As with most things statistical, it depends a) what hypothesis you were setting out to prove / disprove and b) largely as a result of the first question, how you go on to cut the data cake.
With this study
if your hypothesis is that losing over x% of starting weight ( clustered around 15kg), will result in the desired outcome ( ie the specified reduction in hba1c for a specified period of time etc), then all you are really interested in is the group who achieved that weight loss target and the outcome i.e. the 87% figure in this study
If your hypothesis is that following a vlc diet will achieve improved weight loss and glycemic control,then you are interested in how many people lost weight, how much and with what impact on glyceamic markers i.e. in this study, the 46% figure
If your hypothesis is that vlc is an inferior approach to, say, lchf then what you are really interested in is what % of people do not achieve weight loss and / or improved glyceamic control, i.e. 54% -60% figures in this study
Stats can be used ( and abused ) to do many things