The model is more than a guestimate though of course any model will have errors. if you look at the other link I gave then you will see where the researchers think they are likely to be and the predicted variation.
If you reread the abstract you will see that this particular study was comparing it's results with the previous figure of 750, 000. This was in fact higher than found in this study
"In 2004 the prevalence of undiagnosed diabetes, and the proportion of cases of diabetes that were undiagnosed,
appear smaller than in previous studies. This is likely to be due to increased awareness of diabetes and improved clinical care resulting in many of those with previously undetected disease having been diagnosed"
I linked to that study just to give an example but there are others that were used in developing the model.
As for not knowing anyone who has taken part in this sort of study;
Let me ask you whether you know anyone who has taken part in the
Labour Force survey ? It's.one of the biggest surveys that takes place in the UK , far bigger than the type of studies used for diabetes prevalence. I certainly don't know of any friends or acquaintances that have mentioned it but I know it happens because I spent part of my life working for it. (even interviewed a few famous people but the official secrets act means I can't tell you who

)