Safe Blues: The case for virtual safe virus spread in the long-term fight against epidemics

Abstract

Viral spread is a complicated function of biological properties, the environment, preventative measures such as sanitation and masks, and the rate at which individuals come within physical proximity. It is these last two elements that governments can control through social-distancing directives. However, infection measurements are almost always delayed, making real-time estimation nearly impossible. Safe Blues is one way of addressing the problem caused by this time lag via online measurements combined with machine learning methods that exploit the relationship between counts of multiple forms of the Safe Blues strands and the progress of the actual epidemic. The Safe Blues protocols and techniques have been developed together with an experimental minimal viable product, presented as an app on Android devices with a server backend. Following initial exploration via simulation experiments, we are now preparing for a university-wide experiment of Safe Blues.

Figure 1

The 2020 outbreak in Victoria Measured daily new cases of SARS-CoV-2 in the Australian state of Victoria in 2020. Both the first wave and the second wave were mitigated via social-distancing measures. Government imposed various social-distancing directives, with severity labeled stage 2 (lightest), stage 3, and stage 4 (complete lockdown).

Figure 2

The Safe Blues concept Individuals of the population with Safe Blues-enabled devices take part in spreading Safe Blues strands. SARS-CoV-2-infected individuals are in red and others are in green. The Safe Blues system operates independently of the health status of individuals.

Figure 3

Simulation models Model I: at every time point, each of the individuals selects a random number of other individuals to “invite” and this implies physical proximity. In this case, orange and green individuals make invitations. Model II: all individuals traverse a binary tree between their private leaf and the root. At any node, infection follows a continuous-time stochastic SIR model between the individuals who are present. Model III: a spatial model where each individual diffuses either around their base (e.g., their home) or around a center (e.g., a supermarket).

Figure 4

Prediction with Safe Blues Deep Safe Blues: Safe Blues detection of a second wave applied to data generated from three different simulation models. The light-colored lines indicate counts of various Safe Blues strands that are inputs to our predictions. The proportion of infected individuals is only known until the vertical black lines. After that point, only Safe Blues information is available. Nevertheless, Deep Safe Blues (trained up to the black line) is able to accurately predict a second wave of SARS-CoV-2 infections.

Figure 5

Policy projection with Safe Blues Demonstration of policy projection and refinement using Dynamic Deep Safe Blues on model III.

Figure 6

Safe Blues software The control panel, dashboard, and Android app used for the planned campus experiment.