Spatio-temporal analysis of malaria within a transmission season in Bandiagara, Mali

Bandiagara is a town, of approximately 13,364 inhabitants, situated in north-eastern Mali in West Africa (Figure 1) on a rocky plain above the Dogon escarpment and receiving a mean annual rainfall of 600 mm. The rainy season spans from June to October and the dry season from November to May. Bandiagara has a small river, the Yamé, a minor tributary of the Niger which stops flowing during the dry season, and transient post rainfall standing water body during the rainy season. Anopheles gambiae is the principal malaria vector and malaria transmission is highly seasonal meso- to hyperendemic [32].

With less than one infecting bite per person per month at the height of the dry season in March, the transmission season starts in June, peaks at up to 60 infected mosquito bites per person per month in August or September, and ends in December Following transmission fluctuations, malaria incidence is seasonal too, with an intense peak in September to October. In 1999, the clinical malaria incidence was 1.7 episodes per transmission season in children less than 10 years [33, 34]. Plasmodium falciparum represents 97% of malaria infections, Plasmodium malariae 3%, and rare infections are due to Plasmodium ovale.

This study was approved by institutional review boards of the Faculty of Medicine, Pharmacy and Dentistry of the University of Mali and of the University of Maryland School of Medicine. After obtaining permission to work in the community from local officials, elders, and traditional healers as described by Diallo et al.[35], the study was publicized by local radio broadcast, and parents were invited to accompany children aged up to six years to the BMP research clinic to be screened for eligibility. Children in the target age group were eligible for inclusion in the study if they met each of the following inclusion criteria: below six years of age at the time of screening, resident in Bandiagara town, good general health based on clinical evaluation, written informed consent obtained from the parent/guardian, and participation feasible through the 48-month follow up. Exclusion criteria were: simultaneous participation in an interventional clinical trial, chronic medication with known anti-malarial activity (such as trimethoprim-sulphamethoxazole for prevention of AIDS-associated opportunistic infections), or any condition that in the opinion of the principal investigator would jeopardize the safety or rights of a participant in the trial or would render the participant unable to comply with the protocol.

Active and passive surveillance were conducted to capture the incidence of malaria infection and disease. Active surveillance consisted of scheduled monthly visits aimed at detecting asymptomatic malaria infection and anaemia. Clinical examination of the participants was performed by the study physician at enrolment and on a quarterly basis. Following standard protocols, finger-prick blood samples were collected monthly for malaria smears, measurement of haemoglobin level, and parasite genotyping from filter paper. Venous blood was collected quarterly for molecular and immunological analyses. Passive surveillance consisted of continuous availability of free, expeditious, high quality, basic medical care at the BMP research clinic and Bandiagara District Hospital, where parents/guardians were instructed to consult whenever their child was sick. Children were then examined by a physician, and axillary temperature was checked (fever was defined as axillary temperature ≥37.5°C). Blood samples were collected for microscopic examination (thick blood film), haemoglobin level determination, and parasite genotyping. Malaria was treated with artemisinin combination therapy (artesunate + amodiaquine or artemether + lumefantrine) according to the guidelines of the Mali malaria control programme.

At inclusion and in case of relocation, the household of each child (i e, the place where the child slept) was georeferenced using a handheld global position system (Garmin® Personal Navigator; accuracy approximately within 10 m).

Children household and malaria episodes occurrence were mapped and a Geographical Information System was developed for the study area that also included the Bandiagara house blocks and the water bodies of the area based on a satellite image (Quick Bird, August 2004) and field surveys (2010).

With the demographic expansion of the town and a flood in July 2007 that destroyed many houses in quarters 1, 2, and 5 on the right bank of the Yame River, new neighbourhoods have been built in the north and east. Because no updated satellite image was available at the time of the present study, house blocks of arbitrary shape and size were drawn for the few included children living in these new quarters.

Using Quantum GIS™ software (QGIS™) version 1.7.3 [36], children households were plotted according to their geographic coordinates. Numerous children, likely siblings of the same family, shared the same location. For each corrected location, data were subsequently aggregated, and several variables were calculated: initial number of study participant, total and daily number of recorded malaria episodes, daily number of susceptible children (taking into account the excluded children and a three-week refractory period after a malaria episode). After aggregating data at the house block level, the mean number of malaria episodes per child was calculated for each block over the entire year. The mean number of malaria episodes per child-week was also calculated for each successive period.

Using QGIS™, the numbers of included children asnd malaria episodes were mapped using proportional circles. The spatial distribution of malaria risk was illustrated by choropleth mapping at the block level of the mean number of malaria episodes per child, as described above.

In order to better assess the spatial variability of malaria risk, a cluster analysis was performed using Kulldorff’s statistics through the SaTScan® software [25, 37]. This widely applied method [5, 38‐40] moves a circular or elliptic scanning window over the study area and compares observed and expected case numbers inside and outside this window in order to detect clusters and estimate risk ratios. Using daily malaria episodes and susceptible children at each location, a Bernouilli distribution model with 50% of the population at risk, and elliptic scanning windows, high or low risk purely spatial clusters were sought over the whole year and over each study period. The standardized incidence ratio (SIR) was defined as the ratio of observed to expected cases. Cluster significance (P-value) was computed with a likelihood ratio test provided by a Monte Carlo approach using 999 random simulations under the null hypothesis of no cluster. Statistically significant spatial clusters (P-value < 0.05) were subsequently mapped on QGIS™.

Study children lived in 168 locations which have been were geopositioned. Children habitats were mainly distributed in the centre and north-eastern blocks of Bandiagara (Figure 2A).

During the course of the study period, 11 active surveys were carried out, and a total 296 P. falciparum clinical episodes were recorded among 178 children. Fifty-one episodes were documented from active surveys and 245 episodes from passive survey. Episodes mostly concentrated along the right banks of the Yamé River and, to a lesser extent, on the north-eastern blocks of the city near the brickyard (Figure 2B), exhibiting a marked spatial heterogeneity.

In order to analyse the temporal distribution of malaria episodes, data were aggregated by week, and a time series of malaria episodes was plotted together with locally measured rainfalls (Figure 3). In 2009, the first malaria clinical episodes had a lag of four weeks after the onset of the rainy season. Children continued to experience malaria episodes late into the 2009–2010 dry season (from December 2009 to May 2010), a period that accounted for nearly half of the 296 total episodes. With respect to the rainfall time distributions, four successive periods were delimited (Figure 3). Period 1 extended from the beginning of the study (2009 week 23) until the rainfall maximum (2009 week 34; 1 June to 23 August 2009). Period 2 spanned the rest of the rainy season (from 2009 week 35 to week 44; 24 August to 1 November 2009). Period 3 covered the first half of the dry season (2009 week 45 to 2010 week 5; 2 November 2009 to 24 January 2010). Period 4 extended throughout the second half of the dry season (2010 week 6 to week 19). Because this was the start of the next rainy season, the remaining period in the time series (2010 week 20 to week 23) was excluded from further period-based analyses.

In order to better visualize the malaria risk across Bandiagara and to minimize any sampling bias, the mean number of malaria episodes recorded per sampled child was mapped for each block (Figure 4). Over the entire year of follow up, while the greatest risk of malaria transmission seemed to be located on the western blocks, along the northern shore of the Yamé River and on the northern side of the brickyard, the centre of the town exhibited a patchy yet globally low risk pattern.

For each of the four epidemiological periods, as defined in Figure 3, a similar choropleth mapping of the mean number of episodes per child-week was computed, as illustrated in Figure 5. These sequential maps exhibited a contrasting pattern. While malaria incidence appeared globally low during the beginning of the rainy season (Period 1, Figure 5A), especially in the centre of Bandiagara, malaria incidence was much higher during the end of the rains (Period 2, Figure 5B), with a patchier distribution. During the first half of the dry season, incidence became more intense on the edge of the town, particularly along the Yamé River and just north of the brickyard (Period 3, Figure 5C). Malaria risk returned to negligible during the second part of the dry season (Period 4, Figure 5D), except for a few limited foci in the southern part of the town.

To statistically confirm this spatial distribution of malaria risk across Bandiagara, high or low-risk spatial clusters were sought using Kuldorff’s statistics. In the full year of study (Figure 4), the first significant cluster was a low-risk cluster located on the eastern part of the town centre and separated from the brickyard and the Yamé River by a few blocks of houses (SIR = 0.3; P-value <10^-5). A secondary significant high-risk cluster covered the western blocks of Bandiagara located near the bend of the Yamé River (SIR = 2.36; P-value <0.01). During Period 1 (Figure 5A), this method didn’t establish any significant cluster, but a nearly significant low-risk spatial cluster included the south-east part of the town on both sides of the Yamé River (SIR = 0.091; P-value = 0.07). During Period 2 (Figure 5B), no area with a significantly higher or lower incidence could be detected. During the first half of the dry season (Period 3, Figure 5C), a significant low-risk cluster covered the eastern part of the town centre (C1, SIR = 0.3; P-value <10^-5). A secondary high-risk cluster spanned most blocks located along the river (C2, SIR = 2.36, P-value <0.01). Conversely, no significant low-risk cluster was detected during Period 4 (Figure 5D).