INTRODUCTION TO MICROSIMULATION
Course Level: Intermediate
Course Limit: 15
Overview: The course is designed to serve as an intermediate gateway from "usual" expected-value, cohort-based, health state transition (Markov) models to more advanced model structures that rely on individual patient-level simulation (e.g. two-dimensional Monte Carlo-based simulation).
Background: Discrete-time, health state transition (HST) models are commonly used in a variety of scenarios. Traditional HSTs lack ‘memory’. In other words, events occurring prior to the individual entering a state are not considered, implying that individuals inhabiting a state at a particular time are homogeneous with respect to the risk of subsequent events. Representing the subtleties and nuances of clinical or health policy problems often requires a plethora of health states. Microsimulation is an alternative modeling paradigm that allows one hypothetical person to traverse the model at a time. Individuals can be assigned attributes that depend on prior history and allow for the memory that is lacking in traditional HST models. The latter allows for substantial simplification model structure. Apart from solving the memory issue, Microsimulation methods can also be used to handle a number of practical modeling tasks and can serve as a foundation for more complex model structures.
Format Requirements: The class will be divided into three segments each with a 30 minute didactic session followed by a 30 minute hands-on exercise. The exercises will conducted using TreeAge software. Limited-time licenses will be provided to participants who have neither a full or student license. Familiarity with building expected-value, cohort-type, Markov models in TreeAge will be assumed.
Description and Objectives:
Part 1 - The limits of expected value (EV) health state transition (HST) models, Microsimulation theory and an introduction to tracker variables
Part2 - More on tracker variables, reducing the number of health states required by the model via the use of trackers, microsimulation to explore individual-level variability, using microsimulation to ensure internal mdel validity
Part 3 - Reporting frequencies of interim events within models, extension of microsimulation to dynamic models, discrete event simulation and two dimensional simulation
- To understand the limitations of EV, cohort HSTs
- To grasp microsimulation theory
- To simplify the structure of a HST by using microsimulation and tracker variables
- To simulate patient-level variability by sampling from distributions of patient attributes
- To verify internal model function by using tracker variables and global matrices
- To validate model function by comparing intermediate outcomes with observed data
- To extend a microsimulation-based model to two-dimensional simulation
David Naimark, MD, MSc, BSc
University of Toronto
Institute of Health Policy, Management and Evaluation