The main figure from the paper is worth looking at. For legibility, just repeating one panel from it:
Looking at T2D risks for males (other panels show females and CAD risks, but the trends are the same in each case).
The chart on the left is for "healthy" BMI people; overweight in the middle chart; obese on the right.
The bars are show risks by level of different fat deposits adjusted for BMI: red = visceral fat; yellow = subcutaneous abdominal; blue = butt/thigh fat adjusted. Lightest shading for bottom quintile; heaviest shading for top quintile.
Note the distinction between visceral fat and subutaneous abdominal fat: ie fat packed around internal organs vs fat tucked under the skin.
The visceral/subcutaneous/butt&thigh fat measures here are all relative to BMI and together should be read as how body fat gets divided up between these categories (approximately).
So eg in the left most chart, the heaviest red shading applies to healthy-BMI people whose (small amount of) body fat is more highly concentrated than average in the visceral area, and so.
Overall, without looking down to this level of how fat gets divided up, there's a clear trend of T2D risk increasing from healthy-BMI to obese-BMI: the bars get bigger as you move from left chart to right char. Just as you'd expect.
But the picture is more nuanced when you drill down into fat distribution.
The very best place to be in the figure is the lightly-shaded red bar in the left-most chart: "healthy" BMI, low visceral fat relative to BMI => low 1.6% T2D risk.
The very worst place is the heavily-shaded red bar in the right-most chart: "obese" BMI, high visceral fat relative to BMI => high 15.3% T2D risk.
Again, nothing unexpected.
But also note that it's better to be BMI-obese but with low visceral fat relative to BMI, than to be BMI-heatlhy with high visceral fat relative to BMI: 5.3% vs 6.6%.
Anyway, it's cool to see this kind of quantification of relative risks, highlighting that "skinny fat" or whatever you want to call it is a real category which needs the same kind of T2D/CAD screening and attention as obesity.
Even the simplified (and also very cool) AI-based classification system employed by these guys requires an MRI scan to identify the category but maybe that becomes justified as the risk levels become clearer ...