HN user
An interesting idea, but a big problem would be that the uterine environment is shared between twins, and so any contributions from that shared environment will not be disentangled in your proposed model.
This is incorrect. Twin studies typically compare MZ twin similarity against (same sex, usually) DZ twin similarity. Assuming that there is nothing special about MZs for the trait (e.g. in this case if MZ twins lived longer by virtue of being MZ twins), you can estimate heritability free of shared environments.
Based on your own citation, this is untrue. The citation you have linked is a meta analysis of 31 trials, many of which specify no medication use at recruitment.
You are however correct about population being important (which is a big reason meta analyses can be very useful).
I think they mean in the sense of pro-inflammatory. Which it very much is (especially red meat).
Slowing in this context means going down. Basically they look at 'age acceleration', I.e. how old are you epigentically compared to chronologically. They saw a reduction of several years in this measure over a much shorter period, basically meaning their epigenetic ages went down.
Although one of the clocks they used, DunedinPACE, only looks at pace of ageing, so in that case you can only infer that it slowed (as you do not get an 'epigenetic age' figure from DunedinPACE).
However they have no evidence of this. While I would expect this to be the case (as a researcher of ageing epigenetics), this is entirely conjecture.