Forecasting tools of infectious disease epidemicsEpidemic models are increasingly used for forecasting purposes. The predictive performance of mathematical models has been demonstrated through empirical studies of spatiotemporal epidemic prediction, in particular, via global forecasting exercises of emerging infectious diseases [1,2]. Such models are represented by those applied to SARS, influenza A (H1N1-2009) and Middle East respiratory syndrome [1]. The success of global forecasting has been partly supported by recent advances in computational statistics, especially those employing the Markov Chain Monte Carlo technique. Additionally, the epidemiological dynamics of global spread has been shown to be mainly (and mostly) characterized by the global mobility pattern of humans [3]. The time from emergence of a novel infectious disease in a country to arrival at an importing country is linearly predicted by using a simple metric of the airline transportation network (which is referred to as the 'effective distance'). This has not only enabled us to exploit the metric for real-time forecasting of global spread of Middle East respiratory syndrome and Zika virus infection [4,5], but also led us to analyze the effectiveness of travel restriction policies in preventing international spread of Ebola virus disease [6].In addition to the progress in forecasting theory and techniques, epidemiological models have been considerably improved to capture actual transmission dynamics using epidemiological determinants of transmission as an additional input variable (in addition to patient data from surveillance). For instance, the dependence of influenza transmission on humidity has been empirically demonstrated by adding humidity as an explanatory variable of the transmission coefficient to the so-called susceptible-infectious-recovered model [7]. In this instance, humidity acts as an input variable of part of the transmission model parameters, and it is pleasing that both mechanistic understanding of the epidemic dynamics and predictive performance are jointly improved only by replacing a constant transmission coefficient by a simple mathematical function of humidity.Promising mechanistic models for forecasting would require additional data to surveillance information in order to ensure improved performance. However, airline transportation network data and spatiotemporal data of humidity are both regarded as part of the big data, and their incorporation into epidemic modeling would require substantial effort and computational resources for data collection and analysis. It is often practically the case that the input data are very limited, especially in cases where the epidemic is in a country where the civil war is underway. Forecasting in a data-limited setting requires a more parsimonious modeling approach. Here we describe how we overcame the limitation of input data during the cholera epidemic in Yemen, 2017 [8].
Cholera in YemenCholera is a bacterial infection caused by Vibrio cholerae. Through contaminated water or an e...