One health at the last mile: Multi-scale predictors of Schistosoma japonicum infection in southwest china across two decades of control

In China, schistosomiasis is targeted for elimination. As the country approaches elimination, it is critical to evaluate how the dynamics of transmission are changing in remaining pockets of disease. We have been studying areas of schistosomiasis reemergence and persistence in Sichuan, China since 2007. This study used gradient boosting machines to identify key predictors of infection across two periods, 2007-2010, a period when schistosomiasis had reemerged, and 2016-2019, a period when schistosomiasis was approaching elimination. We also evaluated how key predictors of infection have shifted over time and whether combinations of predictors amplified risk. We considered predictors describing agriculture, domestic animals, socio-economic status, water and sanitation infrastructure, and demographics at individual, household and village-level scales. Our re-emergence and elimination models demonstrated strong predictive performances (AUC-PR = 0.92 and AUC = 0.85, respectively). In both periods, a person's age and village level agricultural practices including the average area of dry crops, rice planted, and night soil use, were among the most influential factors. Village-level factors dominated in 2007-2010, while household and individual predictors increased in predictive importance in 2016-2019. Between 2007-2010 and 2016-2019, there were increases in the importance of household agricultural practices such as the area of dry crops and rice cultivated, and household cat and dog ownership, while the importance of factors describing water and sanitation infrastructure decreased. In the elimination period, our models found the combination of high village dry crop cultivation and lack of improved sanitation amplified infection probability. Our findings suggest adding precision interventions targeting high-risk households on top of existing community-wide measures may accelerate schistosomiasis elimination. Practitioners should consider adding agricultural, sanitation and animal infection data to end-game surveillance programs, while researchers evaluate the consistency of these findings in other low-endemic settings and explore causal pathways to inform adaptive, locally tailored strategies.