Malaria risk assessment and mapping using satellite imagery and boosted regression trees in the Peruvian Amazon

Angel Rosas-Aguirre,Morgan Pearcy,Niko Speybroeck,Diamantina Moreno-Gutierrez,Marie-Pierre Hayette,Walter Valdivia,Philippe Lejeune,José Pasapera-Gonzales,Elisa Solano-Villarreal,Catherine Linard,Alejandro Llanos-Cuentas

doi:10.1038/s41598-019-51564-4

Abstract

This is the first study to assess the risk of co-endemic Plasmodium vivax and Plasmodium falciparum transmission in the Peruvian Amazon using boosted regression tree (BRT) models based on social and environmental predictors derived from satellite imagery and data. Yearly cross-validated BRT models were created to discriminate high-risk (annual parasite index API > 10 cases/1000 people) and very-high-risk for malaria (API > 50 cases/1000 people) in 2766 georeferenced villages of Loreto department, between 2010–2017 as other parts in the article (graphs, tables, and texts). Predictors were cumulative annual rainfall, forest coverage, annual forest loss, annual mean land surface temperature, normalized difference vegetation index (NDVI), normalized difference water index (NDWI), shortest distance to rivers, time to populated villages, and population density. BRT models built with predictor data of a given year efficiently discriminated the malaria risk for that year in villages (area under the ROC curve (AUC) > 0.80), and most models also effectively predicted malaria risk in the following year. Cumulative rainfall, population density and time to populated villages were consistently the top three predictors for both P. vivax and P. falciparum incidence. Maps created using the BRT models characterize the spatial distribution of the malaria incidence in Loreto and should contribute to malaria-related decision making in the area.

Highlights

In spite of the investment in control and prevention allocated by the Peruvian government over the last decades, malaria due to both Plasmodium vivax and P. falciparum remains a significant public health issue in the country
Using cross-validated boosted regression tree (BRT) and remote-sensing (RS) satellite data, we modelled the distribution of malaria incidence by Plasmodium species in the Peruvian Amazon at village level between 2010 and 2017, and identified the most critical factors associated with this distribution
BRT models built with predictor data were able to efficiently discriminate the species-specific malaria risk in villages of the same year, and most of these models performed well when predictor data was used to discriminate malaria risk in the following year

Summary

Introduction

In spite of the investment in control and prevention allocated by the Peruvian government over the last decades, malaria due to both Plasmodium vivax and P. falciparum remains a significant public health issue in the country. Loreto is a hypoendemic malaria area[4,5], but malaria transmission in the department is highly heterogeneous, with some areas in remote Amazonia having entomological inoculation rates (EIRs) near those reported in Africa. This heterogeneity creates opportunities for targeted interventions, provided high-risk hotspots are first identified[3,6]. Satellite data[26,27] were used to derive environmental and social potential predictors[28,29,30] of malaria risk in villages of Loreto during the period 2010–2017. Maps identifying villages at highest risk were created to support the malaria decision making in Loreto

Methods

Results

Conclusion