I saw a great illustration of why you should be really careful when reading about diagnostic accuracy:
I have a simple way to test if any random person who walks in has leukaemia. On average, this test is >99% accurate.
The test is as follows: Just declare the person who walks in to be leukaemia-free.
Since (verrrry) roughly, 1 in 5000 people will have leukaemia at a given moment, the chance of any random person having leukaemia is ~0.02%, so a test that never, ever detects leukaemia will be right 99.8% of the time.
Gulp.