That kind of confounding is something that it is well understood how to filter for, by matching cohorts. It is possible when N is in the tens of thousands. But

That kind of confounding is something that it is well understood how to filter for, by matching cohorts. It is possible when N is in the tens of thousands. But their report did not say the effect was more pronounced in any particular selection of cohorts, other than as noted.