The numbers become relevant at different rates, simply because the sample size gets larger at different rates.
For example, preflop you get data each hand. Showdown you only get data from hands that saw the flop and hands that went all the way to showdown.
I don't think you avoid using data that you have, just keep in mind that after 18 hands the preflop is only starting to become accurate, but if your opponent is already 35/30 you can start to guess that he's in the loose-aggressive end. If he's 35/30 after 100 hands then it's more reliable, but even after 100 hands the river bet frequency probably only has a few samples.
I'll use whatever data the HUD gives me; how much I rely on it depends on the sample size. I don't rely on a fixed threshold for "reliability". I think PT used to default to not showing any stats until there were 40 hands in history or something.