MICROSIMULATION AS A TOOL TO MODEL HEALTH CARE DECISIONS

Background: Medical decision models that incorporate significant clinical detail become quite complex: standard state-transition models, burdened by Markovian limitations of path or history independence, may require thousands of states to represent clinically realistic situations. Individual microsimulation, where single individuals are represented and enter the model independently, is a technique that allows the presentation of complex clinical detail without an explosion of states, and allows for an intuitive, clinically realistic modeling framework.

Format Requirements: The class will be a lecture-based description of the use of microsimulation and Monte Carlo methods for modeling complex health care problems. Students will be expected to have a basic understanding of decision modeling, including the construction of basic decision trees, the use of simple state-based (Markov) processes, and should understand the basic concepts of sensitivity analysis. Examples using microsimulation to model liver disease and HIV disease will be presented.

Description and Objectives: The goals of the course are to provide an understanding of the limitations of cohort-based simulations, and illustrate how microsimulation releases many of the limitations implied by using standard state based (Markov) models. The course will:

• Review basic modeling types (branch and node decision trees, state transition models) and describe their limitations
• Describe the technique if individual microsimulation and demonstrate its equivalence to constructing very large state transition model
• Illustrate how individual heterogeneity can be incorporated into microsimulation models
• Provide examples of the use of microsimulation from several disease models
• Describe the computational and interpretive issues inherent in microsimulation models, especially with regards to sensitivity analysis
• Define and illustrate 1^st order and 2^nd order Monte Carlo methods

Throughout the course, recommendations from the SMDM/ISPOR workshop on best modeling practices will be reviewed.

The objectives of this short course are:

• Understand the limitations of cohort-based modeling
• Distinguish types of problems that may require microsimulation from those that do not
• Learn best practices in microsimulation model construction
• Distinguish the types of sensitivity analysis and understand the computational issues involved in probabilistic sensitivity analysis in microsimulation