All it does is normalize the ice time so you can compare players and not have that as a factor...it does not project or tell you anything. If you're extrapolating data you're taking known data, assuming a function which is generally considered linear, and seeing what happens as you go outside the data set. That is highly prone to error.
It is no different than taking a players batting average in baseball. You don't just look at players total number of hits. Then you can't compare guys who have vastly different number of at bats. Or a QB's completion percentage in football. You can't just compare completions if one guy has thrown 100 more passes than another. Or like, and I'm not a basketball fan, but isn't the DEFAULT stat for pretty much everything in basketball per game?