Works in Centralized Data Management, Burlington Hospital, Vt.
Video can be accessed at: https://drive.google.com/drive/folders/10W5q8s-M1NWm4ME97knXCUozPLwgGO1t?usp=sharing
Notes from the Interview:
- COVID looked like a normal distribution at first
- Model was updated based on incoming data from Italy
- Consider what indicators you will actually use [when it comes to models]
- Used SIR from SG2
- Initially interested in ‘big hospital’ resources -beds, ventilators, etc…
- Models differ when looking at hospitalized patients vs. testing vs. deaths.
- Vermont had a large spike in deaths due to COVID in nursing homes
- IHME model (https://covid19.healthdata.org) was inaccurate for Vermont because it looked at deaths
- Found that hospitalizations were more accurate
- Average length of stay:
- Non-ICU: <8 days
- ICU: ~10 ICU+6 extra
- As testing capacity increases, they’re becoming more reliant on it when updating their models
- The hospital saw a base
of 1.5-1.6 (lower than in urban areas)
- Use longer term ensemble models (3-4 models to see which is most accurate)
- Single models need to be care of timeframe (even no more than 2 weeks into the future
Supply Chains and Modeling:
- Lead times on supplies will also determine how long your model should go
- Determine maximum burn rate
- Take into consideration political events– stay-at home orders, outbreaks at colleges, etc…
Modeling and Government Action:
for government action
- Herd immunity v.s. Government Action –Depends on Rep. or Dem administration
- SNS stockpile depends on Presidency/Senate for funding/action
- Would companies ramp up production b/c of the promise of SNS contracts?
- Or will they do so anyways when demand increases over the winter?