Performing global sensitivity analysis on simulations of a continuous-time Markov chain model motivated by epidemiology

TL;DR

This paper applies a global sensitivity analysis methodology to a continuous-time Markov chain epidemiological model, assessing the effects of parameters, randomness, and their interactions using advanced simulation algorithms.

Contribution

It adapts and applies a sensitivity analysis framework from chemical reaction networks to epidemiology, including a novel discussion on simulation algorithm impacts.

Findings

Different simulation algorithms influence sensitivity analysis results

Quantification of parameter and randomness effects on epidemic outcomes

Comparison of three exact simulation algorithms for SARS-CoV-2 model

Abstract

In this paper we apply a methodology introduced in Navarro Jimenez et al (2016) in the framework of chemical reaction networks to perform a global sensitivity analysis on simulations of a continuous-time Markov chain model motivated by epidemiology. Our goal is to quantify not only the effects of uncertain parameters such as epidemic parameters (transmission rate, mean sojourn duration in compartments), but also those of intrinsic randomness and interactions between epidemic parameters and intrinsic randomness. For that purpose, following what was proposed in Navarro Jimenez et al, we leverage three exact simulation algorithms for continuous-time Markov chains from the state of the art which we combine with common tools from variance-based sensitivity analysis as introduced in Sobol (1993). Also, we discuss the impact of the choice of the simulation algorithm used for the simulations on the results of sensitivity analysis. Such a discussion is new, at least to our knowledge. In a numerical section, we implement and compare three sensitivity analyses based on simulations obtained from different exact simulation algorithms of a SARS-CoV-2 epidemic model.