According to my Mullard databook Nixie section, life can be as little as 3000hr (ZM1040, single digit illuminated continuously), the same tube is >20,000hr with digits changing at least every 100hr. At the other end of the spectrum, for a long life tube (ZM1000) it quotes >100,000hr (digits changing at least every 1000hr) at rated current. I'm sure dropping the current a bit would improve on this.
The end of life is normally taken as the point where cathode poisoning causes incomplete digit illumination. I think there are techniques (short term over-drive spluttering?) that can partially reverse this. Continuous long term illumination of a single digit causes poisoning of adjacent non-illuminated digits (cathodes).
As I said for the Soviet ones it can vary between models, a few are troublesome, the IN-1 being the worst (I think) due to inter-cathode short formation. Gas composition and Hg doping can make a big difference.