Stanford Center for Biomedical Informatics Research Colloquia - Shared screen with speaker view
Excuse me, I was accidently moved into the attende list, could I go back to the panel bored please?
Feel free to enter Q&A in to Q&A box
Great thread of questions rolling in. We'll get to them in the next discussion break
Questions pending: Ensuring distribution between training vs. validation sets is the same good enough?
Is external validity essential if going to apply models locally?
Why report sample size if 1M training data?
Amin Adibi: Sometimes clinical journal reviewers ask for non-ideal methods
You can post question to Q&A box
tina hernandez boussard
Moving forward, How do these models account for the drastic changes that we have seen as we learn more about COVID - going from a ~45% mortality rate in the onset of the pandemic to lower than 2% now; data collected from early pandemic can not accurately predict outcomes in 2021 given the changes in treatment etc. Is this ever taken into consideration?
How do you update these models, doesn’t that require hospitals to share real-time data? Will they do that?
Does it even make sense to attempt develop prediction models for COVID? Aren’t prediction models more useful in case of well-understood diseases where we are trying to incrementally improve care? Does it even make sense to attempt prediction models for COVID-19, when we are not very well aware of disease mechanisms, and hardly know how to use predictions to guide treatment?
Do yo uwant me to open your mic, or just post your message/question?