So, I know people involved in COVID vaccine trials (I can't remember which, I think it was Pfizer or Moderna), and also work with people who have been involved with other sorts of trials involving sensitive populations.
Awhile ago, we had a conversation about the issue you're referring to. Basically what was happening was that people in the control group were getting vaccinated, increasingly so, after the trials showed efficacy. It wasn't even so much they were leaving the study, they were just violating the inclusion criteria.
They were very aware of this, and discussed all sorts of ways of trying to get around it, too many to note here, including some weird designs with a new sample that at the time didn't really make logical sense to me from an efficacy RCT standpoint, but did from an ethical standpoint.
There were at least two things on their mind. First, you can't force participants to remain in the trials long-term. So if they want to go get vaccinated, they're going to get vaccinated. You can't keep them from doing that. So if you lose 95% of your participants to that, and your power goes through the floor, what can you do?
Second, there's a strong argument to be made that if you show strong efficacy of an intervention that prevents a world-crippling disease, it is highly unethical to keep the intervention from the control participants. It's not just unethical to encourage them not to get the vaccine, it's unethical to not offer it to them. There are many RCTs my colleagues have been in, of much more inconsequential treatments, where they ended up giving the treatment to controls because of this issue. So what about a COVID vaccine?
Yes, long-term findings are necessary and/or desirable, but this can't really be done while circumventing ethical issues about withholding a highly effective intervention. There's lots of scenarios where this happens all the time -- where you'd like to get more and/or any RCT data but doing so raises ethical obstacles that are difficult to overcome.
Maybe you could scrounge together enough people across trials but then you run into another issue: are the people who choose to remain in the control group long term really comparable to treatments on average anymore? You lose your randomization in an important sense, so it's all moot anyway.
It's a difficult issue.