COVID-19 continues to affect countries around the world, and while the accelerating distribution of vaccines is much welcomed news, it will be many months before enough people are vaccinated to achieve herd immunity. New COVID-19 variants present a continuing threat. In the meantime, we still need a data-driven, scientifically sound way to quantify the risks of infection and reduce its spread.
In the future, Pandemonium's app and also its framework can be applied to new pandemics and outbreaks.
Pandemonium's risk app and framework differ from other risk apps. Pandemonium is:
by incorporating user data so anyone can see their own COVID‑19 risk estimate.
by incorporating data on the city and county level, allowing for more specific risk estimates.
to better represent viral transport between communities.
to more accurately model, especially for locations with smaller case counts and populations.
allowing more detailed data and models to be added where available, making Pandemonium useful for future pandemics and outbreak events.
with an easy-to-use interface, making Pandemonium's risk app a powerful, useful tool even for non-epidemiologists.
While the app appears very simple to the user, it has a sophisticated Stochastic Heterogeneous Hybrid Spatially-Hierarchical/Dynamic and Demographically-structured Regional Susceptible-Exposed-Infectious-Recovered-Dead (SEIRD) Compartmental model built on the Pyro Probabilistic Programming Language (PPL), using Markov Chain Monte Carlo (MCMC) and other statistical inference techniques.
We are excited to announce the launch of our Pandemonium Personal Risk Assessment app for testing. We are starting off with a very limited feature set, which will be gradually expanded on a rolling basis. We hope to use the first phase of testing to receive user feedback, resolve issues promptly and make improvements before we scale it up.
We welcome those interested in being a first‑phase tester to sign up at firstname.lastname@example.org.
Our risk app's interface allows users to input to
input demographic and personal metrics such as age, sex, location, and vaccination status and dates. It also uses behavioral metrics such as mask usage, location history, quarantine periods and planned travel.
One of our sub-projects involves detecting the percentage of people wearing face masks of each type from publicly posted geo-tagged photos to better estimate the face mask usage in each location and at each point in time to be fed into our epidemiological model. Shruthi Ravichandran, one of the very talented Research Science Institute (RSI) students we mentored, developed a machine learning model for us in 2020 as described in her RSI paper. Subsequent work on this sub-project has been focused on mitigating the inherent sampling bias that results from using publicly posted photos.