Changing Time Representation in Microsimulation Models

Eric Wong

doi:10.1177/0272989X251319808

Changing Time Representation in Microsimulation Models

2025-02-25

0

PRE

AI

E

Eric Wong

W

Wanrudee Isaranuwatchai

DOI：10.1177/0272989X251319808

Original

Original request for help

Save

Abstract

En 中文

Background In microsimulation models of diseases with an early, acute phase requiring short cycle lengths followed by a chronic phase, fixed short cycles may lead to computational inefficiency. Examples include epidemic or resource constraint models with early short cycles where long-term economic consequences are of interest for individuals surviving the epidemic or ultimately obtaining the resource. In this article, we demonstrate methods to improve efficiency in such scenarios. Furthermore, we show that care must be taken when applying these methods to epidemic or resource constraint models to avoid bias.Methods To demonstrate efficiency, we compared the model runtime among 3 versions of a microsimulation model: with short fixed cycles for all states (FCL), with dynamic cycle length (DCL) defined locally for each state, and with DCL features plus a discrete-event-like hybrid component. To demonstrate bias mitigation, we compared discounted lifetime costs for 3 versions of a resource constraint model: with a fixed horizon where simulation stops, with a fixed entry horizon beyond which new individuals could not enter the model, and with a fixed entry horizon plus a mechanism to maintain a constant level of competition for the resource after the horizon.Results The 3 versions of the microsimulation model had average runtimes of 515 (95% credible interval [CI]: 477 to 545; FCL), 2.70 (95% CI: 1.48 to 2.92; DCL), and 1.45 (95% CI: 1.26 to 2.61; DCL-pseudo discrete event simulation) seconds, respectively. The first 2 resource constraint versions underestimated costs relative to the constant competition version: $20,055 (95% CI: $19,000 to $21,120), $27,030 (95% CI: $24,680 to $29,412), and $33,424 (95% CI: $27,510 to $44,484), respectively.Limitations The magnitude of improvements in efficiency and reduction in bias may be model specific.Conclusion Changing time representation in microsimulation may offer computational advantages.

Keywords:

microsimulation

discrete event simulation

hybrid models

open parallel models

computational efficiency

Changing Time Representation in Microsimulation Models

Abstract

Journal

Organization

Cited Papers