One difficulty is that it's far from clear what the ideal dropout rate should be. Probably not zero, for two reasons:
Even the best students sometimes find that their interests change over time, in ways nobody could have predicted. Dropping out may become a quite sensible choice, in which case it's harmful for the university to discourage it.
It's reasonable for a university to give someone a chance even if it's not certain that they will succeed. The only way to get a really low drop-out rate is to admit just the applicants that are obviously destined for success. The top few schools can get away with this, but if everyone tried it, then many talented candidates would be shut out from graduate school.
So some attrition is OK, and some is bad but may be a necessary consequence of policies that are on the whole good. The question then becomes how you distinguish these kinds from needless and damaging attrition, and then how you minimize that kind.
Advisors play a key role here, because some are much better than others at being supportive or motivational. However, at least in mathematics, most advisors don't supervise very many students, so the numbers often aren't large enough to see patterns clearly.
I'm sure people have studied this problem, and perhaps identified best practices for addressing it. However, in my experience any studies or solutions are not especially influential (at least in the few math departments I've been in). Most discussions are at the stage of trying to figure out whether there is a real problem and if so why, rather than what to do about it.