While a vaccine is under research, without immediate prospect for success, vector control remains the only way to prevent dengue transmission ( 1 – 3 ). Vector control programs are essentially based on source reduction, eliminating Aedes aegypti larval habitats from the domestic environment, with increasing community involvement and intersectoral action in recent decades ( 4 , 5 ). However, current entomologic indicators do not seem to reliably assess transmission risks, define thresholds for dengue epidemic alerts, or set targets for vector control programs ( 6 , 7 ). Therefore, defining new indicators for entomologic surveillance, monitoring, and evaluation are among the research priorities of the World Health Organization Special Programme for Research and Training in Tropical Diseases. Although only adult female Aedes mosquitos are directly involved in dengue transmission, entomologic surveillance has been based on different larval indices ( 8 , 9 ). The house index (HI, percentage of houses positive for larvae) and the Breteau index (BI, number of positive containers per 100 houses) have become the most widely used indices ( 6 ), but their critical threshold has never been determined for dengue fever transmission ( 9 , 10 ). Since HI 5%) ( 12 ), but these values need to be verified ( 13 ). The vector density, below which dengue transmission does not occur, continues to be a topic of much debate and conflicting empiric evidence. For example, dengue outbreaks occurred in Singapore when the national overall HI was 1 inhabitant was detected with confirmed dengue fever during the September–October 2000 outbreak. "Control" blocks (or neighborhoods) were randomly sampled from those in the study area where no dengue case was reported. Data Collection Dengue Fever Dengue cases were defined as patients with fever and >2 symptoms of dengue fever such as myalgia, arthralgia, headache, and rash, with serologic confirmation by immunoglobulin M–capture enzyme-linked immunosorbent assay ( 1 , 12 ) at the national reference laboratory of viral diseases in the Institute of Tropical Medicine, Havana. During the epidemic, suspected cases were identified through the health services. Additionally, a seroepidemiologic survey was conducted in the study area at the end of October 2000; all family physicians made home visits to families under their responsibility, searching for recent denguelike illnesses. Blood samples were collected from all persons with a history of fever. All confirmed dengue patients (passively and actively found) were interviewed by their family physician, supervised by an epidemiologist of the health area, to determine the exact date of symptom onset and places visited in the 10 preceding days. The completeness of the collected information was verified by epidemiologists of the Institute of Tropical Medicine, and if necessary, patients were revisited. Entomologic Information We used entomologic surveillance data that were independently recorded by the National Vector Control Program. At 2-month intervals, vector control technicians exhaustively inspected every house in the Playa Municipality for larval stages of Ae. aegypti. We used data collected in 3 cycles, July–August 2000 (before the epidemic), September–October 2000 (during the epidemic), and November–December 2000 (after the epidemic). We extracted information on the number of inspected houses, positive containers (with Ae. aegypti pupae or larvae), and houses with >1 positive container. We eliminated 4.8% of the blocks from the study because they were not inspected in the 3 inspection cycles. Data Analysis We related all data collected to geographic coordinates by a unique house block code and introduced it in MapInfo software (MapInfo Corporation, Troy, NY, USA). Case-patients were located by their address in the corresponding block. For the 3 entomologic inspection cycles, HI and BI were calculated at the block, neighborhood, and health area level. Additionally, we identified the BImax, which is the highest or maximum BI at the block level for each neighborhood of the case and control blocks included in the study. This variable is derived with the following equation: ,where BI i is the BI of the ith block belonging to the concerned neighborhood N, and ∀i⊂N indicates that all BI i of N are considered to identify the BI with the highest value as BImax. All data were exported to SPSS (SPSS Inc., Chicago, IL, USA) for analysis. We calculated the Spearman rank correlation coefficient between the different indices in the 3 inspection cycles. The entomologic indices were transformed to approximately normal distributions (by using square root transformation) for calculating means, standard deviations, and 95% confidence intervals. Differences in the distribution of the indices were assessed with the Mann-Whitney test. We assessed the discriminative power of the indices by using receiver operating characteristic (ROC) curves. Their accuracy to discriminate between case and control blocks (and neighborhoods) was classified according to the value of the area under the ROC curve (AUC) ( 24 ) as noninformative (AUC 50% specificity, for discriminating case and control geographic units was taken as the optimal cutoff point. The lower limit of 50% specificity was set to safeguard positive predictive value and decrease the number of units falsely classified at high risk for dengue transmission, which triggers unnecessary action and generates unproductive costs. The association between the entomologic indices and dengue transmission was further explored by logistic regression models. Results During the epidemic, health services assisted 4,679 febrile patients in the 5 health areas included in the study. All patients were serologically tested 5 days after onset of fever, and dengue infection was confirmed in 47. In the seroepidemiologic survey, 82.5% of the families were effectively visited by their family physician. The survey found 7,008 persons with symptoms of fever between September and October 2000 who had not previously attended the health services. Serum specimens were collected from all of them, and dengue infection was confirmed in 22. As a result, 69 (47 passively identified plus 22 actively identified) dengue cases were confirmed, all patients were interviewed, and 4 cases epidemiologically related to outbreaks in other municipalities were excluded from the study. The final sample consisted of 65 confirmed dengue fever patients who lived in 38 different blocks in the 5 health areas included in the study. In the July to August inspection cycle, before the outbreak, the overall municipal BI and HI were 0.92 and 0.87%, respectively (Table 1). The mean values of the indices calculated at the health area level were also ≈1 for areas with or without dengue cases during the subsequent epidemic. However, the mean BI and HI were >1 for case neighborhoods and substantially 1. Even more marked differences existed at the block and neighborhood levels, and after the outbreak the indices returned to average values 0.94, p 4, with a maximum BI of 50. Of the 17 confirmed dengue patients in September, only 3 (18%) lived in a block with BI>4 in the July–August inspection cycle. However, 15 (88%) lived in a neighborhood with at least 1 block with BI>4. The Aedes infestation increased during the second inspection cycle and then decreased again, concurrent with the intensified vector control activities during the epidemic. From November to December, after the outbreak, 71.6% of house blocks were Aedes-free, while 6.3% had BI>4. Figure Spatial distribution of dengue cases and Breteau indices (BI) at the block level before, during, and after the dengue outbreak, Playa Municipality, Havana, 2000. The mean block BI, the mean neighborhood BI, and the mean BImax for case and control blocks are given in Table 2. Before the epidemic, the mean BI values were approximately equal for case and control units. However, the BImax values were significantly higher for neighborhoods of case blocks. While transmission started in neighborhoods with high BImax infestation levels, it spread into blocks and neighborhoods with lower mean BI values in October. Still, during the epidemic, the indices remained systematically and significantly higher in case blocks. After the epidemic, they returned to similar values for case and control units. Table 2 Mean BI for case and control blocks before, during, after the dengue outbreak, Playa Municipality, Havana, 2000* Block July–August 2000 (before epidemic), mean (95% CI) September–October 2000 (during epidemic), mean (95% CI) November–December 2000 (after epidemic), mean (95% CI) BI NBI BImax BI NBI BImax BI NBI BImax September case blocks (n = 9) 0.53 (0.02–1.75) 1.52 (0.76–2.53) 6.28† (3.29–10.23) 11.95† (2.26–29.27) 10.75† (6.73–15.70) 28.4† (16.1–44.1) 0.63 (0.04–1.70) 0.64 (0.37–0.91) 2.94 (1.71–4.83) October case blocks (n = 29) 0.29 (0.05–0.72) 1.01 (0.60–1.54) 4.24 (2.48–6.46) 1.39† (0.50–2.71) 3.16† (1.99–4.61) 12.2† (7.79–17.6) 0.66 (0.06–0.91) 0.76 (0.44–1.06) 2.87 (1.50–4.35) Control blocks (n = 38) 0.20 (0.02–0.58) 0.69 (0.42–1.02) 2.96 (1.71–4.56) 0.42 (0.07–1.05) 1.52 (0.91–2.29) 1.52 (3.57–8.32) 0.33 (0.06–0.82) 0.68 (0.36–1.18) 2.34 (1.43–4.27) *BI, Breteau index; CI, confidence interval; NBI, neighborhood BI; BImax, maximum BI at the block level for each neighborhood.
†Significantly different from corresponding values for control blocks (p 1.30 gave similar results. Block-level BIs were less accurate. Comparable cutoff points for the indices in the September to October inspection cycle discriminate best for predicting transmission in October (data not shown). After the epidemic, in the November to December inspection cycle, the indices had a high specificity: 89.6% for BI 4 was a significant predictor for identifying blocks with a case in September (OR 6.00, p 0 September transmission 2.57 (0.57–11.70) October transmission 1.69 (0.58–4.94) BI per neighborhood >1 September transmission 3.00 (0.66–14.17) October transmission 1.08 (0.40–2.90) BImax>4 September transmission 6.00 (1.09–32.98)‡ October transmission 1.21 (0.45–3.25) September–October 2000 inspection cycle (during epidemic) BI per block >0 October transmission 3.49 (1.20–10.10)‡ BI per neighborhood >1 October transmission 5.06 (1.46–17.38)‡ BImax>4 October transmission 3.44 (1.23–9.63)‡ *OR, odds ratio; BI, Breteau index; CI, confidence interval; BImax, maximum BI at the block level for each neighborhood.
†Optimal cutoff value determined as specified in Methods.
‡p 3% ( 31 ). Recently, Scott and Morrison ( 16 ) showed that traditional larval indices in Peru are correlated with the prevalence of human dengue infections. The variety of thresholds proposed in these and other studies could be partially explained by different methods and geographic levels of analysis used, but other factors influence the relationship between Aedes density and transmission risk, such as herd immunity ( 11 ), population density ( 31 ), mosquito-human interaction ( 34 ), virus strain, and climate, which affects mosquito biology and mosquito-virus interactions ( 16 ). Entomologic indices, however, were strongly associated with transmission, and we used ROC analysis ( 24 ) to assess the potential of these indices to predict in which blocks transmission would occur and to select an operating point that would provide an optimum tradeoff between false-positive and false-negative results ( 35 ). BImax>4 followed by neighborhood BI≈1 during the preceding ≈2 months provides good predictive discrimination. At longer intervals, the sensitivity of these indices becomes too low. More frequent inspection cycles might perform better since Aedes needs only 9–12 days to develop from egg to adult ( 36 ). Care should, however, be taken when extrapolating these findings to communities with other herd immunity levels or different environmental conditions. Our data also show that the geographic level of analysis determines the Aedes indices obtained. Marked heterogeneity is not only found inside Playa Municipality but also inside smaller health areas. Indices at the neighborhood level perform best, followed by indices at the block level. Geographic scale has too often been neglected when dengue transmission is studied. In general, overall indices are calculated for communities (sometimes of different sizes) defined by administrative boundaries, which do not constitute entomologically homogeneous units. Notwithstanding, local variability of larval indices can be inferred from the literature, in which it is sometimes mentioned. Chan et al. ( 27 ) noted that HI in different sections of Singapore's Chinatown varied from 10.2% to 25.0%. However, Goh et al. ( 30 ) reported an overall HI of 2.4% in Singapore, but at the level of 7 blocks taken together (approximately the same scale as our neighborhood), HI up to 17.9% were found. Tran et al. ( 36 ) defined 400 m and 40 days as the spatial and temporal boundaries of maximum dengue transmission in a dengue focus. Perez et al. ( 37 ) identified areas in Havana with heterogeneous risks for vector infestation by using a geographic information system. Spatial heterogeneity has also been observed at the household level for both Aedes populations ( 10 , 38 , 39 ) and dengue transmission ( 26 , 29 , 40 ), but this level seems less suitable for identifying areas for intervention. Blocks or neighborhoods, given the epidemiologic situation in our study area, are a more appropriate scale. The unit of analysis used in our study, the block, is based on manmade boundaries. While these may not describe the ecology of risk, they seem to be useful markers from the perspective of community-based control interventions. In most settings, appropriately sized and locally meaningful geographic units could be similarly defined for entomologic surveillance, but the use of different boundaries or different analytical techniques could produce different results. In our study, BI>1 and BImax>4 seemed to be a suitable action threshold and target, respectively, in community based dengue prevention. However, these results are derived from the analysis of 1 epidemic, and the thresholds identified may not constitute suitable targets in another epidemic or in locations where different ecologic conditions prevail. Similar studies in future epidemics and in other settings are necessary to verify the general applicability of our results.