Metamodeling for simulation-based optimization of strategies in healthcare: an introduction and case study on colorectal cancer screening

ABSTRACT

It can be computationally challenging to perform analyses requiring many simulation model evaluations. For example, studies applying optimization algorithms to optimize healthcare strategies may require tens of thousands to millions of model evaluations. Even on modern computers, the computational burden of such analyses quickly makes them infeasible to perform within acceptable time frames, in particular for patient-level simulation models. Consequently, when optimizing healthcare strategies often only a limited set of pre-defined strategies are evaluated and compared. This may lead to suboptimal decision making if the strategy that actually is best, is not included in this pre-defined set of strategies.

In this seminar I will explain the concept of metamodeling and when its use can support computationally challenging model-based analyses. I will briefly discuss the different practical steps necessary for developing and applying metamodels, such as the identification of metamodeling techniques, the simulation of datasets used for metamodel training and testing and metamodel validation. Furthermore, I will illustrate the use of metamodeling using a case study focusing on the optimization of colorectal cancer screening using a computationally demanding simulation model. Finally, the value and limitations of using metamodeling in this case study will be discussed.

Join Zoom Meeting
https://york-ac-uk.zoom.us/j/92754503812?pwd=L1EybElFYXdtVG9lUkcwY0R1bzBlZz09

Meeting ID: 927 5450 3812
Passcode: 975430