Sensitivity Analysis and Uncertainty Analysis in a Large-scale Agent-based Simulation Model of Infectious Diseases

摘要

The purpose of this study is to develop appropriate statistical methods and procedures for dealing with parameter uncertainty and for improving the computational efficiency of sensitivity analysis in a large-scale agent-based model of infectious disease. An agent-based model is a rule-based computational simulation model that can keep track of the dynamical activities of all agents and their interactions within an environment and analyze the course of a disease through the population and evaluate interventions. Sensitivity analysis is a method for quantifying uncertainty in a complex model by systematically changing inputs (parameters and initial conditions) of the model and quantifying the consequences for the output of the model. Sensitivity analysis and uncertainty analysis are used for agent-based model to analyze the uncertainty in the model.udThe specific aims of the study are to (1) develop specific procedures and criteria to determine important input parameters in the FRED agent-based influenza model; (2) develop specific procedures and criteria to determine high sensitivity parameters in the FRED agent-based influenza model via local sensitivity analysis; (3) improve the computational efficiency of sensitivity analysis by comparing two sampling procedures for probabilistic sensitivity analysis in agent-based models: simple random sampling and Latin Hypercube sampling; and (4) apply uncertainty analysis procedures to evaluate the cost-effectiveness for different school closure intervention strategies as well as the reliability of the uncertainty analysis in the FRED agent-based influenza model.udThis study emphasizes the important role of sensitivity analysis, uncertainty analysis and statistical analysis in making better use of simulation results for decision-making in the control of infectious disease. In this study, the FRED (Framework for Replicating Epidemic Dynamics) influenza model is used to produce all the simulation results from sensitivity analysis. The methods and procedures that are developed in this study can be generalized to all kinds of disease models under the FRED framework. udIn public health practice, this study will help to provide timely responses for decision-making when there is a public health crisis. It also provides important information for public health policy makers about how certainly the FRED framework can provide reliable intervention comparison results for decision-making.