Prediction of the Risk Distributions for Anopheles sinensis, a Vector for Malaria in Shanghai, China

Abstract

Malaria is a parasitic disease caused by Plasmodium, and Anopheles sinensis is a vector of malaria. Although malaria is no longer indigenous to China, a high risk remains for local transmission of imported malaria. This study aimed to identify the risk distribution of vector An. sinensis and malaria transmission. Using data collected from routine monitoring in Shanghai from 2010 to 2020, online databases for An. sinensis and malaria, and environmental variables including climate, geography, vegetation, and hosts, we constructed 10 algorithms and developed ensemble models. The ensemble models combining multiple algorithms (An. sinensis: area under the curve [AUC] = 0.981, kappa = 0.920; malaria: AUC = 0.959, kappa = 0.800), with the best out-of-sample performance, were used to identify important environmental predictors for the risk distributions of An. sinensis and malaria transmission. For An. sinensis, the most important predictor in the ensemble model was moisture index, which reflected degree of wetness; the risk of An. sinensis decreased with higher degrees of wetness. For malaria transmission, the most important predictor in the ensemble model was the normalized differential vegetation index, which reflected vegetation cover; the risk of malaria transmission decreased with more vegetation cover. Risk levels for An. sinensis and malaria transmission for each district of Shanghai were presented; however, there was a mismatch between the risk classification maps of An. sinensis and malaria transmission. Facing the challenge of malaria transmission in Shanghai, in addition to precise An. sinensis monitoring in risk areas of malaria transmission, malaria surveillance should occur even in low-risk areas for An. sinensis.

Prediction of Oncomelania hupensis distribution in association with climate change using machine learning models

“…Machine learning models can better handle multidimensional data and have been widely used for parasitic disease risk prediction and vector spread trends [ 43 , 44 ]. Previous research has shown that presence/absence-based models were superior to presence-based models in terms of predictive performance [ 19 ].…”