Introduction The global incidence of dengue has increased exponentially since 1955 to reach 1–50 million infections per year in 2000–2005 . Dengue has become the most important arboviral disease of humans, and an increasing urban health and economic problem in tropical and subtropical regions worldwide. Classic dengue fever (DF) and its more severe forms, dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS), have expanded worldwide from Southeast Asia since 1946 . The four existing serotypes of dengue virus (DEN-1, DEN-2, DEN-3 and DEN-4) only confer life-long immunity against reinfection by the same serotype, and subsequent infections with a different serotype enhance the likelihood of DHF/DSS ,. There are currently no anti-dengue drugs available, and promising vaccines still await efficacy trials . Aedes aegypti, a domestic mosquito species that develops in water-holding containers, is the main vector of dengue and urban yellow fever. Current dengue control strategies seek to reduce Ae. aegypti abundance; optimize diagnosis and treatment of dengue cases, and decrease the frequency, magnitude and severity of dengue epidemics through integrated control strategies . Dengue became a recognized public health threat in the western hemisphere after the 1981 Cuban epidemic, the first major DHF/DSS outbreak in the region . Eradication programs initiated in 1915 achieved the apparent elimination of Ae. aegypti from most of the region by 1970 utilizing vertically-structured programs that included full-coverage source reduction supplemented with perifocal insecticide spraying with DDT ,. During the 1970s, however, Ae. aegypti reinfested most of the countries from where it had been eliminated and is no longer considered a target for elimination . Despite the transition from eradication to control, very little information on the long-term effects of vector control actions are available , and success stories are limited. Moreover, “Experience of vector control programs at various program levels and of successful and unsuccessful disease and vector surveillance systems needs to be recorded to allow adoption of best practices in other places” . After a contained dengue outbreak by DEN-1 and DEN-4 in Roraima (northern Brazil) in 1981–1982, the Southern Cone countries of South America began to suffer major DF outbreaks in 1986 (Brazil), 1987–1988 (Bolivia), 1988–1989 (Paraguay), 1998 (Argentina) and 2002 (the island of Pascua, Chile) ,. Uruguay and continental Chile have remained free of local dengue outbreaks, though they notified imported cases in 2006–2007. Northern Peru has had dengue activity since 1990, but the southern departments have remained free of Ae. aegypti and dengue so far. In Brazil, the number of reported dengue cases between 1995 and 2007 ranged from approximately 100,000 to 800,000 per year –. The northeastern and southeastern Brazilian states were invaded by DEN-1 in 1986–1987; by DEN-2 in 1990–1991, and by DEN-3 in December 2000. DHF/DSS cases have occurred every year in Brazil since 1994  and in Bolivia since 2002. In Argentina, the last epidemic outbreak of DF before the eradication era occurred in eastern and northeastern provinces in 1926 . Declared eradicated in 1963, the presence of Ae. aegypti was again detected in two northeastearn provinces in 1986, and most provinces north to 35°S were found to be reinfested by 1999 . The number of localities infested currently exceeds those recorded before the eradication era . The first local outbreak of DF (by DEN-2) occurred in the northwestern Province of Salta in 1998 ,, and was most likely linked to an underreported outbreak of DEN-2 emerging in Bolivia in 1996 and 1997 ,. The second outbreak of dengue (by DEN-1) occurred in the northeastern provinces of Misiones and Formosa in 2000, on the border with Paraguay where a massive DEN-1 outbreak occurred in 1999–2000 . The emergence of DEN-3 in Salta in early 2003 (co-circulating with DEN-1 and DEN-2) was shortly followed in 2004 by significant outbreaks of DEN-3 there and in the neighboring provinces of Jujuy and western Formosa , and in 2006 in Salta and Misiones (by DEN-2 and DEN-3). The historical pattern of dengue case notifications in the Southern Cone between 1995 and 2007 shows increasing hyperendemic transmission of several serotypes with multiannual fluctuations in Brazil, Bolivia and Paraguay, with epidemic behavior in northern Argentina. Clorinda, the most affected city in Formosa during the 2000 outbreak, is located in one of the high-risk zones of potential dengue transmission in Argentina . In collaboration with the local Municipality and other health and research institutions, Fundación Mundo Sano (FMS) launched a citywide control program in late 2002 with the objective of “reducing the risk of occurrence of autochthonous cases of dengue in Clorinda and its area of influence through the application of strategies that reduce the population abundance of Ae. aegypti”. The intervention program had a health promotion and education component but relied primarily on the systematic application of larvicides combined with source reduction efforts. Between 2003 and 2007, the program conducted and monitored 14 cycles of focal treatment, with only the first previously reported . As part of a cooperative agreement between FMS and the University of Buenos Aires established in late 2006, we herein describe the implemented intervention program and assess the long-term effects of vector suppressive actions on Ae. aegypti larval indices and on the reported incidence of dengue during the five-year period. We also sought to diagnose major limitations in the control strategy conducted, and prescribe possible solutions for improved vector control. Methods Study site The city of Clorinda (25°17′S, 57°43′W, Formosa, Argentina) is located on the southern margin of Pilcomayo River at 45 km from the city of Asunción (population, 519,000 people in 2005) in Paraguay (Figure S1). Clorinda had 38 identified neighborhoods with 10,752 houses and 47,240 inhabitants in 2001, and approximately 16,000 houses and 49,000 people in late 2007; individual neighborhood size differed greatly from 0.1). Citywide house indices declined sharply from 13.7% at baseline to 3.7% at the second focal cycle conducted mostly through spring of 2003 whereas Breteau indices declined from 19.0 to 4.8 (Figure 2). Larval indices then fluctuated seasonally and peaked every year between summer and early fall, with large variations between neighborhoods within anyone focal cycle as expressed by interquartile ranges. Monthly house and Breteau indices at a citywide scale over the five years were highly positively correlated (r = 0.966, n = 60, P 0.3). Breteau indices showed similar patterns. 10.1371/journal.pntd.0000427.g003 Figure 3 Short-term impact of larviciding and manual operations on larval indices at the same blocks in focal treatment cycles 1–7 in Clorinda, Argentina, March 2003–April 2005. The abundance and infestation of water-holding container types and distribution of larval indices differed largely among types of container. For example, at focal cycle 12 (spring–summer 2006–2007), tanks, barrels and drums for water storage (B type) were the most abundant containers (4,380) and the most likely to be infested (15.2%), accounting for 49% of all infested containers found (Figure 4). The second most important container class was disposable bottles, cans and plastics (E type), which were abundant (1,476) and as frequently infested (15.0%) as B containers, but accounted for only 16% of all infested containers. The frequency distribution of containers per type in cycle 12 was highly overdispersed between neighborhoods, with coefficient of variations that increased from 36% (type B) to 288% (D). 10.1371/journal.pntd.0000427.g004 Figure 4 Distribution of the number of containers inspected and infested by Ae. aegypti according to type of container at focal treatment cycle 12 in Clorinda, Argentina, August 2006–January 2007. The reported incidence of classic DF cases in Clorinda declined from 10.4 per 10,000 inhabitants in 2000 (46 confirmed cases including 20 autochthonous cases by DEN-1, and about 500 suspect cases) to 0 from 2001 to 2006 ,, and then increased up to 4.5 cases (by DEN-3) per 10,000 in January–April 2007  (Figure 5A). Of 267 suspect DF cases in 2007, 21 were confirmed (including only 5 autochthonous cases); 86 were excluded as dengue, and 160 remained without confirmation . Meanwhile in Paraguay, following the 1999 outbreak with 1,164 reported cases, 24,282 cases by DEN-1 were reported in 2000 (incidence, 49.3 per 10,000) though estimates ranged up to 300,000  (Figure 5B). This outbreak was followed by a low-level transmission period with decreasing number of cases from 1,871 in 2002 (by DEN-1, DEN-2 and DEN-3), to 137 and 164 cases in 2003 and 2004 (by DEN-3), and another upward trend with 405 cases in 2005 (by DEN-2) and 4,271 cases (by DEN-3) in 2006 . In Asunción, with high infestation levels, 1,700 DF cases were reported in 2006 , and the incidence of DF was 135.7 per 10,000 inhabitants in January–April 2007 . The 2007 outbreak included 28,130 reported cases by DEN-3 (up to week 21, several with unusually severe manifestations), 55 confirmed cases of DHF/DSS, and 17 deaths probably related to dengue . 10.1371/journal.pntd.0000427.g005 Figure 5 Reported cases of dengue and incidence of dengue in Clorinda, Argentina, and Paraguay, 1988–2007. Discussion The citywide intervention program in Clorinda (i) exerted a significant impact on larval indices with respect to pre-intervention levels but failed to keep them below the desired target levels (especially during summer in some neighborhoods) despite the extended use of temephos and additional actions; (ii) achieved sustained community acceptance; (iii) most likely averted new dengue outbreaks between 2003 and 2006, and (iv) limited to a large extent the 2007 outbreak of DEN-3 in an immunologically naive population. Failure to reach the desired target levels occurred in the absence of significant insecticide resistance in Ae. aegypti populations from Clorinda, where the resistance ratios to temephos steadily increased from 1.31 in 2004 to 3.10 in 2007 . The intervention program exerted a greater, more sustained impact than the time-limited vector control campaigns conducted in late 2001, whose effects were short-lived and undetectable by late 2002. Post-intervention larval indices during every late summer from 2001 to 2005 were much lower in Clorinda than in Tartagal and Orán (Salta), where in spite of larviciding efforts mean house and Breteau indices peaked at 27–35% and 20–130, respectively . The seasonal dynamics of infestation documented in Clorinda, with peaks in late summer is overall consistent with data recorded in Salta , and Iguazú (H. D. Coto, unpublished data). This appears to be a generalized regional pattern despite local differences in the main types of infested containers. The putative reasons for failing to keep larval indices below target levels in Clorinda are multiple: Incomplete surveillance coverage. Vector surveillance and insecticide treatment coverage were high but incomplete in space and time despite the community's high levels of acceptance of suppressive measures. House indices were inversely related to surveillance coverage. Houses lost-to-inspection may have constituted a persistent source of infestations of unknown magnitude. Re-visiting houses at appropriately scheduled times increased surveillance coverage from late 2006 forward. Despite its importance elsewhere , intensified searches for alternative developmental sites in non-residential habitats (tree holes and bromeliads) yielded negative results, and only 3.4% of indoor containers were found to be infested in early fall 2007 (unpublished results); Limited residuality of temephos. The expected duration of temephos residual effects (about three months) and standard control guidelines, further compounded with operational restrictions, dictated that each focal treatment cycle took an average of almost four months. However, the residuality of temephos has recently been shown to vary widely between <1 and 6 months depending on the type of formulation and container –, and is much more limited in the field . In Clorinda, the occurrence of infested containers mostly within 46 days after larviciding operations at the same blocks during focal cycles 1–7, combined with preliminary results of bioassays in treated tanks (Garelli, unpublished) , suggest that the actual residuality of temephos is much shorter and more variable than assumed. The fast turnover rate of potable water stored in tanks during spring and summer may have also contributed to reduced residual effects; Permanent developmental sites. In the absence of major changes in the management of large containers for permanent water storage, these mosquito developmental sites were continuously available, rainfall-independent, and varied little in frequency among neighborhoods. Unfortunately, the intervention program did not seek to identify the most productive containers for targeted control, and the intervention protocol only suffered minor adjustments over time. Pupal surveys conducted in a large neighborhood in 2007 showed that large tanks used for potable water storage were both the most abundant and the most productive type of container ; Suitable climatic conditions. The prevailing local weather conditions are very favorable for Ae. aegypti (as shown by the presence of immature stages on every month over the five years), and would determine fast egg-to-adult developmental times ranging from nearly 6 days in summer up to 10 days in early spring or fall according to experimental data . The gonotrophic cycle length of female Ae. aegypti would have been 3.8 days at local mean temperatures (27°C) over January–March 2003–2007, as interpolated from tables in . Larval indices were highly significantly associated with average minimum temperatures and rainfall one week earlier and average temperatures 4 weeks earlier. These data suggest that recent cool temperatures and rainfall influenced variations in Ae. aegypti larval indices, as other studies also showed for female and larval abundance ,. Fast development rates and the limited residuality of temephos during hot weather jointly create a window of opportunity in which the vector population recovers partially between focal cycles and then spreads by flight dispersal at a local scale ; Limited source reduction efforts. These efforts occurred mainly at the outset of the intervention program. Householders' response to cleanup messages and removal of containers was neither assessed nor coupled with improved solid waste disposal by municipal authorities, which facilitated the persistence of potential developmental sites; Lack of regular perifocal residual spraying with insecticides. Only indoor ULV spraying with partial or minimal coverage was conducted sporadically and therefore probably exerted negligible effects on subsequent larval indices; Lack of adequate, sustained community participation beyond mere acceptance of regular control measures albeit at high levels. Larval indices decreased more sharply both in relative and absolute terms immediately after the multiple, high-coverage control actions executed at focal cycle 1 than at subsequent cycles focusing mainly on larviciding operations and eventually adding ULV sprays. The actual effectiveness of vehicle-mounted ULV sprays has been questioned repeatedly ,, and was not specifically assessed in Clorinda. However, a posteriori comparisons of Breteau indices (Table 1) between successive focal cycles conducted during summer-fall (cycles 1–2, 4–5, 7–8, 10–11 and 13–14) show maximum percent relative reductions between cycles 13–14 (83%, including high larviciding coverage and six citywide cycles of vehicle-mounted ULV sprays) and cycles 1–2 (75%; intense source reduction and incomplete indoor ULV), followed by cycles 4–5 (63%, no indoor ULV) and 10–11 (52%, partial larviciding, blocking, marginal indoor ULV and six citywide cycles of vehicle-mounted ULV sprays), whereas cycles 7–8 showed a 24% relative increase (marginal indoor ULV). In conclusion, maximum impact was achieved through multiple, high-coverage control actions including surveillance and larviciding efforts combined with either intense source reduction or repeated citywide cycles of vehicle-mounted ULV sprays. The upsurge in larval indices at focal cycle 10 reflects the limitations of chemical control and the well-known capacity of population recovery of Ae. aegypti. This upsurge is probably explained by a relative decrease in surveillance and larvicide treatment coverage over the two previous focal cycles 8 and 9 combined with a net increase in rainfall of 106 mm between December 2005 and March 2006, at very similar period temperatures as those recorded over 2003–2005. In contrast, the increasing surveillance and larvicide coverage achieved over focal cycles 11–13 combined with repeated vehicle-mounted ULV sprays at cycle 13 and more sustained cool temperatures in the driest winter 2007 preceded the lowest larval indices ever recorded during the five-year period (1.8% and 2.1). Therefore, several factors likely contributed to the sharp decrease in larval indices at focal cycle 14. Infestations varied largely between neighborhoods throughout the intervention period, with a few neighborhoods having persistently high indices over time. Such heterogeneities were uncovered by the random-effects multiple regression model. More peripheral neighborhoods with more discontinuous water service tended to display larger and more variable larval indices. An unreliable, discontinuous water service has been recognized as a major determinant of persistent developmental sites in Venezuela . Other environmental and socio-demographic determinants may have also contributed to the complex infestation dynamics observed. Heterogeneities in infestation and dengue transmission risk have been uncovered elsewhere , and may lead to developing more cost-effective, targeted control programs. House and Breteau indices gave a consistent picture of infestation levels over time, and in 2007 were positively correlated with pupal counts at the block level, though not very strongly so . Although the Breteau index was once considered the most informative measure for monitoring control operations ,,, both indices have limitations when used to assess the quantitative impact of control actions, partly because they are based on presence-absence of immature stages . Although it is often difficult to show significant intervention effects on larval indices, the post-intervention reductions observed in most focal cycles were substantive and statistically significant. Human infection usually is the crucial outcome for evaluating the impact of vector suppressive actions. The sustained intervention program most likely averted the occurrence of major dengue outbreaks in Clorinda from 2003 to 2006 despite of the vicinity of endemic dengue transmission in Paraguay during the same time period; the large daily movement of people across the border; the invasion of the new DEN-3 serotype into Brazil, Bolivia and Paraguay since 2001–2002, and the very low herd immunity to DEN-1 and DEN-3 in the Clorinda population. When a major DEN-3 outbreak emerged in Paraguay in early 2006, the ongoing larval suppressive actions supplemented by limited source reduction and ULV space spraying averted a local outbreak and also limited the apparent number of DF cases in 2007. These results need to be interpreted with caution and may constitute a reasonable circumstantial case, considering that the apparent incidence of DF in 2007 was both much lower than in 2000 and than in the concurrent epidemic in Asunción. Moreover, most of the 21 confirmed DF cases in Clorinda in 2007 reported recent travel history to Paraguay and were not considered autochthonous cases. In addition, the apparently random spatial distribution of autochthonous or suspect cases throughout the city  argues against intense local viral transmission. Vector control efforts apparently reduced the likelihood of localized outbreaks initiated from individuals who became infected in Paraguay, but Clorinda is rather small and this by itself could have reduced the likelihood of dengue transmission. Whether the several layers of heterogeneity detected in Clorinda reduced the likelihood of an epidemic remains an open question. Dengue transmission risks may be measured by entomologic biting rates ,. At mean ambient temperatures (27°C) in Clorinda during January–March 2003–2007, the expected extrinsic latent period of the virus in the mosquito would be 12.9 days, and the estimated entomologic threshold for dengue transmission (assuming 0% of dengue antibody prevalence) would only range from 0.7 to 0.83 Ae. aegypti pupae per person in January–May 2007 (interpolated from tables in ). A house index <1% and a Breteau index <5 were suggested as dengue transmission thresholds ,, and a maximum Breteau index ≥4 at the block level was considered a suitable predictor of dengue transmission in Cuba . If these transmission thresholds held in our study area, several sections of most neighborhoods would have been well above threshold during every summer, when cases typically peak in the region ,. However, dengue transmission thresholds are potentially modified by several factors. Some of the implicated parameters are hard to measure in the field, and its estimates are affected by large sampling errors and other sources of bias acting multiplicatively (i.e., increased uncertainty around threshold estimates) . Entomologic transmission thresholds need further validation , at various spatial scales before they are used operationally. Some problems limit the interpretation of our results. Quantitative data on manual treatment of containers and container-based coverage were not collected. The study design is a before-and-after community intervention with no internal control arm; the latter was not justified in view of the impending risk of another dengue outbreak in Clorinda. In practice, neighboring Paraguay served as a surrogate external control area for comparison of reported incidence rates given the limited vector suppressive actions conducted there. Before-and-after comparisons are potentially biased by temporal trends in vector density  but our multivariate analysis of intervention effects on larval indices accounted for weather effects. The reported incidence of dengue in Clorinda in 2007 was much lower than during the first local outbreak in 2000, but exactly how much is uncertain because no cohort study was conducted to establish the actual incidence of infection. Intense movement of people across the border with Paraguay seriously undermines the capacity to establish whether an infection was autochthonous or imported. The assessment of intervention effects may be further complicated by underreporting of asymptomatic or mild dengue infections, imported infections, and misclassification of other febrile illnesses with overlapping clinical features, such as influenza. The net effects of different sources of bias running in opposite directions are hard to assess. Febrile cases unrelated to dengue were 2–3 times as frequent as DF cases during the 2000 summer outbreak in Clorinda  and in 2007. Silent transmission of dengue viruses is more likely in populations without previous dengue, especially if caused by strains with low virulence . However, enhanced community awareness and continuous passive surveillance of febrile syndromes after the 2000 outbreak suggest that the absence of reported or suspect DF cases during 2003–2006 may at most represent marginal occurrence of local transmission over such period. A major feature of the five-year intervention program is that it was mostly supported and run citywide in a high-risk area by an external organization in close cooperation with the local Municipality and the Ministry of Health. The program was thus an example of highly desirable intersectoral cooperation . It was initiated and consolidated in the context of a severe socio-economic crisis in a border area, and succeeded in maintaining high levels of community acceptance over time. The latter in part may be attributed to the education and promotion campaign launched at the outset of operations, further enhanced by community awareness of the threat of a dengue outbreak spilling over from Paraguay. The fact that most of the vector control personnel were women from the same community may account for the very low mean fraction of households that denied entry to premises (3%) even in neighborhoods whose residents frequently did not allow the labor of municipal agents inside house premises. House-to-house larval surveys are typically plagued by difficulties of access. Issues of acceptability, coverage and delivery, which frequently compromise the effectiveness of the available vector control tools, were minimized in the Clorinda experience. However, the effective surveillance coverage of closed or vacant houses remains to be addressed. The intervention program trained a significant number of local human resources for future transference of control routines to official health services. Gaps in this transference or interruption of vector control actions are predicted to pave the way to rapid recovery of Ae. aegypti populations and eventual local dengue outbreaks. Implications for dengue vector control and research Gubler  noted that “… The sporadic nature of dengue epidemics and the misguided reliance on using insecticidal space sprays to kill adult mosquitoes prevented most countries from developing and implementing programs that focused on larval mosquito control, which were much more difficult to implement and maintain”. Here we advocate that such programs are still needed; they should be run permanently with high coverage, especially in a hyperendemic context with regional expansion of DHF/DSS cases, and may be mostly maintained with locally available resources, duly coordinated and supervised. This does not preempt the fact that governments must invest much more resources and efforts on scientifically-based vector control programs run by qualified personnel than they have done so far. Although the Clorinda intervention program did not reach target levels, it had a positive impact on public health because it prevented the serious dengue outbreaks that occurred in neighboring countries during the study period. Without the interventions, at the reported incidence rates in Paraguay the situation in Clorinda would have been much worse. The Clorinda experience was successful at some program levels and left some lessons for further improvement. The use of controlled-release insecticide formulations or sachets that are retrievable during cleaning and washing of water-storage containers would extend the residual activity of temephos . There is a great need of new, more effective larviciding products that last longer, but they will also face water management and coverage issues. The upward trend observed in temephos resistance indicates that resistance management schemes should be developed and considered in the near future, because RRs around 3 do not revert spontaneously to pre-intervention susceptibility levels. Resistance to temephos with potential cross-resistance to other insecticides has expanded greatly and caused repeated control failures in Brazil ,, and has been detected elsewhere in Argentina . Pyriproxyphen is a relevant candidate for larvicide replacement . Identification of neighborhoods at increased risk of infestation and transmission are needed for developing more cost-effective, targeted control strategies. Several sources of heterogeneity pose major challenges to the control of Ae. aegypti. The implemented program was born as a community-based intervention by some definitions  and gradually turned into a top-down vector control program suffering from excessive reliance on insecticides. In Clorinda, current interventions should evolve towards a multifaceted integrated program with intensified coverage, source reduction and environmental management measures, such as providing lids or insecticide-treated covers to water storage containers ,. Such integrated program needs also strengthened communication and health education components  and a broad social participation aiming at long-term sustainability. Other strategic solutions include development of the infrastructure for providing potable water and improved disposal of solid waste. Vector control and disease management must remain a regional effort to prevent “spillovers” such as those from Asunción to Clorinda or from Brazil to Paraguay, within the frame of sustainable development rather than being viewed exclusively as a matter of health . These issues are common to other neglected tropical diseases affecting vulnerable populations in the Gran Chaco region over Argentina, Bolivia and Paraguay . The price of not establishing regional control efforts of a more permanent nature has lead to the predicted  and observed expansion of DHF/DSS in the Americas, which reached Paraguay in 2007. Supporting Information Alternative Language Abstract S1 Translation of the Abstract into Spanish by Ricardo E. Gürtler. (0.02 MB DOC) Click here for additional data file. Figure S1 Map of Clorinda, Formosa, Argentina, its neighborhoods and location relative to Asunción, Paraguay. (5.62 MB TIF) Click here for additional data file. Figure S2 Percentage of houses inspected for larval infestations, closed or vacant, and refusing access for inspection at focal treatment cycles 1–14 in Clorinda, Argentina, 2003–2007. (0.26 MB TIF) Click here for additional data file. Text S1 Supplementary methods. (0.03 MB DOC) Click here for additional data file.