Improving spatial prediction of <i>Schistosoma haematobium</i> prevalence in southern Ghana through new remote sensors and local water access profiles

Background Schistosomiasis is a water-related neglected tropical disease. In many endemic low- and middle-income countries, insufficient surveillance and reporting lead to poor characterization of the demographic and geographic distribution of schistosomiasis cases. Hence, modeling is relied upon to predict areas of high transmission and to inform control strategies. We hypothesized that utilizing remotely sensed (RS) environmental data in combination with water, sanitation, and hygiene (WASH) variables could improve on the current predictive modeling approaches. Methodology Schistosoma haematobium prevalence data, collected from 73 rural Ghanaian schools, were used in a random forest model to investigate the predictive capacity of 15 environmental variables derived from RS data (Landsat 8, Sentinel-2, and Global Digital Elevation Model) with fine spatial resolution (10–30 m). Five methods of variable extraction were tested to determine the spatial linkage between school-based prevalence and the environmental conditions of potential transmission sites, including applying the models to known human water contact locations. Lastly, measures of local water access and groundwater quality were incorporated into RS-based models to assess the relative importance of environmental and WASH variables. Principal findings Predictive models based on environmental characterization of specific locations where people contact surface water bodies offered some improvement as compared to the traditional approach based on environmental characterization of locations where prevalence is measured. A water index (MNDWI) and topographic variables (elevation and slope) were important environmental risk factors, while overall, groundwater iron concentration predominated in the combined model that included WASH variables. Conclusions/Significance The study helps to understand localized drivers of schistosomiasis transmission. Specifically, unsatisfactory water quality in boreholes perpetuates reliance on surface water bodies, indirectly increasing schistosomiasis risk and resulting in rapid reinfection (up to 40% prevalence six months following preventive chemotherapy). Considering WASH-related risk factors in schistosomiasis prediction can help shift the focus of control strategies from treating symptoms to reducing exposure.背景 血吸虫病（Schistosomiasis）是一种与水相关的被忽视的热带疾病。在诸多流行地区的中低收入国家中，监测与报告工作的不足导致血吸虫病病例的人口学与地理学分布特征难以被精准刻画。因此，研究人员常依赖建模手段预测高传播区域，为防控策略制定提供参考依据。本研究提出假设：将遥感（Remote Sensing, RS）环境数据与水、环境卫生与个人卫生（Water, Sanitation and Hygiene, WASH）变量相结合，可优化现有预测建模方法。 研究方法 本研究采用从加纳73所乡村学校采集的埃及血吸虫（Schistosoma haematobium）感染率数据，构建随机森林模型，以探究15种源自遥感数据（包括Landsat 8、Sentinel-2以及全球数字高程模型（Global Digital Elevation Model））的精细空间分辨率（10~30米）环境变量的预测能力。研究共测试了5种变量提取方法，用于明确以学校为基础的感染率与潜在传播场所环境条件之间的空间关联，其中包括将模型应用于已知的人类涉水接触地点。最后，将本地取水条件与地下水水质指标纳入基于遥感的模型，以评估环境变量与WASH变量的相对重要性。 主要研究发现 相较于传统的以感染率测量地点的环境特征为基础的建模方法，基于人类接触地表水的特定场所环境特征构建的预测模型表现出一定程度的提升。水体指数（Modified Normalized Difference Water Index, MNDWI）与地形变量（海拔、坡度）为重要的环境风险因子；而在纳入WASH变量的联合模型中，地下水铁浓度整体占据主导地位。 结论与意义 本研究有助于阐明血吸虫病传播的局域驱动因素。具体而言，钻孔水井的水质不佳会导致人群持续依赖地表水，间接提升血吸虫病感染风险，并造成快速再感染（在预防性化疗后6个月内感染率可达40%）。在血吸虫病预测模型中纳入与WASH相关的风险因子，有助于将防控策略的重心从症状治疗转向暴露风险的降低。