> I've never seen anything testing that hypothesis.

Actuarial tables would tell you.

Though there would also be confounding variables e.g. previous "at risk" behaviour could still be in effect.