Abstract
Preventing adverse health outcomes is complex due to the multi-level contexts and social systems in which these phenomena occur. To capture both the systemic effects, local determinants, and individual-level risks and protective factors simultaneously, the prevention field has called for adoption of system science methods in general and agent-based models (ABMs) specifically. While these models can provide unique and timely insight into the potential of prevention strategies, an ABM’s ability to do so depends strongly on its accuracy in capturing the phenomenon. Furthermore, for ABMs to be useful, they need to be accepted by and available to decision-makers and other stakeholders. These two attributes of accuracy and acceptability are key components of open science. To ensure the creation of high-fidelity models and reliability in their outcomes and consequent model-based decision-making, we present a set of recommendations for adopting and using this novel method. We recommend ways to include stakeholders throughout the modeling process, as well as ways to conduct model verification, validation, and replication. Examples from HIV and overdose prevention work illustrate how these recommendations can be applied.
Highlights
Prevention research has made many advances in the last two decades, following a path taken by medical science beginning in the 1960s in building an empirical knowledge with rigorous testing of well-defined preventive interventions against standard conditions or against other competing interventions (Hill, 1961)
Based on our shared experiences in developing the most widely used agent-based models (ABMs) platform (Wilensky, 1999), and over a decade of contributions to the push for reliable ABMs (e.g., Vermeer et al, 2020; Wilensky & Rand, 2007), we identified three major themes that we use to structure the recommendations presented in this manuscript: (1) ensuring model validity, (2) facilitating replication, and (3) acceptance, adoption, ownership, and use by stakeholders
When developing an ABM, it is possible that we discover inaccuracies
Summary
Prevention research has made many advances in the last two decades, following a path taken by medical science beginning in the 1960s in building an empirical knowledge with rigorous testing of well-defined preventive interventions against standard conditions or against other competing interventions (Hill, 1961). As such, building these models and reporting on the process of doing so should be seen as methodological contributions to prevention and modeling sciences that can be scaled up or out with local data from other systems and locales To facilitate these adaptations, we recommend that both the context-related and individual-level input components be included but distinct from the core model code. Given the sheer amount of description needed to capture all nuances and interactions of a high-fidelity model, even the best documentation is likely to fall short in providing the details needed for perfect replication This makes it nearly inevitable that differences will occur during translation from text to code. While the creation of high-fidelity models should go hand-in-hand with the use of models for decision support, it will be policy makers and community representatives, not
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
