2 — Literature Review
2.1 — Typhoid transmission pathways and environmental determinants
Typhoid fever is transmitted primarily via the fecal–oral route, with waterborne and foodborne pathways predominating in endemic settings (Adu-Gyamfi et al., 2025; World Health Organization, 2023). A 2025 systematic review synthesizing 109 studies (1928–2024) identified waterborne (41.3%), foodborne (44%), and sanitation-related (31.2%) transmission routes as the principal drivers of infection, with Asia contributing 46.8% of global evidence (Adu-Gyamfi et al., 2025). Antimicrobial resistance (AMR) among Salmonella Typhi isolates, estimated at 50–70% in high-burden settings, and chronic carriage rates of approximately 10% further sustain endemicity in communities with inadequate WASH infrastructure (Andrews et al., 2019; World Health Organization, 2024).
Environmental contamination driven by flooding plays a critical amplifying role. Wastewater surveillance in urban Indonesia detected S. Typhi positivity in 13% of river samples at flood interfaces (OR = 12.68, p = .007), highlighting that contamination risk is linked more to disrupted sanitation than to rainfall quantity alone (Baker et al., 2025). In Nepal, monsoon-season peaks are well documented: HMIS records show 51,967 cases in Shrawan 2078 concentrated in Terai districts such as Rautahat (1,711 cases per monsoon month), where 25% open defecation rates and recurrent flooding converge with chronic WASH deficits (Central Bureau of Statistics and UNICEF, 2019; Ministry of Health and Population, Government of Nepal, 2023). WHO estimates that unsafe WASH contributes to 1.4 million deaths annually, with 73.9% attributed to diarrheal diseases (World Health Organization, 2025). Nepal’s residual typhoid burden — estimated at 80,000–100,000 cases per year despite 85% national TCV coverage — underscores the need for climate-sensitive surveillance that goes beyond vaccination alone (Coalition Against Typhoid, 2024).
2.2 — Impacts of temperature, rainfall, and humidity on waterborne diseases
Climate variables modulate typhoid and other waterborne diseases (WBDs) by altering pathogen viability, environmental persistence, and dissemination pathways, with effects that are nonlinear and context-dependent (Carlton et al., 2015; Levy et al., 2023). Temperature increases of 1 °C are associated with 5–10% elevated Salmonella risk through enhanced virulence and accelerated bacterial replication, with thresholds above 25 °C linked to amplified typhoid and cholera transmission across Southeast Asian settings (Kim et al., 2023; Trájer et al., 2022). Relative humidity in the range of 80–95% prolongs S. Typhi survival in water bodies and on environmental surfaces, and European studies have identified high streamflow (50% of cases) and heavy rainfall events (19–22%) as primary drivers of waterborne disease outbreaks (Brubacher et al., 2020; Levy et al., 2023).
Rainfall demonstrates a dual, context-dependent effect on typhoid risk. Extreme precipitation events exceeding 50 mm per day have been shown to precede approximately 71% of waterborne disease outbreaks in multiple settings by mobilizing fecal contaminants into water supplies, accounting for 24% of documented waterborne incidents in the United States (Carlton et al., 2015). However, the direction of effect varies by baseline aridity: in humid Vietnam, higher rainfall was associated with lower typhoid incidence (β = −0.0010, p < .001), suggesting dilution effects, while in arid Chad the association was positive (β = 0.0037, p < .001), reflecting contaminant concentration under low-flow conditions (Haque et al., 2025). In Nepal, extreme monsoon precipitation events — such as the 499.85 mm recorded in Sunsari District in July 2016, which coincided with 116 flood events nationally — preceded 15–20% increases in typhoid incidence in affected districts (Bhandari et al., 2022; Ministry of Health and Population, Government of Nepal, 2023). Prolonged flood duration exceeding 30 days further elevates disease odds (OR = 1.15–1.25) in Belt and Road Initiative nations (Li et al., 2025). Climate projections forecast an additional 250,000 waterborne disease deaths globally by 2050 under business-as-usual emissions, with India attributing 70% of typhoid outbreak variance to extreme rainfall events — a pattern that mirrors the dynamics observed in Nepal’s Terai region (India Health Fund, 2025; World Health Organization, 2023).
2.3 — Climate change scenarios and extreme weather events in Nepal
Nepal is among the most climate-vulnerable countries in the world, warming at 0.056 °C per year — twice the global average — with monsoon precipitation intensifying by 20–30% since 1990 under CMIP6 projections (Intergovernmental Panel on Climate Change, 2022; Ministry of Forests and Environment, Government of Nepal, 2021). Under SSP2-4.5 and SSP5-8.5 scenarios, mean temperatures are projected to rise by 1.2–4.2 °C by the 2080s, with corresponding 120% increases in extreme precipitation events between 2071 and 2100, expanding Bagmati Basin flood extents by 342–359 km² (Shrestha et al., 2025). Central Himalayan floods are projected to intensify by 40% under SSP2-4.5 and 79% under SSP5-8.5 by 2060–2099, driven primarily by increased rainfall–runoff (Potutan et al., 2025). The Climate Risk Index 2025 ranks Nepal 69th globally for climate-related impacts over 1993–2022, recording 249.7 annual deaths and 5,667 fatalities from floods and landslides between 2012 and 2024, at a cumulative cost of NPR 49 billion (Germanwatch, 2025; NPDRR Nepal, 2025).
Recent extreme events illustrate the trajectory. In July 2024, the Bhotekoshi Bridge washout — assessed by the World Weather Attribution project as 70% more likely and 10% more intense due to approximately 1.3 °C of accumulated warming, coupled with deforestation and rapid urbanization — displaced communities and overwhelmed local health infrastructure (Jha, 2024; World Weather Attribution, 2024). Glacial lake outburst floods (GLOFs) are projected to threaten over 200,000 people in Nepal by mid-century (United Nations Office for Disaster Risk Reduction, 2025). Nepal’s Country Partnership Strategy (2025–2029) explicitly prioritizes flood-resilient infrastructure, recognizing that the health co-benefits of climate adaptation — including reductions in waterborne disease — are achievable when WASH investments are designed with flood risk in mind (Asian Development Bank, 2025; Ministry of Forests and Environment, Government of Nepal, 2021).
2.4 — Machine learning in infectious disease prediction
Machine learning has demonstrably improved infectious disease forecasting by enabling integration of multimodal data while outperforming traditional statistical models in nonlinear and high-dimensional settings (Choi et al., 2023). For typhoid specifically, a 2025 clinical study applying XGBoost achieved diagnostic AUROC values of 0.9998 (treatment success) and 0.9834 (diagnosis), with SHAP analysis identifying disease severity and biochemical markers as dominant predictive features (Ssemuyiga et al., 2025). Hybrid approaches combining RF, XGBoost, and multilayer perceptrons with lagged climate and demographic features have achieved 77% forecasting accuracy for zoonotic diseases with similar data structures (Dixon et al., 2023; Morin et al., 2025).
Climate-integrated ML approaches are increasingly applied to seasonal disease prediction in tropical settings. RF, support vector machine, and k-nearest neighbor models forecast Philippine typhoid and hand-foot-mouth disease with AUROC exceeding 0.85 using morbidity and environmental inputs (Go, 2025). LSTM networks, designed to capture temporal dependencies across sequences, have shown promise for climate-sensitive disease time-series where lagged predictor effects are important (Go, 2025; Hess et al., 2020). Within Nepal specifically, ANN and XGBoost models integrated with HMIS-like surveillance data have achieved approximately 60% accuracy for combined malaria and typhoid prediction (Ogunleye et al., 2023). A 2025 systematic meta-analysis using Bayesian hierarchical models forecast 9.2 million annual typhoid cases globally, projecting 20–30% declines following TCV rollout under Gavi-supported scenarios, and underscoring the need to incorporate environmental drivers beyond vaccination in future prediction frameworks (GBD 2023 Causes of Death Collaborators, 2025). Despite these advances, no published study has combined Nepal’s HMIS district-level surveillance, DRR-portal flood records, and ERA5-CHIRPS climate grids into an integrated ML predictive model — the specific methodological gap this study addresses.
2.5 — Identified gaps and justification
Three interconnected gaps limit the utility of current typhoid prediction models for climate-vulnerable Nepal:
- Inadequate handling of flood lag and humidity thresholds. Predominant SEIR and ARIMA frameworks emphasize vaccination and AMR dynamics but inadequately account for flood lag effects (1–2 months post-event) and humidity thresholds above 90% relative humidity that sustain S. Typhi viability, leading to underestimation of monsoon-driven case surges by 15–20% (Bhandari et al., 2022; Levy et al., 2023).
- Lack of subnational resolution. National-level surveillance aggregates obscure the Terai hotspots (e.g. Rautahat: 1,711 cases per monsoon month) and the ecological gradients between Terai, Hill, and Mountain zones that are critical for targeting interventions, with fewer than 20% of models stratifying by WASH access (GBD 2023 Causes of Death Collaborators, 2025; Pitzer et al., 2025).
- Limited multimodal forecasting. While ML diagnostic models achieve excellent AUC scores, they rarely forecast epidemic dynamics multimodally or incorporate Nepal’s 1,248 documented flood events in conjunction with ERA5-CHIRPS climate metrics (Go, 2025; Semiu et al., 2025).
This study addresses these gaps by developing a hybrid GLMM and ML framework that integrates HMIS surveillance (1.2 million cases over nine years; N = 7,327 district-month observations), DRR flood records, and ERA5-Land / CHIRPS climate grids for district-stratified, lagged attributions. A four-model machine-learning framework — Random Forest, XGBoost, Multilayer Perceptron, and a Weighted Ensemble — achieves R² = 0.8675 on a strictly chronological 12-month held-out test window. By quantifying the contribution of each climate driver (mean temperature r = 0.50 – 0.63; precipitation r = 0.33 – 0.37; monsoon r = 0.25 – 0.29; flood events r = 0.19) and generating scenario-based projections through 2050, the study supports Nepal’s National Adaptation Plan (2021–2050) and WHO typhoid control targets, while providing scalable, replicable insights for comparable LMICs in South Asia.