Methods

We conducted a comprehensive characterisation of exposure (water contact) within the SchistoTrack Cohort for 2,867 individuals aged 5-90 years in 38 fishing villages in Eastern and Western Uganda and collected detailed biomedical, behavioural, socio-demographic, and environmental data. We used generalised additive models (GAMs) to characterise age-dependent trends in having any water contact, water contact frequency, and duration as well as gender differences in water contact over age. We also used GAMs to describe how water contact was influenced by household distance to water bodies. To select variables for our main regression models predicting water contact and Schistosoma mansoni infection status, we used Bayesian variable selection (BVS) on a candidate set of 30 variables. All variables with marginal inclusion probabilities ≥ 0.5, corresponding to the marginal probability model, were selected for inclusion. Multivariable logistic regressions with standard errors clustered at the household level were used to account for our sampling design. We evaluated model performance based on the area under the receiver operating curve (auROC) obtained from 10-fold cross-validation.

Results

For every 1km increase in household distance to freshwater bodies, water contact decreased by an absolute 24%. Water contact peaked at age 30; whereas infection peaked at age 15—the same year that gender differences in water contact emerged. Among adults (age 18+), males engaged primarily in occupational water contact and females in domestic water contact, which accounted for 82% and 75% of their total water contact duration, respectively. In multivariable regression models, age, gender, occupation, site contamination, drinking water source, and village-level infection prevalence predicted water contact. Predictors of S. mansoni infection status were age, level of education, occupation, type of water site, number of water sites per village, and village-level infection prevalence. Among 12 selected predictor variables of water contact, only five (42%) were also selected as predictors of current infection status. The predictive performance of the water contact model was significantly higher than the performance of the model predicting infection status (auROC of 0.783 versus 0.695, respectively, p<0.01).

Discussion

Trends in age-dependent current exposure did not correspond to age-dependent infection prevalence. Thus, assumptions in mathematical transmission models which stipulate a direct exposure-infection correspondence may not be warranted. Our findings show that high-risk groups for current schistosome exposure differ from the most currently infected individuals. This result is aligned with fundamental schistosome biology as processes governing exposure and infection play out on different timescales. We demonstrate the feasibility of predicting exposure status which could be enable future control strategies explicitly targeted at high-exposure individuals.