This has been mentioned variously, but buried in other arguments - the gold standard would have been a longitudinal, well designed randomised double-blind placebo-controlled trial over several decades. That should have been embarked upon long before gender medicine became a formal practice, but for obvious ideological reasons it was not. So, in the absence of well designed trials, we are left to construct studies using existing clinical data. This is a genuine limitation, but that doesn't mean the results are invalid. It does mean that confounding factors, including causation vs correlation are harder control for and eliminate.
What's important to keep in mind is that the burden of proof actually lies with the treatment, i.e. that evidence for improvement outweighs the statistical noise and methodological uncertainty of no treatment (including the placebo effect) and the risk of harms. Instead what has been demonstrated is that the treatment, far from being beneficial, or having no effect, caused harm. And the numbers are so disproportionate they drowns out statistical and methodological uncertainty. That's astonishing and in fact shows gender medicine is a uniquely harmful pseudo-science and not a science based treatment at all.
It's quite likely the treatment, in fact, is close to 100% harmful (to say nothing about whether it's beneficial) but the statistics that are available are not powerful enough to show that, e.g. there is considerable mental harm that can be done that doesn't rise to the level of requiring psychiatric care and considerable physical harm (including shortened life and serious, debilitating illness) that wouldn't be captured at all.