Looking at the link they give for it, the 98% accuracy isn't for individuals but for aggregate data. That is: they can't predict with 98% accuracy when you or I will die, but they can with a sufficiently large group which averages out all the noise. The phrasing in the article is somewhat unfortunate.