Very interesting question. How valid it is to evaluate “legacy effects” from randomized in-utero exposures? I hope someone with experience in this area answers.
The study in your link looks at morphometric differences between two groups of children, 9 years after their mothers were randomized, during pregnancy, to treatment with either insulin or metformin for treatment of their gestational diabetes. The RCT in which the mothers were enrolled was published in NEJM in 2008 (10.1056/NEJMoa0707193). The 2008 trial specified perinatal complications as primary outcomes of interest and neonatal (not long-term) morphometric measurements as secondary outcomes. As you noted, the two groups showed similar rates of the outcomes of interest at the stated end of the study.
What you are essentially asking is: “If we follow these children over many years after birth and continue to monitor them in various ways, how confident can we be that any between-group differences that we might document at any point in time are related to differences in their randomized in utero exposure?”
I don’t know the answer to your question from a statistical standpoint. Wouldn’t longterm between-group differences be difficult to interpret in the absence of any prespecified intent to look at them (?) There is certainly precedent for looking at legacy effects in patients enrolled in diabetes and statin trials…
In clinical terms, I guess the most pragmatic approach would be:
- to not worry too much about how to interpret any observed long-term between-group differences that aren’t clinically significant (e.g., subtle morphometric differences many years later);
- to take more seriously any potentially clinically important longer-term differences (specifically, between-group differences that might constitute a harm signal) going forward. Everyone remembers the diethylstilbestrol tragedy as an example of how serious consequences of an in utero exposure might not arise until children are older.