Background
Malaria remains one of the biggest health problems within the tropical region despite the improvements in malaria control programmes at a global scale. There were approximately 212 million cases of infection and 429,000 malaria-related deaths in 2015; and more than 90% of these cases occurred in sub-Saharan Africa [
1]. In Zimbabwe, malaria continues to be a major public health threat with an estimated over half of the population of 13.5 million at risk of contracting malaria [
2]. However, by 2010, Zimbabwe had managed to reduce malaria incidence to 45 malaria cases per 1000 inhabitants per year thereby surpassing the Abuja 2010 set target of 68 cases per 1000 inhabitants [
3]. Consistent with the national trend, Gwanda district located in the Matabeleland South Province in Zimbabwe has progressed to malaria pre-elimination phase. A malaria pre-elimination capacity assessment study conducted in Matabeleland South Province in 2011 reported malaria positivity rates of 8% and
Anopheles larvae scoop of four for Gwanda district [
4]. Malaria control in Gwanda district is mainly through indoor residual spraying (IRS), use of long-lasting insecticidal nets (LLINs) and larviciding [
5].
Malaria transmission is heterogeneous at varying geographical scales even in the malaria pre-elimination zones [
6]. Usually malaria endemicity levels, and especially in low incidence areas, malaria tends to cluster in ‘hotspots’ and ‘hot’ populations that become sources of continued infection [
7]. Malaria hotspots have close spatial associations with vector-breeding habitats, and in certain ‘high-risk’ sub-sets of the population, having higher exposure to vector-breeding habitat: ‘hot-pops’ [
8]. Active and timely identification of these hotspots and related factors is important for effective malaria control [
7]. Malaria heterogeneity in time and space has also been attributed to risk factors, including altitude, climate, environmental parameters, and socio-economic factors [
6,
7]. Rainfall, temperature and altitude are key factors in determining the habitat suitability of malaria vectors, including
Anopheles arabiensis which is common in Zimbabwe, as well as determining malaria incidences [
9‐
12]. In most tropical African countries, high habitat suitability of malaria vectors such as
An. arabiensis mostly translates into increased malaria incidences [
11]. Mabaso et al. [
12] used a model to analyse the spatial and temporal role of climate in inter-annual variation of malaria incidence in Zimbabwe for the period 1988–1999 and their study demonstrated that mean values of temperature, rainfall and vapour pressure are strong predictors of malaria incidence. Gwitira et al. [
13] also concluded that annual precipitation, precipitation of the wettest month, isothermality and temperature seasonality combined with altitude, are key predictors of
An. arabiensis habitat suitability in Zimbabwe. In their study, the habitat suitability was significantly and positively correlated with recorded malaria incidences. Based on their observations they inferred that high malaria cases would be expected in areas of high vector habitat suitability. However, they also noted the need to consider malaria interventions in the study region in order to draw meaningful conclusions about the relationship between mosquito habitat suitability and malaria incidence.
Accurately assessing the local risk of transmission is fundamental for the development of malaria control programmes. Differences in malaria transmission exist, not just between different regions but also at local level [
14‐
16]. As Gwanda district is moving towards malaria elimination phase a better understanding of the distribution of malaria cases/incidence at local scale is essential. In this regard, it is critical to understand the key factors for determining variability of malaria cases at micro-spatial scale for improving malaria control strategies in cases of relapses or outbreaks.
An exploratory study to investigate the relationship between the above-mentioned factors (rainfall, temperature, altitude, and other factors, including vegetation cover and wetness) at local scale is required. This paper reports on the spatial distribution of malaria incidence in 2015 based on health facility cases and related risk factors, in order to strengthen control measures in the pre-elimination phase of Gwanda district, Matabeleland South Province, Zimbabwe.
Discussion
Results of this study have shown that the distribution of annual cases or incidence of malaria is heterogeneous even in the malaria pre-elimination zones, as observed in other studies [
6‐
8]. Differences exist not just between different regions but also at local level [
14‐
16]. Identifying the malaria clusters (areas with elevated number of cases) is critical in developing or improving malaria control strategies at local scale. In this study, clusters were detected in rural and urban areas. This indicates that malaria incidence could be high or above average in both rural and urban settings in Gwanda district. However the cluster had a higher relative risk compared to the urban cluster. These clusters may point to the areas that need immediate attention in terms of preparation and implementation of the disease control strategies.
Heterogeneity of malaria cases is driven by a variety of ecological, biological and sociological factors [
6]. As noted by Ehlkes et al. [
35] most studies assume homogeneous influence of exploratory variables [
42‐
45] but this may not always be most appropriate [
23]. In this study, the analysis showed that assuming that other variables vary at local level substantially improves the GWPR model performance. Allowing spatial heterogeneity within the regression model allows clearer interpretation regarding the true nature of potential associations [
35]. That could be because the LLINs are mostly distributed to the wards that are known to have higher incidences in Gwanda district and there was limited data regarding the actual use of these nets. When assessing the associations between environmental variables and malaria cases one must consider the pathways in which these variables under study lie [
35]. For example, the environmental variables: minimum temperature, NDVI, altitude, NDWI which influence the malaria cases were considered in this study as they determine the abundance of mosquito or their breeding habitats. Malaria control strategies (IRS and LLINS coverage) per ward were also considered in this study. These factors tend to reduce the cases of malaria. The interaction between these factors and malaria cases may bring out unexpected results, defying the norms regarding the relationship between environmental factors and malaria. It has been established that transmission potential decreases as the altitude increases [
2,
46,
47]. This was also noted in this study as altitude showed its expected negative relationship with malaria cases in some of the wards but also showed positive relationship in the wards to the southern part of Gwanda district. However, it was not significant in any of the wards. NDWI coefficients showed the expected positive association with malaria cases and were significant in the rural wards located in south and southeastern part of Gwanda district. Altitude and minimum temperature estimated coefficients ranged from negative to positive over Gwanda district. This indicates that GWPR successfully captured the spatially non-stationary of these factors and how the global model can be misleading since averaging these local effects reveals a single impact assumed to hold across all regions [
33]. The estimated NDVI coefficients were negative over the study area, which was against the expectation.
The weak positive and strong negative correlation coefficients between environmental factors and malaria incidences in some of the wards could be due to the protection effect of malaria control factors, such as vector control methods including LLINs and IRS. These malaria interventions contribute significantly to the decline in malaria cases particularly in areas progressing towards malaria elimination [
47]. Gwitira et al. [
13] also noted that in cases where there is effective malaria control, there will be weak correlations between habitat suitability and malaria cases. This was observed in this study based on the proxies for land cover (NDVI), wetness (NDWI). However, there is need to consider mapping the land use and land cover types and relate to malaria cases at micro-scale. Changes in land use and land cover have also been found to be critical in determining the survival of
Anopheles malaria vectors [
35,
48]. Significant land cover and land use changes may lead to increase in the abundance of malaria vectors and consequently increased malaria transmission.
Geographically weighted regression (GWR) has shown its ability to handle the socio-economic variables in relation to disease transmission. For example, the semiparametric-GWR (s-GWR) managed to detect schistosomiasis hot spots based on socio-economic and environmental factors at household level in Ndumo area, uMkhanyakude in South Africa [
35]. Active surveys are required to capture the data on socio-economic factors, including housing structure and entomological data. Previous studies showed that malaria transmission tends to be higher in houses built with mud and thatch than those with asbestos or iron sheets and are built using cement [
9]. All these factors influence the spatial and temporal distribution of malaria incidence. The assessment of the effect of malaria control measures may need to be done in spatio-temporal modelling. The temporal aspect would be able to show the decrease or increase in malaria cases in relation to control measures and environmental factors over time. There was also no information regarding larviciding for 2015 and 2016. In the pre-elimination and elimination phases, interventions have to be targeted to entire villages or towns with higher malaria incidence until only individual episodes of malaria remain and become the centre of attention [
49].
This study has shown the feasibility of using passive surveillance data from health facilities to map malaria cases and detect clusters. The passively identified health facility cases reflected malaria transmission levels in places where malaria cases tend to cluster at ward level as also noted by Rulisa et al. [
7]. These cases were parasitologically confirmed using RDTs or microscopy and were complete both spatially and temporally. The clinical malaria cases were previously used in trend analysis of malaria transmission in relation to climatic and environmental factors in Tubu village, Botswana [
50]. However, these data are usually incomplete [
51], and future studies may consider approaches that adjust for health facility utilization and under-reporting [
52,
53]. Sturrock et al. [
17] noted that mapping malaria in low transmission settings is a challenge, given that as incidence drops, transmission concentrates in hot spots and hot pops. It is challenging to identify hot pops operationally because only a sub-set of febrile individuals may seek treatment at formal health facilities [
53]. In low malaria transmission settings, the treatment-seeking patterns may be determined by individual immune status [
54]. For example, the onset of fever in low-immunity populations may lead to presentation at peripheral health centres, while populations highly exposed (hot pops) are less likely to seek treatment [
55]. These malaria hot spots may serve to perpetuate residual malaria transmission in low-transmission seasons and hinder efforts to eliminate malaria [
56]. Active and timely identification of these hot spots and associated risk factors through active surveys is essential for targeting interventions to optimize malaria control [
49].
This study presents an analysis of a single year of cross-sectional data that was temporally aggregated, hence the temporal dimension was not considered. Using only 1 year of data may only reflect where cases occurred during that single year, but not temporally stable areas of high malaria risk. This ignores and masks any seasonal pattern, which is highly important in areas approaching elimination. However the results of this study will inform further exploratory studies in pre-elimination zones also considering the amount of data which will be collected through DHIS2. The analysis in this study was also restricted by the limited number of wards in Gwanda district which impacted on sample size. Ward is the smallest political administrative unit with reliable data on population size. This might have impacted on the analysis as Paez et al. [
57] advised that GWR may not be used in cases with sample size less than 160 as the small sample size issue could result in inaccurate estimates in statistical modelling [
57‐
60]. However, Li et al. [
61], noted that after considering the spatial heterogeneity in the county-level data, with sample size less than 160, the GWPR outperformed the traditional generalised linear models (GLM) in predicting fatal crashes in individual counties [
61]. Therefore this study contributes to other GWR studies on disease and risk factors, most of which show that global statistical models may produce misleading results [
62]. GWPR models are capable of capturing the spatial heterogeneity of phenomenon compared to global estimates/models [
63]. The local coefficient maps in this study also show the magnitude, significance and direction of the relationships between malaria cases and exploratory variables. For local planning, such as district or ward (as in this case), the local GWR models seem to be more appropriate, since global models may not capture local changes or variation [
63].
Authors’ contributions
TM, MJC and SM conceived the research. TM and MM participated in fieldwork. TM performed data analysis. TM, MJC, SM, and MM contributed to draft manuscript editing/reviewing. All authors contributed to the revisions. All authors read and approved the final manuscript.