The maps and model terms presented in this paper provide important insight into the factors that contribute to increased risk for malaria in certain populations and regions of the DRC. Malaria risk was found to vary geographically and to be dependent on a variety of individual-level and community-level variables. As expected, living further from a town was associated with higher rates of malaria prevalence, indicating parasitaemia is generally endemic to more rural areas. Malaria risk also decreased with increasing altitude [[31–33]]. Younger males were found to have the highest risk of malaria, possibly due to occupational exposures or decreased use of health care services . While negative associations with malaria parasitaemia were found between rainfall and temperature, these variables were not significant after controlling for other factors.
Several factors were protective at the community level but not the individual level. Notably, living in a wealthier community more greatly decreases one's odds of having malaria than individual wealth, suggesting that less impoverished people living in impoverished neighbourhoods are still at increased risk. While individual bed net usage was significant when entered alone into the model, it was no longer significant when community bed net ownership was entered, indicating multicollinearity. Inclusion of the community bed net ownership variable provided better model fit and was thus retained. Therefore, while individual bed net ownership is important, community ownership was a stronger predictor of parasitaemia, indicating that herd immunity may be occurring within communities. Community effects on disease transmission have been reported for a variety of diseases and settings, especially for infectious diseases [[55–58]].
Surprisingly, the use of untreated nets was negatively associated with parasitaemia while the use of treated nets was not. While there is evidence that insecticide-treated nets are more protective against malaria , this finding is likely attributable to the fact that roughly twice the number of respondents slept under untreated nets as compared to treated nets (see Table 1). This is because the survey was not originally intended as a malaria indicator survey, and thus equal sampling of those with treated versus untreated nets was not a primary aim of the DHS.
Most notably, the level of conflict since 1994 occurring within 100 km of one's community was negatively associated with an individual's malaria risk in the majority of the DRC. This relationship persisted even when areas having the greatest potential to contribute to confounding (the Kivu provinces) were excluded. The GWR analysis indicated that the negative direction of the relationship was in fact stationary across the entire country.
While the relationship between conflict and infectious disease has been explored in past research [59, 60], to date no studies have compared localized density of conflict with malaria parasitaemia. The inclusion of conflict variables in the models was intended to determine the outcome of parasitemia in places that have long been characterized by conflict. The nature of these places indeed differ from those areas little-affected by conflict, and this study has shown such differences to be relevant to the understanding of the drivers of parasitemia in the DRC. The inclusion of conflict density in these models is not, however, intended to indicate a direct causal relationship between specific past battle events and individual malaria parasitemia. Possible explanations for this observed association include population migration away from and increased humanitarian efforts in places having experienced large amounts of fighting. Displacement from rural areas due to conflict may lead to less dense human host populations for malaria transmission in zones of insecurity . The focus of humanitarian efforts in war-affected regions on preventing and treating malaria  may also underlie the findings, although further knowledge of the geography and practices of humanitarian agencies in the DRC would be necessary to support this premise.
This study has several limitations. Because the presence of clinical symptoms is unknown in this study, our maps highlight where prevention may be most effective but not necessarily where treatment is most needed. Lack of blood sampling from children is also an important limitation to consider, as they suffer the greatest risk for illness and death from malaria. If children had been included in this study, malaria prevalence values might be expected to change and the significance and magnitude of parameter estimates in our models may have been different. There has only been one published report looking at age stratification of PCR-positive malaria, and the difference between adults and children was minor . Data were also limited in that individuals could only be located to the centre of their community, and not to their actual place of residence. Furthermore, mobility and migration characteristics are unknown for the individuals in our dataset. Consequently, several of the geographic variables we computed (distance to a road, water body, city or town, agricultural land cover, and altitude) may lack precision and could affect the terms of these variables in the multivariate models. Finally, with the exception of Kinshasa, the study was conducted during the dry season, so the annual peak prevalence may be higher than reported here.