Author + information
- Alfred Hallstrom, PhDa
We agree with Goldberger et al. that subgroup analyses are prone to overinterpretation. One must be very cautious in dealing with subgroup analyses, particularly when they are post hoc (ours was a priori), based on essentially randomly defined subgroups (ours was based on a prespecified mechanistic “dose response” relationship), and the result is based on lack of significance (if our null hypothesis had been that there was no treatment effect in the lowest sextile, not being able to reject the null would probably be a low-power issue [Goldberger et al.’s power calculations are, in fact, based on this null hypothesis]—however, our null hypothesis was that the treatment effect would be the same in all sextiles, and we were able to reject this null hypothesis with an, admittedly marginal, significance of 0.05).
Thus, while we would not disagree with Goldberger and colleagues’ caution “don’t believe it,” we might also caution “don’t disbelieve it.”
- American College of Cardiology