Abstract :
[en] Background: Malaria in Loreto department remains a public health problem, accounting for more than 90% of reported cases in Peru. This is the first study in the Peruvian Amazon aimed at assessing the risk of malaria transmission using satellite imagery and Boosted Regression
Trees (BRT).
Methods: Villages with at least one malaria case between 2010 and 2015 from the routine surveillance data in Loreto were georeferenced and their cases aggregated by year and species. Social and environmental variables were derived from Landsat satellite imagery and other spatial data, then included as explanatory variables into a crossvalidated Poisson BRT model for malaria incidence at the local level.
Time-dependent explanatory variables included forest coverage (FC, %), annual forest loss (FL,%), cumulative annual rainfall (CAR, mm), annual-mean land surface temperature (LST, oC), normalised difference vegetation index (NDVI), and normalised difference water index (NDWI).
Other variables were Euclidean shortest distance to rivers (SDR, meters), time to major populated villages/towns (TPV, minutes), and night-time lights (NTL, mean value 2010-2013) as proxy of population density. BRT accounts for nonlinearities and interactions between factors with high
predictive accuracy for disease risk mapping.
Results: A total of 1524 villages were included in the analysis (70% of total Loreto’s villages). More than 90% of relative influence in the overall malaria incidence was explained by five variables: NTL (67.8%), TPV (8.1%), FC (6.5%), CAR (5%) and SDR (4.6%). The analysis
by species showed a higher influence of environmental variables (CAR, LST, NDVI and NDWI) for P. falciparum (18.4%) than for P. vivax incidence (9.7%). Malaria risk maps were generated based on model predictions taking into account the relative influence of variables.
Conclusions: Remotely sensed data analysed using BRT allowed for maps delimiting areas of high malaria risk in Loreto. These maps will help malaria stakeholders to prioritise areas for control interventions.
