Comment by brolover | Hacker News Reader

Comment by brolover | Hacker News Reader

Parent Full thread

This might suffer from the same issue the alcohol consumption study lacked.

People that aren't consuming eggs might do so because of health issues. Meaning that they are more prone to feel the side effects.

Similar effect happened in the alcohol study, where moderate consumption had a lower mortality than those consuming nothing. It turned out that those abstaining from alcohol were more prone to get side effects immediately. After adjusting for these errors in data reporting, the u-curve disappeared and the relationship was linear (more alcohol you consumed higher the mortality).

I was also, embarrassingly, citing the alcohol study, since then I've realized the first reaction to a paper should be doubt and I'll definitely abstain from acknowledging epidemiological diet studies in the future. Even meta-analyses taking them into account.

One of the best examples is that many meta-analyses conclude that dietary cholesterol does not increase serum cholesterol (proven risk in CVD). Disregarding the fact that many of the aggregated studies do not measure baseline cholesterol, and it is assumed that there's a linear response on serum cholesterol to dietary cholesterol when it's a 20 year old information that response is non-linear.

>People that aren't consuming eggs might do so because of health issues.

The authors are aware of this, and excluded people suffering from several common health problems for this reason:

"In the present study, we excluded individuals reporting medical histories of cancer (n=2577), heart disease (n=15 472) or stroke (n=8884), or having prevalent diabetes (n=30 300) defined by self-reported diabetes or on-site plasma glucose testing (fasting blood glucose ≥7.0 mmol/L or random blood glucose ≥11.1 mmol/L). We made these exclusions to avoid a prevalence–incidence bias and minimise the effect of reverse causality led by potential confounders such as lifestyle factors."

This isn't all possible sources of reverse causality, but it does at least make it less likely.

> dietary cholesterol does not increase serum cholesterol

The link between diet and cholesterol seems to be indirect.

Take for example coffee. Non-filtered coffee increases serum cholesterol even when it does not contain cholesterol. Coffee oils (diterpenes like cafestol and kahweol) mess up body’s ability to metabolize and regulate cholesterol.

Coffee contains 1% of diterpenes. Fortunately normal paper filter removes most of the oils. French press and Turkish style are the worst.

Do you have any additional links on this? I'm curious to read more.

> I've realized the first reaction to a paper should be doubt

appreciating the reminder

In sociology, the opposite is almost always true, in psychology, it’s usually false, in medicine, maybe half true, but what about machine learning?

>what about machine learning?

The results are probably real, but maybe not for the reasons the author suggests.

And here we have the clearest demonstration of confirmation bias.

Do you have the source for the alcohol study and the corrections? Didn't hear about this.

Here's the study saying that alcohol has protective effects: https://www.ncbi.nlm.nih.gov/pubmed/28818200

Not the first one of its kind but pretty large sample.

https://www.ncbi.nlm.nih.gov/pubmed/20636661

Here's a study showing that drinking a little alcohol lowers your chance of having liver cirrhosis compared to drinking no alcohol.

https://sci-hub.tw/10.1111/add.13627

Here's the criticism.

> We would like to add to her discussion of the significance of former and occasional drinker biases in this literature and highlight how they can cause both overestimation of cardioprotection and underestimation of cancer risks across the whole drinking continuum. The underlying theory here is that, as a population ages, a selection bias operates whereby individuals with poorer health are more likely to cut down or stop drinking completely. Such individuals are often still classified as ‘abstainers’ and used as a reference against which all current drinkers are compared. In simple terms, they make drinkers at all levels of consumption ‘look good’ by comparison.

https://www.ncbi.nlm.nih.gov/pubmed/26997174

Guys above then adjust for the mistake. The benefits disappear. Mortality risk grows as alcohol consumption increases. Yeah, there are people with protective genes, but on a population scale, recommending to go from 0 to N glasses of alcoholic beverage a day is insane.

I have personally increased my alcohol consumption, which was obviously dumb, given that the data was flawed. After this study I don't think I'll have more than couple of drinks per year.

Reply on news.ycombinator.com