Model Families for Multi-Criteria Decision Support: A COVID-19 Case Study

TL;DR

The paper tackles the challenge of sustaining decision support in long-running, knowledge-evolving crises like the COVID-19 pandemic by promoting a modular model-family approach rather than a single monolithic model. It details four core components—ABEM, IWM, HM, and ASM—each with distinct resolutions and purposes, interconnected via causal loop diagrams to support multi-criteria decision making. Through four success stories, the authors demonstrate improved forecast flexibility, reduced computation and calibration effort, and clearer communication and validation when using interacting submodels. The work argues that this modular framework enables robust scenario analysis and rapid adaptation to new information, with tangible benefits for public health policy and preparedness.

Abstract

Continued model-based decision support is associated with particular challenges, especially in long-term projects. Due to the regularly changing questions and the often changing understanding of the underlying system, the models used must be regularly re-evaluated, -modelled and -implemented with respect to changing modelling purpose, system boundaries and mapped causalities. Usually, this leads to models with continuously growing complexity and volume. In this work we aim to reevaluate the idea of the model family, dating back to the 1990s, and use it to promote this as a mindset in the creation of decision support frameworks in large research projects. The idea is to generally not develop and enhance a single standalone model, but to divide the research tasks into interacting smaller models which specifically correspond to the research question. This strategy comes with many advantages, which we explain using the example of a family of models for decision support in the COVID-19 crisis and corresponding success stories. We describe the individual models, explain their role within the family, and how they are used - individually and with each other.

Figures (10)

• Figure 1: Schematic causal loop diagram of all elements regarded in the four models discussed in this work. The greyed out components are not regarded in the Agent-Based Epidemics Model. The components coloured in black represent the model states, the components coloured in green the model inputs and the components coloured in blue the model outputs.
• Figure 2: Schematic causal loop diagram of all elements regarded in the four models discussed in this work. The greyed out components are not regarded in the Immunity Waning Model. The components coloured in black represent the model states, the components coloured in green the model inputs and the components coloured in blue the model outputs. The dashed arrow indicates a causal link which is implemented inversely in the model.
• Figure 3: Schematic causal loop diagram of all elements regarded in the four models discussed in this work. The greyed out components are not regarded in the Hospitalisation Model. The components coloured in black represent the model states, the components coloured in green the model inputs and the components coloured in blue the model outputs.
• Figure 4: Schematic causal loop diagram of all elements regarded in the four models discussed in this work. The greyed out components are not regarded in the Age Structure Model. The components coloured in black represent the model states, the components coloured in green the model inputs and the components coloured in blue the model outputs.
• Figure 5: Combining different case forecast scenarios from the ABEM [a] with different assumptions for virulence (equal and $30\%$ increased) of the a new variant in the HM [b].