Inference and prediction of malaria transmission dynamics using time series data

Abstract

Background: Disease surveillance systems are essential for effective disease intervention and control by monitoring disease prevalence as time series. To evaluate the severity of an epidemic, statistical methods are widely used to forecast the trend, seasonality, and the possible number of infections of a disease. However, most statistical methods are limited in revealing the underlying dynamics of disease transmission, which may be affected by various impact factors, such as environmental, meteorological, and physiological factors. In this study, we focus on investigating malaria transmission dynamics based on time series data. Methods: A data-driven nonlinear stochastic model is proposed to infer and predict the dynamics of malaria transmission based on the time series of prevalence data. Specifically, the dynamics of malaria transmission is modeled based on the notion of vectorial capacity (VCAP) and entomological inoculation rate (EIR). A particle Markov chain Monte Carlo (PMCMC) method is employed to estimate the model parameters. Accordingly, a one-step-ahead prediction method is proposed to project the number of future malaria infections. Finally, two case studies are carried out on the inference and prediction of Plasmodium vivax transmission in Tengchong and Longling, Yunnan province, China. Results: The results show that the trained data-driven stochastic model can well fit the historical time series of P. vivax prevalence data in both counties from 2007 to 2010. Moreover, with well-trained model parameters, the proposed one-step-ahead prediction method can achieve better performances than that of the seasonal autoregressive integrated moving average model with respect to predicting the number of future malaria infections. Conclusions: By involving dynamically changing impact factors, the proposed data-driven model together with the PMCMC method can successfully (i) depict the dynamics of malaria transmission, and (ii) achieve accurate one-step-ahead prediction about malaria infections. Such a data-driven method has the potential to investigate malaria transmission dynamics in other malaria-endemic countries/regions.

Assessing the spatiotemporal malaria transmission intensity with heterogeneous risk factors: A modeling study in Cambodia

“…Here, we introduce a comprehensive representation of vector density for a location (e.g., a district in our study) by combining the human and vector distributions for different types of land cover to explore the spatial heterogeneity of the vector density for measuring the local malaria transmission risk in an accurate way. Based on the assumed linear relationship between the mosquito population and rainfall ( Ceccato et al., 2012 ; Connor, 2002 ; Shi et al., 2020 ), we integrate the mosquito density, which is based on the volume of rainfall, into a weighted linear function in terms of different types of land cover. Moreover, for each type of land cover, the estimation of vector density can be obtained based on the distribution of vector occurrence and that of the human population density for the selected type of land cover in a gridwise manner.…”

“…In matrix , some parameters are directly calculated using the climate data based on domain knowledge ( Ceccato et al., 2012 ; Shi et al., 2020 ), while the rest are inferred via the standard gradient descent method in optimization. The details of optimization procedure can be found in the Supplementary Material.…”

“…For local transmission risk, the vectorial capacity (VCAP, i.e., the expected number of infectious bites from mosquitoes per day brought by a single infectious case), the entomological inoculation rate (EIR, i.e., the expected number of infectious bites from mosquitoes per day per human), and the basic reproduction number ( , i.e., the expected number of infected cases which is brought by a single infectious case) are widely adopted to evaluate the transmission potential and guide intervention planning ( Brady et al., 2016 ; Routledge et al., 2018 ; Smith & McKenzie, 2004 ). However, these metrics do not fully consider the heterogeneity of vectors and environments ( Ceccato et al., 2012 ; Shi et al., 2020 ), such as the distributions of the dominant vector species and different types of land covers ( Munga et al., 2006 , 2009 ; Mushinzimana et al., 2006 ). When measuring imported risks, there has been a focus on identifying the characteristics of mobile and migrant populations (MMPs) and assessing the risk level of MMPs ( Grietens et al., 2015 ; World Health Organization & Regional Office for South-East Asia, 2017 ).…”

“…As a result, the impact of various risk-related factors on the disease transmission intensity or risk cannot be quantified in a precise way; and it is hard to provide clear guidance for decision-makers to implement time/location/factor-specific measures in response to potential risks or outbreaks according to the modeling results. Recent studies have started to incorporate some domain knowledge, such as the ordinary differential equations ( Kargas et al., 2021 ; Wang et al., 2021 ) or the metrics (e.g., the VCAP and EIR) ( Shi et al., 2017 , 2020 ; B. Yang et al., 2014 ) that characterize the disease transmission process, into the data-driven models.…”

Exploring predictive frameworks for malaria in Burundi

“…In the literature, not only statistical linear models and machine learning based models are used to estimates malaria cases. Dynamics models ( Mandal et al., 2011 ), agent-based models ( Smith et al., 2018 ), and time series ( Hussien et al., 2017 ; Shi et al., 2020 ) are also used. However, if we want to understand better the effects of different factors, GLM models and machine learning based models provide more insights.…”