The annual incidence of dengue in control cities varied more widely than in treatment cities, and the median annual incidence in treatment cities was generally higher (Figure 2, panel A). However, there was a trend of decreased difference in incidence between annual incidence in treatment cities relative to control cities during the years (2010–2011) in which MID was used during the peak dengue season (January–May or June) (Figure 2, panel A). This trend was confirmed by the finding that the RI before and after the time frame of MID was 2.7× higher (a decrease from 4.0 in control cities to 1.3 in treatment cities; 68%), in control cities relative to those that used MID (z = −31.59, p<0.0001) (Figure 2, panel D). The RI for treatment cities for each population group is shown in Figure 4.

The most parsimonious generalized linear model of RI in MID cities included PS and IDFM (Technical Appendix 2 Table 2). PS and IDFM showed a negative correlation with RI, although the correlation of PS was marginally not significant (p = 0.083 for PS and p = 0.0023 for IDFM) (Technical Appendix 2 Table 3). This finding indicates that MID effectiveness was higher in cities with stronger economies and that there is a trend of higher effectiveness in larger populations. In contrast, the most parsimonious model of cost-effectiveness included IMFA and D3L (Technical Appendix 2 Table 2). IMFA and D3L showed a negative correlation with cost-effectiveness (p = 0.0086 and p = 0.032, respectively) (Technical Appendix 2 Table 3). Thus, cost-effectiveness was higher in cities with higher mosquito infestation levels and cities that were farther from cities with large populations.

Under the assumption that dengue could affect 30% of a population, we estimated that the number of cases prevented by MID annually in the largest cities (>130,000 inhabitants) was 2,300–3,900 (Figure 5). In the smallest cities (<40,000 inhabitants), these estimates decreased to 143–182, and the total number in all 21 cities was 27,191. However, these numbers depend on the assumed number of potentially susceptible persons (Figure 3). The average cost-effectiveness was $227/case prevented, which was driven mainly by a few larger values (Figure 6, panel A). The median value was $58, indicating that the average value was higher than the cost-effectiveness value in most cities. The number of cases prevented translated to net total savings of $8,999,406 annually. Savings in health care and vector control costs was $364,517, and savings in lost wages was $7,138,940 (Figure 6, panel B; Technical Appendix 2 Table 4).

Accurate estimates of dengue incidence and its economic effects are more limited (16,19) than are estimates of other infectious diseases that pose similarly serious public health threats. This finding is caused mainly by high variability in clinical disease, high underreporting rates, and lack of studies that directly measure the efficacy of controls. Consequently, only a few studies have demonstrated the cost-effectiveness of vector control activities (19–21). In one of these studies, targeted source reduction was more effective than nontargeted vector control, reducing vector abundance by 52%–82% depending on the country (21). Another study found that targeted vector control reduced the dengue case load by 53% (20). Our estimate of a 68% reduction in incidence caused by targeted control efforts by MID was higher than that in the study by Suaya et al. (20). One reason may be geographic differences in the effects of source reduction methods (the previous study was conducted in Cambodia). Alternatively, our higher estimate may be caused by implementation of MID at a fine spatial scale over a broader area, which produced higher intervention efficacy.

The trend of increased effectiveness in larger populations might not be significant in the multivariable model (which includes IDFM) because PS and IDFM showed a positive correlation (r = 0.53). The single-variable model results suggest that MID may be more effective in larger populations (Technical Appendix 2 Table 1). The fact that MID was more cost-effective in cities with higher mosquito infestation levels emphasizes the power of targeting vector control practices to areas in which gravid mosquitoes are most abundant. A possible reason for higher cost-effectiveness in cities that were farther from large cities could be that proximity to larger cities may enable a higher proportion of cases that were contracted elsewhere (i.e., during travel or commuting to large metropolitan areas) (22,23).

PED, a measure of MID quality, was not correlated with effectiveness or cost-effectiveness of MID. This result suggests that variation in the force of infection between cities overwhelmed differences in PED, the current measure of PED is inaccurate, or both. The relationship between mosquito infestation and human incidence is highly variable in space and time (24–26). Studies of dengue virus serotype circulation in Brazil have found dominance of a single serotype during any given year, and different genotypes within the serotypic groups have caused large, severe outbreaks because of reduced population immunity (5,27–29). Thus, variations between cities in novel genotype dynamics might have affected variation in RI because of PED. In addition, PED pertains only to MID activities and does not evaluate control activities. It is likely that the quality with which cities conduct prescribed control practices varies, which could also explain the lack of relationship between PED and RI. Furthermore, PED is assessed by a yes or no checklist for MID activities, rather than by an evaluation of the quality of each activity. If only minor activities constitute most of the variation in PED, then little meaningful variation in MID quality between cities might be observed. Regardless, the lack of relationship between PED and efficacy of MID suggests that an additional method for assessing MID quality, perhaps through collaboration with city control personnel, might be useful for maintaining, standardizing, and improving quality.

One caveat to our method of estimating cost-effectiveness is that it was necessary to estimate the number of susceptible hosts in the absence of MID (we used 30%) to predict the number of cases that would be prevented. Although 30% is not unreasonable based on previous studies (16), the maximum incidence observed in a given city in Minas Gerais during 2007–2010 was only 8.8%. This discrepancy was partly caused by underreporting, which was not accounted for in our study. Nevertheless, we also provided predictions for lower values of K to understand how it could affect our estimates. When fewer hosts are susceptible, the number of cases prevented is also lower, which decreases the cost-effectiveness of MID. Thus, previous large outbreaks with the same serotype and vaccination programs would be expected to decrease the cost-effectiveness of MID. Another caveat to our assumption is that K varies by city because of historical disease patterns and other factors. Collection of longitudinal serologic data would be useful for more accurate, city-specific predictions of K. Last, our study used previously estimated costs for control activities, health care, and lost wages. These costs were per capita estimates that we assumed could be extrapolated to each city equally. The accuracy of our results could be improved through microcosting analyses within each city.

Although MID showed an average cost-effectiveness value of $227 (median $58) per case prevented in Minas Gerais, the average value increased to $616 in 6 moderately sized cities (population 73,000–117,000) that did not show any savings in direct costs (Technical Appendix 2 Table 4). Three of the cities saved on indirect costs and total costs, but the 3 other cities (Joao Monlevade, Itabira, and Conselheiro Lafeite) had a net loss of up to $81,042 in direct costs and $66,246 in indirect costs because of incorporating MID into their budgets. These 3 cities had relatively low annual dengue incidence in the 2 years before MID implementation (12, 18, and 72 cases per 100,000 population relative to a range of 104–2,014 cases in the other 18 cities except for Paracatu, which had 4 cases). Thus, in general, cities with annual incidences of >72 cases per year were more likely to have higher MID cost-effectiveness.

Furthermore, cities in which MID was implemented had historically high dengue incidences relative to control cities (mean ± SD 2007 and 2008 were 549.1 ± 592 in MID cities and 240.4 ± 567.6 in control cities). Thus, average estimates of cost-effectiveness may be high in cities in which MID was implemented. However, factors determining incidence patterns in a given city, such as population immunity, infrastructure, or human behavior, may not be static over time because high population immunity is not protective against novel serotypes (or genotypes with high forces of infection) and human behavior and infrastructure are continually changing. A predictive model of serotype dynamics across cities formulated on the basis of serologic data would be useful for decisions on which cities should implement MID so that the most cost-effective strategy can be achieved statewide.

Our study showed that MID is generally effective for decreasing case loads and suggested that an MID strategy is theoretically better than other strategies. Although MID cost-effectiveness varied between cities, implementation of MID saved hundreds of thousands of dollars on health care and ≈7 million dollars in lost wages statewide, and half the cities had cost-effectiveness values <$58. Furthermore, these numbers are underestimates because our study did not account for underreporting or additional costs from deaths. Investing more effort into integrating MID strategies and costs with vector control operations, and standardizing the MID-based control system across cities, should help to increase MID cost-effectiveness.

Monitoramento Inteligente da Dengue (Intelligent Dengue Monitoring System), Minas Gerais, Brazil.