Spatiotemporal variation in risk of Shigella infection in childhood: a global risk mapping and prediction model using individual participant data

**Background:** Shigella is a major cause of diarrheal disease and death in children under 5 in low- and middle-income countries (LMICs), with over 212,000 deaths annually. Vaccine development is advancing, making it critical to map infection risk to guide future vaccine trials and campaigns. Previous estimates were limited to national or subnational levels and did not account for subclinical infections or seasonal variation. This study aimed to produce high-resolution, spatiotemporal maps of Shigella prevalence across LMICs using individual participant data and environmental covariates.

**Methods:** Individual participant data from 20 studies across 23 LMICs (Central/South America, sub-Saharan Africa, South/Southeast Asia) contributed 66,563 stool samples from children ≤59 months. Shigella positivity was determined by PCR. Covariates included age, symptom status (asymptomatic, community-detected diarrhea, medically attended diarrhea), household-level factors (sanitation, water source, education, flooring), and environmental/hydrometeorological variables (temperature, precipitation, soil moisture, wind speed, humidity, solar radiation). Time-varying hydrometeorological data were extracted from the Global Land Data Assimilation System (GLDAS). Generalized additive models (GAMs) with splines were fitted to allow non-linear associations. Model performance was assessed via cross-validation, and variable importance was evaluated using Shapley values and accumulated local effects (ALE). Predictions were made for nine age/symptom strata across LMICs.

**Key Results:** Age was the most important predictor: children aged 24–59 months had 4.82 times higher odds of Shigella positivity (OR=4.82, 95% CI 4.43–5.21) compared to infants 0–11 months; those aged 12–23 months had OR=3.42 (3.24–3.61). Symptom status and study design were next in importance. Among hydrometeorological variables, temperature showed a non-linear, inverse U-shaped association: probability of Shigella detection peaked at ~33°C with a 43% probability for uncomplicated diarrhea cases and 26% for asymptomatic children, then declined at higher temperatures. An interaction between precipitation and soil moisture was observed: when both were above average, Shigella probability exceeded 20%. Wind speed had an inverse U-shaped peak at 4 m/s. Relative humidity and solar radiation were directly associated with risk. Improved sanitation reduced odds by 19% (OR=0.81, 0.76–0.86), and open defecation reduced odds by 18% (OR=0.82, 0.76–0.88) compared to unimproved sanitation. Peri-urban residence increased odds (OR=1.18, 1.07–1.29). Moderate-to-severe underweight (OR=1.17, 1.08–1.26) and stunting (OR=1.16, 1.09–1.23) were associated with higher risk. Protective factors included improved floor material (OR=0.83, 0.78–0.88), caregiver primary education (OR=0.85, 0.81–0.89), and improved water source (OR=0.86, 0.79–0.93). Predicted prevalence maps for 12–23-month-olds showed hotspots in sub-Saharan Africa (e.g., Central African Republic, South Sudan), Central America (e.g., Nicaragua), South America (e.g., Bolivia, Colombia), the Ganges–Brahmaputra Delta, and Papua New Guinea. Seasonality varied: e.g., Central African Republic had a boreal spring peak, while Botswana peaked in December–February.

**Clinical Implications:** The study demonstrates that Shigella transmission is highly sensitive to climatological factors, particularly temperature, with a threshold around 33°C. The interaction between precipitation and soil moisture suggests that flood-prone areas may see elevated risk. The protective effect of improved sanitation (19% reduction) and the counterintuitive finding that open defecation also reduced odds (18%) compared to unimproved sanitation highlight the importance of sanitation quality. Undernutrition (stunting, underweight) increases susceptibility, reinforcing the bidirectional link between malnutrition and enteric infections. These high-resolution maps can guide prioritization of populations for phase 3 vaccine trials and eventual vaccine roll-out, especially in identified hotspots. The findings also support the use of seasonal forecasts to time interventions. Limitations include potential bias from pooling heterogeneous study designs and the inability to directly estimate Shigella-attributable diarrheal burden. Nonetheless, the study provides the most detailed spatiotemporal risk assessment for Shigella in LMICs to date.