Malaria is one of the leading causes of morbidity and mortality in the world. Indeed, more than 2.4 billion people are exposed to the risk of malaria [1]. The incidence of malaria worldwide is estimated at 300 to 500 million cases per year, with 90% of these cases occurring in sub-saharan Africa. Malaria kills between 1.1 and 2.7 million people per year, including almost one million children under the age of five years in Sub-Saharan Africa [1, 2]. The impact of this disease, not only in terms of mortality and morbidity, but also in terms of economic and social losses, led the United Nations to make the fight against malaria one of the priorities of its Special Initiative on Africa. The persistence of malaria despite the many control programs is due partly to the very high costs of monitoring, which are particularly difficult for developing countries to bear [3, 4]. In places, the relaxation of monitoring measures has even led to increases in disease levels. The methods of control recommended by the WHO [1] are based on chemical and physicochemical control of the vector (insecticide or larvicide applications, use of mosquito nets impregnated with insect repellent), environmental modification (e.g. draining of backwaters), chemical prophylaxis (essentially in pregnant women and travelers) and the early detection, containment and prevention of epidemics. In addition, major progress in research has led to the development of several candidate vaccines, which are currently in clinical trials [5–7]. However, these control methods are expensive and therefore cannot be implemented on a large scale and in a sustained fashion in the economic context of developing countries [8]. In addition, the large-scale use of anti-vectorial measures and anti-malarial prophylaxis may lead to resistance or adaptations in the vector and in the parasite. The setting up of anti-malaria actions targeting specific zones is therefore a priority. Indeed, since 1984 the WHO has recommended control measures integrated into primary healthcare, favoring local involvement [9]. Anti-malaria actions, whether involving prevention, treatment or epidemiological information, are based at local level. Taking into account the complexity of malaria, precise research studies are needed, such as research on physiopathology, immunology or genetic susceptibility to malaria or such as intervention trials to evaluate treatments or prophylactic measures (drug, vaccine, anti-vectorial devices), in order to improve our understanding of the disease and the control [10]. Particularly, sites for malaria vaccine field trials must be precisely prepared [11]. This requires a precise knowledge of the geographic zones at risk, the levels of risk, the various risk factors and the exposed populations. The highly focal nature of malaria epidemics results in marked heterogeneity, even at the scale of a village [12]. The risk of Plasmodium falciparum infection is highly variable over space and time [13]. An analysis of the local epidemiological situation is therefore essential and such analyses formed one of the priorities of the 18^th WHO Report [3], reiterated in the 20^th WHO Report [1]. The WHO recommends the stratification of malaria risk. This involves an analysis of local variations, making it possible to define high-risk zones on a fine geographical scale, with the aim of increasing the efficacy of anti-malaria measures [14].

The development of geographical information systems (GIS) has been an indispensable asset to this approach [12]. Together with the progress of statistical methods for spatial analysis, GIS have improved studies for the detection of clusters at high risk of diseases over space and time. However, despite the increasing number of studies reporting on temporal or spatial changes in malaria risk, few studies have analyzed this risk at a fine scale (below district level) [4, 12].

Research studies on malaria disease and intervention trials, such as vaccine trials, can be improved by such a precise epidemiological modeling. Before initiating such studies it is necessary to define local patterns and predictors of malaria transmission and infection (in time and in space). This will facilitate the selection of the appropriate study population, the intervention allocation and will enhance the accuracy and the efficiency of the analysis describing the impacts of the studied interventions.

Therefore, this study aimed to identify malaria risk at household level (resolution of 1 to 3 m), and to evaluate changes in this risk over time, in a hyperendemic village in Mali. These efforts to identify high-risk zones in space and time were designed to make it possible to identify the population at risk and local risk factors, in order to plan vaccine trials.

By identifying high-risk zones of malaria, this study made it possibly to stratify local risk temporally and spatially, as recommended by the WHO [1, 2]. Although the entire region is classified as a high-risk zone for malaria (MARA prevalence estimation = 62.27%; 95%CI [56.37%;68.18%]) [31], the identification of clusters demonstrates the high variability of malaria risk over space and time in this village. The use of a GIS made it possible to analyze these variations precisely, at the level of households (resolution of 1 to 3 m), improving our knowledge of the disease in this village, thereby facilitating its control and its understanding, above all to plan anti-malaria intervention trials.

The time series of P. falciparum incidences are consistent with the well-known seasonality of infection (strongly linked to the rainy season), with marked regularity. Indeed, incidence peaked in October or September. We noted a persistence of high incidence into the start of 2000, due to the rains occurring in January 2000. Overall, the re-infection with P. falciparum peaked at a maximum of almost 70% (95%CI [68.1%-73.3%]) of the children studied (October 1996).

The incidence of carriage of P. falciparum gametocytes changed in a much less regular manner over time, notably, peaks in February and December in 1998. The peak in August 1999 was very large, exceeding the upper bound of the 95%CI. No such abrupt change was seen in changes in the P. falciparum incidence. There is presumably a link between this peak in the gametocyte carriage prevalence and the lengthening of the epidemic period in 1999.

A tendency towards decreasing P. falciparum incidence has been reported in other studies at the same site [15, 16]. This tendency is unlikely to be due to natural changes in the frequency of P. falciparum in the region. It is also unlikely to be due to changes in the village, particularly as the proportion of dwellings with thatched roofs remained constant (about 47%). Similarly, this tendency almost certainly does not result from changes in the number of children included over time as the number of children included was already large at the start of the study and this infection is hyperendemic in this region. The tendency of P. falciparum re-infection to decrease is probably linked to the presence of the medical team in a population already highly aware of the problem of malaria, and also to the treatment of all infected children. Correct usage of chloroquine as the first line drug for malaria treatment has reduced significantly the self medication in the village of Bancoumana. The proportion of malaria self medication went from 6.5% in 1997 to 3.8% in 1998, 3.7% in 1999, and 0.8% in 2000 [18]. This was able to reduce chloroquino-resistant malaria parasites at the study site of Bancoumana. By contrast, we observed much more erratic changes in incidence with P. malariae and P. ovale, not consistent with seasonal transmission.

This pattern, although similar to pattern obtained for other geographic locations [4], describe an average over the entire area studied and does not take into account the geographic heterogeneity that exists, even at the small scale of a village. At household level, we can therefore call into question the globally seasonal pattern of transmission with a tendency towards decreasing incidence.

The transmission of P. falciparum is linked to local factors that must be identified before initiating control programs or research studies. The change in clusters over space and time is presumably linked to spatial and temporal changes in local factors, such as temporary backwaters in particular. Note that cluster 5 for P. falciparum infection is located on a recent site of adobe brick production. This process involves the removal of earth for the production of bricks by local craftsmen and the resulting excavations create breeding sites for mosquitoes.

These results showed that the analysis of mean changes over time at the level of an entire village is of too low a resolution whereas the search for high-risk clusters makes it possible to find a suitable interpretation for spatial and temporal changes in P. falciparum infection.

We observed proximity in time and space between clusters of P. falciparum infections and of gametocyte carriage. These observations should alert epidemiologists in the field to the existence of this zone of particularly high risk. Similarly, the extreme proximity of the two clusters of gametocyte carriage also indicates particularly high risk of transmission to the Anopheles mosquitoes. The first of the two clusters of P. malariae infections occurred close in space and time to clusters of P. falciparum infection, indicating the existence of local risk factors common to these two types of infection such as breeding sites for mosquitoes. The presence of space-time clusters of P. malariae infection should serve as an additional alarm signal.

The detection of clusters with a high risk of infection extending over several rainy seasons suggests that if that cluster had been identified when it first appeared, it might have been possible to control the risk by means of surveys on the ground to identify risk factors and the implementation of targeted control measures for this geographic zone. Although some publications have reported epidemiological analyses at district level, few have considered finer scale analyses [32–37] and only rarely has a spatial or spatio-temporal statistical model been used [12, 38–40].

In this work, we studied three endemic species of Plasmodium. The relationships between these species are complex [41–43], particularly as P. falciparum is largely predominant in Mali. However, the environmental risk factors for infections with these species are largely similar. Thus, mapping the risk of infection with P. ovale or P. malariae also provides useful information concerning the risk of infection with P. falciparum [44]. Furthermore, space-time analysis of these different species of Plasmodium could improve the understanding of their relationships. Similarly, an analysis of blood infection with P. falciparum gametocytes provides an indication of spatial and temporal variations in malaria transmission and improves the understanding of the transmission process.

Kulldorff 's permutation model was chosen because it has several advantages: it uses only the number of cases and their localization, with no need for population at risk data; it adjusts for confounding variables; there is no pre-selection bias since the clusters are searched with no prior hypothesis on their location, size or time period; the test statistic takes into account multiple testing and delivers a single p-value [29]. Unfortunately, it is not possible to estimate confidence intervals for the rate ratios of clusters detected by scan statistics, because of the multiple testing part of the many circles evaluated.

The environmental measures recommended by the WHO [1] provide selective and targeted means of malaria control. In particular, the specific management of an environment favoring the proliferation of vectors can significantly decrease transmission [14]. The choice of interventions and their relative importance are determined by our understanding of environmental heterogeneity [8, 45–47] at a sufficiently fine scale. Furthermore, in front of the high complexity of malaria transmission and infection, study populations and study environment have to be precisely evaluated before planning research studies and intervention trials [11]. The development of GIS has made it possible to increase this so-called "micro-epidemiological" knowledge [48]. This understanding and management of the environment could be applied in large African cities. The towns of Sub-Saharan Africa are growing very rapidly. The urbanization is associated with poverty, and leads to an increase in the number of malaria cases. Indeed, the new quarters created tend to lack basic sanitation structures, have high-density, poor-quality housing and there are often no drains, all of which results in the emergence of Anopheles breeding sites [1, 14, 48–50]. This situation favors large increases in the number of malaria epidemics. The detailed mapping of malaria infections in these quarters at high-risk is therefore a matter of urgency, to guide targeted interventions and studies [48, 49].

We must remember that trends indicating a decrease in incidence describe an average over the entire area studied. This marginal analysis should not be allowed to mask the heterogeneous distribution of malaria. Indeed, despite the overall trend, high-risk zones may persist in villages, as shown here. Even at this scale, changes are heterogeneous and probably depend on changes in the number of mosquito breeding sites (creation and destruction, whether spontaneous or due to human activities).

Analysis at the level of households, using GIS, makes it possible to determine precisely the pattern of heterogeneity in the risk of P. falciparum infection and transmission. The micro-epidemiological modeling makes it possible to orient control programs, treating the high-risk zones identified as a matter of priority, and to improve the planning of intervention trials or research studies on malaria. Warranting the use of such data analysis approach, in 2006 the Malaria Vaccine Development Branch (MVDB) at the NIAID/NIH, and the Malaria Research and Training Center (MRTC) at the Department of Epidemiology of Parasitic Diseases (DEAP), University of Bamako, have set up a malaria vaccine trials site (phase I, II, III) at Bancoumana.