Actually you need to train every clinician on how to spot problems with cross-validation, how to compare ML algorithms, why supervised learning alone doesn't give you individualized predictions, how not to fit ML models per clinical trial arm, etc... #StatSky
Why repeated cross-validation is important, when feature attribution is not a casual effect, etc... Statistical thinking intuitively makes you develop these solutions from scratch. People who don't learn this in stats101 then think it doesn't matter for whatever the next hottest thing is.
Comments
With handy PDF table!
For my thoughts on a one-semester stats education for clinician-scientists, see my post today.