Predicting Schistosomiasis Intensity in Africa: A Machine Learning Approach to Evaluate the Progress of WHO Roadmap 2030

Xinyue Chen; Jiaxu Le; Yi Hu

doi:10.4269/ajtmh.23-0751

. 2024 May 21;111(1):73–79. doi: 10.4269/ajtmh.23-0751

Predicting Schistosomiasis Intensity in Africa: A Machine Learning Approach to Evaluate the Progress of WHO Roadmap 2030

Xinyue Chen ¹, Jiaxu Le ¹, Yi Hu ^1,^*

PMCID: PMC11229641 PMID: 38772355

ABSTRACT.

The World Health Organization (WHO) 2030 Roadmap aims to eliminate schistosomiasis as a public health issue, targeting reductions in the heavy intensity of infections. Previous studies, however, have predominantly used prevalence as the primary indicator of schistosomiasis. We introduce several machine learning (ML) algorithms to predict infection intensity categories, using morbidity prevalence, with the aim of assessing the elimination of schistosomiasis in Africa, as outlined by the WHO. We obtained morbidity prevalence and infection intensity data from the Expanded Special Project to Eliminate Neglected Tropical Diseases, which spans 12 countries in sub-Saharan Africa. We then used a series of ML algorithms to predict the prevalence of infection intensity categories for Schistosoma haematobium and Schistosoma mansoni, with morbidity prevalence and several relevant environmental and demographic covariates from remote-sensing sources. The optimal model had high accuracy and stability; it achieved a mean absolute error (MAE) of 0.02, a root mean square error (RMSE) of 0.05, and a coefficient of determination (R²) of 0.84 in predicting heavy-intensity prevalence for S. mansoni; and an MAE of 0.02, an RMSE of 0.04, and an R² value of 0.81 for S. haematobium. Based on this optimal model, we found that most areas in the surveyed countries have not achieved the target of the WHO road map for 2030. The ML algorithms used in our analysis showed a high overall predictive power in estimating infection intensity for each species, and our methods provided a low-cost, effective approach to evaluating the disease target in Africa set in the WHO road map for 2030.

INTRODUCTION

Schistosomiasis, a complex water-borne parasitic disease caused by blood flukes, includes five species that parasitize humans globally: Schistosoma haematobium, Schistosoma mansoni, Schistosoma japonicum, Schistosoma intercalatum, and Schistosoma mekongi. In 2019, it was estimated that schistosomiasis infected over 140 million people worldwide, causing approximately 24,000 deaths and imposing a burden of 2.5 million disability-adjusted life-years (DALYs) in 2016.¹ Predominantly found in tropical and subtropical regions, schistosomiasis ranks as the second most common parasitic disease impacting public health after malaria.² More than 90% of cases occur in sub-Saharan Africa.³ The predominant species in these regions are S. mansoni, transmitted via feces and responsible for intestinal and hepatic schistosomiasis, and S. haematobium, transmitted via urine, leading to urogenital schistosomiasis.⁴

To mitigate the health and economic damage caused by schistosomiasis, the WHO has made a number of recommendations and measures to eliminate schistosomiasis, including preventive chemotherapy; water, sanitation, and hygiene (WASH) (more precisely, access to safe water, improved sanitation and management of excreta across communities, and individual hygiene education), and vector control.⁵ In Africa, by 2020, 67% of school-age children have been covered with preventive chemotherapy in some regions with moderate to severe transmission.⁶ Nonpharmacological intervention measures are also promoted actively under the leadership of communities and schools. For example, comprehensive interventions such as environmental sanitation, WASH, health education, economic empowerment, and teachers from the formal education system and religious institutions (such as madrassas in schistosomiasis control programs) help in reducing the prevalence of schistosomiasis.⁵^,⁷ Despite the combined approach of drug treatment and nonpharmacological interventions being adopted in Africa, the disease remains severe on the continent. In some endemic areas, the prevalence reached 50%, and the proportion of the population requiring preventive chemotherapy was >10% in 2019.⁸^,⁹ In this context, the WHO set an ambitious goal, in January 2021, to eliminate schistosomiasis as a public health problem (currently defined as <1% proportion of heavy-intensity schistosomiasis infections) in all 78 endemic countries by 2030, which was included in the WHO’s 2021 to 2030 road map for neglected tropical diseases.⁸ Unfortunately, only morbidity prevalence was produced in most studies that evaluated the progress of schistosomiasis control programs,⁹^–¹³ prohibiting the accurate assessment of target achievements set by the WHO road map. Hence, timely and accurate estimates of the prevalence of infection intensity categories, particularly heavy intensity, are urgently needed.

Previous studies¹⁴^,¹⁵ showed there were associations between prevalence of infection intensity categories and morbidity prevalence for S. haematobium, but no associations for S. mansoni.¹⁶ In addition, prevalence of infection intensity categories of schistosomiasis was found to be influenced by a combination of biological factors (e.g., pathogens, hosts and vector organisms) and socioeconomic factors (e.g., gross domestic product, population and under 5 mortality rate).¹⁷^,¹⁸ These indicated that morbidity prevalence and biological and socioeconomic covariates are potential predictors for prevalence of infection intensity categories. In this study, we use a series of machine learning (ML) algorithms to investigate the potential link between prevalence of infection intensity categories for schistosomiasis and morbidity prevalence, as well as some environmental and socioeconomic covariates in Africa. Machine learning techniques have found extensive application in disease diagnosis,¹⁹^,²⁰ assessment of disease progression,²¹ disease early warning,²² and disease risk mapping²³ as a result of their capability to detect nonlinear and interrelated relationships among different variables.

MATERIALS AND METHODS

Prevalence and intensity data.

Survey data on the prevalence and infection intensity of schistosomiasis for S. mansoni and S. haematobium were obtained from the Expanded Special Project to Eliminate Neglected Tropical Diseases (ESPEN; https://espen.afro.who.int/) developed by the WHO Africa Regional Office in 2016. We collected the following information from the ESPEN database: number of people assessed, morbidity prevalence (or number of infected), percentage of different intensities, schistosome species investigated, year of the survey, and name and coordinates of the survey site. Data points with a missing survey year, those that could not be geocoded to a specific implementation unit, those that lacked intensity data, and those with a prevalence that was zero were disregarded in our analyses. Finally, data from 3,364 georeferenced surveys from 2004 to 2018 were incorporated, including 656 for S. mansoni and 2,708 for S. haematobium, spanning 12 countries: Burkina Faso, Burundi, Cote d’Ivoire, the Democratic Republic of the Congo, Guinea, Liberia, Madagascar, Malawi, Niger, South Africa, Tanzania, and Togo (Figure 1). More information about the survey can be found in the Supplemental Table 1.

Figure 1. — Location of the sample point. (A) The sample point of *Schistosoma mansoni*. (B) The sample point of *Schistosoma haematobium*.

The WHO’s current widely used criteria for classifying the prevalence of infection intensity categories of schistosomiasis are based on the number of schistosome eggs in the urine or feces of infected people. Schistosoma haematobium infection intensity has consistently been characterized by the number of schistosome eggs per 10 mL of urine, with 1 to 49 eggs per 10 mL of urine defining a light infection and >50 eggs per 10 mL of urine indicating a heavy infection.²⁴ For S. mansoni, infection intensity is measured as the number of schistosome eggs per gram (EPG) of stool. Infection intensity is now commonly split into three categories: 1 to 99 EPG for a light infection, 100 to 399 EPG signifying a moderate infection, and >400 EPG for a heavy infection.²⁴

Environmental, socioeconomic, and mass drug administration data.

Environmental data included the distance to water bodies, daytime land surface temperature, average annual precipitation, normalized digital vegetation index (NDVI), and vapor pressure; socioeconomic data contained gross domestic product (GDP), population, and lack of access to sanitation based on latent project assessment. The environmental data were all calculated in ArcGlS 10.0 (ESRl, Redlands, CA), with distance to water bodies calculated specifically by the distances from sample points to major rivers and their branches in Africa. Detailed sources of these data, as well as time and space resolution, are provided in Supplemental Table 2. The number of mass drug administrations (MDAs) was obtained from the ESPEN database, which counts the number of MDAs performed at surveillance sites since 2013.

STATISTICAL ANALYSES

We used six ML algorithms to predict the prevalence of infection intensity categories: multivariate adaptive regression splines (MARS), least absolute shrinkage and selection operator (LASSO), random forest (RF), support vector machine (SVM) with a polynomial kernel, back-propagation artificial neural network (BP-ANN), and an Ensemble model that integrates these five models using the generalized linear model approach. Details on the ML algorithms are included in Supplemental Appendix 1. We also conducted an analysis of importance ranking to identify effects of covariates on infection intensity.

For model validation, we divided all data randomly into three subsets: 60% for training, 20% for validation, and 20% for testing. The training and validation subsets were used to train the models, fine-tune parameters, rank the relevant covariates, and evaluate model performance using 10-fold cross-validation. The independent test subset was used to assess the generalization capability of the models. Three indicators were used to select the optimal model: the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Details on the model validation can be found in Supplemental Appendix 1. All model fitting was performed using R software (R v. 4.3.0 Foundation for Statistical Computing, Vienna, Austria), specifically, the caret package.

To assess the progress of WHO’s 2030 road map in the study area, we first used inverse distance-weighted²⁵ interpolation to map morbidity prevalence at a 5- × 5-km resolution. We then used the optimal ML model to map the morbidity intensity across the study area.

RESULTS

In Figure 2, we show the model-fitting results of heavy infection intensity on training data for each ML model for S. mansoni (Figure 2A) and S. haematobium (Figure 2B). It shows that the best-fitting model for S. mansoni was the SVM model, with an average MAE of 0.02, an RMSE of 0.05, and an R² value of 0.75, and each model has a small result range, except for the BP-ANN model. For S. haematobium, the best-fitting model was the RF model, with an average MAE of 0.02, an RMSE of 0.04, and an R² value of 0.77, and each model has a smaller result range than S. mansoni. Results of model comparisons for light and moderate infection intensities for S. mansoni and light infection intensity for S. haematobium are included in Supplemental Figures 1 and 2.

Results of the ranking of covariate importance based on all surveyed data show that morbidity prevalence was the most crucial predictor, regardless of the schistosome species (Figure 2C and D), holding the highest level of importance in influencing the model’s outcomes. The remaining covariates, such as GDP, population, and some environmental factors, albeit slightly less impactful than morbidity prevalence, still demonstrate relevance as determinants for the outcomes.

Table 1 shows the model accuracy based on test data for both species. For S. mansoni, the SVM model outperformed others in predicting the light intensity of infection, with an MAE of 0.04, an RMSE of 0.06, and an R² value of 0.83. The Ensemble model outperformed others in predicting the moderate intensity of infection, with an MAE of 0.03, an RMSE of 0.04, and an R² value of 0.65. The Ensemble model also outperformed others in predicting the heavy intensity of infection, with an MAE of 0.02, an RMSE of 0.05, and an R² value of 0.84. For S. haematobium, the RF model outperformed others in predicting the light intensity of infection, with an MAE of 0.02, an RMSE of 0.04, and an R² value of 0.91. The RF model also outperformed others in predicting the heavy intensity of infection, with an MAE of 0.02, an RMSE of 0.04, and an R² value of 0.81. The scatter plot of the model results in the test set for S. haematobium and S. mansoni can be viewed in Supplemental Figures 3 through 7.

Table 1.

Statistical comparison of prediction accuracy among different models in test data

Infection Intensity	Schistosoma mansoni				Schistosoma haematobium
Infection Intensity	Model	R ²	MAE	RMSE	Model	R ²	MAE	RMSE
Light	MARS	0.767	3.321E-02	5.952E-02	MARS	0.896	2.653E-02	4.774E-02
	LASSO	0.635	4.850E-02	7.527E-02	LASSO	0.902	3.039E-02	4.707E-02
	BP-ANN	0.764	4.138E-02	6.269E-02	BP-ANN	0.876	2.489E-02	5.379E-02
	SVM ^*	0.827	3.779E-02	5.711E-02	SVM	0.868	3.045E-02	5.543E-02
	RF	0.822	4.005E-02	6.356E-02	RF	0.914	2.378E-02	4.377E-02
	Ensemble	0.824	3.304E-02	5.777E-02	Ensemble	0.895	2.032E-02	4.834E-02
Moderate	MARS	0.611	3.015E-02	4.121E-02	–	–	–	–
	LASSO	0.591	2.878E-02	4.205E-02	–	–	–	–
	BP-ANN	0.413	3.529E-02	5.547E-02	–	–	–	–
	SVM	0.617	2.881E-02	4.518E-02	–	–	–	–
	RF	0.662	2.878E-02	4.303E-02	–	–	–	–
	Ensemble	0.654	2.642E-02	3.970E-02	–	–	–	–
Heavy	MARS	0.700	3.018E-02	6.076E-02	MARS	0.631	2.867E-02	5.588E-02
	LASSO	0.649	3.697E-02	7.045E-02	LASSO	0.506	3.074E-02	6.388E-02
	BP-ANN	0.725	3.041E-02	6.879E-02	BP-ANN	0.753	2.313E-02	5.002E-02
	SVM	0.731	2.934E-02	5.722E-02	SVM	0.802	2.058E-02	4.014E-02
	RF	0.812	2.652E-02	5.499E-02	RF	0.805	2.059E-02	3.940E-02
	Ensemble	0.840	2.227E-02	4.550E-02	Ensemble	0.800	1.974E-02	4.195E-02

Open in a new tab

BP-ANN = back-propagation artificial neural network; LASSO = least absolute shrinkage and selection operator; MAE = mean absolute error; MARS = multivariate adaptive regression splines; R² = the coefficient of determination; RF = random forest; RMSE = root mean square error; SVM = support vector machines.

Bold font denotes the best result of this statistical indicator.

The predicted morbidity prevalence of schistosomiasis in these 12 surveyed countries from 2004 to 2018 are shown in Figure 3A and B. The morbidity prevalence of S. haematobium is >20% in the northeastern region of Burkina Faso, the southwestern region of Niger, and the central region of Côte d’Ivoire. In most areas of Madagascar, it exceeds 20%, and in the southern and northwestern regions, it surpasses 50% (Figure 3A). The risk of S. mansoni infection is notably less in these countries compared with S. haematobium. For example, the morbidity prevalence exceeds 20% in the vast majority of areas in Liberia, the western region of Côte d’Ivoire, and the central and southern regions of Madagascar. The morbidity prevalence is >50% in some areas of southwestern Madagascar (Figure 3B). The heavy-intensity prevalence based on the Ensemble model is shown in Figure 3C and D. We found that a significant proportion of regions fall short of meeting the WHO 2030 road map target (i.e., a heavy-intensity schistosomiasis infection of <1%), although the majority of regions exhibit a heavy-intensity prevalence <5%. Specifically, the central region of Madagascar and the northern region of South Africa stand out as areas with heavy-intensity prevalences exceeding 5% for S. haematobium (Figure 3C). The heavy-intensity prevalence of S. mansoni is generally less than S. haematobium, with only a small portion of northern South Africa and central–southern Madagascar exceeding 5% (Figure 3D). Figure 3E and F illustrate the uncertainty associated with our model’s predictions for the prevalence of high intensity, as depicted by the magnitude of the 95% confidence interval range. The results show that areas with greater predicted heavy-intensity prevalence also have greater uncertainty in the model predictions. Specifically, the model predictions for S. haematobium show greater uncertainty in central Congo, northern South Africa, and central Madagascar (Figure 3E). For S. mansoni, the model predictions exhibit elevated uncertainty only in the northern part of South Africa and central Madagascar (Figure 3F).

DISCUSSION

The prevalence of heavy-intensity infection for schistosomiasis, representing a significant health and economic burden, has emerged as a new indicator of concern and a crucial criterion for the WHO to judge the elimination of schistosomiasis.⁵ Previous studies have seldom focused on this indicator. Moreover, endeavors to explore the prevalence of infection intensity categories have been limited to localized surveys. Consequently, determining the precise prevalence of these categories on a national scale or across the African continent poses a formidable challenge.²⁶^,²⁷ Our study integrated data on morbidity prevalence and infection intensity categories obtained from small-scale surveys, along with local climate, and economic and other relevant variables. Using ML algorithms, we established models capable of predicting the prevalence of different-intensity infections based on morbidity prevalence and environmental variables.

Our study indicates that SVM and RF generally outperformed the linear-based MARS and LASSO models, indicating that the relationship between prevalence of infection intensity categories and predictors is more complex than a simple linear correlation. The relationship between schistosomiasis prevalence of infection intensity categories and morbidity prevalence was more significant for S. haematobium than for S. mansoni.¹⁴ In our study, all models for S. haematobium outperformed those for S. mansoni. Schistosoma haematobium infection intensity categories were divided into two levels, as opposed to three levels for S. mansoni, which may result in a more stable and accurate model for S. haematobium.

We found that morbidity prevalence was the most influential predictor of prevalence of infection intensity categories, with other predictors having small effects on it (Figure 2C and D). We took these environmental and socioeconomic covariates into account, because they can affect the distribution, density, and activity of schistosomiasis vectors,¹⁶^,¹⁷ which in turn can affect indirectly the prevalence of schistosomiasis. The influence of the distance to water bodies was found to be notably more pronounced for S. mansoni compared with S. haematobium, probably because the link between the intermediate hosts of S. mansoni and the distance to water bodies is more significant compared with S. haematobium.²⁸ In our model, factors such as the daytime land surface temperature, NDVI, GDP, and precipitation were also among the top-ranked predictive variables. Although there is no established relationship among these factors and infection intensity, they do exert a significant influence on the morbidity prevalence of the disease.¹⁰ As a result, these factors also play a relatively important role in our model.

Figure 3C and D illustrates that most areas in the surveyed countries have not achieved the target of the WHO’s 2030 road map. In the majority of regions with a high proportion of heavy-intensity prevalence, there is also a high morbidity prevalence of schistosomiasis infection. For example, in the central region of Madagascar, the morbidity prevalence for both S. haematobium and S. mansoni exceeds 20%, and the prevalence of heavy-intensity infection exceeds 5%, exemplifying the urgent need to meet the WHO’s targets in these areas. To address this challenge effectively, an imperative measure involves the control of schistosomiasis morbidity prevalence. Strategies include escalating the frequency of MDA campaigns and broadening their scope to encompass the adult population. Simultaneously, the implementation of comprehensive health education programs emphasizing WASH practices is advocated, which may result in a decline in schistosomiasis morbidity prevalence.²^,⁸ In regions exhibiting low morbidity prevalence but a high heavy intensity, multiple factors, including population composition and mobility, and environmental conditions, might play a role. For example, South Africa is developed economically, but it confronts challenges associated with significant population mobility and inequitable access to sanitation resources.²⁹ These factors potentially contribute to elevated heavy-intensity prevalence. In developing countries, such as the Democratic Republic of the Congo and Tanzania, inadequate health-care resources and limited access to clean water sources are likely contributing factors to the high prevalence of heavy-intensity cases.³⁰^,³¹ Consequently, there is an urgent need to enhance health-care services and expand the coverage of clean water resources in these areas.

Timely identification of the prevalence of infection intensity categories is of great importance, not only in achieving the WHO’s road map targets for 2030, but also in guiding the allocation of medical resources in resource-constrained areas. Numerous studies have demonstrated a high incidence of symptoms such as bloody urine, bladder fibrosis, and ureter and kidney damage in cases of S. haematobium heavy infection,³²^,³³ whereas S. mansoni heavy infection has been associated with increased rates of liver and spleen enlargement, fibrosis, portal hypertension, and peritoneal fluid accumulation.²⁸ The irreversible damage to these organs poses significant risks to health, including potential fatality. Prompt intervention in areas with high rates of heavy-intensity infection has the potential to improve disease prognosis, enhance local population health, and mitigate health and economic burdens.

Two limitations of this study need further discussion. First, previous studies³⁴^,³⁵ indicated that other individual-level factors such as the age, sex, and area of residence of the infected person affect the intensity of infection in patients, but in our study we were unable to include this information on an individual level in our models because our study was based on an ecological study. Second, although our model was highly accurate in predicting the intensity of infection of S. mansoni and S. haematobium, it is unknown how well it would have performed in predicting the intensity of infection of other species of schistosome, such as S. japonicum or S. intercalatum. Additional studies are needed to validate the generalizability of the ML algorithms. After further refinement of our model, we plan to simplify it into a more user-friendly tool for program managers to use.

CONCLUSION

In summary, the ML algorithms proposed in our study will enable national control programmers to estimate the prevalence of infection intensity categories starting from the total prevalence of schistosomes. The use of our model provides an updated and efficient tool for researchers, policymakers, and program implementers to prioritize areas for assessment, evaluate local infection status, and target interventions strategically. This advancement enhances the possibility for countries to work effectively toward achieving the WHO’s road map targets by 2030.

Supplemental Materials

tpmd230751.SD1.pdf^{(2.4MB, pdf)}

DOI: 10.4269/ajtmh.23-0751

Note: Supplemental material appears at www.ajtmh.org.

Data Availability

All data used in this study are available on request from the corresponding author.

REFERENCES

1. Hotez PJ. et al. , 2014. The Global Burden of Disease Study 2010: Interpretation and implications for the neglected tropical diseases. PLoS Negl Trop Dis 8: e2865. [DOI] [PMC free article] [PubMed] [Google Scholar]
2. WHO , 2023. Schistosomiasis: Key Facts. Available at: https://www.who.int/news-room/fact-sheets/detail/schistosomiasis. Accessed April 17, 2019.
3. Karagiannis-Voules DA. et al. , 2015. Spatial and temporal distribution of soil-transmitted helminth infection in sub-Saharan Africa: A systematic review and geostatistical meta-analysis. Lancet Infect Dis 15: 74–84. [DOI] [PubMed] [Google Scholar]
4. Gryseels B, Polman K, Clerinx J, Kestens L, 2006. Human schistosomiasis. Lancet 368: 1106–1118. [DOI] [PubMed] [Google Scholar]
5. World Health Organization , 2011. Helminth Control in School-Age Children: A Guide for Managers of Control Programmes, 2nd edition. Geneva, Switzerland: WHO. [Google Scholar]
6. Hürlimann E. et al. , 2018. Effect of an integrated intervention package of preventive chemotherapy, community-led total sanitation and health education on the prevalence of helminth and intestinal protozoa infections in Côte d’Ivoire. Parasit Vectors 11: 115. [DOI] [PMC free article] [PubMed] [Google Scholar]
7. Muhumuza S, Olsen A, Katahoire A, Nuwaha F, 2013. Uptake of preventive treatment for intestinal schistosomiasis among school children in Jinja district, Uganda: A cross sectional study. PLoS One 8: e63438. [DOI] [PMC free article] [PubMed] [Google Scholar]
8. World Health Organization , 2020. Ending the Neglect to Attain the Sustainable Development Goals: A Road Map for Neglected Tropical Diseases 2021–2030. Available at: https://www.who.int/publications/i/item/9789240010352. Accessed April 27, 2024.
9. Lai YS. et al. , 2015. Spatial distribution of schistosomiasis and treatment needs in sub-Saharan Africa: A systematic review and geostatistical analysis. Lancet Infect Dis 15: 927–940. [DOI] [PubMed] [Google Scholar]
10. Luo C, Wang Y, Su Q, Zhu J, Tang S, Bergquist R, Zhang Z, Hu Y, 2022. Mapping schistosomiasis risk in Southeast Asia: A systematic review and geospatial analysis. Int J Epidemiol 52: 1137–1149. [DOI] [PubMed] [Google Scholar]
11. Hu Y, Xia C, Chen Y, Lynn H, Zhang T, Xiong C, Chen G, He Z, Zhang Z, 2017. Assessment of the national schistosomiasis control program in a typical region along the Yangtze River, China. Int J Parasitol 47: 21–29. [DOI] [PubMed] [Google Scholar]
12. Hu Y, Bergquist R, Chen Y, Ke Y, Dai J, He Z, Zhang Z, 2021. Dynamic evolution of schistosomiasis distribution under different control strategies: Results from surveillance covering 1991–2014 in Guichi, China. PLoS Negl Trop Dis 15: e0008976. [DOI] [PMC free article] [PubMed] [Google Scholar]
13. Ekpo UF. et al. , 2013. Mapping and prediction of schistosomiasis in Nigeria using compiled survey data and Bayesian geospatial modelling. Geospat Health 7: 355–366. [DOI] [PubMed] [Google Scholar]
14. Wiegand RE. et al. , 2021. Associations between infection intensity categories and morbidity prevalence in school-age children are much stronger for Schistosoma haematobium than for S. mansoni. PLoS Negl Trop Dis 15: e0009444. [DOI] [PMC free article] [PubMed] [Google Scholar]
15. Lwambo NJ, Savioli L, Kisumku UM, Alawi KS, Bundy DA, 1997. The relationship between prevalence of Schistosoma haematobium infection and different morbidity indicators during the course of a control programme on Pemba Island. Trans R Soc Trop Med Hyg 91: 643–646. [DOI] [PubMed] [Google Scholar]
16. Ruan D, Shi Y, Jin L, Yang Q, Yu W, Ren H, Zheng W, Chen Y, Zheng N, Zheng M, 2021. An ultrasound image-based deep multi-scale texture network for liver fibrosis grading in patients with chronic HBV infection. Liver Int 41: 2440–2454. [DOI] [PMC free article] [PubMed] [Google Scholar]
17. Jiang H. et al. , 2021. Machine learning algorithms to predict the 1 year unfavourable prognosis for advanced schistosomiasis. Int J Parasitol 51: 959–965. [DOI] [PubMed] [Google Scholar]
18. Tran NK, Albahra S, May L, Waldman S, Crabtree S, Bainbridge S, Rashidi H, 2021. Evolving applications of artificial intelligence and machine learning in infectious diseases testing. Clin Chem 68: 125–133. [DOI] [PMC free article] [PubMed] [Google Scholar]
19. Peiffer-Smadja N, Rawson TM, Ahmad R, Buchard A, Georgiou P, Lescure F-X, Birgand G, Holmes AH, 2020. Machine learning for clinical decision support in infectious diseases: A narrative review of current applications. Clin Microbiol Infect 26: 584–595. [DOI] [PubMed] [Google Scholar]
20. Lynch CM, Abdollahi B, Fuqua JD, de Carlo AR, Bartholomai JA, Balgemann RN, van Berkel VH, Frieboes HB, 2017. Prediction of lung cancer patient survival via supervised machine learning classification techniques. Int J Med Inform 108: 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Zhang Y, Chen K, Weng Y, Chen Z, Zhang J, Hubbard R, 2022. An intelligent early warning system of analyzing Twitter data using machine learning on COVID-19 surveillance in the US. Expert Syst Appl 198: 116882. [DOI] [PMC free article] [PubMed] [Google Scholar]
22. Jiang D, Hao M, Ding F, Fu J, Li M, 2018. Mapping the transmission risk of Zika virus using machine learning models. Acta Trop 185: 391–399. [DOI] [PubMed] [Google Scholar]
23. Deol AK. et al. , 2019. Schistosomiasis: Assessing progress toward the 2020 and 2025 global goals. N Engl J Med 381: 2519–2528. [DOI] [PMC free article] [PubMed] [Google Scholar]
24. Isaaks EH, Srivastava RM, 1989. An Introduction to Applied Geostatistics. New York, NY: Oxford University Press. [Google Scholar]
25. Mewamba EM. et al. , 2022. Fine-scale mapping of Schistosoma mansoni infections and infection intensities in sub-districts of Makenene in the Centre region of Cameroon. PLoS Negl Trop Dis 16: e0010852. [DOI] [PMC free article] [PubMed] [Google Scholar]
26. Clements AC, Moyeed R, Brooker S, 2006. Bayesian geostatistical prediction of the intensity of infection with Schistosoma mansoni in East Africa. Parasitology 133: 711–719. [DOI] [PMC free article] [PubMed] [Google Scholar]
27. Jones IJ. et al. , 2021. Schistosome infection in Senegal is associated with different spatial extents of risk and ecological drivers for Schistosoma haematobium and S. mansoni. PLoS Negl Trop Dis 15: e0009712. [DOI] [PMC free article] [PubMed] [Google Scholar]
28. De Boni L, Msimang V, De Voux A, Frean J, 2021. Trends in the prevalence of microscopically-confirmed schistosomiasis in the South African public health sector, 2011–2018. PLoS Negl Trop Dis 15: e0009669. [DOI] [PMC free article] [PubMed] [Google Scholar]
29. Madinga J, Linsuke S, Mpabanzi L, Meurs L, Kanobana K, Speybroeck N, Lutumba P, Polman K, 2015. Schistosomiasis in the Democratic Republic of Congo: A literature review. Parasit Vectors 8: 601. [DOI] [PMC free article] [PubMed] [Google Scholar]
30. Donohue RE, Mashoto KO, Mubyazwi GM, Madon S, Malecela MN, Michael E, 2017. Biosocial determinants of persistent schistosomiasis among schoolchildren in Tanzania despite repeated treatment. Trop Med Infect Dis 2: 61. [DOI] [PMC free article] [PubMed] [Google Scholar]
31. Sacko M, Magnussen P, Deita AD, Traoré S, Landouré A, Doucouré A, Madsen H, Vennervald BJ, 2011. Impact of Schistosoma haematobium infection on urinary tract pathology, nutritional status and anaemia in school-aged children in two different endemic areas of the Niger River Basin, Mali. Acta Trop 120: S142–S150. [DOI] [PubMed] [Google Scholar]
32. Mushi V, Zacharia A, Shao M, Mubi M, Tarimo D, 2022. Persistence of Schistosoma haematobium transmission among school children and its implication for the control of urogenital schistosomiasis in Lindi, Tanzania. PLoS One 17: e0263929. [DOI] [PMC free article] [PubMed] [Google Scholar]
33. Manzella A, Ohtomo K, Monzawa S, Lim JH, 2008. Schistosomiasis of the liver. Abdom Imaging 33: 144–150. [DOI] [PubMed] [Google Scholar]
34. Ayabina DV, Clark J, Bayley H, Lamberton PHL, Toor J, Hollingsworth TD, 2021. Gender-related differences in prevalence, intensity and associated risk factors of Schistosoma infections in Africa: A systematic review and meta-analysis. PLoS Negl Trop Dis 15: e0009083. [DOI] [PMC free article] [PubMed] [Google Scholar]
35. Mazigo HD, Nuwaha F, Dunne DW, Kaatano GM, Angelo T, Kepha S, Kinung’hi SM, 2017. Schistosoma mansoni infection and its related morbidity among adults living in selected villages of Mara Region, North-Western Tanzania: A cross-sectional exploratory study. Korean J Parasitol 55: 533–540. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Materials

tpmd230751.SD1.pdf^{(2.4MB, pdf)}

DOI: 10.4269/ajtmh.23-0751

Data Availability Statement

All data used in this study are available on request from the corresponding author.

[b1] 1. Hotez PJ. et al. , 2014. The Global Burden of Disease Study 2010: Interpretation and implications for the neglected tropical diseases. PLoS Negl Trop Dis 8: e2865. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b2] 2. WHO , 2023. Schistosomiasis: Key Facts. Available at: https://www.who.int/news-room/fact-sheets/detail/schistosomiasis. Accessed April 17, 2019.

[b3] 3. Karagiannis-Voules DA. et al. , 2015. Spatial and temporal distribution of soil-transmitted helminth infection in sub-Saharan Africa: A systematic review and geostatistical meta-analysis. Lancet Infect Dis 15: 74–84. [DOI] [PubMed] [Google Scholar]

[b4] 4. Gryseels B, Polman K, Clerinx J, Kestens L, 2006. Human schistosomiasis. Lancet 368: 1106–1118. [DOI] [PubMed] [Google Scholar]

[b5] 5. World Health Organization , 2011. Helminth Control in School-Age Children: A Guide for Managers of Control Programmes, 2nd edition. Geneva, Switzerland: WHO. [Google Scholar]

[b6] 6. Hürlimann E. et al. , 2018. Effect of an integrated intervention package of preventive chemotherapy, community-led total sanitation and health education on the prevalence of helminth and intestinal protozoa infections in Côte d’Ivoire. Parasit Vectors 11: 115. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b7] 7. Muhumuza S, Olsen A, Katahoire A, Nuwaha F, 2013. Uptake of preventive treatment for intestinal schistosomiasis among school children in Jinja district, Uganda: A cross sectional study. PLoS One 8: e63438. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b8] 8. World Health Organization , 2020. Ending the Neglect to Attain the Sustainable Development Goals: A Road Map for Neglected Tropical Diseases 2021–2030. Available at: https://www.who.int/publications/i/item/9789240010352. Accessed April 27, 2024.

[b9] 9. Lai YS. et al. , 2015. Spatial distribution of schistosomiasis and treatment needs in sub-Saharan Africa: A systematic review and geostatistical analysis. Lancet Infect Dis 15: 927–940. [DOI] [PubMed] [Google Scholar]

[b10] 10. Luo C, Wang Y, Su Q, Zhu J, Tang S, Bergquist R, Zhang Z, Hu Y, 2022. Mapping schistosomiasis risk in Southeast Asia: A systematic review and geospatial analysis. Int J Epidemiol 52: 1137–1149. [DOI] [PubMed] [Google Scholar]

[b11] 11. Hu Y, Xia C, Chen Y, Lynn H, Zhang T, Xiong C, Chen G, He Z, Zhang Z, 2017. Assessment of the national schistosomiasis control program in a typical region along the Yangtze River, China. Int J Parasitol 47: 21–29. [DOI] [PubMed] [Google Scholar]

[b12] 12. Hu Y, Bergquist R, Chen Y, Ke Y, Dai J, He Z, Zhang Z, 2021. Dynamic evolution of schistosomiasis distribution under different control strategies: Results from surveillance covering 1991–2014 in Guichi, China. PLoS Negl Trop Dis 15: e0008976. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b13] 13. Ekpo UF. et al. , 2013. Mapping and prediction of schistosomiasis in Nigeria using compiled survey data and Bayesian geospatial modelling. Geospat Health 7: 355–366. [DOI] [PubMed] [Google Scholar]

[b14] 14. Wiegand RE. et al. , 2021. Associations between infection intensity categories and morbidity prevalence in school-age children are much stronger for Schistosoma haematobium than for S. mansoni. PLoS Negl Trop Dis 15: e0009444. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b15] 15. Lwambo NJ, Savioli L, Kisumku UM, Alawi KS, Bundy DA, 1997. The relationship between prevalence of Schistosoma haematobium infection and different morbidity indicators during the course of a control programme on Pemba Island. Trans R Soc Trop Med Hyg 91: 643–646. [DOI] [PubMed] [Google Scholar]

[b16] 16. Ruan D, Shi Y, Jin L, Yang Q, Yu W, Ren H, Zheng W, Chen Y, Zheng N, Zheng M, 2021. An ultrasound image-based deep multi-scale texture network for liver fibrosis grading in patients with chronic HBV infection. Liver Int 41: 2440–2454. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b17] 17. Jiang H. et al. , 2021. Machine learning algorithms to predict the 1 year unfavourable prognosis for advanced schistosomiasis. Int J Parasitol 51: 959–965. [DOI] [PubMed] [Google Scholar]

[b18] 18. Tran NK, Albahra S, May L, Waldman S, Crabtree S, Bainbridge S, Rashidi H, 2021. Evolving applications of artificial intelligence and machine learning in infectious diseases testing. Clin Chem 68: 125–133. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b19] 19. Peiffer-Smadja N, Rawson TM, Ahmad R, Buchard A, Georgiou P, Lescure F-X, Birgand G, Holmes AH, 2020. Machine learning for clinical decision support in infectious diseases: A narrative review of current applications. Clin Microbiol Infect 26: 584–595. [DOI] [PubMed] [Google Scholar]

[b20] 20. Lynch CM, Abdollahi B, Fuqua JD, de Carlo AR, Bartholomai JA, Balgemann RN, van Berkel VH, Frieboes HB, 2017. Prediction of lung cancer patient survival via supervised machine learning classification techniques. Int J Med Inform 108: 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b21] 21. Zhang Y, Chen K, Weng Y, Chen Z, Zhang J, Hubbard R, 2022. An intelligent early warning system of analyzing Twitter data using machine learning on COVID-19 surveillance in the US. Expert Syst Appl 198: 116882. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b22] 22. Jiang D, Hao M, Ding F, Fu J, Li M, 2018. Mapping the transmission risk of Zika virus using machine learning models. Acta Trop 185: 391–399. [DOI] [PubMed] [Google Scholar]

[b23] 23. Deol AK. et al. , 2019. Schistosomiasis: Assessing progress toward the 2020 and 2025 global goals. N Engl J Med 381: 2519–2528. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b24] 24. Isaaks EH, Srivastava RM, 1989. An Introduction to Applied Geostatistics. New York, NY: Oxford University Press. [Google Scholar]

[b25] 25. Mewamba EM. et al. , 2022. Fine-scale mapping of Schistosoma mansoni infections and infection intensities in sub-districts of Makenene in the Centre region of Cameroon. PLoS Negl Trop Dis 16: e0010852. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b26] 26. Clements AC, Moyeed R, Brooker S, 2006. Bayesian geostatistical prediction of the intensity of infection with Schistosoma mansoni in East Africa. Parasitology 133: 711–719. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b27] 27. Jones IJ. et al. , 2021. Schistosome infection in Senegal is associated with different spatial extents of risk and ecological drivers for Schistosoma haematobium and S. mansoni. PLoS Negl Trop Dis 15: e0009712. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b28] 28. De Boni L, Msimang V, De Voux A, Frean J, 2021. Trends in the prevalence of microscopically-confirmed schistosomiasis in the South African public health sector, 2011–2018. PLoS Negl Trop Dis 15: e0009669. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b29] 29. Madinga J, Linsuke S, Mpabanzi L, Meurs L, Kanobana K, Speybroeck N, Lutumba P, Polman K, 2015. Schistosomiasis in the Democratic Republic of Congo: A literature review. Parasit Vectors 8: 601. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b30] 30. Donohue RE, Mashoto KO, Mubyazwi GM, Madon S, Malecela MN, Michael E, 2017. Biosocial determinants of persistent schistosomiasis among schoolchildren in Tanzania despite repeated treatment. Trop Med Infect Dis 2: 61. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b31] 31. Sacko M, Magnussen P, Deita AD, Traoré S, Landouré A, Doucouré A, Madsen H, Vennervald BJ, 2011. Impact of Schistosoma haematobium infection on urinary tract pathology, nutritional status and anaemia in school-aged children in two different endemic areas of the Niger River Basin, Mali. Acta Trop 120: S142–S150. [DOI] [PubMed] [Google Scholar]

[b32] 32. Mushi V, Zacharia A, Shao M, Mubi M, Tarimo D, 2022. Persistence of Schistosoma haematobium transmission among school children and its implication for the control of urogenital schistosomiasis in Lindi, Tanzania. PLoS One 17: e0263929. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b33] 33. Manzella A, Ohtomo K, Monzawa S, Lim JH, 2008. Schistosomiasis of the liver. Abdom Imaging 33: 144–150. [DOI] [PubMed] [Google Scholar]

[b34] 34. Ayabina DV, Clark J, Bayley H, Lamberton PHL, Toor J, Hollingsworth TD, 2021. Gender-related differences in prevalence, intensity and associated risk factors of Schistosoma infections in Africa: A systematic review and meta-analysis. PLoS Negl Trop Dis 15: e0009083. [DOI] [PMC free article] [PubMed] [Google Scholar]

[b35] 35. Mazigo HD, Nuwaha F, Dunne DW, Kaatano GM, Angelo T, Kepha S, Kinung’hi SM, 2017. Schistosoma mansoni infection and its related morbidity among adults living in selected villages of Mara Region, North-Western Tanzania: A cross-sectional exploratory study. Korean J Parasitol 55: 533–540. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Predicting Schistosomiasis Intensity in Africa: A Machine Learning Approach to Evaluate the Progress of WHO Roadmap 2030

Xinyue Chen

Jiaxu Le

Yi Hu

ABSTRACT.

INTRODUCTION