Leptospirosis is a zoonotic disease caused by the bacterium Leptospira, transmitted primarily through contact with soil or water contaminated by urine of infected rodents or animals. It causes acute febrile illness and is endemic across tropical regions worldwide, yet it remains understudied, largely due to diagnostic difficulty. Hydrometeorological conditions are widely understood to modulate leptospirosis risk, but the specific drivers and lag structures remain poorly characterized, particularly in data-sparse settings. This study first evaluated satellite and reanalysis precipitation and temperature products (IMERG, ERA5, MERRA-2, and Amazon LDAS) against one another and against in situ station data in Loreto, Peru. Because rainfall in the region varies strongly over small spatial scales, remotely sensed products were hypothesized to capture its spatial variability better than the sparse station network. IMERG performed best for precipitation and ERA5 for temperature. The study then examined hydrometeorological drivers of leptospirosis risk in the two largest cities in Loreto: Iquitos (pooled across the districts of Iquitos, Belén, Punchana, and San Juan Bautista) and Yurimaguas. A separate Bayesian negative binomial model was fit to weekly leptospirosis case counts for each city. Candidate predictors were first evaluated individually across seven lag windows to identify each predictor's optimal lag: basin-scale precipitation (Amazon, Marañón, Itaya, Nanay, and Huallaga basins), district-scale precipitation, weekly maximum and minimum temperature, the 1- and 3-month Standardized Precipitation Evapotranspiration Index (SPEI), and the Oceanic Niño Index (ONI). Basin-scale precipitation outperformed district-scale precipitation as a predictor. Spearman correlation analysis of the predictors at their optimal lags revealed strong collinearity among all rainfall variables, so rainfall was reduced to a single basin predictor per city: Amazon basin precipitation for Iquitos and Huallaga basin precipitation for Yurimaguas. Optimal accumulation windows for each basin were then identified within the same Bayesian modeling framework. For the Amazon basin, two windows performed strongly, a 5-month accumulation at a 4-month lag and a 3-week accumulation at a 1-month lag, and both were retained as candidates because their coefficients had opposite signs. For the Huallaga basin, the optimal window was a 3-month accumulation at a 4-month lag. Projection predictive variable selection was then used to identify, for each city, the smallest subset of predictors whose predictive performance was comparable to that of the full reference model. The final Iquitos model retained Amazon basin precipitation (5-month accumulation at a 4-month lag) and ONI (6-month lag), while the Yurimaguas model retained minimum temperature (5-month lag) and ONI (1-month lag). These results show that hydrometeorological drivers of leptospirosis risk differ substantially between the two cities, with most predictors operating at multi-month lags. This lag structure suggests that freely and regularly available satellite and climate index data could support early warning systems, though such systems would need to be tailored to local hydrological conditions rather than applied uniformly.
Leptospirosis is a zoonotic disease caused by the bacterium Leptospira, transmitted primarily through contact with soil or water contaminated by urine of infected rodents or animals. It causes acute febrile illness and is endemic across tropical regions worldwide, yet it remains understudied, largely due to diagnostic difficulty. Hydrometeorological conditions are widely understood to modulate leptospirosis risk, but the specific drivers and lag structures remain poorly characterized, particularly in data-sparse settings. This study first evaluated satellite and reanalysis precipitation and temperature products (IMERG, ERA5, MERRA-2, and Amazon LDAS) against one another and against in situ station data in Loreto, Peru. Because rainfall in the region varies strongly over small spatial scales, remotely sensed products were hypothesized to capture its spatial variability better than the sparse station network. IMERG performed best for precipitation and ERA5 for temperature. The study then examined hydrometeorological drivers of leptospirosis risk in the two largest cities in Loreto: Iquitos (pooled across the districts of Iquitos, Belén, Punchana, and San Juan Bautista) and Yurimaguas. A separate Bayesian negative binomial model was fit to weekly leptospirosis case counts for each city. Candidate predictors were first evaluated individually across seven lag windows to identify each predictor's optimal lag: basin-scale precipitation (Amazon, Marañón, Itaya, Nanay, and Huallaga basins), district-scale precipitation, weekly maximum and minimum temperature, the 1- and 3-month Standardized Precipitation Evapotranspiration Index (SPEI), and the Oceanic Niño Index (ONI). Basin-scale precipitation outperformed district-scale precipitation as a predictor. Spearman correlation analysis of the predictors at their optimal lags revealed strong collinearity among all rainfall variables, so rainfall was reduced to a single basin predictor per city: Amazon basin precipitation for Iquitos and Huallaga basin precipitation for Yurimaguas. Optimal accumulation windows for each basin were then identified within the same Bayesian modeling framework. For the Amazon basin, two windows performed strongly, a 5-month accumulation at a 4-month lag and a 3-week accumulation at a 1-month lag, and both were retained as candidates because their coefficients had opposite signs. For the Huallaga basin, the optimal window was a 3-month accumulation at a 4-month lag. Projection predictive variable selection was then used to identify, for each city, the smallest subset of predictors whose predictive performance was comparable to that of the full reference model. The final Iquitos model retained Amazon basin precipitation (5-month accumulation at a 4-month lag) and ONI (6-month lag), while the Yurimaguas model retained minimum temperature (5-month lag) and ONI (1-month lag). These results show that hydrometeorological drivers of leptospirosis risk differ substantially between the two cities, with most predictors operating at multi-month lags. This lag structure suggests that freely and regularly available satellite and climate index data could support early warning systems, though such systems would need to be tailored to local hydrological conditions rather than applied uniformly.
Hydrometeorological Drivers of Leptospirosis Risk in the Peruvian Amazon
BELL, ELIZABETH GRACE
2025/2026
Abstract
Leptospirosis is a zoonotic disease caused by the bacterium Leptospira, transmitted primarily through contact with soil or water contaminated by urine of infected rodents or animals. It causes acute febrile illness and is endemic across tropical regions worldwide, yet it remains understudied, largely due to diagnostic difficulty. Hydrometeorological conditions are widely understood to modulate leptospirosis risk, but the specific drivers and lag structures remain poorly characterized, particularly in data-sparse settings. This study first evaluated satellite and reanalysis precipitation and temperature products (IMERG, ERA5, MERRA-2, and Amazon LDAS) against one another and against in situ station data in Loreto, Peru. Because rainfall in the region varies strongly over small spatial scales, remotely sensed products were hypothesized to capture its spatial variability better than the sparse station network. IMERG performed best for precipitation and ERA5 for temperature. The study then examined hydrometeorological drivers of leptospirosis risk in the two largest cities in Loreto: Iquitos (pooled across the districts of Iquitos, Belén, Punchana, and San Juan Bautista) and Yurimaguas. A separate Bayesian negative binomial model was fit to weekly leptospirosis case counts for each city. Candidate predictors were first evaluated individually across seven lag windows to identify each predictor's optimal lag: basin-scale precipitation (Amazon, Marañón, Itaya, Nanay, and Huallaga basins), district-scale precipitation, weekly maximum and minimum temperature, the 1- and 3-month Standardized Precipitation Evapotranspiration Index (SPEI), and the Oceanic Niño Index (ONI). Basin-scale precipitation outperformed district-scale precipitation as a predictor. Spearman correlation analysis of the predictors at their optimal lags revealed strong collinearity among all rainfall variables, so rainfall was reduced to a single basin predictor per city: Amazon basin precipitation for Iquitos and Huallaga basin precipitation for Yurimaguas. Optimal accumulation windows for each basin were then identified within the same Bayesian modeling framework. For the Amazon basin, two windows performed strongly, a 5-month accumulation at a 4-month lag and a 3-week accumulation at a 1-month lag, and both were retained as candidates because their coefficients had opposite signs. For the Huallaga basin, the optimal window was a 3-month accumulation at a 4-month lag. Projection predictive variable selection was then used to identify, for each city, the smallest subset of predictors whose predictive performance was comparable to that of the full reference model. The final Iquitos model retained Amazon basin precipitation (5-month accumulation at a 4-month lag) and ONI (6-month lag), while the Yurimaguas model retained minimum temperature (5-month lag) and ONI (1-month lag). These results show that hydrometeorological drivers of leptospirosis risk differ substantially between the two cities, with most predictors operating at multi-month lags. This lag structure suggests that freely and regularly available satellite and climate index data could support early warning systems, though such systems would need to be tailored to local hydrological conditions rather than applied uniformly.| File | Dimensione | Formato | |
|---|---|---|---|
|
Bell_ElizabethGrace.pdf
embargo fino al 13/10/2027
Dimensione
6.19 MB
Formato
Adobe PDF
|
6.19 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/119051