Having read the paper it seems that what they done is to develop something that will (at +95% efficiency) detect existing conditions which they call diabetes and pre-diabetes, but I assume that means simply "people who do not have normal blood glucose". I guess given the population studied there are big advantages in being able to do a test there and then, and not having to do blood draws and lab tests. As the authors say, this can help in early detection.
I don't see how this can be described as "predicting" anything. The people detected already have problems. It's a diagnostic test.