Changing Time Representation in Microsimulation Models.

Area of Science:

• Computational epidemiology
• Health economics modeling
• Microsimulation methodology

Background:

• Microsimulation models often use fixed short cycles, which can be computationally inefficient for diseases with distinct acute and chronic phases.
• This inefficiency is particularly relevant in epidemic or resource constraint models where long-term economic consequences are analyzed.

Purpose of the Study:

• To demonstrate methods for improving computational efficiency in microsimulation models with varying state durations.
• To illustrate techniques for mitigating bias when applying dynamic cycle length adjustments in epidemic or resource constraint models.

Main Methods:

• Compared model runtimes across three microsimulation versions: fixed cycle length (FCL), dynamic cycle length (DCL), and a hybrid DCL with discrete-event features.
• Assessed bias by comparing discounted lifetime costs in resource constraint models using fixed horizon, fixed entry horizon, and fixed entry horizon with constant competition.

Main Results:

• Dynamic cycle length (DCL) and hybrid DCL models significantly reduced runtime compared to fixed cycle length (FCL): 2.70 and 1.45 seconds vs. 515 seconds.
• Resource constraint models with fixed horizons underestimated costs compared to the constant competition model, highlighting potential bias.

Conclusions:

• Adjusting time representation, such as using dynamic cycle lengths and hybrid discrete-event features, can substantially improve microsimulation model efficiency.
• Careful implementation is crucial to avoid bias in economic evaluations, particularly in resource constraint or epidemic models.