They certainly are hard, because dementia is something that develops over years if not decades. Anytime it takes that long to do a single iteration of your experiment, it's hard!
If medical records were public, the answers could even be in those for any medicines that are prescription based. You could then make statements like "A% of patients who took drug B within C years did/didn't develop disease D".
An RCT doesn’t prove causation, it only might observe another correlation that contributes to our causal inferences based on the convergence of evidence.
We don’t have RCTs for most things that you take for granted.