> They actually adjusted for age, smoking (never, former, current), pack-years of smoking, energy intake using the USDA's HEI, alcohol intake, physical activity (some/none), BMI, and waist circumference.
All of which are self-reports (food and energy intake as a self-report, when it's well known in the field that self-reports are flattering? and you have the gall to imply that this is good reliable science) or based on weight, just as I said. Please think a little more about this.
> I'm not sure what you're trying to imply with this one though.
What...? OK, I'll charitably assume you're just ignorant here of problems with dietary research and explain (hopefully I won't be wasting my time and receive a response of 'I WANT TO BELIEVE'): whenever a large regression like this is run, there are always many correlations with various behaviors and food consumption; for example, people who eat nuts or blueberries or who exercise regularly or take statins are also people who tend to live longer on average or have other good outcomes. It is tempting to implicitly assume that eating blueberries is a random practice in the population and so this shows that blueberries probably cause you to live longer; however, it is also true that blueberries are widely believed by the general population to be healthy foods, and so the sort of people who pay for and eat blueberries (or claim to eat them, anyway) are also the sort of people who are in good health, are health-conscious, will follow their doctors' orders, comply with drug dosing regimens or regularly test their blood glucose, etc. This bias is universal, and this is why, for example, in the infamous Nurses's Study, hormones seemed so great when they turned out to be murderous in randomized practice: because hormones were believed to be healthy, it was healthy compliant nurses who self-selected into use of hormones and because they were healthy and complaint, had good outcomes anyway.
So, what is a warning sign of healthy user bias? It's when the common trait among harmful/beneficial correlates is whether they are generally believed to be harmful/beneficial even when the supposed causal chemical mechanism is inconsistent. In this case, soda is considered by the general population to be unhealthy, diet soda to be healthier (it's got diet in the name!), and fruit juice the healthiest of all. The former seems to be bad, and the latter two good or at least neutral. And why would soda be so harmful? Well, generally you'd suggest the acidicness or the sugar content; except... either of them should cause either diet soda or fruit juice to be harmful (acidicness or very high sugar content, respectively). But there is an explanation resolving the lack of harm for either: people with better health habits or interest will preferentially consume diet soda or fruit juice and avoid awful soda... and if we correlate consumption with outcomes, we'll find the soda drinkers have the worst outcomes. Voila: healthy user bias.
(You see something similar in studies of maternal smoking or smoking by pregnant women; the effect seems dramatic and harmful to the child, but consider this: given that everyone believes only the worst mothers in the world would smoke while pregnant, what sort of woman would over the past decades smoke while pregnant? Probably not one who plans ahead, is self-disciplined, health-conscious, rich, or intelligent, all of which are heritable, and are difficult if not impossible to completely measure in all their manifestations & remove as confounds.
If you wanted a more accurate analysis, taking this possibility into mind, you might do something like find women who smoked while having one child and didn't smoke while having a second child, and compare pairs of children; then you'll find that the harmful effects mostly disappear.)
This cannot be adjusted away simply by a few self-reports because it is a systematic subtle effect on all life outcomes - for example, to name just 2 of the lurking confounds, in other studies one can find longevity boosts to IQ and Conscientiousness for what is suggested to be the same exact reason (more likely to understand or carry out good health practices, respectively), but neither of those was measured by this study.
> This is good science and generic insults are just noise when everything is actually addressed: http://www.chc.ucsf.edu/ame_lab/pdfs/Leung%20et%20al%202014a....
Healthy user biases are never 'actually addressed' by a study of this type, that's the point.