It's actually much worse than this.
You essentially have, in the numerator, a highly selected sub-population: trans people [I can't remember the specific nuance of it -- whether it was those with a GRC or not, I forget] who happened to be in prison on some day in March 2021. The figure in the denominator is one the ONS explicitly warned against using in the way the graphic uses it, due to inaccuracies.
So you have what is essentially a really rather poor prison-prevalence ratio estimate, which does not consider any covariates or uncertainty quantification, and which is then misrepresented as some sort of generalisable offending rate.
It's a very annoying graphic I see quite often these days, because most people tend to just take these things at face value rather than thinking about whether the inferences the graphic is clearly trying to lead you towards are actually reasonable.