- Open Access
Inclusion of edaphic predictors for enhancement of models to determine distribution of soil-transmitted helminths: the case of Zimbabwe
Parasites & Vectorsvolume 11, Article number: 47 (2018)
Reliable mapping of soil-transmitted helminth (STH) parasites requires rigorous statistical and machine learning algorithms capable of integrating the combined influence of several determinants to predict distributions. This study tested whether combining edaphic predictors with relevant environmental predictors improves model performance when predicting the distribution of STH, Ascaris lumbricoides and hookworms at a national scale in Zimbabwe.
Geo-referenced parasitological data obtained from a 2010/2011 national survey indicating a confirmed presence or absence of STH among school children aged 10–15 years was used to calibrate ten species distribution models (SDMs). The performance of SDMs calibrated with a set of environmental and edaphic variables was compared to that of SDMs calibrated with environmental variables only. Model performance was evaluated using the true skill statistic and receiver operating characteristic curve.
Results show a significant improvement in model performance for both A. lumbricoides and hookworms for all ten SDMs after edaphic variables were combined with environmental variables in the modelling of the geographical distribution of the two STHs at national scale. Using the top three performing models, a consensus prediction was developed to generate the first continuous maps of the potential distribution of the two STHs in Zimbabwe.
The findings from this study demonstrate significant model improvement if relevant edaphic variables are included in model calibration resulting in more accurate mapping of STH. The results also provide spatially-explicit information to aid targeted control of STHs in Zimbabwe and other countries with STH burden.
Soil-transmitted helminthiases are a group of neglected tropical diseases (NTDs) caused by intestinal parasites that are transmitted through faecal contaminated soil. They include Ascaris lumbricoides, Trichuris trichiura, Necator americanus and Ancylostoma duodenale [1,2,3,4]. These helminths are of a major concern to public health in tropical and sub-tropical countries where their infection is associated with devastating morbidity rates [5, 6]. About 4.5 billion are at risk of infection worldwide [7, 8] and more than 2 billion people are infected by STH .
The disease burden caused by these parasitic worms is enormous. In 2014, Pullan et al.  estimated the global numbers of people infected with hookworm, A. lumbricoides, and T. trichiura, to be 438.9 million, 819.0 million and 464.6 million, respectively. Previous estimates in 2003 by de Silva et al.  showed these numbers to be 740 million, 1221 million and 795 million people, respectively. In 2010, the World Health Organization (WHO) estimated that 875 million children required annual treatment with preventive chemotherapy . The burden of the disease is known to be highly concentrated among the poorest socio-economic groups [12,13,14]. Previous estimates showed that more than 44 million pregnant women had clinical effects from hookworm-associated anaemia . Hookworm-associated anaemia is known to result in the loss of 39 million disability-adjusted life years per year .
Based upon on the public health significance of STH, the WHO has urged member states to ensure access to essential drugs for treating STH infections in all health services in endemic areas and groups at high risk of morbidity. Such high risk groups include women and children. A goal was set to attain a minimum target of the regular administration of chemotherapy to at least 75% of all school-age children at risk of morbidity by 2010 . However, to date this target has not been achieved. This is partly due to limited number of medicines and failure to precisely map the affected populations requiring treatment coupled with poor sanitation coverage and lack of a safe water supply. Global milestones for eliminating STH as a public-health problem in children were drawn by the WHO to guide efforts of member states in the fight against STH . These milestones included completion of country mapping of STH by 2015. Annual mass drug administration achieving a global coverage of at least 75% by 2020 was stipulated . Considering how widespread STH infection is globally, it is therefore surprising that the disease still remains neglected.
In sub-Saharan Africa, STHs have been found to be widely distributed [19,20,21]. However, spatially explicit information on the distribution of specific parasitic nematodes at country level remains scarce. Previous research has provided insight into the spatial epidemiology of the STHs [22, 23]. It is known that the infective stage of these nematodes is found in faecal contaminated environments especially moist and warm soils . Regarding A. lumbricoides, fertilised eggs are known to undergo maturation in the soil for them to become infective. Hookworm eggs also hatch in the moist soil and the larvae moult twice to become infective larvae  that move up to the upper layers of soil to infect human hosts . People typically become infected after ingesting a fully developed A. lumbricoides egg and/or after their skin is penetrated by third-stage hookworm larvae [25, 26]. It follows that the density of infective eggs and larvae in the soil correlates with STH exposure and risk. Thus, accurate modelling and mapping of the spatial distribution of STHs should consider edaphic variables that drive egg development for A. lumbricoides and are suitable for the survival of hookworm larvae.
Previous work used species distribution models (SDMs) to explore the distribution of common STH parasites in various countries including Sierra Leone , Kenya , Nigeria , China , Bolivia  and Brazil . While most SDMs used a combination of several bioclimatic and social-economic variables as co-determinants [23, 27, 32], edaphic variables were overlooked, despite playing an important role in STH ecology and infection. There are, however, a few studies which included edaphic variables to model STHs [22, 29, 30]. In Zimbabwe, Chandiwana . Described the distribution of soil-transmitted helminths (STH) using samples collected for the parasitological diagnosis of Schistosoma mansoni. The study reported a prevalence of 1.6% for hookworms and of 0.5% for A. lumbricoides. Trichuris trichiura was not reported . The study further observed that the majority of infected children were found in the Northeast, the Zambezi Valley, the Central and Southeast low-veld areas of the country. It was, however, highlighted that the data needed to be considered with caution since the stool specimens had been collected for S. mansoni diagnosis and the methodology might not have been suitable for STH . A recent study by Midzi et al.  indicated a combined prevalence of 5.5% for STH. At the species level, hookworms, A. lumbricoides and T. trichiura had the prevalence of 3.2%, 2.5% and 0.1%, respectively. The distribution of STH followed the trend as described previously .
Although these studies represent important progress with regard to linking the ecological theory with SDM techniques to better understand STH distribution, the studies did not report on the relative importance of edaphic variables, nor did they assess and quantify how model performance changed with the inclusion of edaphic variables. This is an important research gap that needs to be filled as a preamble to generating spatially explicit information showing in-country variability in the distribution of STHs to aid disease control and safeguard public health.
Therefore, our study tested the hypothesis that the inclusion of edaphic variables such as soil moisture in SDMs increases model performance in the spatial prediction of the distribution of STHs. Using Zimbabwe as a case study, the performance of ten SDMs comprising a set of environmental plus edaphic variables was compared to that of SDMs calibrated with environmental variables only. Since spatial prediction varies depending on the choice of variables and modeling method selected , it was therefore necessary to run multiple SDMs in search for the evidence for and against the above hypothesis.
The parasitological data used in this study were collected from primary school age children (age range 10–14 years) living in 71 districts distributed among Zimbabwe’s eight rural and two metropolitan provinces . A sample of 15,818 children was calculated using EPI Info 6 statistical package (Epi Info version 6, Centers for Disease Control and Prevention, Atlanta, GA 30333) using 37% as the assumed mean prevalence of schistosomiasis and the error margin of 0.75% (see  for detailed information about study areas, subjects and sampling).
To optimise health delivery, the Ministry of Health and Child Care (MOHCC) classifies 63 of the country’s 89 administrative districts as rural-based districts. The remainder are contained in the two metropolitan provinces, Harare and Bulawayo. However, it should be noted these 63 rural districts are part of the 89 districts recognised by the Government of Zimbabwe as political boundaries for enhancing local governance. Thus, the parasitological data used in this study was collected in almost all rural districts which comprise the spatial planning domain for disease surveillance and management at national scale. When writing this manuscript, the authors considered all the 89 administrative districts in order to demonstrate the important role of remote sensing and GIS technology in predicting the risk of transmission/infection with STH in which case the parasitological data could be ascribed to 71 districts where it was collected in the previous study . Zimbabwe stretches from latitudes 15°37′–22°24′S and lies between longitudes 25°14′–33°04′E (Fig. 1). The country is 390,575 km2 in area. It borders with Zambia, Mozambique, Botswana and South Africa in the north, east, west and south, respectively. The total population was estimated at 13,061,239 in the recent census survey . Altitude ranges from 300 m to 2500 m above sea level .
Zimbabwe has a subtropical climate, with mean monthly maximum temperature ranging from 15 °C in July to 24 °C in November. Total annual rainfall ranges from 400 mm to 1000 mm . The country has assortment variety of soil types ranging from sodic and salliatic soils in the north, ferrialistic soils in the south, paraferrallistic and ortheferrilitic in the east, to regosols and Kalahari sands in the west [38, 39]. The vegetation is dominated by dry miombo woodlands in the central and east regions of the country . Mopane woodlands dominate in the lowveld regions located in the northern and southern areas .
STH occurrence data
Geo-referenced data for A. lumbricoides and hookworms collected during a national cross-sectional survey at randomly selected schools in Zimbabwe during 2010–2011  were used to calibrate the SDMs. The survey targeted primary schools located in 71 of the recognised 89 administrative districts in Zimbabwe including the major urban centres of Harare, Chitungwiza and Bulawayo . The prevalence of STHs was determined using the formol ether concentration and the Kato-Katz smear techniques as explained in . A positive result for A. lumbricoides and hookworm eggs from either of the two techniques was used as an indicator for presence of these parasites among sampled school children .
Environmental and socio-economic variables
A total of six environmental and demographic variables were used to model the spatial distribution of A. lumbricoides and hookworms in Zimbabwe. These environmental variables were: the moderate resolution imaging spectroradiameter (MODIS) monthly daytime and night-time land surface temperature (LST), annual average precipitation (AVP), MODIS normalised difference vegetation index (NDVI), human population density (HPD) and the distance from perennial water bodies (DPW). These environmental variables were selected as they have been found useful for predicting the distribution of STH [22, 29].
In brief, monthly LST daytime and night-time datasets were derived from infrared radiances measured with the MODIS aqua and terra sensors for the period January to December (both years). The datasets were accessed from the Land Processes Distributed Active Archive Centre (LP DAAC) operated by the United States Geological Survey (USGS) at https://lpdaac.usgs.gov/. Monthly LST daytime and night-time datasets were separately clipped by the polygon map of Zimbabwe, added together and divided by 12 to obtain the annual average monthly LST. AVP was calculated from gridded monthly rainfall data for the years 2010 and 2011. These rainfall data were downloaded as raster grids from the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) archive at http://chg.geog.ucsb.edu/data/chirps/. The rainfall data were available at a 5 km spatial resolution. To capture the potential effect of vegetation on STH parasites distribution, MODIS monthly NDVI was used as a proxy for vegetation cover . MODIS monthly NDVI (MOD13A3) was in the format of cloud-free imagery and was downloaded for the months January to December for 2010 and 2011 from LP DAAC at https://lpdaac.usgs.gov/. The monthly NDVI images covering the whole of Zimbabwe were averaged by year to match the temporal window at which STH parasitological data were collected in the field during the national survey.
The distance from perennial water bodies was used as a proxy for moisture availability . Spatial data layers indicating the distribution and spatial extend of surface water bodies were downloaded from the Diva GIS website (diva-gis.org). These layers were projected from a geographical coordinate system (WGS 84) to a metric coordinate system (WGS 84/UTM zone 35). Then, DPW was calculated using the built in Euclidean distance function in ArcMap 9.3 . The output map was projected back to a geographical coordinate system (WGS 84) to match the map projection used by other environmental variables. HPD was used to represent the potential influence of the distribution of human population (the host) on the occurrence of STH parasites . The gridded human population density (version 4) for the year 2010 was downloaded from the Socioeconomic and Data Application Center (SEDAC) accessible at http://sedac.ciesin.columbia.edu/data/ . The population density was mapped at a spatial resolution of 1 km.
A suite of edaphic variables which included soil organic carbon, soil pH and soil moisture, was also used to further characterise the environmental niche of STH. The selection of these edaphic variables was based on previous literature on STH distribution as well as their relative importance to the biology of STH parasites [22, 30]. Data for organic carbon, bulk density, clay content, soil pH for the topsoil (0–30 cm) were downloaded from the ISRIC-WISE soil database as spatial layers . These edaphic variables were made available at a spatial resolution of 5 km . Long-term average soil moisture data with a coarse spatial resolution of 30 km were downloaded from Africa Soil Information Services website . The information in Table 1 indicates the units, spatial resolution and sources of data for the environmental and edaphic variables used to predict STHs throughout Zimbabwe. Prior to modelling, all variables were re-sampled from their native resolution to a common 1 km spatial resolution using the nearest neighbour technique so that they could be overlayed. Thus, the distribution of STHs was modelled and mapped at a spatial resolution of 1 km.
Modelling distribution of STHs
To test for collinearity, pairwise correlations between predictor variables in raster data format were calculated in the R statistical package (Studio, 2012) using Pearson’s product moment correlation test. The folklore threshold value of r > 0.7 between predictor variables was used to eliminate correlated variables and to create a parsimonious model . Elevation and bulk density were dropped from the modelling exercise because the latter was negatively correlated with organic carbon (r = −-0.80) and the former was also negatively correlated with night-time LST (r = -0.74).
Ten species distribution modelling techniques, namely the random forest (RF), gradient boosted model (GBM), surface range envelope (SRE), artificial neural network (ANN), generalised linear model (GLM), generalised additive models (GAM), classification tree analysis (CTA), multiple adaptive regression splines (MARS), flexible discriminant analysis (FDA) and MAXENT were used to separately predict the geographical distribution of A. lumbricoides and hookworms in Zimbabwe. All the models were run in the R statistical package using the BIOMOD2 package . Each model was run twice, first as a full model containing all eight predictors and secondly, as a reduced model comprising five variables without the edaphic variables.
BIOMOD2 was tuned to split presence data with 80% being used for model calibration while 20% were set aside for model validation . Each SDM model was evaluated using the true skill statistic (TSS) and the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. A model’s performance was considered poor if the ROC value was less than 0.6, good if ROC was within the 0.61–0.80 range and excellent if ROC value was > 0.80 . ROC and TSS values were plotted against each other on a scatterplot to visualise variations in model performance under different sets of variables. Models that included and excluded edaphic variables were annotated as 1 and 2, respectively. The change in ROC and TSS model evaluation scores following the inclusion of edaphic predictors was separately calculated as a percentage for all the ten SDMs.
The TSS and ROC values for the ten modelling techniques were tested for normality using the Shapiro Wilk’s test. TSS and ROC scores for A. lumbricoides followed a normal distribution whilst those for hookworms did not follow a normal distribution. Therefore, to test for significant differences in model performances under different variable sets, an independent t-test was used for A. lumbricoides, whereas the Mann-Whitney U-test was used for hookworm data. The TSS and ROC were the response variables and model type was the categorical explanatory variable. Category (1) models comprised of model evaluation scores obtained using a set of variables which included the edaphic predictors. Category (2) comprised of model evaluation scores obtained from a variable set that excluded edaphic predictors.
Consensus modelling of STH
Models with TSS and ROC greater than 0.5 and 0.7 respectively, were identified and used to build a consensus model for predicting the continuous distribution of A. lumbricoides and hookworms throughout Zimbabwe. Specifically, for each species, a consensus model was created by combining the predictions of the top three performing models with ROC > 0.7 and TSS > 0.5. The spatial predictions of the consensus distribution model were exported to geographical information system software (Arc Map 9.3) to display the distribution throughout Zimbabwe as a continuous map. The continuous probability of presence map was classified into five distinct thematic classes based on the natural breaks in the data to enhance visual contrast. To zoom in on potential presence, a threshold value of TSS ≥ 0.5 was used to generate a binary map showing potential presence of A. lumbricoides and hookworms for ease of communication and to aid the management of STH in Zimbabwe.
Assessing variable importance
BIOMOD2 was calibrated to automatically compute variable importance. Variable importance was assessed only for the top three performing models. The goal was to check whether the inclusion of edaphic variables was as hypothesised. A variable was considered to be important when its value was > 0.10.
Prevalence of STH in Zimbabwe
Results used in preparing this manuscript were obtained from the national survey conducted by Midzi et al. . Of the estimated sample size (n = 15,818) for the national survey, 12,252 (77.5%) participants were screened for infection with any of the soil-transmitted helminthes (hookworms, Trichuris trichiura and Ascaris lumbricoides). Results from the study by Midzi et al.  showed the overall combined prevalence of STH of 5.5%, ranging between 0 and 18.3% in provinces, 0–45% in districts and 0–78.7% in schools. There was no significant difference in the prevalence of STH between males (7.5%) and females (6.9%) (Fisher’s exact test, P = 0.231). The prevalence of STH was highest in Binga district (45.5%, 95% CI: 38.46–52.67%) followed by Mutoko (43.5%, 95% CI: 35.55–51.72%) and Murehwa district (40.6%, 95% CI: 34.07–47.46%). Overall, STHs were predominantly distributed in the northern, northeastern and eastern regions and scantly distributed in the western region of Zimbabwe .
Performance of SDMs for predicting STHs distribution in Zimbabwe
Data based on modelling of edaphic variables
Model performance varied among the ten modelling techniques as illustrated in Figs. 2 and 3. Models which included edaphic variables performed better in predicting the distribution of A. lumbricoides compared to models that excluded edaphic variables (Fig. 2). The same pattern was observed for hookworms as illustrated in Fig. 3. Specifically, the results reveal that for both A. lumbricoides and hookworms, models that contained environmental plus edaphic variables yielded superior results (TSS > 0.5 and ROC > 0.75) compared to those which had environmental predictors only.
Figure 2 also illustrates that GBM, GLM, SRE outperformed other modelling techniques in predicting the distribution of A. lumbricoides with TSS and ROC values greater than 0.50 and 0.75, respectively. By contrast, ANN, GAM and CTA performed poorly. For hookworms, the GLM, MAXENT and GBM were the best performing models. The ANN, SRE and RF performed poorly (Fig. 3). Thus for both A. lumbricoides and hookworms, the GLM and GBM consistently performed well whereas the ANN performed poorly for both species with TSS < 0.3.
For A. lumbricoides, the results of the t-test confirmed significant differences in model performance between the two sets of models, i.e. the models trained with environmental variables only versus those trained with environmental plus edaphic variables (TSS: t(18) = 3.1, P = 0.006 and for ROC: t(18) = 2.48, P = 0.023). Similarly, hookworms results for the Mann-Whitney U-test indicated significant differences in model performance between these two sets of SDMs (TSS: U = 17.5, P = 0.01 and for ROC: U = 21.5, P = 0.029).
Changes in model performance
The percentage change in model performance varied among the ten modelling techniques as summarised in Table 2. The largest improvement in model performance was obtained for SRE following the inclusion of edaphic variables with a percentage increase of 160 and 53% for TSS and ROC evaluation techniques, respectively. By contrast, the lowest percentage change in model performance was obtained for GLM and GBM with the former recording a 20% change when evaluated using TSS whilst the latter recorded 1.3% change using the ROC evaluation technique.
Results in Table 3 also show that percentage change in model performance varied amongst the ten SDMs used to model the distribution of hookworms. The SRE recorded the largest percentage increase in model performance (9900%) following the inclusion of edaphic predictors when evaluated using the TSS. The ANN was also characterised by the largest increase in model performance (5000%) when evaluated using the ROC. The lowest percentage change in model performance was recorded for RF with values of 5% and 2.6% for TSS and ROC, respectively.
Predicted geographical distribution of STHs in Zimbabwe
The predicted probability of the presence of A. lumbricoides varied among the 89 administrative districts of Zimbabwe. The districts characterised by the highest probability of presence were located in the eastern parts of the country with a probability > 0.8. These included Chimanimani (3), and Mutasa (10) shown in Fig. 4. The districts located in the western, southern and the central watershed regions such as Harare (4), Gokwe South (5), Insiza (7), Masvingo (8) and Chikomba (2) were characterised by moderately high probabilities of presence. In contrast, districts at the southern, western, and northern extents of the country which included Beitbridge (1), Hwange (6) and Mbire (9) were characterised by low predicted probabilities of the presence for A. lumbricoides.
Similar to A. lumbricoides, the predicted distribution pattern for hookworms indicated the highest probabilities of presence for districts in the eastern areas of the country. The districts characterised by highest probabilities included Rusape (10), Murehwa (8), Chitungwiza (5), Guruve (6) and Bulawayo (3) as illustrated in Fig. 4. The districts situated in the western parts of the country including Binga (2), Nkayi (9) and Umzingwane (11) were characterised by moderately high probabilities of presence. Low probabilities of presence for hookworms were predicted for districts located in the eastern and southern regions of the country such as Mudzi (7), Chiredzi (4) and Beitbridge (1).
Spatial pattern of STHs occurrence in Zimbabwe
Ascaris lumbricoides was predicted to be occurring in 66 districts stretching from the northern to the eastern parts of the country (Fig. 5). Districts characterised by high A. lumbricoides presence included Chipinge (1), Zvimba (7) and Harare (3). The predicted presence for hookworms was more widespread in the country compared to A. lumbricoides with the species predicted as occurring in 74 districts, predominantly districts in the northern, eastern and southern parts of the country, particularly Shamva (7), Shurugwi (8), Chivi (3) and Bulawayo (1).
The results in Table 4 reveal that NDVI was consistently identified as the most important predictor for all the three top performing models. Soil pH was also an important variable for GLM and GBM followed by HPD selected as important by GBM and SRE. It was observed that for each of the top three performing models, at least one of the three edaphic variables was considered an important predictor for modelling the distribution of A. lumbricoides.
With regard to hookworms, soil organic matter was identified as the most important variable for predicting hookworms by GLM, GBM and MAXENT. Similar to the results for A. lumbricoides, at least one of the three edaphic variables was considered important for modelling the distribution of hookworms in Zimbabwe. HPD was selected twice as an important variable for both STHs modelled. Thermal variables, in particular LST (day) and LST (night), also appeared to be influential in predicting both A. lumbricoides and hookworms. NDVI was also a key variable for predicting hookworms when using GLM and GBM.
The results of this study provide empirical support to the hypothesis that the inclusion of edaphic variables improves model performance when predicting the distribution of STHs at country scale. A consistent improvement in model performance was achieved among a wide variety of modelling techniques when edaphic variables such as organic matter content were combined with other environmental variables to make spatial predictions of A. lumbricoides and hookworms presence in Zimbabwe. Furthermore, the observed statistically significant percentage increases in model performance demonstrate that inclusion of edaphic predictors enhances models to determine the distribution of soil-transmitted helminths. While the inclusion of edaphic variables in modelling STH occurrences has been undertaken in China , Bolivia  and Nigeria , this study is the first (to our knowledge) to report superior results when comparing models calibrated using environmental plus edaphic variables to those that exclude the latter. Thus, studies that exclude edaphic variables could be either under- or over-estimating the distribution of STHs [27, 28, 32]. Although model performance consistently improved following the inclusion of edaphic predictors on all the ten SDMs and for both STH parasites, the level of improvement varied with each modelling technique. This result confirms the widely observed discrepancy among different modelling techniques and justifies the need to run several SDMs to better characterise the niche space of a target species.
In this study DPW, soil moisture, soil pH, HPD, AVP, NDVI, daytime LST and night-time LST were found to be important variables for predicting the distribution of A. lumbricoides. This result corroborates a previous study which documented the important role that moist and warm conditions play in promoting quick embryonation of A. lumbricoides . The high importance attached to NDVI in this study suggests that the occurrence of A. lumbricoides is also influenced by vegetation cover. This may not be surprising as previous research reported that eggs of A. lumbricoides die when exposed to direct sunlight . The observation that at least one of the edaphic variables proved to be important for each of the top performing models implies that edaphic variables are critical when modelling the distribution of A. lumbricoides. Similarly, previous studies [22, 27, 30] noted that soil pH, HPD, AVP, LST (day) and LST (night) were relatively important in predicting the distribution of A. lumbricoides after factoring in collinearity among predictor variables.
The observed consistency of high importance values for soil organic matter in all the top performing models are in line with the ecology of hookworms as the parasites feed on organic matter [53,54,55]. Thus, leaving out this edaphic variable in modelling the distribution of hookworms, likely leads to under-representation of the environmental niche within which these parasites thrive. Considering that with the advances in GIS and remote sensing technology, spatial data layers of organic matter content and other edaphic variables are now available in the public domain to modellers, the findings of this study open up opportunities to increase the accuracy of STH mapping at country scale. It is also important to note that DPW, HPD, NDVI, LST day and LST night were identified as important variables. This is in concurrence with previous studies which reported their importance in predicting the distribution of hookworms in different regions of the world [29, 30]. What makes this study different from others is the emphasis on edaphic variables, particularly soil pH and soil organic content, when predicting the distribution of hookworms in different geographical regions of the world.
From a disease management perspective, results of our study indicate a wide geographical distribution of A. lumbricoides and hookworms in Zimbabwe. High probabilities of presence values for A. lumbricoides were found in the northern and eastern districts in the country characterised by warm and moist conditions for the greater part of the year, which give rise to high vegetation cover if other factors, such as anthropogenic disturbance that change land cover, remain constant. In the case of hookworms, a wider distribution compared to that of A. lumbricoides was presented with highest probabilities of presence being reported in the northern, eastern and central districts of the country. Low probabilities of A. lumbricoides presence were found for districts in the southernmost, westernmost and northernmost districts. Since the parasitological results from Midzi et al.  were used in our study, it is not surprising that the findings in we observed some similarities in the distribution trend with the previous observations made at a national scale, i.e. that STH were predominantly distributed in the northern, northeastern and eastern regions, and scantly distributed in the western and south-western regions of Zimbabwe . The parasitological data used by this study were from primary school children aged 10–15 years .
Overall, this work underlines the importance of modelling for policy decisions as this can assist in risk assessment at low cost whilst producing quick results. Specifically, geospatial technology used in this study facilitated the production of the first continuous distribution maps for two problematic STHs in Zimbabwe. These continuous distribution maps have an advantage of showing variations within and across districts in the distribution of STH parasites including some of the districts which were not sampled during the 2010/11 national survey namely Gweru, Kwekwe, Chegutu, Shurugwi, Sanyati and Mhondoro-Ngezi. Thus, the current results complement previous work in which STH prevalence was mapped using point data . The results also show that the districts of Chimanimani, Nyanga, Mhondoro-Ngezi, Epworth, and Chitungwiza need to be added to the list associated with high A. lumbricoides prevalence. Likewise, in the case of hookworms, seven districts including Rusape, Hwedza, Nyanga, Chegutu, Mberengwa and a metropolitan province, Bulawayo, could be considered as high prevalence areas.
Although the findings from our study appear stable considering that ten modelling techniques were employed and model evaluation was based on two metrics, a limitation of the study is that other common STH species which are prevalent in Zimbabwe were not considered due to a lack of geo-referenced occurrence data. Thus, as these spatial data become available, it would be worthwhile to also test the effect of including edaphic variables on model performance when predicting the distribution of other STH such as Trichuris trichiura. This study was also conducted at a national scale with the aim to bolster policy formulation and hence fine scale variations in the distribution of STHs could have been missed. For instance, only distance from permanent water bodies was used to characterise the aquatic habitat of STHs but at the local scale, there are areas that get wet during parts of the year and depending on soil type and livelihoods activities (such as vegetable gardening) can provide suitable conditions for hookworms, especially in areas with poor sanitary conditions.
Another limitation of this study is that whilst the comparison of population densities in urban areas vs rural areas would act as a proxy of for the other related variables including sanitation and access to clean water, in this study we did not choose to analyse for these aspects for the following reasons: (i) a better analysis could have been accomplished if the data on these variables had been collected at the time of the study, and (ii) in Zimbabwe there are several development partners undertaking health development projects in some districts including water and sanitation provision. It is, however, unknown how these facilities are used by the communities of diverse cultures.
This study has shown that inclusion of edaphic predictors enhances model performance when predicting the geographical distribution of STHs. In addition, the study produced the first continuous distribution maps for two widely occurring STHs in Zimbabwe thus, confirming their wider distribution than previously thought.
Nokes C, Grantham-McGregor SM, Sawyer AW, Cooper ES, Bundy DAP. Parasitic helminth infection and cognitive function in school children. Proc R Soc Lond B Biol Sci. 1992;247:77–81.
Hall A, Hewitt G, Tuffrey V, De Silva N. A review and meta-analysis of the impact of intestinal worms on child growth and nutrition. Matern Child Nutr. 2008;4:118–236.
Uneke CJ. Soil-transmitted helminth infections and schistosomiasis in school age children in sub-Saharan Africa: efficacy of chemotherapeutic intervention since world health assembly resolution 2001. Tanzan J Health Res. 2010;12:86–99.
World Health Organization. Accelerating work to overcome the global impact of neglected tropical diseases: a roadmap for implementation. Geneva: WHO/HTM/NTD; 2012.
Augusto G, Magnussen P, Kristensen TK, Appleton CC, Vennervald BJ. The influence of transmission season on parasitological cure rates and intensity of infection after praziquantel treatment of Schistosoma haematobium-infected schoolchildren in Mozambique. Parasitology. 2009;136:1771–9.
World Health Organization. The prevention and control of schistosomiasis and soil-transmitted helminthiasis. Report of a WHO expert committee. WHO technical report series no.912. Geneva: WHO; 2002.
Bethony J, Brooker S, Albonico M, Geiger SM, Loukas A, Diemert D, Hotez PJ. Soil-transmitted helminth infections: ascariasis, trichuriasis, and hookworm. Lancet. 2006;367:1521–32.
Hotez PJ, Brindley P, Bethony JM, King CH, Pearce EJ, Jacobson J. Helminth infections:the great neglected tropical diseases. J Clin Invest. 2008;118:1311–21.
World Health Organization. Soil-transmitted helminthiases: eliminating soil-transmitted helminthiases as a public health problem in children progress report 2001–2010 and strategic plan 2011–2020. Geneva: WHO/HTM/NTD/PCT; 2012.
Pullan RL, Smith JL, Jasrasaria R, Brooker SJ. Global numbers of infection and disease burden of soil-transmitted helminth infections in 2010. Parasit Vectors. 2014;7:37.
de Silva NR, Brooker S, Hotez PJ, Montresor A, Engels D, Savioli L. Soil-transmitted helminth infections: updating the global picture. Trends Parasitol. 2003;19(12):547–51.
Anonymous. Soil-transmitted helminthiases: number of children treated in 2010. Wkly Epidemiol Rec. 2012;87(23):225–32.
WHO. Conducting a school deworming day: a manual for teachers. Geneva: World Health Organization; 2013.
Gabrielli A, Montresor A, Engels D, Savioli L. Preventive chemotherapy in human helminthiasis: theoretical and operational aspects. Trans R Soc Trop Med Hyg. 2011;105:683–93.
Bundy DAP, Chan MS, Savioli L. Hookworm infection in pregnancy. Trans R Soc Trop Med Hyg. 1995;89:521–2.
Utzinger J, Keiser J. Schistosomiasis and soil-transmitted helminthiasis: common drugs for treatment and control. Expert Opin Pharmacother. 2004;5:263–85.
Fifth-fourth World Health Assembly. 2001; http://apps.who.int/gb/archive/pdf_fi les/WHA54/ea54r19.Pdf accessed 30 July 2017.
WHO. Investing to overcome the global impact of neglected tropical diseases: third WHO report on neglected tropical diseases. Geneva: WHO; 2015.
Midzi N, Sangweme D, Zinyowera S, Mapingure MP, Brouwer KC, Munatsi A, et al. The burden of polyparasitism among primary schoolchildren in rural and farming areas in Zimbabwe. Trans R Soc Trop Med Hyg. 2008;102:1039–45.
Midzi N, Mduluza T, Chimbari MJ, Tshuma C, Charimari L, Mhlanga G, et al. Distribution of schistosomiasis and soil-transmitted helminthiasis in Zimbabwe: towards a national plan of action for control and elimination. PLoS Negl Trop Dis. 2014;8:e3014.
Harhay MO, Horton J, Olliaro PL. Epidemiology and control of human gastrointestinal parasites in children. Expert Rev Anti-Infect Ther. 2010;8:219–34.
Oluwole AS, Ekpo UF, Karagiannis-Voules DA, Abe EM, Olamiju FO, Isiyaku S, et al. Bayesian geostatistical model-based estimates of soil-transmitted helminth infection in Nigeria, including annual deworming requirements. PLoS Negl Trop Dis. 2015;9(4):e0003740.
Brooker S, Clements AC, Bundy DA. Global epidemiology, ecology and control of soil-transmitted helminth infections. Adv Parasitol. 2006;62:221–61.
Chiodini PL, Moody AH, Manser DW, Jeffrey HC. Atlas of medical helminthology and protozoology. Edinburgh: Churchill Livingstone; 2001.
Luong TV, MCIWEM, Water, Environment and Sanitation (WES) Programme. Prevention of intestinal worm infections through improved sanitation and hygiene. Thailand: UNICEF East Asia and Pacific Regional Office Bangkok; 2002. p. 1–26.
Ayanda OS, Ayanda OT, Adebayo FB. Intestinal nematodes: a review. Pac J Sci Tech. 2010;1:466–77.
Koroma JB, Peterson J, Gbakima AA, Nylander FE, Sahr F, Magalhães RJS, et al. Geographical distribution of intestinal schistosomiasis and soil-transmitted helminthiasis and preventive chemotherapy strategies in Sierra Leone. PLoS Negl Trop Dis. 2010;4:e891.
Pullan RL, Gething PW, Smith JL, Mwandawiro CS, Sturrock HJ, Gitonga CW, et al. Spatial modelling of soil-transmitted helminth infections in Kenya: a disease control planning tool. PLoS Negl Trop Dis. 2011;5:e958.
Lai Y-S, Zhou X-N, Utzinger J, Vounatsou P. Bayesian geostatistical modelling of soil-transmitted helminth survey data in the People’s republic of China. Parasit Vectors. 2013;6:359.
Chammartin F, Scholte RG, Malone JB, Bavia ME, Nieto P, Utzinger J, Vounatsou P. Modelling the geographical distribution of soil-transmitted helminth infections in Bolivia. Parasit Vectors. 2013;6:152.
Scholte RGC, Schur N, Bavia ME, Carvalho EM, Chammartin F, Utzinger J, Vounatsou P. Spatial analysis and risk mapping of soil-transmitted helminth infections in Brazil, using Bayesian geostatistical models. Geospat Health. 2013;8:97–110.
Karagiannis-Voules D-A, Biedermann P, Ekpo UF, Garba A, Langer E, Mathieu E, et al. Spatial and temporal distribution of soil-transmitted helminth infection in sub-Saharan Africa: a systematic review and geostatistical meta-analysis. Lancet Infect Dis. 2015;15:74–84.
Chandiwana SK. 1989. The problem and control of gastrointestinal helminths in Zimbabwe. Eur J Epidemiol. 1989;5(4):502–15.
Elith J, Leathwick JR. Species distribution models: ecological explanation and prediction across space and time. Ann Rev Ecol Evol Syst. 2009;40:677–97.
Zimbabwe Statistical Agency (ZIMSTAT). Census: National Report. Harare; 2012.
Gwitira I, Murwira A, Zengeya FM, Masocha M, Mutambu S. Modelled habitat suitability of a malaria causing vector (Anopheles arabiensis) relates well with human malaria incidences in Zimbabwe. Appl Geogr. 2015;60:130–8.
Shekede MD, Murwira A, Masocha M, Zengeya FM. Decadal changes in mean annual rainfall drive long-term changes in bush-encroached southern African savannas. Austr Ecol. 2016;41:690–700.
Nyamapfene KW. The soils of Zimbabwe. Harare: Nehanda Publishers; 1991.
Scoones I. The dynamics of soil fertility change: historical perspectives on environmental transformation from Zimbabwe. Geogr J. 1997;163(3):161–9.
Mapfumo RB, Murwira A, Masocha M, Andriani R. The relationship between satellite-derived indices and species diversity across African savanna ecosystems. Int J Appl Earth Obs Geoinfor. 2016;52:306–17.
Masocha M, Dube T. Relationship between native and exotic plant species at multiple savannah sites. Afr J Ecol. 2017; https://doi.org/10.1111/aje.12420.
Ngui AN, Apparicio P, Fleury MJ, Lesage A, Gregoire JP, Moisan J, Vanasse A. Spatio-temporal clustering of the incidence of schizophrenia in Quebec, Canada from 2004 to 2007. Spat Spatiotemporal Epidemiol. 2013;6:37–47.
Esri I. ArcGis version 9.3. Redlands: ESRI; 2008.
Socioeconomic Data and Applications Center. Data Center in NASA's Earth Observing System Data and Information System (EOSDIS). Hosted by CIESIN at Columbia University. 2013. sedac.ciesin.columbia.edu. Accessed 30 July 2017.
ISRI. Keep up with ISRIC - World soil information resource centre. 2012.www.isric.org. Accessed 30 July 2017.
Batjes NH. ISRIC-WISE global data set of derived soil properties on a 0.5 by 0.5 degree grid (version 3.0). Wageningen: ISRIC-World Soil Information; 2005.
Africa Soil Information Service. AfSIS newsletter; 2012. p. 2. africasoils.net. Accessed 30 July 2017
Dormann CF, Elith J, Bacher S, Buchmann C, Carl G, Carré G, et al. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. Ecography. 2013;36:027–46.
Thuiller W, Georges D, Engler R. biomod2: ensemble platform for species distribution modelling. R package version 3.0.3. 2013; http://CRAN R project Orgpackage Biomod2.
Thuiller W, Lafourcade B, Engler R, Araujo MB. BIOMOD - a platform for ensemble forecasting of species distributions. Ecography. 2009;32:369–73.
Phillips SJ, Anderson RP, Schapire RE. Maximum entropy modelling of species geographic distributions. Ecol Model. 2006;190:231–59.
Katz N, Chaves A, Pellegrino J. A simple device for quantitative stool thick-smear technique in Schistosoma mansoni. Rev Inst Med Trop Sao Paulo. 1972;14:397–400.
Goldberg WM, Lymburner R. Strongyloidiasis. Can Med Assoc J. 1951;65:152.
Donaldson RJ. Parasites and western man. Springer Science & Business Media: Lancaster; 2012.
Mabaso MLH, Appleton CC, Hughes JC, Gouws E. Hookworm (Necator americanus) transmission in inland areas of sandy soils in KwaZulu-Natal, South Africa. Tropical Med Int Health. 2004;9:471–6.
We would like to express our gratitude to the Permanent Secretaries for the Ministry of Health and Child Care of Zimbabwe, Brigadier General (Dr) G. Gwinji, and the Secretary for Primary and Secondary Education, Dr. S Utete-Masango, for the support and encouragement during the mapping of schistosomiasis and STH in Zimbabwe. Special acknowledgements go to the National Institute of Health Research, formerly the Blair Research Laboratory, for the expertise and commitment to quality research demonstrated by the research output in this manuscript. Our acknowledgements also go to parents and children for their approval and participation in the National schistosomiasis and STH survey respectively. Acknowledgements are also due to the World Health Organization for the technical support during the implementation of Phase 1 of the National Plan of Action for the Control of Schistosomiasis and STH in Zimbabwe whose results have been useful in conducting additional analysis performed in this study.
This study was funded by UNICEF, Helen Keller Foundation and the Ministry of Health and Child Care.
Availability of data and materials
Data supporting the conclusions of this article are included in the article. The datasets used and/or analysed during the current study are available from the corresponding author upon reasonable request.
Ethics approval and consent to participate
The proposal to conduct the national schistosomiasis and STH survey was approved by the national ethical review board, the Medical Research Council of Zimbabwe. The ethical approval number for the study MRCZ/A/1207 dated 11th March 2010. The Secretary for Education Sport Arts and Culture also approved the study. Written informed consent was sought from the parents/guardian of the study participants. UNICEF delivered parental/guardian informed consent forms addressed to each school by the Secretary for Education Sport Arts and Culture throughout the country in advance to allow school heads sufficient time to liaise with parents/guardians for their consent. On the day of sample collection, only the assenting children whose consent forms were signed by their parents/guardians participated in the study. Enrollment into the study was voluntary and participants were free to withdraw from the study at any time.
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.