We are seeking additional industrial, academic, and government partners!
For local governments navigating the COVID19 pandemic, uncertainty reduction and policy planning are top objectives. This is especially true due to the clinical and social nature of pandemic policy which resides outside of the forecasting toolbox of most localized governments and federal agencies have split focus and limited modeling capacity throughout epidemics.
The microstructure of local (municipal, county, and state level) COVID social distancing policies requires model inputs from stochastic dynamical epidemiology models combined with localized empirical observations of policy efficacy in order to answer a variety of high value questions until herd immunity thresholds are surpassed or a vaccine can be produced and circulated (estimated 12-18 months).
We would like to provide these forecasts and forecast tooling for data professionals, epidemiologists, and county officials to formulate improved policy decisions.
What it does
Our new tool PySEIR provides sophisticated, but easy to use and configure, epidemiological models configured based on the most recent COVID literature, and will continue to be updated as new data becomes available. Our models track hospital bed and ventilator utilization through the crisis.
Expanding upon existing models, we track asymptomatic (but infectious) segments of the population and are nearing the completion of integration of POLYMOD age structures into the simulation. Most importantly, our system can be used to generate mortality-minimizing social distancing policies at a local level.
Here are some example outputs demonstrating the well known "flattening of the curve".
In the next 1-2 days we will complete the incorporation of county level case, death, and recovery information, hospital capacity measures, and epidemiological inference. Our automated system will output forecast data on a daily basis for a variety of scenarios.
What's next for County Level COVID Forecasting and Policy Optimization
- We have partnered with Unacast to obtain county-level social distancing measures based on cell data, allowing policy efficacy measurement.
- Localized inference tooling to measure parameters like R0 at the local level with uncertainty bands (specifically Bayesian inferences using MCMC and dynamical simulations to reproduce the data).
- Daily best estimates for forecasted time series within for worst, nominal, best case bands. These will be updated automatically as new county level covid data flows in.
- Completion of age structured modeling to incorporate age dependent contact-mixing, hospitalization and mortality rates.
- Longer term, studying interacting distancing policies to exploit neighboring county hospitals by attempting to produce "out-of-phase" outbreaks that allow relief valves.