SENSITIVITY ANALYSIS USING LINEAR REGRESSION METAMODELING

Format Requirements: This course consists of lectures interspaced with “hands-on” experience applying metamodeling in simulation models. Participants will work through structured examples using their own computers. Data sets and files needed for the course will be distributed during the course session. A basic level experience with probabilistic sensitivity analysis and linear regression are preferred, but not required.

Format Requirements: Sensitivity analysis reveals the confidence in a certain course of action and is an important component of analyses involved in medical decision making. However, conducting sensitivity analysis at various stages of model development is often a laborious and time-consuming task. In addition, it is often challenging to document and reproduce the results of prior sensitivity analyses. This course teaches the basics of simulation metamodeling as a technique to improve the practice of sensitivity analysis. Metamodels have been used for nearly half a century in many scientific fields that adopt simulation models such as physics and engineering. A metamodel is generally a regression model that can reveal various model characteristics, including sensitivity analysis.

This course consists of lectures interspaced with “hands-on” experience applying metamodeling in simulation models. Participants will work through structured examples using their own computers. Data sets and files needed for the course will be distributed during the course session. A basic level experience with probabilistic sensitivity analysis and linear regression are preferred, but not required.

The purpose of this course is to familiarize students with the concepts and methods of linear regression metamodeling to improve the reporting and documentation of sensitivity analyses results. This course will also examine the desirable statistical properties of metamodels.

By the end of this course, participants will

 *   be familiar with the theory and basics of linear regression metamodeling,
 *   learn the statistical advantages of using metamodeling,
 *   gain hands-on-experience implementing linear regression metamodeling in Microsoft Excel,
 *   understand how to use R to generate a panel of sensitivity analyses, including one-way and two-way parameter sensitivity analysis, and threshold analyses, and
 *   produce publication-quality figures and tables to communicate and document simulation model findings.