> First, the use of lipid-lowering medication was unaccounted for. The risk associated with high cholesterol might have been underestimated. However, in Korea, IHD mortality accounted for only approximately 5% of all-cause mortality, and only 10% of people with hypercholesterolemia received lipid-lowering therapy[39]. Therefore, the impact of not considering medication use is likely to be modest, and the TC levels in this study generally reflect levels without lipid-lowering medications
> only 10% of people with hypercholesterolemia received lipid-lowering therapy
Wouldn't this be much higher in the UK cohort? Also,
> The sex- and age- specific levels of TC in the current study of Koreans were lower than those reported in other high-income countries, including Japan, England, and the US
Which would reduce the effect (relative to what was found in the analysis of UK data) even further.
I don't think it adds a whole lot to talk about the confounding variables of this nature - it's an observational study, so of course it will have all sorts of confounding variables. If you are going to make a strong case that controlling for a specific confounding variable would reverse the relationship though, you should quantify the impact. It's unclear to me that statin usage would make a large difference, given that the study notes it wasn't a popular treatment.
My general feeling is that these types of studies tend to end up measuring overall health - given that most of the factors that lead to high cholesterol level were controlled for (BMI, physical activity, smoking status, drinking, etc) - and the average person in any population is relatively healthy and well-adjusted to the common diet, any large deviation from the norm in any direction statically makes it likely that the subject has some health issues.
What would be interesting to see in a meta-study is whether the optimal level of cholesterol either was consistent across populations that had very different averages or it tends to be near the median in each population. The former would be evidence that it's an independently meaningful biomarker; the latter would be evidence that it's just measuring how normal you are.
Why would you need to account for it? Trying to understand the logic here.
The study goes to showing the association between high cholesterol and all-cause mortality (which is what we really care about right?) is weak.
This would seem to diminish the benefits of statin use, all other things being equal.
With regard to the point of the study, they start by saying, "It is unclear whether associations between total cholesterol (TC) levels and all-cause mortality and the optimal TC ranges for lowest mortality vary by sex and age."
So, interpreting their statement, I would say that their goal with this study was to investigate whether the relationship between total cholesterol and mortality was consistent across age and sex. That is a different point from assessing the strength of the relationship between cholesterol and mortality.
I do agree, however, that there is no single biomarker aside from age that is a particularly powerful predictor of mortality in the general population.
[Edit] I just read another one of your comments that answers my question, and I think you're right: https://news.ycombinator.com/item?id=20138685
I didn't get enough from the tweets to understand how that was accounted for in the data. But I have questions.
By "inverts the causal relationship" I assume you threw "causal" in there by mistake. Nobody is going to think mortality causes high cholesterol. More importantly, if ignoring statin use reverses the slope to weaken the relationship "high cholesterol causes mortality" doesn't that suggest a lack of efficacy for statins? We know they lower cholesterol, but if that also reduces mortality the relationship (between cholesterol and mortality) should hold weather they are accounted for or not. Right?
Prevalent cardiovascular disease, or having a high risk for cardiovascular disease, causes low cholesterol. It does so by your doctor recognizing your elevated risk and putting you on statins.
Statins, like all drugs for prevention, do not prevent all subsequent disease. Perhaps most importantly, they only reduce cholesterol from the time you start taking them (as opposed to a genetically-driven low cholesterol throughout life). All of the drugs for chronic disease prevention tend to have a pretty small effect on the disease risk. Therefore, detecting that someone is taking a statin is a great signal that the person is at high risk for disease. They would be at an even higher risk if they weren't on a statin, but the statin does not erase all risk. That is why failing to account for statin use will lead to low cholesterol looking protective, if there are people in the population taking statins.
If you mean there are no reliable bio markers, period, it would be more clear to just say that.
We actually can measure the biological age. There are clinical biomarkers for that. It is not commonly done. The best one, epigenetic signature, is yet to be clinically approved or widely available.
Now, in that paper, they use broad range of age as a descriptor, confounding many age-associated phenotype and genotypes, which also translate into risk factors. You can definitely say that the 18-34 cohort is different 45-54, and that the later as more risk factors, the same a specific mutation might be associated with higher or even guaranteed chance of developing a disease, and hence will be use as a biomarker.
That being said, it is not typical to measure the biological age of healthy individuals. Actually, we don't know what an healthy individual is. We know a great deal more about diseased individuals. The current status seems to be that if you are not diseased you are healthy. I don't agree with that. That's how science is done. And that is how we end up with half-baked studies in one of the most impactful journal .
What is the precision and accuracy of the best biomarker? ie. if I gave you the necessary sample, how close can you get to the correct age and what are the error bars?
Also, I created those graphs and sent them to him, so (1) he didn't create the graphs or influence how they were created, and (2) and I have no financial interests in that company. His company plans to produce a molecule that disrupts PCSK9 production, thereby reducing LDL-cholesterol.
If you think that the conflict of interest influenced the analysis in a way that you would have done differently, what is the alternative approach by which you suggest analyzing the data?
If they failed to account for statin use, then they will get the wrong answer due to confounding by indication.
https://twitter.com/skathire/status/1110909549811322880
Edit: Found it - they are not accounting for statin use! None of their conclusions seem warranted.