Application of sensitivity- and uncertainty-based techniques for the assessment of epidemiological models in real-life study cases
Fecha
2019
Autores
Rojas Díaz, Daniel
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad EAFIT
Resumen
Uncertainty analysis (UA) and sensitivity analysis (SA) are tools to assess and to quantify the uncertainty spread from the input factors (parameters and initial states) to the model output, taking into account the effect of the interactions among those factors. Throughout the following works, I treat UA as a graphical assessment of uncertainty propagation based on Monte Carlo simulation, which makes it possible to state a range for the model output in cases where it is considered relevant. On the other hand, I privilege the global approach for SA instead of the local one, since the first attempts to quantify the uncertainty contribution of the model factors in their entire distribution range while the second one is only informative for a single locus in the distribution. In this way, when applying global UA/SA on a model, it is possible to identify those factors that mostly determine the model behavior. Furthermore, I have noticed that the concepts and principles of UA/SA are associated with other main tasks in modeling, as factors estimation and confidence intervals achievement: Briefly, those non-identifiable factors in a model (factors whose value can not be estimated uniquely from some information about output data) should belong to the categories of non-sensible or sensitive but correlated from SA; and, the sub-space of the space of factors where the factors may jointly exist producing a model output that fits, in some extent, to a given output data, could be approximately estimated with UA-based approaches, constituting a new kind of confidence interval. Thus, in this compendium, I present five works related to the applications of UA/SA techniques as well as its relevance. The objective of those applications evolves from the most logically immediate to some derived and more complex ones, though still preserving the model pertinence as a central topic.