rockitect

I also got something like 12 rolls of 3 in a row. I would definitely check the RNG routine to ensure that it is not biased to the mean (i.e., verify that you are drawing from a uniform, rather than normal, distribution).

Another comment in this thread points out that a huge confounder could be income. Indeed, as @robwwilliams points out, there could be many potential confounders, including genetics and environment. All that means is that, if drilled into, researchers might find that X (where X is whatever confounder you can think of) might be a better predictor of mortality than protein intake.

It seems to me that the parent comment doesn't aim to "hand wave away" this paper, but instead to suggest that it is a singular datum whose findings should be tested by larger, more in-depth, research (which could test for a whole set of confounding variables that might come to mind). The authors of the original study state the same thing right in their abstract!

I'm not sure why you think that point is not meaningful. IMO, that is exactly how science should be done ---interesting correlations should be stress-tested. If the findings hold up, then we can be especially confident in whatever choices we make. If the findings don't hold up, we then have new avenues to research.