Uncertainty Aware Decision Support with Computationally Expensive Simulation Models: A Case Study of HIV Intervention...

Area of Science:

• Computational epidemiology
• Health policy modeling
• Simulation-based decision support

Background:

• Agent-based models (ABMs) are valuable for policy support but face computational challenges and complex uncertainty.
• Systematic scenario analysis for ABMs is difficult due to high computational costs.
• Existing methods often lack a coherent framework for uncertainty-aware decision support.

Purpose of the Study:

• To present a novel data-driven, uncertainty-aware decision support (DDUADS) workflow.
• To enable effective use of stochastic simulation models for policy decisions under computational constraints.
• To integrate established techniques into a unified pipeline for uncertainty-aware policy analysis.

Main Methods:

• Combines sensitivity screening, Bayesian calibration via simulation-based inference, and multi-surrogate model integration.
• Utilizes a coherent pipeline for uncertainty-aware policy analysis.
• Generates a posterior distribution over model parameters for flexible forward projections.

Main Results:

• The DDUADS workflow was demonstrated on the INFORM-HIV agent-based model.
• Evaluated potential disruptions in antiretroviral therapy (ART) and pre-exposure prophylaxis (PrEP) use.
• Enabled uncertainty-aware projections for HIV transmission scenarios.

Conclusions:

• The DDUADS workflow provides a robust method for decision support using complex simulation models.
• The approach is applicable to various simulation studies beyond HIV modeling.
• Facilitates uncertainty-aware policy analysis in domains with limited computational budgets.