The targeted area is located in Iguhu village (0°17'N, 34°74'E, 1420-1580 m above sea level), Kakamega district, western Kenya. The detailed description of the study area has been included in an earlier study 23 . Briefly, the study area is 6.0 km by 4.5 km and is nearly equally bisected by the Yala river (Figure 1). The population is around 32,000 with 6,060 households (identified from 1 m Ikonos colour image). The targeted intervention area is within 500 m on both sides of the river with a length of 3.5 km from west to east (Figure 1). The study area was split into an intervention zone and a nonintervention zone that are separated by a 1 km buffer to ensure minimal adult mosquito dispersal between sites. For convenience, the actual sprayed valley area in the intervention zone is defined as "intervention valley" and the rest of the intervention zone is named as "intervention uphill", the corresponding areas in the nonintervention zone are defined as "nonintervention valley" and "nonintervention uphill" (Figure 1).

All houses in the targeted intervention valley area were sprayed with lambda-cyhalothrin (ICON), which is among the insecticides recommended by the World Health Organization and National Malaria Control Board of Kenya. All houses in the intervention valley (total of 1,100 houses) were identified and numbered. The interior walls and roofs of the targeted houses were sprayed during the last week of April 2005. This period was prior to the long rainy season of May that triggers a high density of malaria vectors and the start of the peak malaria transmission period in this highland area. There was no other IRS or systematic use of ITNs existing in the study area. Field surveys in 2004 showed that household bed net ownership was about 13% (49/382) and bed net coverage was about 5% (98/1958) of the surveyed population (assuming that one bed net covers two individuals). No other bed net data was available for the study area from 2004 until the mass distribution of free bed nets and free artemisinin combination therapy (ACT) for children under five years of age in the study area commenced in July-September 2006.

The sample size for the baseline survey was estimated in advance to have 90% power to detect a 10% reduction in malaria prevalence in children aged between one and five years, assuming a loss to follow-up of 15% during the 16 week survey, a two-sided type I error probability of 0.05, and a 1:1 ratio of intervention to control (Table 1).

A cross-sectional survey was conducted during the third week of April, 2005, before the IRS, to obtain the baseline malaria prevalence data. Finger prick blood samples were taken from 1,058 randomly selected children aged 1-5 yrs for detection of the Plasmodium falciparum antigen (Pf HRP-2) using the Optimal (DiaMed AG, Switzerland) rapid whole blood immunochromatographic test. Children found to be positive for malaria infection were given free anti-malarial treatment according to Kenyan guidelines. A cohort malaria incidence study was started after 14 days of IRS and a blood sample test was performed using the same Optimal rapid test kit every other week until week 16.

A monthly parasitological survey was carried out in the same area from May, 2004 to March 2006 with an average of 386 (range from 129 to 661) school children aged 6 to 14 yrs randomly surveyed each month. The detailed blood sample collection method including the blood smear preparation, parasite density determination and quality controls can be found in Munyekenye et al 20 . The number of children surveyed each month is shown in table 1.

Children were recruited into the study only with the informed consent of their parents or guardians. Scientific and ethical clearance was given by the Kenya Medical Research Institute (KEMRI) and State University of New York at Buffalo ethical review boards.

Only children under 14 yrs were included in this study because this group has been shown to be the most vulnerable to malaria parasite infection 20 .

The required mosquito adult sample size has been estimated by an earlier study 23 . Indoor resting mosquitoes were collected from 300 houses in the study area by pyrethrum spray collection. Houses were selected to ensure maximal spatial coverage. The number of Anopheles mosquito females was recorded in the intervention zone and control zone. The indoor resting density was determined as the average number of Anopheles females collected for each house. The timings of four mosquito surveys were selected to represent the vector population during the different seasons both before IRS (May, August and November 2004 and February 2005) and after IRS (May, August, and November 2005 and February 2006). Sample sizes of each strata are shown in table 1.

Anophelines were morphologically identified and classified as Anopheles gambiae and Anopheles funestus using morphological keys 25 . Since earlier studies had shown that all An. gambiae sensu lato (s.l.) and An. funestus s.l. specimens were identified as An. gambiae sensu stricto (s.s.) and An. funestus s.s. by using the rDNA-polymerase chain reaction (PCR) method 25 26 , no further classification work was conducted.

Changes in seasonal mean vector densities and monthly parasite prevalence before and after intervention were tested by the use of planned comparison of GLM, using location (valley and uphill) and survey period (before and after intervention) as independent variables and using outcomes from the corresponding nonintervention area as contrast for the dependent variable. The differences in biweekly new infections and parasite prevalence between intervention and nonintervention valleys (or uphill) were tested by using Mantel-Cox test 26 . STATISTICA version 8.0 (StatSoft, Tulsa, OK) was used for data analysis.

Relative risk of Plasmodium parasite infection (RR) during the cohort study is defined as the ratio of Plasmodium parasite prevalence in the intervention valley (or uphill) over the parasite prevalence in the nonintervention valley (or uphill). The protective effectiveness of IRS in reducing malaria infection is measured as (1-RR) and the asymptotic confidence intervals were calculated as RR(1 ± z_α/2 ), where z_α/2 is the upper α/2 percentile of the standard normal distribution and u is the variance of the ratio 27 .

To assess changes in mosquito densities and Plasmodium parasite prevalence associated with the implementation of IRS, the percentage reductions (PR) in parasite prevalence and mosquito densities in targeted intervention valley and uphill areas was calculated using the following formula:

where C ₁ and C ₂ (T ₁ and T ₂) describe the average densities of mosquitoes or parasite prevalence in the control (targeted) zone during baseline (subscript 1) and intervention (subscript 2) periods. This formula takes into account that changes in the mosquito populations and parasite prevalence are taking place at the same level and rate in both targeted and control zones, i.e., the reductions were adjusted for the background differences.

During the last week of April 2005, 1,058 children were recruited for a baseline malaria prevalence test and 1,033 of them were included in the final data analysis. The remaining children have at least one missing record and were not used for the cohort analysis. Among those who were included in the cohort study, 113 were from the intervention valley (Table 1) with 83 (73.5%) P. falciparum positive slides and 112 from the corresponding non-intervention valley with 49 (43.8%) parasite positive slides (Yates χ² = 12.5, d.f. = 1, P < 0.001). The relative risk was 1.5 (95% CI [1.2, 2.0]) and the odds ratio was 2.7 (95% CI [1.6, 4.5]). Similarly, the parasite prevalence was 61.7% (95/154) in the intervention uphill and it was significantly higher than the prevalence of 31.4% (77/249) in the nonintervention uphill (Yates χ² = 34.1, df = 1, P < 0.0001). The relative risk was 2.0 (95%CI [1.6, 2.5]) and odds ratio was 3.5 (95%CI [2.3, 5.4]).

Among parasites detected, P. falciparum accounted for 97.2% of positive samples, Plasmodium malariae comprised 3.9% and Plasmodium ovale less than 2%. There were a few cases of mixed infections of two parasite species. The longitudinal follow up study shows that malaria parasite prevalence in children in both the intervention and nonintervention areas have decreased since May 2005 (Figure 2, Table 2), but the decrease was much higher in the targeted intervention area than in the nonintervention area. Parasite prevalence in the intervention valley dropped significantly from 63.6% before intervention to 16.4% after the intervention (GLM planned comparison, F _1,24 = 309.29, P < 0.0001), the adjusted relative reduction in parasite prevalence was 65.4%. The relative risk of malaria parasite infection dropped in the intervention valley from an average of 1.7 (range from 1.2 to 2.8) before the intervention to 0.6 (range from 0.4 to 0.8) after the intervention. Concordantly, parasite prevalence in the intervention uphill area decreased from 44.9% before intervention to 18.5% after intervention (Figure 2, Table 2), representing an adjusted reduction of 46.4%, and relative risk of parasite infection decreased from an average of 1.9 (range from 1.4 to 3.0) to 1.0 (range from 0.6 to 1.4).

Figure 3 shows that baseline parasite prevalence (week 0) and cumulative incidence rate in the subsequent biweekly surveys during the 14 weeks cohort study period. Parasite infections rebound significantly faster in the nonintervention valley than that in the intervention valley (Figure 3) (Mantel-Cox test, Z = 4.94, P < 0.001). On average, there were 4.5% fewer new infections (range from 1.8% to 9.9%) in the intervention valley than in the nonintervention valley. By week 14, the cumulative incidence rate reached 40.7% and 72.3% in the intervention valley and nonintervention valley, respectively, compared to the baseline prevalence of 73.4% and 43.8% in the corresponding areas. The adjusted reduction by week 14 was 65.4%.

While the cumulative incidence rate was significantly higher in the intervention uphill than nonintervention uphill (Figure 3) (Mantel-Cox test, Z = 7.30, P < 0.001), the average difference in biweekly new infection rates was insignificant (0.9%, range from -3.7% to 5.1%). By week 14, the cumulative incidence rate was 50.0% in the intervention uphill area, which was significantly lower than the baseline prevalence of 61.7%, whereas cumulative incidence rate was 46.9% in the nonintervention uphill area by week 14 compared to a baseline prevalence of 31.4%. A resulting adjusted reduction of 46.4% was calculated for the cumulative incidence rate in the intervention uphill.

Anopheles gambiae density in the intervention valley was reduced more than ten-fold from an average of 3.4 female/house/night (f/h/n) before intervention to 0.2 f/h/n after the intervention (Figure 4) (GLM planned comparison, F _1,14 = 7.63, P = 0.02), which was a 96.8% relative reduction (Table 2). The adjusted reduction in An. gambiae densities in the intervention uphill area was 51.6% (Table 2). Average density of An. funestus in the intervention valley was also significantly reduced from 0.78 f/h/n before intervention to 0.06 f/h/n after the intervention (Table 2) (GLM planned comparison, F _1,14 = 9.16, P = 0.01), and the adjusted reduction was 85.3%. The relative reduction in An. funestus densities in the intervention uphill area was 69.2% (Table 2).