Malaria transmission risk is projected to increase in the highlands of Western and Northern Rwanda

The study investigates how climate change may alter the spatial and seasonal dynamics of malaria transmission in Rwanda, a country that saw malaria cases rise dramatically despite sustained control funding. Prior models often rely on temperature and precipitation to drive mosquito and human infection dynamics, but typically assume linear population growth and insufficiently incorporate altitude and demographic change. Given Africa’s rapid population growth, urbanization effects, and Rwanda’s complex topography—where cooler highlands historically limit mosquito development—the authors hypothesize that warming and changing rainfall will increase malaria risk at higher elevations and modify seasonality. The purpose is to assess spatiotemporal variability in malaria incidence (2010–2015) using an ensemble learning approach and to project future risk under SSP2-4.5 and SSP5-8.5 at district level, informing early warning and control strategies.

Previous research has developed climate-driven malaria models incorporating precipitation and temperature, including lag effects, to simulate vector and human infection dynamics. Studies have highlighted shifts in seasonal and spatial transmission patterns under climate change scenarios and emphasized the importance of non-linear relationships and lagged climate effects. While some argue climate influences are modest compared to health infrastructure, drug resistance, control coverage, population growth, and migration, others find non-negligible climate impacts—especially in highland regions of East Africa and parts of South America. Altitude-driven temperature gradients reduce vector development at higher elevations, historically limiting transmission. Projections suggest increased susceptibility and potential emergence of endemicity in previously malaria-free African highlands by mid-century. However, many models lack explicit integration of rapidly changing population patterns and elevation, underscoring a gap this study addresses.

Study area: Rwanda (26,338 km²), with elevations 1000–4500 m, features bimodal rainy seasons and historically higher malaria risk in lower-elevation eastern areas. Data: Monthly malaria cases (2010–2015) from Rwanda’s Health Management Information System (HMIS); monthly Tmin, Tmax, rainfall for 30 districts from the Rwanda Meteorological Agency. HMIS data undergo annual data quality audits by PMI, Global Fund, and RBC’s MOPDD. Endemicity classes were referenced via prevalence rate (PR) thresholds; in the absence of child-specific PR, all-age PR >1% denoted endemic areas for duration assessment. Modeling: A Random Forest Model (RFM) simulated monthly malaria incidence using predictors: Tmax (lags 0–3 months), Tmin (lags 0–3), rainfall (lags 0–3), geographic coordinates (latitude, longitude), month of year, elevation, and total human population. Ten-fold cross-validation assessed performance; variable importance was based on percent increase in mean squared error upon permutation; partial dependence plots characterized non-linear effects. Statistical analyses included linear regressions, Pearson correlations, and trend analyses. Climate and population projections: Eight CMIP6 GCMs (ACCESS-CM2, CanESM5, IPSL-CM6A-LR, GFDL-ESM4, MIROC6, MPI-ESM1-2-HR, CMCC-ESM2, BCC-CSM2-MR) provided historical (2010–2015) and future (2030–2100) monthly Tmin, Tmax, rainfall under SSP2-4.5 and SSP5-8.5. A delta (change) method downscaled district-level future climate: future site climate = observed (2010–2015) + GCM change for 2030–2035, 2050–2055, 2090–2095. Population projections (0.5° SSP2/SSP5) were area-weighted to districts and combined with historical data to obtain district totals (annual; monthly values held constant within a year). The RFM trained on 2010–2015 then projected future malaria incidence for each scenario and period, analyzing seasonal and elevational patterns and peak shifts.

Historical patterns (2010–2015): • High malaria incidence clustered in eastern and southern Rwanda; western and northern highlands had low incidence, consistent with temperature constraints and topographic rain shadow effects. • Bimodal seasonality with peaks in June (post-long rains) and December (end of short rains). • Interannual coherence: 2011 had lowest incidence with lowest Tmax/Tmin and highest rainfall; 2015 had highest incidence with highest temperatures and lowest rainfall. Model performance and drivers: • RFM achieved R² = 0.79, RMSE = 0.644, MRE = 24.74% (n=2154). • Most important predictors: human population and geography (lat/long), followed by elevation and temperature; precipitation variables contributed least overall. • Rainfall exhibited strong lag effects, especially 2-month lag (positive association), while same-month rainfall showed a gradual negative association. • Tmin had accelerating positive effects on incidence; Tmax effects were more gradual (except at 3-month lag). Future climate changes: • Both SSP2-4.5 and SSP5-8.5 project substantial increases in Tmin and Tmax; Tmin commonly exceeds thresholds for P. vivax (~16°C) and P. falciparum (~18°C), enhancing parasite development. • Dry-season rainfall increases, especially under SSP5-8.5, aligning with positive lagged rainfall effects on incidence. Projected malaria changes: • Overall incidence increases under both scenarios, with notable rises during historical low-incidence months (Feb–Apr, Aug–Oct). • Seasonal fluctuations weaken from January to June; pattern shifts toward a single peak, highest in November–December. • Spatial shift of higher relative changes toward higher elevations; western and northern highlands become increasingly suitable for transmission due to warming. • Under SSP2-4.5, central plateau districts (e.g., Muhanga, Nyamagabe, Rubavu) show positive relative changes by 2030–2035; by 2050–2055, Ruhango and Kamonyi also increase; by 2090–2095, three additional districts (Nyamagabe, Nyamasheke, Muhanga) exceed annual prevalence >1% compared to 2010–2015. • Variability shifts from ~1800 m to 1800–2000 m by the 2090s. • Transmission season lengthens in eastern and southern Rwanda; by 2090–2095 under SSP5-8.5, ~15 districts may experience a 6-month transmission season and 12 districts could become year-round endemic. Peak timing shifts: • Western highlands: peak occurs earlier (to November). • Southeastern Rwanda: peak shifts later (to December). • Eastern region: peak remains in December. Overall, temperature increases set broader suitability while changing precipitation patterns (with lags) dominate seasonal and spatial variability.

The findings support the hypothesis that warming and altered rainfall patterns will expand and shift malaria transmission into Rwanda’s higher-elevation western and northern regions, historically constrained by cool temperatures. By explicitly incorporating elevation, population, and lagged climate effects in a machine learning framework, the study links observed historical dynamics to projected future shifts, explaining both increased suitability at altitude and a transition from bimodal to a more pronounced single-peak seasonality. The results emphasize that climate change could lengthen the transmission window and advance peak timing in highlands, challenging existing control strategies concentrated in traditionally high-burden lowlands. Recognizing rainfall’s lagged role clarifies why dry-season increases in precipitation translate to elevated incidence weeks to months later, especially under SSP5-8.5. These projections are relevant for early warning systems and prioritization of interventions, and align with broader evidence that climate change is already influencing malaria in East African highlands. Integrating these insights can guide resource allocation, temporal targeting of interventions, and surveillance expansion into emergent risk zones.

This work develops and validates a Random Forest modeling framework linking lagged meteorology, elevation, geography, and population to malaria incidence in Rwanda, then projects district-level risks under SSP2-4.5 and SSP5-8.5. It shows substantial future increases in malaria risk, especially at higher elevations in western and northern Rwanda, a likely shift toward a single-peak season with weakened seasonality early in the year, and earlier peaks in highlands. Policy-relevant recommendations include extending ITN distribution and indoor residual spraying throughout the year, adjusting insecticide rotations to pre-peak months (Aug–Oct), and maintaining priority in high-risk zones while adding stringent measures in potential surge areas (e.g., west-central Rwanda). Future research should incorporate quantifiable indicators of mosquito resistance, intervention coverage, and funding allocation; conduct longitudinal assessments of resistance; and expand modeling to global scales as higher-resolution health and climate data become available.

Key limitations include absence of explicit variables on mosquito insecticide resistance and malaria interventions (e.g., LLIN coverage, IRS schedules), which could improve predictive accuracy; lack of longitudinal resistance assessments; limited availability of detailed district-level health and intervention datasets; and reliance on downscaled CMIP6 outputs with inherent uncertainties, especially for precipitation. Population inputs are annual (no intra-annual variation), and some simplifying assumptions (delta method) may not capture all local climate nuances.