11th December 2019
Several people have asked me to comment on a recent Lancet paper ‘Application of non-HDL cholesterol for population-based cardiovascular risk stratification: results from the Multinational Cardiovascular Risk Consortium.’ which made headlines around the world. Here – for example – from the BBC website:
What did the researchers find?
People should have their cholesterol level checked from their mid-20s, according to researchers. They say it is possible to use the reading to calculate the lifetime risk of heart disease and stroke.
The study, in The Lancet, is the most comprehensive yet to look at the long-term health risks of having too much “bad” cholesterol for decades. They say the earlier people take action to reduce cholesterol through diet changes and medication, the better.
They analysed data from almost 400,000 people from 19 countries and found a strong link between bad-cholesterol levels and the risk of cardiovascular disease from early adulthood over the next 40 years or more.
They were able to estimate the probability of a heart attack or stroke for people aged 35 and over, according to their gender, bad-cholesterol level, age and risk factors such as smoking, diabetes, height and weight, and blood pressure.
Report co-author, Prof Stefan Blankenberg, from the University Heart Center, Hamburg, said: “The risk scores currently used in the clinic to decide whether a person should have lipid-lowering treatment only assess the risk of cardiovascular disease over 10 years and so may underestimate lifetime risk, particularly in young people.” 1
The Daily Mail in the UK was a bit more excitable in its reporting
‘Adults ‘should have their cholesterol checked at 25’ because slashing it in the mid-30s can drastically reduce the risk of heart attacks and strokes
Researchers predicted huge 30-year risk profiles for heart disease and stroke, they found higher cholesterol in under-45s is more dangerous than in over-60s Even young people with healthy lifestyles ‘may benefit from knowing their risk.’ 2
My first thought, as always, is to look for the conflict of interest statement, just so you know how independent the researchers may be and make an estimate of potential bias.
My second thought was that this study did not look at cholesterol levels, or HDL levels, it looked at non-HDL cholesterol. An interesting thing to study. This is every form of liver derived lipoprotein that is not HDL, otherwise known as ‘good cholesterol’ or simply high-density lipoprotein.
Part of the reason for not looking at LDL, is that LDL is very rarely measured, or reported. Because the only way to measure LDL accurately is through ultracentrifuge, which is time consuming and expensive. Normally, the LDL levels are simply estimated using the Friedwald equation. To quote from the UK GP Notebook
‘… the ultracentrifugal measurement of LDL is time consuming and expensive and requires specialist equipment. For this reason, LDL-cholesterol is most commonly estimated from quantitative measurements of total and HDL-cholesterol and plasma triglycerides (TG) using the empirical relationship of Friedewald et al.(1972).
[LDL-chol] = [Total chol] – [HDL-chol] – ([TG]/2.2) where all concentrations are given in mmol/L (note that if calculated using all concentrations in mg/dL then the equation is [LDL-chol] = [Total chol] – [HDL-chol] – ([TG]/5))’ 3
*TG = triglyceride
This means that the researchers will not have had any data on LDL levels for most people. The difficulty of directly measuring LDL is the reason why the risk calculators used in the UK and US do not even include LDL. These calculators are Qrisk3 https://qrisk.org/three/ and cvriskcalculator http://www.cvriskcalculator.com/
So, it is important to note that this was not a study on LDL levels. Instead, it was a study on non-HDL levels. Which changes it into something completely different than was reported. Obviously, non-HDL levels bear some relationship to LDL, in that a higher LDL level will tend to raise the overall non-HDL cholesterol level.
However, and very importantly, non-HDL also includes the triglyceride (TG) level. Or at least the TG level divided by 2.2. This is important because a high triglyceride (TG) level, divided by 2.2 or not, is a strong indicator of insulin resistance, which leads to type II diabetes. Here is what WebMD has to say on the matter
‘High TG’s signals insulin resistance; that’s when you have excess insulin and blood sugar isn’t responding in normal ways to insulin. This results in higher than normal blood sugar levels. If you have insulin resistance, you’re one step closer to type 2 diabetes.’ 4
Insulin resistance, whether or not it has developed into type II diabetes, greatly increases your risk of both CVD and overall mortality, as outlined in the paper. ‘Triglyceride–to–High-Density-Lipoprotein-Cholesterol Ratio Is an Index of Heart Disease Mortality and of Incidence of Type 2 Diabetes Mellitus in Men.’
‘This study shows that a high TG/HDL-C ratio in men is a predictor of mortality from CHD and CVD. The TG/HDL-C ratio had a significant and higher HR [hazard ratio] for mortality from CHD and CVD than was found for the TyG index [fasting blood sugar]. These 2 measures, TG/HDL-C ratio and TyG index, similarly predicted incidence of type 2 diabetes, but the HR associated with a high TG/HDL-C seems to make the ratio a preferred single parameter of measurement.’ 5
You will get no argument from me that a high triglyceride level is going to indicate the underlying metabolic catastrophe that is insulin resistance. This, in turn, is going to greatly increase the risk of CVD and early death. But this will have nothing to do with the LDL level. So, it has nothing to do with ‘bad’ cholesterol. Instead is to do with triglycerides.
Therefore, this study is like looking at people who smoke, and who eat red meat, then stating that red meat consumption and smoking cause lung cancer. You have arbitrarily rammed two things together without making any effort to decided which causes what. Scientific nonsense.
There also some massive statistical problems with this study. Where, for example is overall mortality? Not mentioned. Not mentioned means it will not have been significant. Also, the use of a very wide and fuzzy ‘combined end-point.’ I have written about this many times, in many different places. It is a game played to claim statistical significance, where none really exists.
To try and explain as quickly as possible. The most powerful end-point is overall mortality i.e. how many people were dead in either group. Or, to be more positive, how many people were alive in either group.
After this come end-points of decreasing importance. For example, how many people died of CVD. This is clearly important, but if more people died of CVD in one group, yet they were less likely to die of cancer, the overall mortality could remain the same in both groups – even if CVD mortality were lower in one.
- CVD deaths 150
- Cancer death 150
- Total deaths/mortality 300
- CVD deaths 180
- Cancer death 120
- Total deaths/mortality 300
Net benefit = zero. But such results can often be hailed as a massive success for, say, a drug. For example, the ‘FOURIER’ study on Repatha (injectable LDL lowering agent) was hailed as a great success, despite overall mortality being higher in the Repatha arm. How, you may think, was this possible?
Well, the Fourier study had five end-points. Known as a ‘combined end-point’. [Mortality was not one of them.] The primary end point was the combined total of:
- Cardiovascular death
- MI (myocardial infarction)
- Hospitalization for unstable angina
- Coronary revascularization
How can you have five different end-points as a primary end point? Well, you just can… apparently. 6
What you may notice, or maybe not, is that three of these are clinical events: cardiovascular death, MI and stroke. Two of them are clinical decisions. To admit someone to hospital for unstable angina, and to carry out a coronary revascularisation. Revascularisation is, essentially, putting in a stent to keep a coronary artery open.
So, the second two end-points are potentially subject to significant clinical bias. If someone has a low non-HDL cholesterol level, the decision may well be to not admit to hospital for unstable angina, and to not carry out coronary revascularisation. Why, because the physicians think they are protected by their low cholesterol.
[Guess which end-point dragged the FOURIER study into statistical significance.]
You think that clinical decisions are all objective. Then ask yourself why all clinical trials, wherever possible, are double-blinded (neither the patient or the doctor knows who is taking the drug, or the placebo)? This double blinding is considered essential to remove clinical bias. No blinding, bias introduced.
Even if you look at MIs and strokes, this diagnosis is less certain than you might wish. I have had many patients where it is entirely unclear if they have actually had a stroke or a heart attack. With a small stroke it is often, simply a guess. Low cholesterol, I guess not a stroke. High cholesterol, I guess that it is.
Just in case you think I am now talking nonsense, as I was writing this blog, I was sent a BBC report of a clinical trial done on stents in the US, which stated that stents were as safe as bypass surgery, with regard to MIs. However, the researchers decided to use a completely different system for diagnosing MI…
‘The trial called Excel started in 2010 and was sponsored by big US stent maker, Abbott. It was led by eminent US doctor Gregg Stone and aimed to recruit 2,000 patients. Half were given stents and the other half open heart surgery. Success of the treatments was measured by adding together the number of patients that had heart attacks, strokes, or had died.
The research team used an unusual definition of a heart attack, but had said that they would also publish data for the more common “Universal” definition of a heart attack alongside it. There is debate around which is a better measure and the investigators stand by their choice.
In 2016, the results of the trial for patients three years after their treatments were published in the prestigious New England Journal of Medicine. The article concluded stents and heart surgery were equally effective for people with left main coronary artery disease.
But researchers had failed to publish data for the common, “Universal” definition of a heart attack. Newsnight has seen that unpublished data and it shows that under the universal definition, patients in the trial that had received stents had 80% more heart attacks than those who had open heart surgery.
The lead researchers on the trial have told Newsnight that this is “fake information” 7
When is a heart attack not a heart attack? When it is measured by investigators in the clinical study – who have financial conflicts of interest.
Another problem is that, if you carry out a coronary artery revascularisation, there is a fifty per cent chance of triggering a heart attack. Usually pretty small and not clinically significant – but an MI, nonetheless. So, for each two additional revascularisations, you may get one more MI. Which further skews the statistics. [Not an issue in the FOURIER study where only the first event was counted].
Anyway, I hope you are getting the general message that a quintuple combined endpoint is, primarily, nonsense. Full of potential bias, particularly in an observational study. As the Lancet study was.
Finally, because I have run out of energy to spend another minute looking at this study, there is the issue of Lipoprotein(a). Otherwise known as Lp(a). This too forms part of the non-HDL cholesterol measurement.
Lp(a) and LDL are identical apart from the fact that Lp(a) has an additional protein attached to the side called apolipoprotein(a). This protein has a critical role in blood clotting and therefore Lp(a) can be viewed as a pro-coagulant agent – makes the blood clot bigger and more difficult to break down. Higher levels have long been linked to an increased risk of CVD.
Just to choose one quote from many thousands of studies about Lp(a): ‘Lipoprotein (a) and the risk of cardiovascular disease in the European Population: results from the BiomarCaRE consortium.’
‘Elevated Lp(a) was robustly associated with an increased risk for MCE (major cardiovascular events) and CVD in particular among individuals with diabetes. 8
Yes, you will have spotted the link with diabetes a.k.a. insulin resistance. So, a higher triglyceride level, added to raised Lp(a), further increases the risk of CVD.
So, with non-HDL cholesterol and Lp(a) we have another massive confounding factors built into the measurement. Again, I absolutely cannot disagree that raised Lp(a) increases the risk of CVD. I have written about it many, many times. Non-HDL cholesterol is a measure that contains Lp(a) within it…. You probably get the drift by now.
The whole paper, in my opinion, is complete nonsense. Assumptions, built on bias, built on a measure that has nothing much to do with LDL, or ‘bad’ cholesterol. Zoe Harcombe did a more forensic dissection of the paper. I like her ending:
‘The researchers assumed that a 50% reduction in non-HDL cholesterol could and would be achieved. The researchers assumed that a mathematical formula for risk reduction could be applied to that assumed 50% reduction in non-HDL cholesterol. The researchers’ assumed formula included the variable “number of years of treatment” and hence the formula produced a higher number, the earlier treatment started. The assumptions made it so.
The final paragraph of the paper stated: “However, since clinical trials investigating the benefit of lipid-lowering therapy in individuals younger than 45 years during a follow up of 30 years are not available, our study provides unique insights into the benefits of a potential early intervention in primary prevention.”
No, it doesn’t.’
Yet, there was no controlled clinical trial data to back this all up. There are only models and assumptions. Yet it made headlines around the world, as such stuff always does. Not only that, it made the wrong headlines.
As the BBC website stated: ‘The study, in The Lancet, is the most comprehensive yet to look at the long-term health risks of having too much “bad” cholesterol for decades’ Bad cholesterol is the bonkers, unscientific term that is used to describe LDL. This study did not look at ‘bad’ cholesterol… Scientific journalism at is finest.
My analysis. Crumple, throw, bin… forget.