Create dataset of CHIME model predictions per US county

[lossyrob]

To do this, I am combining county level cases, county population and the CHIME model.

This dataset will include numbers for 3 patient "dispositions" (hospitalized, ICU, and ventilated) over time. There are 3 datasets produced by the CHIME model for each of the dispositions, each per day after initial date:

• Total number: the cumulative total number of patients.
• Admitted: the patients admitted on that day
• Census: The total number occupying hospital beds on that given day.

I will use default parameters for the SIR model, using count population and latest confirmed cases, as well as a estimated detection probability as described in https://github.com/CodeForPhilly/chime/pull/272. The code should be such that at a future time point we can collect and use better parameters per-state or per-county from people who know how to estimate their own region better than the defaults.

This dataset will add a time component to the CovidCareMap data, which has been static so far. This include:

• Adding a dimension (time)
• Adding a need for continually updating data (as new confirm case counts come in)

Data deployment will be considered outside of the scope of this issue for now. The data format should optimize for ease of use for visualization and analytics.

The addition of this dataset will move the CovidCareMap.org (CCM) data to have the following information:

• Licensed Beds, Staffed ICU Beds, Staffed All Beds per facility, county and state (with per capita numbers for regional aggregates.
• For County:
  • Number of confirmed cases (as of a specified date)
  • CHIME model output over X duration (6 weeks?), with each future date forecasting:
  • Total number of hospitalized, needing ICU care, and needing ventilation
  • New patients admitted to the hospital, ICU, or requiring ventilation
  • Total number of hospitalized, ICU, and ventilated patients in the region.
• For State:
  • All data per country aggregated by state.
  • 2010 Ventilator Estimates

With the PPE models from #45 we'll may be able to include PPE estimated needs over days based on CHIME model output. This may require staffing estimations, work for which was started here.

This dataset should unblock Care Gap analysis.

[lossyrob]

Potential data to validate against and ideas to integrate: http://www.healthdata.org/research-article/forecasting-covid-19-impact-hospital-bed-days-icu-days-ventilator-days-and-deaths

[lossyrob]

Fixed in #87

This does not save off a dataset but creates a method for dynamically generating it for any dataframe of region information.