Generate visualization and reporting of new metrics (such as capacity gaps)

[abarciauskas-bgse]

A lot of great work has gone into estimating and forecasting supply and demand of hospital resources from COVID 19.

The existing visualizations represent ventilator and hospital bed supplies per state, healthcare referral region (HRR), county and facility.

Additional data discovery and analysis are producing the following metrics which could also be added to these existing maps or produced in new maps or reporting:

• Estimated staffed beds under different scenarios under CONVENTIONAL, CONTINGENCY and CRISIS scenarios (see Generate_CCM_CareModel_Facility_Data.ipynb)
• County-level total PPE needs estimation (see PPE_needs_for_confirmed_covid-19_at_county_level.ipynb)
• Forecasted demand: See 02_Working_with_demand_data.ipynb#Forecasts
  • This notebook includes a number of different forecasting models: IMHE, CHIME, HGHI
• Difference between supply and demand on any given date for different resources. There are multiple models as noted above and a better understanding is needed of the variables being generated to detail what a difference (i.e. "capacity gap") makes sense. For example hospitalized_total in 04_Example_Analysis.ipynb from chime_on_date - does that total represent an aggregation of the forecast since "day 0" or total number to be forecasted to be hospitalized on that day (hopefully the former)?

I think each of these (and other metrics) could be a separate ticket, this ticket is just to start to brainstorming and understand the data being generated.

Next steps will be to:

• [ ] Understand how to generate one sensible reasonable between supply and demand by understanding the different variables and then generating a state / county and / or facility level difference.
• [ ] Run the existing visualization and understand what format the data needs to be in to add a layer
• [ ] Add the layer

[abarciauskas-bgse]

@lossyrob @daveluo @Geoyi let me know if you have any thoughts on this topic.

[Geoyi]

Totally, Aimee. These are already multiple tickets out, PPE needs/demands, and supply. Generate visualization and reporting of the new metrics you listed above it's going to be super helpful. From PPE side, we don't have data from supply-side yet, I could not track down any such data.

Rob has given me some instruction on generating data visualization:

If you are building a map, another template to take a look at is viz/hghi-vents. That app is React and laid out in a way we probably want to move towards; the viz/us-healthcare-system-capacity site is vanilla javascript. I think some good frontend-dev would be to merge those codebases into a single React framework so that we have one frontend codebase to understand/work with/modify

But I'm afraid that it's going to take me some time to get it out with React and don't have much Javascript scripting skill. Though, if someone can help me with the data visualization part of work that'd be awesome. I've done a lot of data visualization and online interactive maps in the past with R, flexdashboardr, shiny and plotly, example 1- Mapping rubber and biodiversity and example 2 - UNICEF school. If Rob, Dave or Jeff are too stretched on current tasks, I was thinking I will start to visualize the PPE burn rate in my old way with R. I will keep you updated on how it goes.

[abarciauskas-bgse]

Thanks @Geoyi for the additional references! I wasn't think you take on building the visualizations to be clear. I was thinking either me or someone else try and pick this up. I think you have your hands full with PPE modeling. But of course you should work on this if you have the time and interest! I also have to learn both about the data and the visualizations technologies. Let's stay in touch with any progress!

[abarciauskas-bgse]

Some notes on how to estimate a "capacity gap" in supply and demand for hospital beds.

It seems clear that number of cases reported by usafacts and nytimes is a cumulative number https://github.com/nytimes/covid-19-data#united-states-data. I would assume not all accumulated cases require hospitalization as of the current date, either because these cases are recovered, are not severe cases, or worst case, have passed. It's impossible to know from data we have here and I think without hospital reporting, what proportion of all reported cases require hospitalization. It's my understanding this data is not being generated and shared yet (which makes sense, reporting may very well not be a priority and hospitals may also require a platform to share this data even if they have it).

So a very rough estimate of number of hospitalizations required I think would be cumulative reported cases on any given day minus any that should have been recovered by that date. The default CHIME recovery parameter is 14 days. So for example, if King County has reported cumulatively 1830 cases as of March 27, we can subtract the number of cases which should have recovered by that date which is the number of cases reported by March 13 (329 cases) - we have 1501 cases as of March 27.

We can compare our estimate for "active / not-yet-recovered" cases number with the number of beds available according to us_healthcare_capacity-county-CovidCareMap.csv:

• Staffed all beds: 4217
• Staffed ICU beds: 541
• Licensed all beds: 4636

The difference between number of "Staffed ICU" and "Staffed All" beds is dramatic. So it seems the real information we need reported is also how many reported cases result in the need for a staffed ICU bed.

[Geoyi]

Go back to what I've learned so far, @abarciauskas-bgse. Instead of using the accumulated confirmed cases, we can treat each confirmed case as an independent case, which:

• 80% of confirmed cases end up under home care;
• 60% of the last 20% positive cases end up in in-patient care, they will need 7- 14 days treatment and recover;
• 40 % of the last 20% positive case will end up in critical care, they will need 10 - 20 days treatment and recover.

I think if we deal with day to day independent cases, it'll make the math and assumption much easier (unfortunately it's far away from accurate). The above conclusions/stats were learned from physicians I talked to and will link back to the original diagram they provide.

[abarciauskas-bgse]

That's super helpful @Geoyi thank you!

[daveluo]

But I'm afraid that it's going to take me some time to get it out with React and don't have much Javascript scripting skill. Though, if someone can help me with the data visualization part of work that'd be awesome. I've done a lot of data visualization and online interactive maps in the past with R, flexdashboardr, shiny and plotly, example 1- Mapping rubber and biodiversity and example 2 - UNICEF school. If Rob, Dave or Jeff are too stretched on current tasks, I was thinking I will start to visualize the PPE burn rate in my old way with R. I will keep you updated on how it goes.

@Geoyi, I'm similar to you in being slow at building visualizations and found Kepler to be a pretty fast and intuitive way to drop geojson files in and style a map viz into roughly something I want it to look like or for others to explore. And it can easily export to a static html file that can be hosted on the repo or anywhere else. Check it out: https://kepler.gl/demo