In Theorem 1.4 of [this paper][1] or of its [preprint version][1], it is shown that, for any measure $m$ on any algebra $\mathcal A$ of subsets of a set $X$, a subset $E$ of $X$ is locally approximable by sets in $\mathcal A$ in the sense of pseudo-metrics $d_A$ if and only if $E$ is Carathéodory-measurable, where $d_A(E,F):=m^*(A\cap(E+F))$, $A$ is a member of $\mathcal A$ with $m(A)<\infty$, $m^*$ is the outer measure corresponding to $m$, and $E+F$ is the symmetric difference between subsets $E$ and $F$ of $X$. 

In particular, if $m(X)<\infty$, then one can use just the approximation relative to one pseudo-metric $d_X$ instead of the local approximation relative to the family of pseudo-metrics $d_A$. 

(We have to say "pseudo-metrics", because $d_A(E,F)=0$ does not in general imply $E=F$.)

It is also shown in that theorem that the completion $\sigma$-algebra is always contained in the Carathéodory $\sigma$-algebra, and the completion $\sigma$-algebra coincides with the Carathéodory $\sigma$-algebra if $m$ is $\sigma$-finite (and this $\sigma$-finiteness condition cannot be dropped).  


  [1]: https://arxiv.org/abs/1702.01142