If you're in a class teaching medical studies, and you tell people "go replicate the original experiment instead of doing meta analysis", you're doing your students a disservice.
Instead say: "It's always important to be able to replicate experiments and I encourage everybody in the class who is going on to do science or medicine spend at least some time in the lab replicating a basic experiment. But medical studies are so large and complicated that we can't typically repeat them in a reproducible way. Instead, we use statistics and probability to make reasonable decisions based on the data we have. Sometimes that means removing a suspect paper from a meta-analysis because we lack confidence in its reproducibility."
For some of the most sophisticated medical science experiments we have today, the only way to replicate an experiment is to be a postdoc and join the lab that does the research, master the technique there, then take your reagents and other material to your professor job, and then get your local set up to replicate what you did in the original lab. This is how it normally works, for example in labs like Mina Bissell's where few people in the work even have the skills to replicate her experiments.
I don't think we have to worry about the file-drawer effect much with controversial COVID treatments, though. Do you?
