One of the things that was brought out in the thread is that there is simply not enough communication between the math/data people and the medical people in medical studies.
The medical people often know the flaws in the data, but don't have the math to correct for them (i.e. weighting the samples). The math people know how to correct, but they don't know where the biases are.
One example that was brought up, if you got that far, was that Covid severity correlated with font. That was because a large hospital with high Covid caseload used a particular font.
I don't think it's quite right to say it's a correlation/causation problem. It's a valid correlation vs spurious correlation problem and finding the right data to tell the difference