Cardiovascular geneticist/statistician and physician-in-training here: This is a completely bogus and scientifically illiterate review. This small group of people in Sweden keep publishing crap like this and it's frankly kind of annoying.
It's hard to even know where to start with the overwhelming amount of evidence that elevated levels of LDL cholesterol causes cardiovascular disease.
Since I'm a geneticist, I'll point out just a handful of strong genetics-guided experiments:
1) Familial hypercholesterolemia - Mutations in the LDLR Gene causing basically only elevated LDL lead to 10x increased odds for coronary artery disease
It simply locates the article as being anchored in the philosophy of science championed by Popper. I should hope any meta-critique would be similarly rooted.
1) Bad stuff sneaks in everywhere if you get lucky with two reviewers and an editor asleep at the wheel. Even good journals. Not usually this bad, though.
2) This isn't a particularly good journal and isn't a cardiovascular journal, so they probably got two reviewers who are not knowledgeable about the field.
3) It might also have been an invited paper. These medoum-low journals often solicit reviews, and it's pretty rare to reject a paper that you've solicited.
No, everything you mentioned says that LDL-C levels correlates with cardiovascular disease rather than causes.
Mutation in LDLR gene is a cause. LDL-C level is a correlate. The mutation may or may not have additional repercussions other than just higher level of LDL-C. Like with many metabolic issues - the problem sometimes truly is the high level of (toxic) product, sometimes overload of the compensatory mechanism (competitive inhibition etc.), sometimes the product is innocent but the mutation has other effects.
PCSK9 inhibitors... I would be very careful. These might be just as inconsistent as statins with equivocal evidence.
For now, Cochrane agrees. (Cochrane cancelled 2018 rereview of statins due to methodology update, the 2013 one is somewhat suspect. Later fluvastatin review shows it's not potent enough etc. But this is all tangential.)
At least they will shine more light on the LDL-C hypothesis than statins which have more complex effects.
They are still not sensitive enough to rule out other mechanisms. The drugs only recently exited phase III trials.
To accurately conclude LDL-C causes cardiovascular disease you will have to figure out a way to increase LDL-C in absence of other relevant factors. Or perhaps strongly control for those factors. This is actually hard.
(among them: obesity, smoking, diet and excercise, gene variants)
This should be quite possible to verify in animal model by pumping them with relevant lipids regularly. Not entirely sure how to extend this to human trials, but a way can be found.
There are literally hundreds to thousands of papers where they have done exactly this: Feed animals a high-cholesterol diet and compare the results to control animals fed a normal diet. Usually they will also compare different drugs in separate experimental arms. The LDL diet is a positive control, the normal diet is a negative control, and the drug arms are usually the intervention being tested. How much evidence do you want? This (molecular imaging of atherosclerosis) is an entire subfield of medicine with tons of work.
I myself have done this in rabbits, followed by by molecular imaging (Tc-99m bound to various ligands) which directly image atherosclerosis in the animal. We use single photon emission computed tomography (SPECT) as well as position emission tomography (PET) to quantitatively measure atherosclerotic burden. We also sacrifice the animals and remove their arteries–upon unmagnified visual inspection of the plaques in the arteries, I can tell you which animals get which diet.
I'm not going to debate all of your points. Sure, it's possible there are other potential explanations for FH or PCSK9. However, they are not likely. Also, mendelian randomization studies are use a causal inference strategy via instrumental variable analysis, getting around the idea that we are measuring only correlations. It's a very nice technique which mimics a randomized controlled trial via the random assignment of genetic material during gametogenesis. So I don't agree with you.
Last thing: The equivocal evidence of statin therapy is only in primary prevention, not in secondary prevention. No one is really challenging that statins work, the question is what population is of high enough risk that they should get statins. This is widely misreported in the popular media and pretty harmful to patients who see a headline and stop taking their drugs.
High cholesterol diet is not the same as high blood level of lipids. There are many steps inbetween. It is additionally not the same as the high occurence of atheroma. Even more steps.
You've introduced an additional super complex variable (diet) where there should be none (blood level of LDL-C).
I do not dispute accuracy of atheroma measurements at all. This is easy enough technicality.
What is this "LDL diet" anyway? Do you feed animals with lipids?
If you mean high saturated fat diet, then that's a different beast. It does many more things than just increase LDL-C levels - it also increases TG levels, LDL levels in total, places metabolic burden on liver and more interesting things.
Even there, the trials are rightly calling these levels correlates. Not causes. And it is not a pedantic point, it is critical and informs the way we should be dealing with related CVD.
Secondary prevention wrt statins, hmm, I see one good USPTF study. It includes primary prevention too and has well defined results which are mildly positive.
It actually says you're not allowed to attribute statin results to LDL-C levels as LDL-C dose/response strategies were not superior to plain fixed dose.
More importantly, this area has not been evaluated with enough power.
Key point 1b. No dose response, likely no direct causation.
Suspected mechanism is thus different from direct LDL-C levels.
Most importantly: "Benefits of statins did not appear to be restricted to patients with severely elevated lipid levels, because similar effects were observed in subgroups stratified according to baseline levels."
We have barely enough statistical power to say they're superior, not even figure out the dosage in 3 levels.
Interesting fact there being that biggest trials reported smallest relative risk reduction. (~15% with pretty big error bars on this.) Absolute risk values are even more funny, like ~0.5% reduction... in high risk groups.
We feed the animals regular chow mixed with a percentage of pure cholesterol (0.5% by mass, if I recall, but I might be wrong). I do not mean a high saturated fat diet.
This is not the introduction of complex "diet" variables as in humans whatsoever. I am quite aware of the differences in various lipids. Feeding animals pure cholesterol is about as close as we can get. There might be studies that have literally infused LDL in animals, I am not exactly sure. It sounds like a huge pain to do experimentally since we would need to obtain vascular access at least each day and risk infection, stress and pain in the animals, etc.
We measure animal blood using the same machines used on humans and see primarily increases in total cholesterol and LDL cholesterol following the diet.
There is not "one good" study for secondary prevention of statins, there are dozens.
There is very good evidence for the effectiveness of dose-response in statins. Again, I don't even know where exactly to start with this body of literature. Look up "high intensity statin therapy." This review for primary prevention, which is probably why it was underpowered in this meta-analysis of 19 trials in primary prevention (where I said the evidence was more equivocal; this study actually shows pretty good evidence for the effectiveness of statins. in 1° prevention). A RR of 0.86 in only 1-6 years is pretty damn good for a chronic disease.
Finally, I think you misunderstand statistics when you say that this review (in which there were only three studies which could be used to analyze dose-response) implies there is no dose response.
When they analyze dose-response (which is done through increasing statin dosage until the LDL-C levels are low enough), I quote: "RR for cardiovascular mortality, 0.61 [95% CI, 0.37 to 1.02]" Yes, this is underpowered because the CI includes 1 but the point estimate of 0.6 is pretty impressive. This is a poor conclusion by the authors via incorrect application of the NHST framework, and not evidence that there is no dose-response effect. The best evidence is actually pretty strong (RR of 0.6 is big!). I personally doubt it's even that big.
Furthermore, RR for composite cardiovascular outcomes, 0.63 [95% CI, 0.53 to 0.76] was very statistically significant and again with a pretty strong signal. So I don't think this paper says what you say it does.
You didn't feel the need to address mendelian randomization studies in 500,000 plus people?
I'm not disputing effectiveness of statins one bit. They're effective, though not as hugely effective as we're led to believe by marketing. (15% RRR is probably worthwhile even if we don't know how these work.)
I'm disputing that the effect is related to LDL-C. There were no important differences in trials that titrated based on LDL-C or used fixed dose. Compare RR 0.63 vs 0.71 and full error bar overlap. There were also no relevant differences of outcome based on stratification by LDL-C levels.
I did not feel the need to address "genetic"/"metabolomic" studies. (Also known as data sifting.)
Standard MR is essentially n-D crosscorrelation of LDL-C levels with LDLR genetic variant and CHD occurences. We know that beforehand. It doesn't tell us anything of causation nor anything we don't know. The authors like to misuse R as "explains".
The stronger MR variant results suggest adiposity causes LDL-C levels. Which we already know, again, does not give us a clear treatment target.
The other advanced studies pinpoint specific lipids as the cause of CHD, but cannot say where they are from. And no, they're not quite LDL-C.
I'm not entirely sure how the authors got the rationale from what they're writing about.
The better MR-like methods as mentioned are like stepwise crosscorrelation and median clustering. Not entirely sure how they can actually show causality.
When the statistical guesswork method (MR is one kind of) does not match the observations... it is likely that either the method is incorrect or that our treatment is misapplied and reality does not match it. MR suggests RRR 3.5 for all LDL-C lowering drugs. We see 1.15 on statins and none on PCSK9 inhibitors vs CVD mortality.
Sorry, but I actually understand precisely what mendelian randomization and instrumental variable estimation are.
You apply nonlinear least squares estimation to every selected suspected variable which is a bunch of linearization/polynomial approximating cross correlations with normalized by estimated covariances.
That is nonlinear IV method.
MR is this based on an assumption that genomes are random fields of genes with equal independent probabilities of occurence (known to be false, partly corrected by population statistics which is not enough) giving direct results on marker conditions/correlates but not on endpoints (known to be false, partial attempts to correct for it are made to check for pleiotropy but those checks are weak and linear) which markers are nonlinearly covariate with endpoints (hope you used the right mapping) and that highest magnitudes of the "correlations" say something about causation related to certain endpoints. (not in general)
MR shines as a negative method to rule out genetic causation or susceptibility, not positive method to confirm it. It can be used to confirm internal validity of a marker - invariance of it vs genetics.
The choice of variables (dimensions) is an educated guess too.
The assumptions that are verified is that genetic sequence is not heterogenous, heteroskedastic, there are no hidden linkages etc.
The assumptions are not typically met or verified for complex system such as lipid metabolism alone, much less endpoints such as CVD. Polygenetic phenotypes are common and typical, not to say anything about epigenetics.
Nothing is said about the choice of the checked variables which is critical, and trying too many gives the problem of multiple measurement reducing power.
The strongest assumption is that the variables are not covariant with estimation error. This is not possible to check with MR statistics alone unless you can somehow check those variants experimentally directly. (e.g. via gene splicing or mutation) Good luck with that.
You wrote a completely wrong thing, edited it when I pointed out it was incorrect, and then 4 hours later you read up about MR on wikipedia or something and came back here and posted a partially correct answer with a lot of buzzwords and nice-sounding phrases, but with even more straw-man arguments and outright errors.
What would you say are the weakest aspects of the scientific consensus on LDL-C? Not doubting your 3 main points, just curious what you would identify as the weakest links in the reasoning as someone within the profession.
From my perspective, cardiology for a while was misunderstanding the contribution made to CVD from exogenous cholesterol (what you eat) vs. endogenous cholesterol (what your body makes/transports around). Initially there was a lot of focus on reducing dietary cholesterol. This isn't "Wrong" with a capital "W", but emerging evidence seems to indicate that dietary cholesterol doesn't play as big of a result as a person's endogenous cholesterol processes. However, it's still good to eat a lower cholesterol diet.
A lot of the resistance to LDL-C evidence comes from people who want patients to eat less refined sugar. This is 100% a great idea; refined sugar is bad for you metabolically for a number of reasons. However, it's not an either-or proposition and I can't understand why people keep making this case.
Finally, I think we are still working out the role chronic inflammatory processes play in cardiovascular disease. They are an important component of the atherosclerotic process and one of the three main "knobs" we have available to twist with medication (the major other ones being blood pressure via antihypertensives and LDL via statins/PCSK9is).
However, our typical anti-inflammatory drugs don't really work for cardiovascular disease yet the most common ones work on the COX pathway, which (I don't want to get into it, but it's pretty well studied) in chronic use actually causes increased risk for things like heart attack via upregulation of various prostaglandins that you don't want upregulated.
There's some really interesting work on monoclonal antibodies targeting different inflammatory pathways that seems promising, but it's (1) still a ways in the future and (2) seems to be less powerful (so far) than is managing BP and LDL.
Edit: Also, if you had asked me five years ago, I'd say that the evidence for the pleiotropic effects of statins (e.g., effects of statins on things other than LDL cholesterol) were pretty interesting. And they were, 5 years ago! However, a lot of good evidence has come out since then showing that most of the effects of statins can be attributed by their action on LDL. It seems like the pleiotropic effects of statin therapy are getting less and less significant as we study them more. For me, this emerging evidence was actually kind of disappointing as I had been planning a few studies to look at statin pleiotropy.
Actually I've dug out a recent (2016) analysis that actually reasonably strongly suggests mechanism different from direct lipid lowering effects. See thread above.
More importantly, PCSK9 inhibitors have not shown clear mortality benefits in best analysis we have and they're much more specific to cholesterol...
And yes, this is after and including the big trials. There are some benefits shown, but they do not translate to mortality endpoint, despite not translating to obvious other controlled morbidity. It looks like an odd and unexpected result, but partly predicted by studies correlating higher total cholesterol with lifespan.
Everyone would like it to be as simple as turning down LDL-C knob and fixing the world. Unfortunately, reality is apparently not as accomodating. We truly do not know the exact mechanisms.
Again, this paper does not support what you're saying it does at all. You are either misreading or intentionally distorting what it says.
1) Patients had a baseline cholesterol of 106 mg/dL. So already pretty good control of lipids. This implies they were already on intensive statin therapy, and so the effect of PCSK9is can be only incremental improvements. Given that you say PCSK9is are more specific to cholesterol (although this is debatable), we can thus probably attribute most of the effect to reduction in LDL.
2) I quote:
Treatment with a PCSK9 inhibitor was associated with a lower rate of
a) myocardial infarction (2.3% versus 3.6%; odds ratio [OR]: 0.72 [95% confidence interval (CI), 0.64-0.81]; P<0.001)
b) stroke (1.0% versus 1.4%; OR: 0.80 [95% CI, 0.67-0.96]; P=0.02)
c) coronary revascularization (4.2% versus 5.8%; OR: 0.78 [95% CI, 0.71-0.86]; P<0.001).
d) All cause mortality was only marginally significant: all-cause mortality (OR: 0.71 [95% CI, 0.47-1.09]; P=0.12).
This does not mean PCSK9is don't reduce mortality, it just means that we haven't followed enough people for long enough. Again, you have to understand how null hypothesis testing works to interpret this statement. The reason that they reduces MI and Stroke without the same effect on reducing all-cause mortality is that we treat MI and Stroke pretty well typically, so you usually don't die from them. The event rate for these is higher, which means if you run a trial for them as the outcome instead of all-cause mortality you can use fewer patients and get it done faster and cheaper, which is ultimately good for patients. However, we obviously know that MI and Stroke are associated with increased risk death. Furthermore, you can die from a lot of other reasons than cardiovascular disease, which makes the trial require even more people to separate out the signal.
No one is arguing that PCSK9 inhibitors are a magic bullet (well maybe someone is, but not most people), but they are clearly pretty good even when added to already-pretty-good therapy.
This is an interesting definition of marginal significance, with P=0.12 (more than double the standard 0.05) and 10%+ OR overlap.
You seem to be focusing on OR means too much.
Absolute risk adjustments are surprisingly low considering certain predictions.
I agree with the conclusion that 106 mg/dL mean could've reduced the power of the experiment.
I also agree that a good followup is relevant and important. It could change results a lot, including the initial positive results. However, in the meantime, we have no way to actually evaluate the risks and are essentially making educated guesses.
Note that the patients treated were on average 60 years old. This reduces followup time considerably but also reduces potential effect size.
The results are mostly based on ODYSSEY LONG TERM and FOURIER trials. The former verified PCSK9i alongside high dose statin therapy. The latter used a monotherapy in some cases but not defined the subgroup.
FOURIER alone is quite damning. ODYSSEY LONG TERM does not consider mortality endpoints.
Such interesting tidbits like "In terms of individual outcomes, evolocumab had no observed effect on cardiovascular mortality, and hence P values for other outcomes should be considered exploratory."
The inhibitors in question were allegedly not mainly tested here as an adjunct to statin therapy or replacement for resistant cases, but it so happened anyway.
You're supposed to be comparing it to current best known treatment which is statins, which is a very valid comparison. Any other comparison would only be of educational value but should be done regardless. It is even more suspect that it was not done but instead it was used as an adjunct. Not even canadian cross design was used...
It seems like your argument strategy is to throw out 10 different ideas. I refute them, then you throw out 10 more. Meanwhile you ignore any evidence I provide. This is not an honest conversation and no evidence is going to change your mind. I feel like a global warming scientist.
As someone who is not expert, what's annoying is having to play the game of "Who do I believe?" You have your points, and they have their meta-study. I don't want to dismiss their study just because you don't like their journal, which is a bit of an ad hominem. And the problem is you are making your case, but not actually addressing their study.
So my request to you: If you were a journal referee and this paper landed on your desk for review, what would your response to the editor be? I think many here would like to hear your criticism of the actual study.
To give an analogy, I feel a bit like a global warming scientist (I don't actually know anything about global warming-I just understand it's a scientific consensus with a really vocal but marginal minority).
It's not just me, it's me and 99% of cardiovascular scientists vs. a few people scattered around the world.
Extraordinary claims require extraordinary evidence, and this paper glosses over literally hundreds of contradictory studies, cherry picks a few that sort of support it, and wraps it all with some rhetorical flourishes. It's not science.
I usually take at least two Or three hours to review a paper. I'm not sure if I want to waste my time writing out line by line all of the mistakes. something like this will require even more time since I would feel the duty to to dig up dozens and dozens and dozens of papers which would contradict them.
If that would change your mind, I'll think about doing it.
Finally, I said it's "annoying" because I work really hard on all my reviews and papers to make sure what I am publishing is right. The goofballs slap some crap together with only a passing understanding of the evidence and get another line on their CV. Feel free to ignore this ad hominem but hopefully it helps to understand my perspective.
26 comments
[ 4.0 ms ] story [ 206 ms ] threadIt's hard to even know where to start with the overwhelming amount of evidence that elevated levels of LDL cholesterol causes cardiovascular disease.
Since I'm a geneticist, I'll point out just a handful of strong genetics-guided experiments:
1) Familial hypercholesterolemia - Mutations in the LDLR Gene causing basically only elevated LDL lead to 10x increased odds for coronary artery disease
2) PCSK9 inhibitors - essentially only reduce LDL concentration, dramatically reduce risk
3) Mendelian randomization techniques - statistical causal inference in massive datasets shows pretty conclusively LDL increases CVD
Yes, LDL cholesterol causes CVD. Yes, there are other factors as well (especially hemodynamics and chronic inflammation).
1) Bad stuff sneaks in everywhere if you get lucky with two reviewers and an editor asleep at the wheel. Even good journals. Not usually this bad, though.
2) This isn't a particularly good journal and isn't a cardiovascular journal, so they probably got two reviewers who are not knowledgeable about the field.
3) It might also have been an invited paper. These medoum-low journals often solicit reviews, and it's pretty rare to reject a paper that you've solicited.
Mutation in LDLR gene is a cause. LDL-C level is a correlate. The mutation may or may not have additional repercussions other than just higher level of LDL-C. Like with many metabolic issues - the problem sometimes truly is the high level of (toxic) product, sometimes overload of the compensatory mechanism (competitive inhibition etc.), sometimes the product is innocent but the mutation has other effects.
PCSK9 inhibitors... I would be very careful. These might be just as inconsistent as statins with equivocal evidence. For now, Cochrane agrees. (Cochrane cancelled 2018 rereview of statins due to methodology update, the 2013 one is somewhat suspect. Later fluvastatin review shows it's not potent enough etc. But this is all tangential.)
At least they will shine more light on the LDL-C hypothesis than statins which have more complex effects. They are still not sensitive enough to rule out other mechanisms. The drugs only recently exited phase III trials.
To accurately conclude LDL-C causes cardiovascular disease you will have to figure out a way to increase LDL-C in absence of other relevant factors. Or perhaps strongly control for those factors. This is actually hard. (among them: obesity, smoking, diet and excercise, gene variants)
This should be quite possible to verify in animal model by pumping them with relevant lipids regularly. Not entirely sure how to extend this to human trials, but a way can be found.
If you have such results, I'd like to read them.
I myself have done this in rabbits, followed by by molecular imaging (Tc-99m bound to various ligands) which directly image atherosclerosis in the animal. We use single photon emission computed tomography (SPECT) as well as position emission tomography (PET) to quantitatively measure atherosclerotic burden. We also sacrifice the animals and remove their arteries–upon unmagnified visual inspection of the plaques in the arteries, I can tell you which animals get which diet.
I'm not going to debate all of your points. Sure, it's possible there are other potential explanations for FH or PCSK9. However, they are not likely. Also, mendelian randomization studies are use a causal inference strategy via instrumental variable analysis, getting around the idea that we are measuring only correlations. It's a very nice technique which mimics a randomized controlled trial via the random assignment of genetic material during gametogenesis. So I don't agree with you.
Last thing: The equivocal evidence of statin therapy is only in primary prevention, not in secondary prevention. No one is really challenging that statins work, the question is what population is of high enough risk that they should get statins. This is widely misreported in the popular media and pretty harmful to patients who see a headline and stop taking their drugs.
You've introduced an additional super complex variable (diet) where there should be none (blood level of LDL-C). I do not dispute accuracy of atheroma measurements at all. This is easy enough technicality.
What is this "LDL diet" anyway? Do you feed animals with lipids?
If you mean high saturated fat diet, then that's a different beast. It does many more things than just increase LDL-C levels - it also increases TG levels, LDL levels in total, places metabolic burden on liver and more interesting things.
Even there, the trials are rightly calling these levels correlates. Not causes. And it is not a pedantic point, it is critical and informs the way we should be dealing with related CVD.
Secondary prevention wrt statins, hmm, I see one good USPTF study. It includes primary prevention too and has well defined results which are mildly positive.
It actually says you're not allowed to attribute statin results to LDL-C levels as LDL-C dose/response strategies were not superior to plain fixed dose. More importantly, this area has not been evaluated with enough power.
https://jamanetwork.com/journals/jama/fullarticle/2584057
Key point 1b. No dose response, likely no direct causation. Suspected mechanism is thus different from direct LDL-C levels. Most importantly: "Benefits of statins did not appear to be restricted to patients with severely elevated lipid levels, because similar effects were observed in subgroups stratified according to baseline levels."
We have barely enough statistical power to say they're superior, not even figure out the dosage in 3 levels.
Interesting fact there being that biggest trials reported smallest relative risk reduction. (~15% with pretty big error bars on this.) Absolute risk values are even more funny, like ~0.5% reduction... in high risk groups.
This is not the introduction of complex "diet" variables as in humans whatsoever. I am quite aware of the differences in various lipids. Feeding animals pure cholesterol is about as close as we can get. There might be studies that have literally infused LDL in animals, I am not exactly sure. It sounds like a huge pain to do experimentally since we would need to obtain vascular access at least each day and risk infection, stress and pain in the animals, etc.
We measure animal blood using the same machines used on humans and see primarily increases in total cholesterol and LDL cholesterol following the diet.
There is not "one good" study for secondary prevention of statins, there are dozens.
There is very good evidence for the effectiveness of dose-response in statins. Again, I don't even know where exactly to start with this body of literature. Look up "high intensity statin therapy." This review for primary prevention, which is probably why it was underpowered in this meta-analysis of 19 trials in primary prevention (where I said the evidence was more equivocal; this study actually shows pretty good evidence for the effectiveness of statins. in 1° prevention). A RR of 0.86 in only 1-6 years is pretty damn good for a chronic disease.
Finally, I think you misunderstand statistics when you say that this review (in which there were only three studies which could be used to analyze dose-response) implies there is no dose response.
When they analyze dose-response (which is done through increasing statin dosage until the LDL-C levels are low enough), I quote: "RR for cardiovascular mortality, 0.61 [95% CI, 0.37 to 1.02]" Yes, this is underpowered because the CI includes 1 but the point estimate of 0.6 is pretty impressive. This is a poor conclusion by the authors via incorrect application of the NHST framework, and not evidence that there is no dose-response effect. The best evidence is actually pretty strong (RR of 0.6 is big!). I personally doubt it's even that big.
Furthermore, RR for composite cardiovascular outcomes, 0.63 [95% CI, 0.53 to 0.76] was very statistically significant and again with a pretty strong signal. So I don't think this paper says what you say it does.
You didn't feel the need to address mendelian randomization studies in 500,000 plus people?
I'm disputing that the effect is related to LDL-C. There were no important differences in trials that titrated based on LDL-C or used fixed dose. Compare RR 0.63 vs 0.71 and full error bar overlap. There were also no relevant differences of outcome based on stratification by LDL-C levels.
I did not feel the need to address "genetic"/"metabolomic" studies. (Also known as data sifting.)
Standard MR is essentially n-D crosscorrelation of LDL-C levels with LDLR genetic variant and CHD occurences. We know that beforehand. It doesn't tell us anything of causation nor anything we don't know. The authors like to misuse R as "explains".
The stronger MR variant results suggest adiposity causes LDL-C levels. Which we already know, again, does not give us a clear treatment target.
The other advanced studies pinpoint specific lipids as the cause of CHD, but cannot say where they are from. And no, they're not quite LDL-C.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4816855/
I'm not entirely sure how the authors got the rationale from what they're writing about.
The better MR-like methods as mentioned are like stepwise crosscorrelation and median clustering. Not entirely sure how they can actually show causality.
When the statistical guesswork method (MR is one kind of) does not match the observations... it is likely that either the method is incorrect or that our treatment is misapplied and reality does not match it. MR suggests RRR 3.5 for all LDL-C lowering drugs. We see 1.15 on statins and none on PCSK9 inhibitors vs CVD mortality.
I'd bet on MR being invalid/confused.
I'm not even sure what the methods you propose are, they're definitely not popular ways to do causal inference in medicine or epidemiology.
You apply nonlinear least squares estimation to every selected suspected variable which is a bunch of linearization/polynomial approximating cross correlations with normalized by estimated covariances. That is nonlinear IV method.
MR is this based on an assumption that genomes are random fields of genes with equal independent probabilities of occurence (known to be false, partly corrected by population statistics which is not enough) giving direct results on marker conditions/correlates but not on endpoints (known to be false, partial attempts to correct for it are made to check for pleiotropy but those checks are weak and linear) which markers are nonlinearly covariate with endpoints (hope you used the right mapping) and that highest magnitudes of the "correlations" say something about causation related to certain endpoints. (not in general) MR shines as a negative method to rule out genetic causation or susceptibility, not positive method to confirm it. It can be used to confirm internal validity of a marker - invariance of it vs genetics.
The choice of variables (dimensions) is an educated guess too.
The assumptions that are verified is that genetic sequence is not heterogenous, heteroskedastic, there are no hidden linkages etc.
The assumptions are not typically met or verified for complex system such as lipid metabolism alone, much less endpoints such as CVD. Polygenetic phenotypes are common and typical, not to say anything about epigenetics.
Nothing is said about the choice of the checked variables which is critical, and trying too many gives the problem of multiple measurement reducing power.
The strongest assumption is that the variables are not covariant with estimation error. This is not possible to check with MR statistics alone unless you can somehow check those variants experimentally directly. (e.g. via gene splicing or mutation) Good luck with that.
You wrote a completely wrong thing, edited it when I pointed out it was incorrect, and then 4 hours later you read up about MR on wikipedia or something and came back here and posted a partially correct answer with a lot of buzzwords and nice-sounding phrases, but with even more straw-man arguments and outright errors.
Peace out, I'm done with you.
I'll list a few in no particular order:
From my perspective, cardiology for a while was misunderstanding the contribution made to CVD from exogenous cholesterol (what you eat) vs. endogenous cholesterol (what your body makes/transports around). Initially there was a lot of focus on reducing dietary cholesterol. This isn't "Wrong" with a capital "W", but emerging evidence seems to indicate that dietary cholesterol doesn't play as big of a result as a person's endogenous cholesterol processes. However, it's still good to eat a lower cholesterol diet.
A lot of the resistance to LDL-C evidence comes from people who want patients to eat less refined sugar. This is 100% a great idea; refined sugar is bad for you metabolically for a number of reasons. However, it's not an either-or proposition and I can't understand why people keep making this case.
Finally, I think we are still working out the role chronic inflammatory processes play in cardiovascular disease. They are an important component of the atherosclerotic process and one of the three main "knobs" we have available to twist with medication (the major other ones being blood pressure via antihypertensives and LDL via statins/PCSK9is).
However, our typical anti-inflammatory drugs don't really work for cardiovascular disease yet the most common ones work on the COX pathway, which (I don't want to get into it, but it's pretty well studied) in chronic use actually causes increased risk for things like heart attack via upregulation of various prostaglandins that you don't want upregulated.
There's some really interesting work on monoclonal antibodies targeting different inflammatory pathways that seems promising, but it's (1) still a ways in the future and (2) seems to be less powerful (so far) than is managing BP and LDL.
Edit: Also, if you had asked me five years ago, I'd say that the evidence for the pleiotropic effects of statins (e.g., effects of statins on things other than LDL cholesterol) were pretty interesting. And they were, 5 years ago! However, a lot of good evidence has come out since then showing that most of the effects of statins can be attributed by their action on LDL. It seems like the pleiotropic effects of statin therapy are getting less and less significant as we study them more. For me, this emerging evidence was actually kind of disappointing as I had been planning a few studies to look at statin pleiotropy.
More importantly, PCSK9 inhibitors have not shown clear mortality benefits in best analysis we have and they're much more specific to cholesterol...
https://www.ncbi.nlm.nih.gov/pubmed/29223954
And yes, this is after and including the big trials. There are some benefits shown, but they do not translate to mortality endpoint, despite not translating to obvious other controlled morbidity. It looks like an odd and unexpected result, but partly predicted by studies correlating higher total cholesterol with lifespan.
Everyone would like it to be as simple as turning down LDL-C knob and fixing the world. Unfortunately, reality is apparently not as accomodating. We truly do not know the exact mechanisms.
1) Patients had a baseline cholesterol of 106 mg/dL. So already pretty good control of lipids. This implies they were already on intensive statin therapy, and so the effect of PCSK9is can be only incremental improvements. Given that you say PCSK9is are more specific to cholesterol (although this is debatable), we can thus probably attribute most of the effect to reduction in LDL.
2) I quote:
Treatment with a PCSK9 inhibitor was associated with a lower rate of
a) myocardial infarction (2.3% versus 3.6%; odds ratio [OR]: 0.72 [95% confidence interval (CI), 0.64-0.81]; P<0.001)
b) stroke (1.0% versus 1.4%; OR: 0.80 [95% CI, 0.67-0.96]; P=0.02)
c) coronary revascularization (4.2% versus 5.8%; OR: 0.78 [95% CI, 0.71-0.86]; P<0.001).
d) All cause mortality was only marginally significant: all-cause mortality (OR: 0.71 [95% CI, 0.47-1.09]; P=0.12).
This does not mean PCSK9is don't reduce mortality, it just means that we haven't followed enough people for long enough. Again, you have to understand how null hypothesis testing works to interpret this statement. The reason that they reduces MI and Stroke without the same effect on reducing all-cause mortality is that we treat MI and Stroke pretty well typically, so you usually don't die from them. The event rate for these is higher, which means if you run a trial for them as the outcome instead of all-cause mortality you can use fewer patients and get it done faster and cheaper, which is ultimately good for patients. However, we obviously know that MI and Stroke are associated with increased risk death. Furthermore, you can die from a lot of other reasons than cardiovascular disease, which makes the trial require even more people to separate out the signal.
No one is arguing that PCSK9 inhibitors are a magic bullet (well maybe someone is, but not most people), but they are clearly pretty good even when added to already-pretty-good therapy.
You seem to be focusing on OR means too much. Absolute risk adjustments are surprisingly low considering certain predictions.
I agree with the conclusion that 106 mg/dL mean could've reduced the power of the experiment.
I also agree that a good followup is relevant and important. It could change results a lot, including the initial positive results. However, in the meantime, we have no way to actually evaluate the risks and are essentially making educated guesses. Note that the patients treated were on average 60 years old. This reduces followup time considerably but also reduces potential effect size.
The results are mostly based on ODYSSEY LONG TERM and FOURIER trials. The former verified PCSK9i alongside high dose statin therapy. The latter used a monotherapy in some cases but not defined the subgroup.
FOURIER alone is quite damning. ODYSSEY LONG TERM does not consider mortality endpoints. Such interesting tidbits like "In terms of individual outcomes, evolocumab had no observed effect on cardiovascular mortality, and hence P values for other outcomes should be considered exploratory."
The inhibitors in question were allegedly not mainly tested here as an adjunct to statin therapy or replacement for resistant cases, but it so happened anyway.
You're supposed to be comparing it to current best known treatment which is statins, which is a very valid comparison. Any other comparison would only be of educational value but should be done regardless. It is even more suspect that it was not done but instead it was used as an adjunct. Not even canadian cross design was used...
So my request to you: If you were a journal referee and this paper landed on your desk for review, what would your response to the editor be? I think many here would like to hear your criticism of the actual study.
Thanks.
It's not just me, it's me and 99% of cardiovascular scientists vs. a few people scattered around the world.
Extraordinary claims require extraordinary evidence, and this paper glosses over literally hundreds of contradictory studies, cherry picks a few that sort of support it, and wraps it all with some rhetorical flourishes. It's not science.
I usually take at least two Or three hours to review a paper. I'm not sure if I want to waste my time writing out line by line all of the mistakes. something like this will require even more time since I would feel the duty to to dig up dozens and dozens and dozens of papers which would contradict them.
If that would change your mind, I'll think about doing it.
Finally, I said it's "annoying" because I work really hard on all my reviews and papers to make sure what I am publishing is right. The goofballs slap some crap together with only a passing understanding of the evidence and get another line on their CV. Feel free to ignore this ad hominem but hopefully it helps to understand my perspective.