Geospatial distribution and predictive modeling of onchocerciasis in Ogun State, Nigeria

Onchocerciasis caused by infection with Onchocerca volvulus is a disease of public health importance and is highly associated with disability. As Nigeria is aiming at eliminating onchocerciasis by 2030, there is a need to develop newer tools to map disease prevalence and identify environmental factors driving disease prevalence, even in places that have not been previously targeted for preventive chemotherapy. This study produced predictive risk-maps of onchocerciasis in Ogun State. Georeferenced onchocerciasis infection data obtained from a cross-sectional survey at 32 locations between March and July 2015 together with remotely-sensed environmental data were analyzed using Ecological Niche Models (ENM). A total of 107 field occurrence points for O. volvulus infection were recorded. A total of 43 positive occurrence points were used for modelling. ENMs were used to estimate the current geographic distribution of O. volvulus in Ogun State. Maximum Entropy distribution modeling (MaxEnt) was used for predicting the potential suitable habitats, using a portion of the occurrence records. A total of 19 environmental variables were used to model the potential geographical distribution area under current climatic conditions. Empirical prevalence of 9.3% was recorded in this study. The geospatial distribution of infection revealed that all communities in Odeda Local Government Area (a peri-urban LGA) showed remarkably high prevalence compared with other LGAs. The predicted high-risk areas (probability > 0.8) of O. volvulus infection were all parts of Odeda, Abeokuta South, and Abeokuta North, southern part of Imeko-Afon, a large part of Yewa North, some parts of Ewekoro and Obafemi-Owode LGAs. The estimated prevalence for these regions were >60% (between 61% and 100%). As predicted, O. volvulus occurrence showed a positive association with variables reflecting precipitation in Ogun State. Our predictive risk-maps has provided useful information for the elimination of onchocerciais, by identifying priority areas for delivery of intervention in Ogun State, Nigeria.

Onchocerciasis caused by infection with Onchocerca volvulus is a disease of public health importance and is highly associated with disability. As Nigeria is aiming at eliminating onchocerciasis by 2030, there is a need to develop newer tools to map disease prevalence and identify environmental factors driving disease prevalence, even in places that have not been previously targeted for preventive chemotherapy. This study produced predictive risk-maps of onchocerciasis in Ogun State. Georeferenced onchocerciasis infection data obtained from a cross-sectional survey at 32 locations between March and July 2015 together with remotely-sensed environmental data were analyzed using Ecological Niche Models (ENM). A total of 107 field occurrence points for O. volvulus infection were recorded. A total of 43 positive occurrence points were used for modelling. ENMs were used to estimate the current geographic distribution of O. volvulus in Ogun State. Maximum Entropy distribution modeling (MaxEnt) was used for predicting the potential suitable habitats, using a portion of the occurrence records. A total of 19 environmental variables were used to model the potential geographical distribution area under current climatic conditions. Empirical prevalence of 9.3% was recorded in this study. The geospatial distribution of infection revealed that all communities in Odeda Local Government Area (a peri-urban LGA) showed remarkably high prevalence compared with other LGAs. The predicted high-risk areas (probability > 0.8) of O. volvulus infection were all parts of Odeda, Abeokuta South, and Abeokuta North, southern part of Imeko-Afon, a large part of Yewa North, some parts of Ewekoro and Obafemi-Owode LGAs. The estimated prevalence for these regions were >60% (between 61% and 100%). As predicted, O. volvulus occurrence showed a positive association with variables reflecting precipitation in Ogun State. Our predictive risk-maps has provided useful information for the elimination of onchocerciais, by identifying priority areas for delivery of intervention in Ogun State, Nigeria.

Introduction
Onchocerciasis also known as river blindness is a tropical disease caused by the filarial nematode O. volvulus and transmitted through repeated bites of Simulium blackflies. The epidemiology of the disease is greatly influenced by the presence of adequate vector breeding sites [1]. The vector of onchocerciasis breeds in fast-flowing rivers where they attach to rocks, plants and suspended particles, or smaller river habitats of freshwater crab (Potamonautes spp.) [2,3]. Members of the Simulium damnosum complex have however been implicated as the only vectors of human onchocerciasis in West Africa [4]. Clinical manifestation of the diseases could either be ocular or dermal with symptoms such as nodules, itching, skin thickening, visual impairment, and blindness [5]. Onchocerciasis is of public health concern in 31 sub-Saharan African countries. However, Yemen, South America, Ecuador, Colombia, and Mexico have been verified by World Health Organization (WHO) to have successfully interrupted the transmission of O. volvulus [1]. Nigeria has one of the heaviest burdens of onchocerciasis in the world, accounting for almost one-third of the global prevalence [6,7]. Nodules removal (nodulectomy), vector control and mass drug administration (MDA) with ivermectin are some of the interventions that have been implemented in the control of morbidity of onchocerciasis and interruption of transmission of parasite with varying degrees of success [1,8]. In most areas of Africa, annual community-directed treatment with ivermectin (CDTI) is the primary intervention for control of onchocerciasis, with the exception of a few foci where semi-annual treatment is implemented [1]. In Ogun state, Nigeria, eight local government areas (LGAs) have been reported to be endemic for onchocerciasis, with two suitable vector habitats (Ogun and Osun rivers) with submerged rocks and vegetation, traversing several of the communities in these LGAs [7]. Following rapid epidemiological mapping of onchocerciasis (REMO), control strategy using CDTI have been implemented in Ogun state since 2004 [9], however, despite the intensified global efforts to eliminate onchocerciasis and the success achieved with MDA in Ogun state, the presence of microfilaria (mf) observed amongst adult population is a reflection that mass treatment may not really be effective [7], or a result of varying degrees of acceptability of medicines in the communities [10]. In line with the elimination targets set by WHO, there is a need to produce a model-based estimates of level of risk of Onchocerciasis across the communities or regions of Ogun state and this will invariably help policy makers to know where intervention strategies may require modifications and intensification. In this study, the geographical information system (GIS) and remote sensing (RS) technology were employed utilizing O. volvulus infection data from a cross-sectional survey across 8-endemic LGAs in Ogun State Nigeria, to produce model-based estimates of infection risk for the whole of the State.
Advancement in geospatial technology have made it possible to conduct disease mapping, surveillance, outbreaks prediction and evaluating infectious diseases spread pattern, particularly epidemic or pandemic prone-diseases [11][12][13]. Geospatial technology is now being adopted by experts in the public health field and decision makers to make visualization and analytical tools in order to carry out disease control programs in affected and/or suspected regions. It can also be used to perform analysis and predictions that were previously technologically impossible [13]. This technology is valuable as it is helpful in generating maps which are useful in comprehending the geographical distribution of disease for case frequency study, disease mapping and disease correlation with environmental factors and so on. These geospatial tools can forecast infection across a vast geographical zone by integrating a range of possible disease drivers which include remotely sensed climatic and environmental data in addition to the applicable socio-demographic information to advance model predictions [14].
In Nigeria, onchocerciasis programme has progressed from control to elimination in line with the objective of the WHO-Neglected Tropical Diseases (NTD) 2030 roadmap. To achieve elimination, there is a need to develop newer tools to map disease prevalence, identify sociodemographic and environmental factors driving disease prevalence and to forecast estimates of infection risk even in places that have not been previously targeted for preventive chemotherapy. In this study, we mapped the distribution of onchocerciasis (based on rapid diagnostic test (RDT) and skin snip data for mf prevalence) at community and local government level in Ogun state while also broadening the geostatistical methods to consider and integrate important environmental inferential covariates. The objective of this study was to present a geospatial map of onchocerciasis infection in Ogun state, identify the major environmental factors that correlate with onchocerciasis infection and to produce model-based estimates of infection risk of onchocerciasis for the whole of Ogun state.

Study area, design and population
This study was conducted in Ogun State, Nigeria (Fig 1). Details of the study area, design and population surveyed have been described elsewhere [7]. In brief, the study was carried out between March and July 2015 during the wet season. We designed a cross-sectional survey, and employed a systematic grid sampling method in the selection of communities to ensure an unbiased representation across the selected endemic LGAs [7]. A total of 3,895 children and adults, from 32 spatially selected communities participated in the study [7]. In each community, all households and their occupants were considered eligible for participation and invited to participate in the study. Only participants who gave their consent and were willing to participate were included in this study.

Onchocerciasis infection data
The field and laboratory procedures have been previously described in [7]. In brief, both RDT and skin snip methods were employed to determine prevalence of onchocerciasis. For the RDT, participants were serially arranged in a census form and the corresponding number was written on each Alere™ SD Bioline IgG4 test kit (Abbott Laboratories, Abbott Park, IL, USA) cassette using a permanent marker for easy identification. The kit was used according to the manufacturer's instruction [6]. For the skin snip examination, two bloodless skin biopsies of each participant were obtained from the left and right posterior iliac crest with the aid of a 2 mm corneoscleral punch (Holth and Modified Walser) and placed on a glass slide with a drop of saline water. The tissues in the slides were examined under a x40 binocular microscope after 30 min for micro-filarial manifestation. Each skin tissue was incubated in a 96-well microtitration plate containing a normal saline solution and when the column was full the wells were covered with a transparent adhesive film and kept for re-examination within 24 h. Mf results were expressed as positive (mf present) or negative (mf absent); mf prevalence was expressed as number of persons positive divided by the total number of persons examined [7,15].

Environmental data
The climatic variables such as temperature and precipitation influenced global and meso scales and topographic variables such as altitude and aspect that likely affects species distributions at meso and topo-scales while land-cover variables like percent canopy cover influence distributions at the micro-scale [16,17]. Hence, the use of the climatic and topographic variables in the prediction of distributions of O. volvulus infection was used in Ogun State.
To determine the possible distribution of O. volvulus infection in Ogun State using the prediction model, a total of 107 positive occurrences for O. volvulus infection were recorded from the field study. The autocorrelation problems were addressed by eliminating redundant presences on the scale of the bioclimatic variables used in each 1 × 1 km grid [18]. In addition, records for spatial autocorrelation were screened in ArcGIS 10.7.1 using average nearest neighbor analyses to remove spatially correlated data points [19,20]. After this selection, a total of 43 O. volvulus positive occurrence points were used to create a prediction model. We considered 19 environmental variables as potential predictors of the target species habitat distribution [21,22]. These variables were chosen based on their biological relevance to the target species distributions and other habitat modeling studies [17,[23][24][25][26]. Nineteen bioclimatic variables, biologically more meaningful to define eco-physiological tolerances of a species [27,28], with a 30 arc-second spatial resolution (about. 1 km 2 ) were downloaded from the WorldClim database (http://www.worldclim.org/) [29].
Then we utilized the standard geographic coordinates in decimal degrees (to four decimal places) in WGS 84. Then we spatialized and checked the geographic coordinates on Google Earth. After downloading the climatic files (covering the period 1950-2000), the Nigeria layer was extracted by using a boundary mask. After that, extracted files were converted to ASCII format via using ArcGIS 10.7.1 software to be use later with Maxent software.
All combinations of the 19 environmental variables have been tested for multi-collinearity through the calculation of R-squared in linear regression analysis in R software ver. 4.1.2. In this study, because some of these bioclimatic variables were strongly correlated (R 2 � 0.7), only those variables that showed little correlation with other predictors were retained; following [30,31]. A total of 8 environmental variables were selected in this study (R 2 < 0.7); Annual Mean Temperature (bio 1), mean diurnal range (max. temp-min. temp) (bio2), Isothermality (bio 3), min temperature of coldest month (bio 6), mean temperature of wettest quarter (bio 8), Annual Precipitation (bio12), Precipitation seasonality (Coefficient of variation) (bio 15), and precipitation of coldest quarter(bio19) ( Table 1).

Modeling procedure
The modeling technique maximum entropy distribution or Maxent were used in this study; which has been found to be the most effective among several different modeling methods [22,32,33], and may continue to be effective even with small sample sizes [21,25,[34][35][36]. For the study area, it only requires species presence data (not absence) and environmental variable (continuous or categorical) layers. We used the freely available Maxent software, version 3.3.3, which generates an estimate of the probability of the presence of the species that varies from 0 "unsuitable" to 0.99 "best habitat suitability". ASCII files of the 8 selected environmental variables and a CSV file of species presence coordinates in decimal degrees were used to create the module. Maxent's performance was assessed using a threshold independent Receiver-Operating Characteristic (ROC) analysis and Area Under Receiver-Operating Characteristic Curve (AUC) values (0.5 = random to 1 = perfect discrimination). The algorithm either runs 1000 iterations of these processes or continues until convergence is reached (threshold 0.00001).
For the model, the relative importance of each environmental predictor was evaluated using the percentage contribution of the Jackknife test, which is the best index for small sample sizes [25]. The default logistic output format was chosen, i.e. related to the probability of suitable conditions, ranging from 0 to 1. A total of 75% of the location point data were used for training, and the remaining 25% to test the predictive ability of the model, in addition 10 replicates were considered. Average and Standard deviation values for training and test AUC for the 10 models were extracted from the Maxent text result output. The ASCII output map for the average model for the target species was loaded in ArcGIS 10.7.1 where the prediction models of habitat suitability were divided based on Choudhury et al. (2016) into 5 classes; very low (0-0.1), low (>0.1-0.2), moderate (>0.2-0.4), high (>0.4-0.6) and very high (>0.6) using natural breaks in the symbology tools to produce the habitat suitability model picture [37].

Data summaries
A total of 3,895 infection data was included in the survey. The demographic characteristics of the study population have been described elsewhere [7]. However, Table 2

Geospatial distribution of onchocerciasis by age and sex across the selected LGAs
The distribution of O. volvulus infection by age and sex is presented in Fig 3. The results showed that there was no regular pattern of infection with respect to age across all the LGAs. For instance, prevalence was generally higher (p<0.05) among adults (25 years and above) compared with those below 25 years at the two highly onchocerciasis endemic LGAs (Odeda and Abeokuta North), while there seems to be no difference (p>0.05) in prevalence of O. volvulus infection with respect to age across the other LGAs (Fig 3).  In the same vein, there was no observable pattern in prevalence of O. volvulus infection with respect to gender across all the LGAs. Prevalence was higher among females in some LGAs (Imeko-Afon and Odeda) while it was higher in males in other LGAs (Abeokuta North, Ewekoro, Abeokuta South, Ifo and Obafemi-Owode) (Fig 3).

Logistic regression model for O. volvulus infection risk
Except for sex, all other covariates, in  LGAs respectively compared with Ifo LGA, while risk of infection was 8 and 10 times higher among people from Abeokuta South (COR = 9.2, AOR = 7.8) and Ewekoro (COR = 8.7, AOR = 9.7) ( Table 3).
Those who are within the age range of 5-14years were 0.3 times likely to be infected with O. volvulus, while risk of infection was 0.8, 1.3, and 1.1 times higher in those who are within the age range of 15-24years, 25-34years and 35-44years respectively compared with those who are above 44years. Furthermore, risk of infection was 2-folds higher (COR = 0.2, AOR = 1.5) among those who did not receive preventive chemotherapy (ivermectin) compared with those who received (Table 3).

Predicted risk of onchocerciasis in Ogun State
The predicted high-risk areas (probability > 0.8) of O. volvulus infection were all parts of Odeda, Abeokuta South, and Abeokuta North, southern part of Imeko-Afon, a large part of Yewa North, some parts of Ewekoro and Obafemi-Owode LGAs. The estimated prevalence for these regions were >60% (between 61% and 100%). The prediction showed that some part of Yewa North was expected to have high prevalence of onchocerciasis in contrast to the geospatial distribution with zero percent prevalence from the sampled location (which differs from the area predicted for high prevalence) (Fig 4).

Model performance and influencing factors
The average percent contribution (PC) and permutation importance (PI) of the 8 variables used in the modeling of onchocerciasis in this study were also assessed. In this study, precipitation seasonality (coefficient of variation) had the highest PC and PI of 53.6 and 49.7 respectively, followed by annual precipitation with PC of 43.5 and PI of 43.0 and precipitation of coldest quarter with PC of 1.7 and PI of 7.3. The results showed that both precipitation seasonality (coefficient of variation) and annual precipitation are strong predictors of O. volvulus distribution in Ogun State, accounting for 97.1% of the variations in distribution observed ( Table 4). The ROC curve obtained as an average of the 10 replications runs is shown in Fig 5, and specificity was calculated. The average and standard deviation of the AUC for the 10 replicate The jackknife test also showed that bio15 (precipitation seasonality (coefficient of variation) and Bio 12 (annual precipitation) are the two variables that will affect the prevalence of onchocerciasis when used alone. The jackknife test also showed that bio15 (precipitation seasonality (coefficient of variation) decreased the gain the most when removed from the model. Figs 6 and 7 shows the main highest estimated environmental variables (contributions) that determines the prevalence of O. volvulus infection in Ogun State. Spatial distribution analysis was done to determine the geographical variability with regards to the selected environmental variables in the State. The response curves of three variables to O. volvulus habitat suitability are shown in (Fig 8). We found out that precipitation seasonality (bio15) ranged from 61 to 77 mm, annual precipitation (bio12) ranged from 1901 to 2014 mm while precipitation of the coldest quarter (bio19) ranged from 323 to 704 mm. The response curves showed that between annual precipitation of 1100 and 1200 mm favors the potential transmission of O. volvulus. Similarly, precipitation seasonality of 60 to 64 mm significantly and potentially favoured the distribution of O. volvulus while precipitation of the coldest quarter ranging between 400 and 450 mm significantly favoured the transmission of O. volvulus in Ogun State (Fig 8).

Discussion
In this study, we utilized O. volvulus infection data from a cross-sectional survey across eightendemic LGAs in Ogun State Nigeria, to produce model-based estimates of infection risk for the whole State. To the best of our knowledge, this is the first predictive geospatial-model for prevalence of O. volvulus infection in Ogun State, using empirical data.
The empirical prevalence of 9.3% recorded in this study, with higher prevalence in certain LGAs is an indication that elimination of onchocerciasis may still be a problem in Ogun State if control efforts are not geared up. The geospatial distribution of infection revealed that all communities in Odeda LGA (a peri-urban LGA) showed remarkably high prevalence compared with other LGAs having either low or zero prevalence at the community level. Our

PLOS ONE regression model further suggested an incredibly high risk of infection in Odeda
LGA. This has serious implications for elimination. MDA has been on-going for over ten years in the state [7,38], it is quite worrisome that prevalence among children 5-14 years is still incredibly high or at par with adults across different LGAs in the State. This result showed that transmission is still actively on-going in the affected LGAs despite on-going MDA. This may be as a result of poor uptake of preventive chemotherapy with ivermectin observed among the study population. Our regression model also indicated a double-fold increase in risk of infection among those who did not take ivermectin compared with those who did. Studies have reported the efficacy of ivermectin in reducing mf load and subsequently reduce transmission [39,40]. Therefore, there is a need to investigate the reasons for this shortfall and how to subsequently improve geographical coverage if we are to meet the elimination target.
Modelling the baseline endemicity of infection is very germane in determining where and how to scale up control programmes. In order to raise adequate awareness and develop proper framework for advocacy in areas of community/institutional engagement, there is dire need to have good knowledge of the extent of disease burden [41]. In view of this, we used our model to predict the hotspots for onchocerciasis in Ogun State. It is important to note that predicted prevalence in locations that are not currently occupied by humans can increase risk of transmission when people with infection establish communities in such places [42].
Our model showed varying degrees of predicted prevalence across Ogun State. The predicted high-risk areas (probability >0.8) of onchocerciasis infections were all parts of Odeda, Abeokuta South, and Abeokuta North, southern part of Imeko-Afon, a large part of Yewa, some parts of Ewekoro and Obafemi-Owode LGAs. One thing that these LGAs had in common and could possibly explain the high prevalence recorded in this study is the availability of fast flowing rivers which is a potent breeding sites responsible for consistent bites by the S. damnosum s.l [7]. The higher risk of infections in Odeda communities, Abeokuta South, and Obafemi-Owode can be attributed to their proximity to Ogun river, while that of Abeokuta North and Yewa North is the proximity to Arakanga and Yewa river respectively. It is therefore expected to have higher risk of onchocerciais prevalence in communities along river system due to the preponderance of the Simulium vectors in these areas [43].
The prediction showed that some part of Yewa North was expected to have high prevalence of onchocerciasis in contrast to the geospatial distribution with 0 prevalence from the sampled location (which differs from the area predicted for high prevalence). This is quite possible and this calls for proper evaluation. Eight LGAs are known to be endemic for onchocerciais in Ogun State, however, a prevalence of 26.4% was reported by [44] in a community in Ijebu North East, a LGA that was thought to be non-endemic for onchocerciasis. It is therefore important to extend control programmes across all the LGAs if we want to effectively disrupt onchocerciais transmission in Ogun State. We also noticed in our predictive maps that some LGAs (that were previously considered as non-endemic) with high infection risk were from areas sharing boundaries with adjoining highly endemic LGAs. This suggest that when planning for control programmes, communities sharing boundaries with highly endemic LGAs should be given as much attention as possible if we are to drive the elimination of onchocerciasis in Ogun State.
It is equally important to stress, that our model revealed that transmission could have spread to additional eight LGAs in Ogun State, with low to moderate infection risk. This finding corroborates that of [45], who recently reported a prevalence of 1.64% in Yewa South LGA, which was presumably considered as non-endemic. The findings from this study should be taken as warning alert, requiring drastic and deliberate actions in other not to frustrate the existing and limited control efforts within the state.
Three variables; precipitation seasonality (coefficient of variation), annual precipitation and precipitation of coldest quarter were strongly associated with prevalence of onchocerciasis in this study. A study by [46] also stressed the association between precipitation and prevalence of onchocerciasis. This finding suggests that optimum condition for breeding of blackflies and transmission of onchocerciais may be present in areas which the majority of precipitation occurs in the cooler parts of the year.
This study has predicted the prevalence of O. volvulus infection in Ogun State using a robust geospatial modeling approach, and as well the spatial pattern of disease spread. The empirical and predicted prevalence for O. volvulus infection was very high at some LGAs while infection may have spread to previously thought non-endemic LGAs. The empirical data used for this model was limited to 8 LGAs (32 communities) with 107 occurrence data points, hence may have present a limitation to this rigorousness of the model if the value for AUC is not up to 0.75. The MaxEnt modeling used in this study is a general-purpose method for making predictions of inferences from incomplete information [16,27]. The model used in this study is a present only modeling algorithm (i.e. absence data are not required). More so, the performance has been reported to be relatively better than other modeling methods [17,32]. The report that the model has been hardly influenced by small sample sizes with relatively robust prediction hence putting it among the top performing modeling tools [32] made it the choice model for our study.

Conclusion
In Nigeria, the focus of onchocerciasis programs has shifted from control to elimination of O. volvulus transmission, prompting the need to employ newer tools to aid in efficient prioritization of decisions regarding elimination mapping and interventions [42] especially in areas that were not previously considered for intervention. As a result, we have generated a baseline prevalence map for Ogun State using geospatial modeling technique. Knowing that elimination mapping of any disease can be expensive, hence the map presented here could be used by the state onchocerciasis elimination programs to direct resources for elimination mapping. Coupled with that fact the method described here may be an inexpensive first step that can extrapolate state-wide prevalence from existing data, the national onchocerciasis elimination programs can also employ this mean in providing a better situational analysis of onchocerciasis in Nigeria from known prevalence data.