Temporal evolution of the malaria incidence
During 2008-2012, the malaria incidence rate for the entire area showed an annual resurgence dependent on rainfall (Figure 1). The incidence rate peaks of epidemic periods ranged from 26.4 cases/100,000 person-weeks in 2009 (October) to 115.34 cases/100,000 person-weeks in 2012 (October). Low to very low incidences of malaria were recorded even during the driest and hottest seasons.
Identification of malaria transmission periods
The change-point analysis helped to detect 5 LTPs and 5 HTPs (Figure 1).
The HTPs (except for the last one) overlapped 2 consecutive years. Annual epidemics began in July or August and ended in January or February of the following year. The median HTP duration was 28 weeks. The 2012 HTP had the highest cumulative malaria incidence rate (38.1 cases/100,000 person-weeks). The 2009-2010 HTP had the lowest cumulative malaria incidence rate (7.53 cases/100,000 person-weeks) (Table 1).
LTPs began in January or February and ended between June and August. The median LTP duration was 27.5 weeks. The 2011 LTP had the highest cumulative malaria incidence rate (2.73 cases/100,000 person-weeks), and the 2010 LTP had the lowest cumulative malaria incidence rate (0.83 cases/100,000 person-weeks) (Table 1).
High Transmission Periods (HTP) and Low Transmission Periods (LTP) with cumulative incidence rate, start and end dates, and duration (in weeks); hotspot status of villages ( hotspot or non-hotspot); number of hotspot and non-hotspot villages; cumulative incidence rate in hotspot and non-hotspot villages; number and percentage of hotspot and non-hotspot villages that received seasonal malaria chemoprevention (SMC); weekly average rainfall and standard deviation in hotspot and non-hotspot villages; dominant vegetation type (open shrublands, grasslands, croplands, mixed vegetation) in hotspot and non-hotspot villages for each period.
* Cumulative incidence rate (cases/100,000 person-weeks)
** Number and percentage of villages that received SMC (seasonal malaria chemoprevention)
*** Standard deviation
**** Dominant vegetation type for each period
Hotspot characterization during HTPs
The cluster analysis helped to detect 356 villages (out of 575) that were malaria hotspots at least once during HTPs (Table 2).
During HTPs, the median malaria incidence in hotspot and non-hotspot villages was 33.94 and 7.53 cases/100,000 person-weeks, respectively (Table 1).
The 2012 HTP had the largest number of hotspot villages (147). These villages were mostly located in the northeast of the study area (Figure 2) and were dominated by grasslands (representing 59.9% of villages). By contrast, the 428 non-hotspot villages were dominated by mixed vegetation (representing 51.2% of villages). This HTP showed the highest cumulative malaria incidence rate in both hotspot and non-hotspot villages (119.24 and 25.74 cases/100,000 person-weeks, respectively) (Table 1). It also showed the highest weekly average rainfall in both hotspot and non-hotspot villages (31.71 and 30.49 mm/week, respectively).
The 2009-2010 HTP was the least affected HTPs by malaria, with only 62 hotspot villages (37.1% of which received SMC intervention) compared to 513 non-hotspot villages (39.18% of which received SMC intervention). Hotspot villages were mainly located in the southeast of the study area (Figure 2). The cumulative malaria incidence rate in hotspot and non-hotspot villages was 27.0 and 5.17 cases/100,000 person-weeks, respectively. The weekly average rainfall was low (but not the lowest) at 20.61 and 18.21 mm/week, respectively. Both hotspot and non-hotspot villages were dominated by mixed vegetation (representing 93.55% and 83.04% of villages, respectively).
Hotspot characterization during LTPs
The cluster analysis helped to detect 82 villages (out of 575) that were malaria hotspots at least once during LTPs (Table 2).
During LTPs, the median malaria incidence in hotspot and non-hotspot villages was 12.65 and 0.87 cases/100,000 person-weeks, respectively (Table 1).
The 2011 LTP had the longest duration (30 weeks) and showed the highest number of hotspot villages (43). These villages were located mainly in the south of the study area (Figure 2). This LTP showed a high cumulative malaria incidence rate in hotspot villages and the highest cumulative malaria incidence rate in non-hotspot villages (12.69 and 1.57 cases/100,000 person-weeks, respectively). The weekly average rainfall was fairly high at around 9 mm/week in both hotspot and non-hotspot villages. Hotspot villages were dominated by mixed vegetation (representing 72.09% of villages), whereas non-hotspot villages were dominated by grasslands (representing 40.79% of villages).
The 2010 LTP had the shortest duration (21 weeks) and 22 hotspot villages located in the northwest and west-central parts of the study area (Figure 2). The cumulative malaria incidence rate in hotspot villages was 12.61 cases/100,000 person-weeks, compared to a very low cumulative malaria incidence rate of 0.41 cases/100,000 person-weeks in the 553 non-hotspot villages. The weekly average rainfall was low at around 3 mm/week in both hotspot and non-hotspot villages.
The descriptions of the other transmission periods are available in additional file 2.
Factors associated with the spatio-temporal variation of malaria hotspots
According to the multivariate analysis (GAMM, 38% deviance explained), villages receiving SMC intervention were protected from the risk of being a hotspot (OR=0.48, 95%CI: (0.33, 0.68)). The random effect of health posts was significant (τ=0.53, 95%CI: (0.31, 0.88)).
For villages dominated by open shrublands, the risk of being a hotspot did not vary over time (Figure 3, panel A). A non-linear association was found between rainfall and the risk of being a hotspot (p=0.0002; Figure 4, panel A). When rainfall was not very abundant, these villages were relatively protected from the risk of being a hotspot. However, this risk became significant from 15 mm/week rainfall (Smoothed value = 1.26, 95%CI: (0.09, 2.43)) and continued to increase before stabilizing at a maximum rainfall of around 22 mm/week (Smoothed value = 2.47, 95%CI: (1.24, 3.7)).
For villages dominated by grasslands, the risk of being a hotspot varied significantly over time (p<0.0001; Figure 3, panel B). This risk became significant and increased from late December 2009 (Smoothed value = 0.76, 95%CI: (0.12, 1.40)), peaked in early November 2010 (Smoothed value = 2.13, 95%CI: (1.45, 2.82)), and then decreased until late September 2011 (Smoothed value = 0.46, 95%CI: (0.03, 0.9)). These villages were protected from the risk of being a hotspot from early January 2012 (Smoothed value = -0.65, 95%CI: (-1.2, -0.1)) to late December 2012 (Smoothed value = -2.04, 95%CI: (-3.22, -0.84)). Moreover, a non-linear association was found between rainfall and the risk of being a hotspot (p<0.0001; Figure 4, panel B). When rainfall was not very abundant, these villages were relatively protected from the risk of being a hotspot. However, this risk became significant from 19 mm/week rainfall (Smoothed value = 0.52, 95%CI: (0.01, 1.04)) and increased with rainfall.
For villages dominated by croplands, the risk of being a hotspot varied little over time (p=0.0013; Figure 3, panel C). This risk became significant from late April 2010 (Smoothed value = 0.58, 95%CI: (0.07, 1.08)), peaked in mid-November 2010 (Smoothed value = 1.08, 95%CI: (0.48, 1.68)), and then decreased until mid-May 2011 (Smoothed value = 0.7, 95%CI: (0.01, 1.42)). These villages were relatively protected from the risk of being a hotspot from late January 2012 to late December 2012 (Smoothed value = -0.63, 95%CI: (-1.2, 0.05) to Smoothed value = -1.17, 95%CI: (-2.12, -0.25)). Moreover, a non-linear association was found between rainfall and the risk of being a hotspot (p<0.0001; Figure 4, panel C). When rainfall was not very abundant, these villages were relatively protected from the risk of being a hotspot. However, this risk became significant from 18 mm/week rainfall (Smoothed value = 0.72, 95%CI: (0.09, 1.36)) and continued to increase roughly with rainfall.
For villages dominated by mixed vegetation, the risk of being a hotspot varied over time (p<0.0001; Figure 3, panel D). This risk increased from mid-June 2008 (Smoothed value =0.42, 95%CI: (0.14, 0.7)), peaked in early April 2011 (Smoothed value =2.16, 95%CI: (1.61, 2.71)), and then decreased until mid-October 2011 (Smoothed value =0.48, 95%CI: (0.09, 0.87)). These villages were protected from the risk of being a hotspot from late November 2011 (Smoothed value = -0.4, 95%CI: (-0.76, -0.03)) to late December 2012 (Smoothed value = -2.65, 95%CI: (-3.22, -1.97)). Moreover, a non-linear association was found between rainfall and the risk of being a hotspot (p<0.0001; Figure 4, panel D). Once again, when rainfall was not very abundant, the villages were relatively protected from the risk of being a hotspot. This risk became significant from 21 mm/week rainfall (Smoothed value = 0.29, 95%CI: (0.02, 0.56)) and continued to increase linearly with rainfall.
According to the spatial interpolation obtained with the multivariate GAMM for the entire study (Figure 5), 2 zones (red colour) located in the southwest and southeast of the study area had the highest risk of being a hotspot (Smoothed value min = 0.64, 95%CI: (0.02, 1.27); Smoothed value max = 4.1, 95%CI: (3.29, 4.92)). Moreover, villages located in 2 geographically restricted areas—one in the extreme northwest and the other in the east-central part of the study area (blue colour)—were relatively protected from this risk (Smoothed value min = -8.99, 95%CI: (-12.96, -5.02); Smoothed value max= -0.8, 95%CI: (-1.57, -0.03)).