The Data of Disease
In our Zoom for Thought on April 28th, 2020 , UCD Discovery Directory Prof. Patricia Maguire spoke to Scott Rickard, Chief Data Scientist, Citadel, and Adjunct Professor, UCD Institute for Discovery about the Data of Disease.
In case you missed it, here are the top Takeaway Thoughts!
Normalize the data
- Scroll down to table and click (twice) on the arrows in the "Deaths/1M pop” to sort highest to lowest — better indication of ‘risk’ than raw numbers in each country
- Also here: https://www.worldometers.info/coronavirus/worldwide-graphs/ — informative comparisons of Europe vs. USA (infection growth similar, but more deaths in Europe)
Embrace model uncertainty
- Red shadow indicating model uncertainty - don’t just consider the ‘forecast’ but the range of reasonable possible outcomes
The trend is your friend
- Looking at other countries as example outcomes (countries ahead of where we are, and normalizing to the start point) we can get a ‘model-free’ view of what our future might hold
Understand what is being measured might not be what you want to model
- Everyone is modeling deaths based on reported deaths - but we want to model actual deaths so should be looking at actual Covid-19 deaths, not just those occurring in the right location (hospitals) or at the right time (when sufficient testing is available)
- This site tells us which countries are doing a good job of measuring their actual Covid-19 deaths vs. this which are probably undercounting
- Also - report deaths are often not even occurring on the day they are reported — https://twitter.com/CMOIreland/status/1252344273938219011/photo/1 — most of the deaths reported on April 20, for example, occurred a week or more earlier