those who were negative in the first survey (60.8%) (Table 3). At area level, negative correlations were found between incidence of infection and level of education, mean income and population density (Table 1). However, the correlation was statistically significant only with education level (r = −0.50; P < 0.01).

Table 3.  Incidence of infection with the dengue virus in individuals residing in the 30 sampling areas with respect to their previous immunological status, Salvador, Brazil 1999
Previous serological stateState at second surveyNo.Incidence (%)
Negative (N = 331)Positive for one  serotype 7723.3
Positive for two  serotypes12437.5
Subtotal 20160.8
Positive for one  serotype (N = 264)Positive for two  serotypes21983.0

The incidence of infection varied among study areas from 50% to 90%. It was higher in areas with lower initial seroprevalence.

Contrary to the seroprevalence survey, seroincidence presents a trend towards increase in the direction of the areas with best living conditions. As a consequence, the highest incidence rates were found in those areas that had been classified as having the highest standard of living. However, this trend was not statistically significant (χ2 = 1.332; P = 0.25; Table 2).

The PI for A. aegypti ranged from 0.27% to 25.6% between areas (mean = 7.4% and median = 5.2%). There was a weak positive correlation between PI and incidence of infection, which was not statistically significant (r = +0.17; P = 0.36). When the sampling areas were re-grouped according to PI level, in the areas with the lowest PI, the observed seroincidence (adjusted for age and mean seroprevalence) was the lowest but nevertheless very high (54.6%, Table 4). In the areas with higher PI levels, the seroincidences were similar. The only statistically significant difference in the seroincidence level occurred between the first and second PI group (RR = 1.38; 95% CI 1.07–1.77), and the chi-square test did not show a statistically significant trend (inline image = 0.14; P < 0.71). As in the areas with PI≤3 asignificantly lower seroincidence was observed, PI = 3 was considered as the threshold level below which the transmission rate starts to decrease. On this basis, the PF in the areas of Salvador with PI equal or below 3 was calculated at 29.7%.

Table 4.  Incidence adjusted for the initial seroprevalence and mean age of the residents of the 30 sentinel areas, for level of Premise Index (PI) for Aedes aegypti, Salvador, Brazil 1999
Premise Index (%)Adjusted (%)Crude (%)RR


The results of this study provide evidence that there was a silent dengue epidemic in the city of Salvador between 1998 and 1999, undetected by the official surveillance system. By extrapolation of the dengue incidence rate in the sample to the city population, it can be estimated that during a 12-month period, around 560 000 individuals were infected by one or two serotypes of the dengue virus. In the same period, the official system of notification had recorded only 360 cases. It is worth noting that in previous years, mainly 1995 when the first dengue epidemics occurred in Salvador, and 1996, a total of 26 446 cases were registered by the city epidemiological surveillance system (Teixeira et al. 2001).

Moreover, this survey showed that the virus still has a surprisingly high rate of transmission 4 years after its introduction into the city. Even considering the deficiencies of the epidemiological system it was capable to detect previous epidemic, suggesting that after the first epidemic cycles the virus circulation continues intensively, but the clinical manifestations must be far less severe in most cases. Facilitating factors for this high transmission of the virus include the tropical climate, high population density and high levels of infestation by A. aegypti (Kuno 1995). There is no evidence that a major part of these cases is caused by migration or travel, as the immigration to Salvador is currently low and between the two surveys there was no register of large epidemics in other cities of Bahia State.

The high seroprevalence recorded in the first survey indicates that approximately 1.5 million people were infected with dengue virus. Nevertheless, incidence observed among susceptible individuals was no less than the initial seroprevalence, which represents the cumulative incidence in the first 3 years since the introduction of the virus.

The incidence of infection is higher than 55% in each group of sampling areas by PI, indicating that viral transmission remains high even when the PI levels are low in spite of a partial herd immunity for the two circulating serotypes (43.2% as found in the seroprevalence survey). This finding agrees with the theoretical model of dengue transmission developed by Newton and Reiter (1992) from experiences in Singapore, according to which transmission can be maintained at high levels in situations of low vector density and partial herd immunity. These authors also concluded that in order to eliminate viral transmission, the vectorial indices should be very close to zero and/or the herd immunity should be at least 90%.

As the PI levels found in different areas of Salvador must be, in part, a consequence of the vector control activities implemented in the city, and as the lowest level of incidence was found in areas with PI≤3, it follows that vector control activities will start to have some effectiveness when PI reaches a level below 3.

The high risks of infection observed in all three categories of area, and the lack of a relationship between incidence and the standard of living, suggest that the risk of being infected by the dengue virus is now similar in all areas of Salvador. The observed associations between seroprevalence and population density, and between incidence and the PI, should not be interpreted as indications that the greatest risk of infection is in the poorer areas, as these two factors are present not only in the poorest areas but also in areas with the highest standards of living.

These findings differ from other Brazilian data originating from notified cases in that the latter indicate that people from lower socio-economic backgrounds are more affected and at greater risk (Medronho 1995; Costa & Natal 1998). These differences can be explained by the fact that the official notification system only registers cases which are reported by the public health care system, which is used more commonly by the poor. The results of this study are similar to others based on serological surveys in random populational samples (Vasconcelos et al. 1998), whose results are likely to be closer to the real situation of viral circulation.

A significantly higher incidence was found among those who were positive for one serotype in the first survey. This suggests a degree of heterogeneity of exposure which was not apparent in the ecological analysis. This is striking in view of the reasoning by which differences in exposure at the individual level should be amplified in ecological analyses (Koopman & Longini 1994). Further studies are required to identify the behavioural and environmental risk factors (both in the public and private domains), which have the greatest influence on transmission.

In the period between the first reported case in Salvador (January 1995) and our first survey (April 1998), there were nearly 30 000 notified cases of dengue. All these were classical dengue cases and the mortality burden was practically zero as no cases of haemorrhagic dengue had been reported. No doubt the morbidity burden had an impact upon work and school absenteeism and on out-patient health care demand, although these were not quantified. During the study period, despite the great number of individuals with new infections, there were very few reported cases of classical dengue and no cases of haemorrhagic dengue.

There is no basis to expect a decrease in the severity of the disease. This finding therefore suggests that after some years of the virus' introduction, transmission continues to occur but with a decrease in the perception of the problem by the individuals and health professionals and hence reduced sensitivity of the surveillance system. On the other hand, the increase in herd immunity to two different serotypes of dengue, and the maintenance of the environmental conditions necessary and sufficient for the transmission of new serotypes, create the conditions for the occurrence of epidemic haemorrhagic dengue (Halstead 1981) with all its consequences for morbidity and mortality (Gubler 1998).


The authors acknowledge the financial and technical support of the National Center of Epidemiology, Ministry of Health, Brazil, especially its Director Dr Jarbas Barbosa da Silva Junior.