Article Preview
Top1. Introduction
The epidemiological field has been greatly enhanced by the use of computational and mathematical models, e.g. the studies of Anderson and May (1991), Weber et al, 1997; Keeling and Rohani (2008), Amouroux et al. (2010) and Hamami and Atmani (2013). Such models are considered indispensable both to understand the pathophysiology of human disease and to follow the spread of disease. The latter in particular allows public health policies to be developed by using predictive models to explore suitable disease control strategies.
For any modelling, the main goal is to provide accurate disease representation and realistic long term prediction; at least, as far as possible given that “the real world is undeniably replete with many complications; economic and social as well as biological” (Anderson and May, 1991). Capturing the complex, dynamic and variable nature of disease spread depends on strong partnership working between epidemiologists and modellers, to achieve careful refinement, elaboration and optimisation of models. Even so, the developed models (Anderson and May, 1991; Frost, 1995; Oaken et al., 2014) rely heavily on the experience of the experts and developers, and a degree of speculation and inspiration regarding identification of pertinent model features or accurate parameter estimation. Keeling and Rohani (2008) confirm this point of view: “The feasibility of model complexity is compromised by computational power, the mechanistic understanding of disease natural history, and the availability of necessary parameters. Consequently, the accuracy of any model is always limited”. However, relying on expert knowledge and assumptions is not enough to ensure model accuracy when this depends on knowledge or features unknown to the expert/developer team.
In this context, many works (vynnycky and Fine, 1997; Debanne, 2000; Geisweiller, 2006; Prandi, 2010; DeEspíndola et al. 2011; Oaken REF, Goeyvaerts, 2015) focus on optimisation, as it becomes as a natural step in the modelling process. Optimisation has grown in recent years from considering simply parameter values, to refining model structure. Of great help in this process is the availability of massively complex datasets on epidemics, containing quantitative, qualitative, textual, Boolean, etc., information (Maumus et al., 2005). Our conclusion is that to decrease uncertainty in epidemic modelling, providing rigorous model descriptions containing the most important system features so parameters can then be correctly estimated, it is urgent to devise a solution to assisting experts/developers in acquiring only the most pertinent information from a dataset, and allow them to review their reasoning about the underlying epidemic system (Moundalexis and Nag, 2013).