Current water contact and *Schistosoma mansoni* infection have distinct determinants: a data-driven population-based study in rural Uganda

Environmentally mediated pathogens disproportionately affect rural populations in low- and middle-income countries. Schistosomiasis transmission depends on human water contact with infested freshwater, snail ecology, access to safe water/sanitation, occupations, and behavior. Because direct measurement of cercarial exposure is difficult, water contact is often used as a proxy, but prior studies have been limited by self-reporting, crude metrics, sparse environmental data, and methodological shortcomings. Repeated mass drug administration (MDA) complicates inference by altering infection-immunity dynamics, and standardized exposure tools are lacking. This study focuses on Schistosoma mansoni in Uganda to (1) comprehensively characterize current water contact patterns and determinants across individual, household, and village levels; (2) compare determinants of exposure (water contact) with determinants of current infection; and (3) assess whether age trends in exposure mirror those in infection. The aim is to inform targeted interventions beyond MDA, including WASH, environmental management, and behavior change.

The paper summarizes prior evidence indicating: (a) water contact is widely used as a proxy for exposure but typically via cross-sectional, self-reported measures; (b) integrating environmental variables (e.g., snail abundance) has yielded inconsistent improvements in infection prediction; (c) exposure metrics often lack granularity (immersion levels, timing), and studies rarely employ systematic variable selection or validation; (d) immunity, shaped by prior infection and treatment, may decouple current exposure from current infection; and (e) distance to waterbodies predicts infection in some studies, though whether this reflects exposure gradients remains unclear. These gaps motivate a comprehensive, data-driven assessment distinguishing determinants of water contact from determinants of infection.

Design and setting: Cross-sectional analysis using baseline (January–February 2022, dry season) data from the SchistoTrack cohort in rural Uganda. Thirty-eight villages within ~3 km of Lake Albert, Lake Victoria, or the River Nile were sampled across Pakwach, Buliisa, and Mayuge districts. All districts had received ≥13 annual rounds of praziquantel MDA since 2003, with the most recent >1 year prior to the study. Sampling and participants: From village registers/MDA records, ~40 households per village were randomly sampled (n=1459 targeted; n=1444 with complete household data). Eligibility required ≥1 child and ≥1 adult resident for ≥6 months. From each household, one adult (18+) and one child (5–17) were selected for clinical assessments, yielding 2867 individuals aged 5–90 years with complete data. Informed consent/assent obtained; all participants treated with praziquantel after exams. Data collection: - Household survey: socio-demographics, occupation, religion/tribe, biomedical history, WASH, GPS location, and water contact behaviors. - Clinical diagnostics: Kato-Katz (duplicate thick smear from a single stool sample) for eggs per gram (EPG); POC-CCA performed and used in sensitivity analyses (trace classified negative). - Environmental/malacology: malacologists mapped all village water sites, recorded GPS, site type (river, marsh, beach, swamp, pond), evidence of human fecal contamination, and collected snails (30 min per site; scooping/hand-picking). Snail infectivity determined by cercarial shedding under sunlight the day after collection. Village infrastructure survey captured public latrines, taps/boreholes, and primary school GPS locations. Exposure measurement: - Water contact activities (11): getting drinking water; washing clothes with/without soap; bathing with/without soap; washing jerry cans/household items; fishing; fishmongering; collecting papyrus; collecting shells; swimming/playing. For each person: weekly frequency, typical duration per trip (categorical; minimum 30 min), and typical time of day (eight time windows). - Primary exposure outcome: any water contact (binary) defined as ≥1 water-contact activity per week. Secondary: activity type (domestic, occupational, recreational), total weekly frequency (sum across activities), and total weekly duration (frequency × duration per activity, summed). Infection outcomes: - Infection status: S. mansoni infection defined as >0 EPG by Kato-Katz. - Heavy infection: ≥400 EPG (per WHO guidelines). Spatial variables: - Distances (Euclidean): household to closest water site; village center to closest water site; school to closest water site; distance to closest public latrine; distance to closest tap/borehole. Indicators for closest water site type and whether the closest site to the household is within the village. Counts: number of water sites per village; numbers of latrines/taps per village. Village-level prevalence categorized as 10–49% (moderate) vs ≥50% (high) by Kato-Katz (no ≤10% villages). - Snail variables (for extended models): number of water sites with infected snails per village; number of snails within 1 km of the household; total snails and infected snails per village; presence of any infected snails; distance to closest site with infected snails; whether the closest infected site is within the village. Statistical analysis: - Descriptive and non-linear trends over age and distance modeled with generalized additive models (mgcv), with knots chosen via k.check. - Variable selection: Bayesian variable selection (BVS) via BAS with Jeffreys–Zellner–Siow priors; model space sampled using MCMC (2×10^7 iterations); median probability model threshold pr ≥ 0.5. Alternative selection via likelihood ratio tests (LRTs) at p<0.05 for comparison. - Main regressions: generalized linear models (binomial family; log link) for any water contact and infection outcomes; standard errors clustered at household level; district fixed effects included. Durbin–Wu–Hausman tests informed fixed vs random effects; fixed effects used for consistency and low residual clustering. - Activity-type models: separate logistic regressions for domestic, occupational, and recreational contact, excluding individuals with multiple activity types (6.4%). - Frequency/duration and infection intensity: negative binomial models; checked zero inflation; multicollinearity assessed via gVIFs (none > sqrt(10)). - Validation/sensitivity: compared self-reported vs direct observations in 12 villages; reclassified Kato-Katz negatives as positive when POC-CCA positive (trace negative) for sensitivity analyses. - Performance: 10-fold cross-validation with 100 repeats; auROC comparisons across models; Moran’s I for residual spatial correlation.

- Water contact prevalence and patterns: - 46.7% (1339/2867; 95% CI 44.9–48.5%) reported ≥1 weekly water contact. - Median weekly frequency: 6 trips (IQR 3–11); median weekly duration: 8 h (IQR 3.5–17.5); overdispersed distributions. - Most common activities: getting drinking water 17.3% (497/2867); washing clothes with soap 16.8% (481/2867); fishing 12.3% (354/2867). - Time of day: overall peak 6–9 am (23.6% of contacts). Fishing most common 5–7 pm (24.3%). - Distance gradients: - 80% of individuals with water contact lived within 0.43 km of a water site; among ≤0.43 km residents, 50.3% had contact vs 36.3% beyond 0.43 km. - For household distance 0–1 km, each additional 100 m associated with a 3.4 percentage-point decrease in water contact; village-center distance showed a weaker 1.9 pp decrease per 100 m; school distance showed no gradient. - Age and gender patterns: - Modeled any water contact over age: 16% at age 5 (95% CI 14–18%), peak 70% at age 30 (95% CI 66–74%), ~28% at age 70 (95% CI 20–36%). - Females: 49.3% with contact; males: 43.6%. Females had higher frequency (median 7 vs 5 trips/week) but shorter duration (7 vs 10 h/week). After adjusting for covariates, gender differences in frequency/duration were not significant. - Activity types: males reported fishing 24.3% vs females 2.5%; females collected drinking water 22.1% vs males 11.5% and washed clothes with soap 23.4% vs males 8.7% (all p<0.01). Adults: female contact predominantly domestic (75.3% of duration), males predominantly occupational (81.8%). - Determinants of any water contact (logistic regression): - Age: OR 1.18 (95% CI 1.15–1.21); age^2: OR 0.9978 (0.9975–0.9982). - Female: OR 1.41 (1.17–1.68); effect became non-significant when occupation removed (OR 1.05; 0.89–1.24). - Occupation vs none/other: fishing OR 6.67 (4.06–10.95); fishmongering OR 2.10 (1.12–3.95). - Environmental/WASH: fecal contamination at closest site OR 0.76 (0.63–0.92); distance to closest public latrine per km OR 0.87 (0.81–0.94). - Closest site type vs river: beach OR 2.49 (1.75–3.55); swamp OR 1.57 (1.04–2.37); pond OR 3.24 (2.05–5.13); marsh ns. - Access: number of water sites per village OR 1.13 (1.05–1.21); closest site within village OR 1.49 (1.25–1.78). - Village-level prevalence: ≥50% vs 10–49% associated with lower odds of water contact OR 0.70 (0.57–0.87). - No adjusted district differences; weak residual spatial correlation (Moran’s I=0.064, p<0.01). - By activity: females had higher odds of domestic contact (OR 2.43; 1.97–3.00) but lower recreational (OR 0.28; 0.11–0.73) and occupational (OR 0.17; 0.11–0.27). - Infection burden and age trends: - Overall infection prevalence: 43.3% (1240/2867); heavy infection: 8.2% (236/2867). - District prevalence: Pakwach 50.4% (477/947); Buliisa 44.1% (422/958); Mayuge 35.4% (341/962). - Modeled infection prevalence by age: 29% at age 5 (95% CI 25–33%), peak 63% at age 15 (95% CI 59–67%), 31% at age 50 (95% CI 24–36%). Water contact peaked later at age 30, remaining higher than infection until ~65 years. - Water contact vs infection correlations: - Practically no individual-level correlation between current water contact and infection (p≈0.03; n=2867); household adult-child pairwise correlations weak for both outcomes; village-level water contact proportion uncorrelated with village-level prevalence (p≈0.98; n=38). Infection did not exhibit the linear household-distance decline observed for water contact. - Determinants of infection (logistic models): - Shared selections with water contact: age, age^2, occupation, number of water sites per village, village prevalence; gender not selected. - Significant predictors for infection status: age OR 1.05 (1.03–1.08); primary education vs none OR 1.53 (1.20–1.94); number of water sites per village OR 0.88 (0.82–0.94); village prevalence ≥50% vs 10–49% OR 3.26 (2.73–3.90); landing site in village associated with lower odds OR 0.51 (0.37–0.68). Fishing occupation associated with heavy infection only (OR 1.72; 1.03–2.87), not infection status (OR 1.16; 0.81–1.64). District effects: Pakwach associated with higher odds of heavy infection (OR 1.65; 1.05–2.61). - No water contact or household-level WASH variables met inclusion thresholds for infection models. - Model performance: - Water contact model auROC 0.777 vs infection status auROC 0.693 (p<0.01). BVS outperformed LRTs for water contact (0.777 vs 0.522); smaller gap for infection (0.696 vs 0.648). Adding snail variables or granular exposure variables did not materially improve auROC. - Validation/sensitivity: - Self-reported water contact age patterns closely matched direct observations in 12 villages. Reclassifying infection using POC-CCA positivity for Kato-Katz negatives (trace negative) produced similar infection results.

The study demonstrates that current water contact and current S. mansoni infection have distinct determinants and age profiles. Water contact increases into adulthood and peaks around age 30, whereas infection peaks in adolescence (~15 years), suggesting that acquired immunity, prior treatment, and other host factors decouple current exposure from current infection. Determinants of water contact span socio-demographic (age, occupation, gender), environmental (site type, site access, fecal contamination), and WASH (proximity to public latrines) domains, with household-level and environmental contexts playing large roles. In contrast, infection status is more strongly associated with individual-level socio-demographics and village-level endemicity and less with household-level factors; notably, granular water contact metrics and WASH variables did not improve infection prediction. The absence of strong correlations between water contact and infection across individuals, households, and villages challenges assumptions in models that infer exposure directly from age-specific infection patterns. Programmatically, water contact concentrates within ~0.43 km of water sites and differs by gender and occupation: domestic activities dominate for women, and occupational activities (e.g., fishing) dominate for men, often at different times of day. These profiles can guide targeted exposure-reduction strategies distinct from infection-driven MDA targeting. Household distance to water sites shows clear exposure gradients, while school-based distance measures do not, implying that school sentinel sites may miss fine-scale exposure variation. Self-reported water contact appears valid at population level when benchmarked against direct observations, supporting cost-effective surveillance approaches.

This study provides a comprehensive, data-driven profile of water contact behaviors and determinants alongside infection patterns in rural Uganda. Key contributions include: (1) documenting the decoupling of age-specific exposure and infection peaks; (2) identifying multi-level, environment-linked determinants of water contact distinct from infection determinants; (3) demonstrating strong household-distance gradients in exposure but not infection; and (4) validating self-reported water contact for population-level surveillance. Findings support spatially targeted exposure-reduction interventions tailored to group-specific behaviors (domestic for women/children; occupational for men) and implemented at smaller spatial scales (e.g., within 0.5 km buffers) than MDA units. Future work should generalize methods to other countries and schistosome species, employ longitudinal designs integrating exposure histories with immunological measures, and leverage wearable sensors for finer temporal-spatial exposure quantification. Guidelines may incorporate water contact metrics when determining treatment frequency, and standardized tools for measuring water contact should be developed for program use.

- Cross-sectional design precludes causal inference between exposure and infection and cannot capture cumulative exposure histories. - Immunological factors and infection history (timing of first infection, longevity of infections, co-infections, age-specific immune responses) could not be fully assessed and may confound exposure–infection relationships. - Self-reported water contact, while validated against observations in 12 villages, remains subject to recall and reporting bias; direct observations were only available for ~32% of villages. - Kato-Katz has limited sensitivity for light infections; although sensitivity analyses with POC-CCA were conducted, misclassification may persist. - Environmental measurements reflect dry-season conditions; seasonal variability in contact and snail dynamics was not assessed. - Few households lived >1 km from water sites, limiting inference on longer-distance gradients. - No low-prevalence (≤10%) villages were included, limiting generalizability across the full endemicity spectrum.