In countries of sub-Saharan Africa, resources are often constrained. Assessment of differential risk within epidemic-prone highland areas is therefore important for targeting interventions. Studies of malaria in highland areas have generally concentrated on broad assessment of large areas, in an effort to develop much-needed early warning systems for malaria in these areas [12–15]. Such studies typically examined temporal trends and were not spatially referenced at a local scale. Other studies on spatial variation of malaria incidence in highland areas were conducted either over a large area, with relatively coarse spatial resolution , or on a finer scale, but over only a single season of malaria transmission . Literature searches did not reveal studies that have assessed whether finer scale spatial patterns of malaria incidence persist over time, and whether specific environmental factors are consistently associated with malaria risk. The present study addresses this gap in knowledge and demonstrates that in at least one highland area, there are persistent "hotspots" of increased malaria incidence and consistent ecological risk factors associated with increased risk. These findings suggest that focused interventions in the high-risk areas of such sites have the potential to significantly reduce malaria risk in the area overall.
Variation in malaria incidence within a highland area of this size (~16 km2) has not previously been described. The magnitude of the differences (15-fold to 39-fold differences in incidence between the subunit areas of highest and lowest incidence) was striking and consistent, suggesting that even within this small area, risk of malaria ranges from very low to quite high. The high attack rates in the areas of highest incidence (672 cases/1,000 persons/year in the subunit area of highest incidence during the epidemic year) demonstrate that even in areas of significant elevation (all households were at >1,900 m), high levels of malaria transmission may occur.
The four ecological factors independently associated with increased malaria risk in this study – lower altitude, proximity to swamp, proximity to forest, number in household and presence of a metal roof – are all biologically plausible contributors to increased malaria risk in this community. Moreover, the associations of these risk factors with malaria incidence were strong, consistent, similar over the years, and highly significant. In Kipsamoite, the range of altitude for areas on which households are built is only 150 m, but altitude remained a significant independent correlate of malaria risk. Lower altitude within a highland area has been described in several studies as a risk factor for malaria [3, 17]. The lower incidence of malaria at higher altitudes is most likely due to the decreased temperatures at these altitudes, and the adverse effects of decreased temperature on the Anopheline vector [18–20] and on development of P. falciparum within the vector . Proximity to forest and swamp have both been associated with increased vector density. Vector density has been shown to cluster in low-lying flat swampy areas . Much of the swamp land in Kipsamoite has been reclaimed for agricultural use, and reclaimed swamps and swamp cultivation have been associated with increased vector density [4, 22]. Proximity to forest was also a risk factor in the highland Kenya study by Brooker et al , and may reflect the consequences of scattered deforestation. Clearing of trees in the forest surrounding Kipsamoite has small soil depressions that retain water and could serve as breeding sites. Such deforestation may also facilitate survival of immature stages of A. gambiae in highland areas . Higher numbers of occupants may increase the attractiveness to Anopheles spp. and has been associated with higher Anopheles spp. densities . The association of metal roofing with increased malaria risk during the epidemic year is contrary to findings from studies in lowland areas . Preliminary evidence from a case-control study conducted in 2004 indicates houses with metal roofing also tend to have open eaves (perhaps due to the low level of nuisance insects) and separate kitchens as compared to those with thatched roofing (Ernst K, unpublished data). Open eaves allow easy accessibility for mosquitoes, and smoke from a kitchen fire may be a deterrent to the vector. Further exploration of this finding will be made in the analysis of case-control data. Fewer malaria cases during the non-epidemic years may have affected the power to detect this association in those years. Alternatively, household level risk factors such as housing construction may be of greater importance during an epidemic, when ecological conditions favor a wider distribution of the vector population. Spatial analysis of a cohort of ~250 individuals from randomly selected households who were under active surveillance for malaria in 2003 and 2004 demonstrated the same malaria hotspot area, suggesting that numbers from passive surveillance provided accurate information for this study.
There was no spatial autocorrelation in the residuals of the individual-level or household-level models. This indicates that our predictors explain the global level clustering seen in the data. However, there were some households at lower elevations in close proximity to swamp and forest that were malaria free even during the epidemic in 2002. Additional factors likely contribute to malaria risk in the study area. More detailed ecology and entomology studies in this area are underway, to assess for additional or alternative risk factors. The present study was based on passive surveillance of cases and controls. Although active surveillance tends to generate higher incidence figures, it is very difficult to do risk factor analysis in an active surveillance cohort in a highland area, for the simple reason that there are usually too few cases in individuals in the cohort to provide adequate sample size for the study of rsk factors. For this reason, passive surveillance, though it may miss some cases and possibly mislabel others as controls, is the most practical method of conducting such analyses. In addition, our group's findings in this area demonstrate a very strong correlation between active and passive surveillance incidence figures, suggesting that passive surveillance does provide an accurate reflection of true incidence in the community.
It is unclear what initiates the foci of transmission. It may be that people "seed" transmission each season. Previous studies in this area by our group did not document a large reservoir of asymptomatically infected individuals in this study site during periods of low transmission . However, re-analysis of these data demonstrated that asymptomatic parasitaemia was disproportionately frequent among individuals living in the two villages, which cover most of the hotspot area (John CC, unpublished data). Examination of cases identified in the very low season of transmission (September – November) also showed clustering of cases in the same area. These findings suggest that sustained transmission may occur even during the dry season in this area, and further investigations are planned. this possibility. It is also possible that residents traveling to high transmission areas introduce infection [26–28]. However, very little travel outside the study area was documented. Transport of infected vectors by vehicles arriving in the area is an additional possibility, though this is difficult to establish.