Prediction of dengue fever outbreaks using climate variability and Markov chain Monte Carlo techniques in a stochastic susceptible-infected-removed model

Abstract

The recent increase in the global incidence of dengue fever resulted in over 2.7 million cases in Latin America and many cases in Southeast Asia and has warranted the development and application of early warning systems (EWS) for futuristic outbreak prediction. EWS pertaining to dengue outbreaks is imperative; given the fact that dengue is linked to environmental factors owing to its dominance in the tropics. Prediction is an integral part of EWS, which is dependent on several factors, in particular, climate, geography, and environmental factors. In this study, we explore the role of increased susceptibility to a DENV serotype and climate variability in developing novel predictive models by analyzing RT-PCR and DENV-IgM confirmed cases in Singapore and Honduras, which reported high dengue incidence in 2019 and 2020, respectively. A random-sampling-based susceptible-infected-removed (SIR) model was used to obtain estimates of the susceptible fraction for modeling the dengue epidemic, in addition to the Bayesian Markov Chain Monte Carlo (MCMC) technique that was used to fit the model to Singapore and Honduras case report data from 2012 to 2020. Regression techniques were used to implement climate variability in two methods: a climate-based model, based on individual climate variables, and a seasonal model, based on trigonometrically varying transmission rates. The seasonal model accounted for 98.5% and 92.8% of the variance in case count in the 2020 Singapore and 2019 Honduras outbreaks, respectively. The climate model accounted for 75.3% and 68.3% of the variance in Singapore and Honduras outbreaks respectively, besides accounting for 75.4% of the variance in the major 2013 Singapore outbreak, 71.5% of the variance in the 2019 Singapore outbreak, and over 70% of the variance in 2015 and 2016 Honduras outbreaks. The seasonal model accounted for 14.2% and 83.1% of the variance in the 2013 and 2019 Singapore outbreaks, respectively, in addition to 91% and 59.5% of the variance in the 2015 and 2016 Honduras outbreaks, respectively. Autocorrelation lag tests showed that the climate model exhibited better prediction dynamics for Singapore outbreaks during the dry season from May to August and in the rainy season from June to October in Honduras. After incorporation of susceptible fractions, the seasonal model exhibited higher accuracy in predicting outbreaks of higher case magnitude, including those of the 2019–2020 dengue epidemic, in comparison to the climate model, which was more accurate in outbreaks of smaller magnitude. Such modeling studies could be further performed in various outbreaks, such as the ongoing COVID-19 pandemic to understand the outbreak dynamics and predict the occurrence of future outbreaks.

Dengue Fever: From Extreme Climates to Outbreak Prediction

“…The general idea is that a DF outbreak in a province has a link to significant climate change events in that province rather than small climate perturbations. Thus, instead of directly feeding raw climate data into prediction models like all existing works [6], [8], [11], [22], we discretize climate data into events of extreme (abnormal) weather conditions and use them as inputs for our prediction model. In this way, we also flatten the differences in climate value ranges of different provinces and thus create a unified view for all provinces based on climate events.…”

“…E.g., [23] found a positive correlation between rainfall and DF incidence with a lag time from 0-3 months in Hanoi, Vietnam, while [3] found no significant correlation. Based on these factors, a wide range of Machine Learning (ML) models have been employed to predict DF incidence rates/cases or outbreaks for many different areas e.g., Queensland in Australia [11], Guangzhou in China [8], Singapore [6], Honduras [6], Brazil [24], Bangkok in Thailand [18], Selangor in Malaysia [25], and Vietnam [13]. These models range from traditional to recent deep learning methods, e.g., Seasonal Autoregressive Integrated Moving Averaged (SARIMA) [11], Poisson regression [26], Support Vector Regression (SVR) [8], Gradient Boosting Machine (GBM) [7], [8], Generalized Additive Models (GAMs) [8], Generalized Linear Mixed Models (GLMMs) [13], Artificial Neural Networks (ANNs) [9], Back-propagation neural network (BPNNs) [7], Long-short term memory (LSTM) [7], Convolution Neural Networks (CNNs) [10], and Transfomer [10].…”

“…Most of these techniques are heavily tailored to specific data (which are not publicly available) and specific areas e.g., [13], [18], which thus limits their applicability in wider contexts. Most techniques only predict DF for a single area (city or province) [6], [18]. Others study multiple areas, e.g., [7], [10], [13].…”

Analyzing hydro-meteorological data from Yamuna river basin, western Himalaya: Using a Markov Chain and LSTM approach to forecast future disasters

“…Because of this purpose, rainfall data was modelled using the Markov Chain approach in the Yamuna River basin. For a problem to be modelled as a Markov process, a series of states have to be defined, and the time interval over which these transitions will occur has to be established, respectively (Martheswaran et al, 2022).…”

Prediction of dengue incidents using hospitalized patients, metrological and socio-economic data in Bangladesh: A machine learning approach

“…Martheswaran et al . [ 19 ] analyzed case report data from 2012 to 2020 to estimate dengue incidence in Singapore and Honduras using the random-sampling-based susceptible-infected-removed (SIR) model. Aside from that, the suggested model was fitted using the Bayesian Markov Chain Monte Carlo (MCMC) technique.…”