Table 1.
Description of candidate predictors for S. japonicum infection by year, including variables that describe agricultural practices, potential animal reservoirs, individual characteristics, occupation, socio-economic status, and access to water, sanitation and hygiene (WASH) infrastructure.
Table 2.
Predictive performance of the reemergence (2007-2010) and elimination models (2016-2019) with 95% confidence intervals across 20 bootstrapped iterations.
Fig 1.
Change in the ranked importance of predictors from 2007-2010 (reemergence period) to 2016-2019 (elimination period).
Red lines indicate a decrease in ranked importance from the reemergence period to the elimination period. Blue lines indicate an increase.
Fig 2.
Partial dependence plots (PDP) of the six most important predictors of human schistosomiasis infection risk in 2007-2010 (reemergence model).
The PDPs display the change in the average predicted infection risk as predictors vary over their marginal distribution while holding all other variables constant. Fitted curves (dashed lines), smoothing splines (solid blue lines) and 95% confidence intervals based on 1000 bootstrap replicates of the data set (shading) are shown. The full distribution of the predictors is displayed as rug ticks on the top of the plot.
Fig 3.
Partial dependence plots (PDP) of the top six predictors of human schistosomiasis infection risk in 2016-2019 (elimination model).
The PDP (blue line) displays the change in the average predicted infection risk as predictors vary over their marginal distribution while holding all other variables constant. Fitted curves (dashed lines), smoothing splines (solid blue lines) and 95% confidence intervals based on 1000 bootstrap replicates of the data set (shading) are shown. The full distribution of the predictors is displayed as rug ticks on the top of the plot.
Table 3.
Pairwise interactions that were the strongest predictors of S. japonicum infection (interaction size > 5) in the reemergence period, 2007-2010.
Fig 4.
Three-dimensional partial dependence plots of the three most important pairwise interactions in 2007-2010.
Each panel shows the GBM’s fitted value (z-axis; higher values indicate higher predicted S. japonicum infection risk) as a function of two predictors (x- and y-axes) with all other covariates averaged over their observed distributions. Left-to-right: (A) village-level dry crop area and village-level improved sanitation; (B) village-level night soil use to rice fields and dry crop area; (C) village-level bovine presence and village-level well-water use.
Table 4.
Pairwise interactions that were the strongest predictors of S. japonicum infection (interaction size > 5) in the elimination period, 2016-2019.
Fig 5.
Three-dimensional partial dependence plots of the three most important pairwise interactions in 2016-2019.
Each panel shows the GBM’s fitted value (z-axis; higher values indicate higher predicted S. japonicum infection risk) as a function of two predictors (x- and y-axes), with all other covariates averaged over their observed distributions. Left-to-right: (A) village-level dry crop area and household improved sanitation; (B) household dog presence and individual age; (C) village-level dry crop area and individual age.