Background

In 2021, with COVID-19 sweeping through the United States, Virginia’s Department of Health approached the Virginia Modeling and Simulation Center (VMASC) for strategic and policy support in modeling and optimizing the rollout of vaccines across the state. Facing urgent time pressure and a recommendation from colleagues at Boston University, VMASC looked to HASH’s first browser-based simulation IDE, hCore, to quickly spin up a custom simulation model.

Requirements

VMASC needed to develop a simulation they could trust, and fast. Existing tools were non-starters.

With HASH, the team’s researchers were able to confidently:

test that real-world dynamics were accurately modelled and reliably simulated;
securely combine sensitive internal data with generally-available public data and statistics;
update parameter estimates relating to the virus’ characteristics continuously, whenever new information became available.

The team were also able to derive insights from these simulations:

automatically recomputing scenarios when parameter estimates or other underlying data updated;
running large numbers of simulations in parallel as part of experiments;
producing informative visualizations to communicate impacts over time and according to various vaccine distribution policies;
easily sharing the model with end users on non-specialized mobile and desktop devices;
allowing non-technical users to experiment with the model, vary assumptions, and create large scale experiments themselves, to run on high-performance computing infrastructure.

Alternatives

Speed was of the essence in VMASC’s choice to use HASH. Being users of other simulation tools, the team knew that they would need something faster, more modern, and more capable if the simulation they produced was to be of any use in the real-world. It would need to be built and calibrated within hours or days, as opposed to weeks and months, and prove itself capable of turning around valuable metrics and insights in that timeframe as well. At the same time, it would need to remain fast and accurate throughout the model’s lifetime, and the team would need to be able to clearly communicate its results to a wide variety of stakeholders, with different levels of technical sophistication.