But what is the mystery number which sentences you to cardiac death? Obviously the original Framingham study did not use a random scale. But the scale used in the study was highly non linear. Here is the same graph with the real numbers added.
Once you have realised quite how unlinear the scale is I would just like to draw your attention to that number on the bottom right written in RED. No, it's not a typo. That really is 1124mg/dl. Okaaaaaaaaay. Hands up if you think a TC of 1124mg/dl is part of a normal distribution of cholesterol values in a "normal" population. Are there any conditions which elevate cholesterol and increase the risk of heart disease? Answers on a postcard to any endocrinology department you happen to know the address of.
Anyway, next thing to do is to linearise the scale. I've been kind and left the upper data point in the middle, not placed it at the upper end, of the range quoted:
Now here is my main cheat, and I admit it's a cheat. There must be a population towards the upper end of that last data point who have medical problems to give them an increased risk of heart disease. Cushings Syndrome and hypothyroidism are two for starters. So I would argue that it is only fair to the representatives of this subpopluation that the risk scale is extended up for their benefit. 100 "events" per 1000 in this group seems possible, hence the extended scale:
So let's go back and look at the initial random points and leave the top cholesterol data point where it belongs, about a yard to the right of your screen:
Ah, that's better. Some semblance of honesty now. But again, not quite as convincing as the first graph. BTW did anyone notice that the left hand scale was events, not deaths?
Did the Framingham investiators look at deaths? Hahahahahahahahahahahaha bonk. Sorry, that's me laughing my head off. Of course they did. There is no association between elevated cholesterol and increased cardiac deaths, but the trend is that high cholesterol is protective. Luckily for the Framingham researchers they were underpowered to detect this. A whiff of the low powered studies we see nowadays.
Now let's just look at the MRFIT screenees. These are the many, many people who were looked through before cherry picking the victims for the MRFIT intervention trial which, incidentally, killed more people in the intervention group than died in the the usual care group. Luckily (again) this did not reach statistical significance, though it may have been of some biological significance to those extra people who ended up dead in the intervention group.
The original study had non linear group sizes (like Framingham) to specifically, oh, I mean accidentally, obscure the effects on all cause mortality in the low cholesterol individuals. The graphs above are taken from a later sub-analysis by a rather more honest and objective lead researcher. Such people do exist in cardiology.
The other hysterical aspect of these graphs is that the original data presented from the MRFIT screenees wasn't, wait for it......
It wasn't corrected for smoking!
Yet another re-analysis shows a marked association between TC and cardiac deaths, but only in smokers. About what you would expect if LDL cholesterol was doing its best to repair smoking induced damage and failing.
The association is still present in non smokers and is statistically significant but so much weaker that the biological significance is highly debatable. It probably represents sucrose intake.
So in summary, the original MRFIT screenees study (presented as the clincher for cholesterol causing heart disease) obscured the scary aspects of low TC and, err, forgot to correct for smoking!
Don't you love the foundations of modern cardiology! Can't sum up MRFIT better than Dr Werko.
For a breath of fresh air it's worth going to Norfolk in the UK.
This innocent little graph is plotted from the EPIC Norfolk data. Now I'd hate to suggest that being hyperglycaemic has anything to do with heart disease, but you can read the graph as well as I can. Association not causation. And of course we know that eating fat causes hyperglycaemia, just ask any diabetologist. Shrug.
The weird thing about UK researchers is that they give you the raw data, it's not a matter of a little table of regression coefficients corrected for this, that and the other. You can read the results tables and plot the graphs. They even give you raw smoking data. And of course you get both TC and that evil killer, the LDL level. Let's plot them on the same graph as the HbA1c vs relative risk of cardiac events.
Personally I'd be looking to minimise my HbA1c rather than my LDL-C.
But then I would say that.
Peter
