Search This Blog

Tuesday, 26 April 2016

Mediterranean diet score in stable heart disease, and, more thoughts on Ramsden et al.

This news article that made the rounds yesterday demonstrates how confirmation bias keeps the diet-heart hypothesis afloat.

Healthy eating key to heart disease

After 3.7 years' follow-up, a heart attack, stroke or death - termed a major adverse cardiac event - had occurred in 10.1 per cent of the participants. Such events occurred in 7.3 per cent of the people in the highest Mediterranean-diet bracket, 10.5 per cent in the next bracket down and 10.8 per cent in those who ate smaller quantities of the healthier foods.

"After adjusting for other factors that might affect the results we found that every one unit increase in the Mediterranean diet score was associated with a 7 per cent reduction in the risk of heart attacks, strokes or death from cardiovascular or other causes in patients with existing heart disease," Mr Stewart said.

The elements of the Mediterranean Diet Score can be found in the full paper and supplementary tables. It turns out foods like dairy, eggs, and tofu were also found to be protective but weren't included in the Med diet score; whereas lower meat intake wasn't protective but was included; and wholegrains weren't protective, but were included. Go figure.

Auckland University heart disease researcher, Professor Rod Jackson noted that the authors did not report on saturated fat consumption or fat consumption at all because they stated it had not been recorded reliably.

"However, the findings are quite consistent with the standard diet-heart hypothesis. A Mediterranean diet is low in saturated fat and was associated with lower risk of CHD [coronary heart disease].

"The Western diet score was based on consumption of refined carbohydrates, sweets and desserts, sugared drinks and deep-fried food. None of these foods except deep-fried foods, and only if the fat was saturated, are associated with CHD. They are associated with overweight/obesity and diabetes but the pro-fat lobby have always confused the issue by wrongly lumping obesity and diabetes with CHD.

" ... they are very different conditions and are trending in opposite directions."

This association of saturated fat and CHD seems to be a bit imaginary, but what is the explanation for junk foods having no association with CHD?
Well firstly, this was a very crude data collection effort, even by diet epidemiology standards. Many foods either weren't measured or were tucked away in the nearest category.
Secondly, because it depends on "diet scores" to aggregate non-significant associations, the non-significant association between deep fried food (the biggest source of omega-6 PUFA here) and CHD has been overlooked.
Thirdly, if you're going to eat less junk food, it is possible to replace it with "healthy" foods that aren't associated with benefit here. Namely wholegrain products, which are the densest calorie source in the Med diet score category. Imagine someone eating fewer biscuits and replacing that with wholemeal muffins. You could replace sugar-sweetened soft drink with fruit juice too - I'm not sure if that fits anywhere in these scores.Fourthly, this survey took place during another massive failed drug trial. A drug supposed to protect those with stable heart disease did diddly-squat. This data was salvaged from the wreckage. That's not a confounder that I can see, but I do find it interesting that this bit of context didn't make the papers.
Fifthly, there are some huge differences in smokers, BMI, education and income between the higher vs lower Med diet score groups. If these are associated with junk food intake and you're correcting for them, then you're correcting for a large association in the hope of leaving a smaller one intact. It's a wonder, with all its flaws, that this study arrived at any result resembling a plausible reality. But it did, in my opinion.


I wrote a letter to the Herald about this yesterday, but it wasn't published today, so here it is.

Dear sir,

     The standard diet-heart hypothesis says that saturated fat in the diet causes heart disease by raising LDL cholesterol. This notion has taken a bit of a drubbing recently, so it is understandable that Professor Rod Jackson interprets yesterday’s study, about a higher Mediterranean diet score protecting against heart attacks, strokes, and deaths in those with stable heart disease, in its favour. 
     However, this ignores two findings from this study; firstly, that the mean LDL cholesterol level was not significantly different (2.3 vs 2.2 mmol/L) across the “Mediterranean diet score” categories, and secondly, that the two traditional food sources of saturated fat measured, meat and dairy, were not associated with increased risk; in fact dairy was associated with reduced risk.
     Although wholegrains were included in the Mediterranean diet score, they were not associated with benefit by themselves, and it would, for instance, be possible from this data to show that a “Paleolithic diet score” of eggs, meat, fruit, vegetables and fish, but no grains, was associated with as much benefit as the Mediterranean diet score. Furthermore, the two Mediterranean foods which the earlier PrediMed intervention identified as being most beneficial, olive oil and nuts, were not even measured in the new study.
     The one reliable finding from this study is that, the more minimally processed, nutrient-dense foods you include in your diet, the healthier it is. Maybe this should be the new diet-heart hypothesis until a better one comes along.

yours, etc

Ramsden et al has been the gift that keeps on giving. I had some more thoughts about the kind of problems a high omega-6 intervention might run into which I appended to Steven Hamley's analysis of the MCE study here. The FADS2 polymorphism study I refer to is this one.

I notice that those defending omega 6 interventions in the BMJ rapid responses have cited the Farvid et al meta-analysis of observational studies. However Farvid et al did not control for omega 3 fatty acids at all and this is quite clearly stated, so cannot be cited to refute any Ramsden et al meta-analysis.
Further, this is a bizarre procedure. If experiments don't confirm observations from population studies, you can't just cite another population study to refute the experiments. Prof Brunner does this in the rapid responses using quite a minor observational study that used "dietary pattern" analysis, with a healthy "dietary pattern" including margarine, to refute the experiments. If this is the procedure of epidemiologists, no wonder we are where we are with this zombie hypothesis.

Edit: I dug up the Whitehall II study that Prof Brunner cited, and which he co-authored.

Increased CHD risk (hazard ratio for top quartile: 2.01, 95%CI 1.41-2.85, adjusted for age, sex, ethnicity and energy misreporting) was observed with a diet characterised by high consumption of white bread, fried potatoes, sugar in tea and coffee, burgers and sausages, soft drinks, and low consumption of French dressing and vegetables."
This was dietary pattern 1.
A higher score on dietary pattern 1 was associated with higher total cholesterol, lower HDL cholesterol and higher triglycerides. Dietary pattern 2 was characterised by higher consumption of red meat, cabbage, brussels sprouts and cauliflower, and lower consumption of wholemeal bread, jam, marmalade and honey, tofu and soy, buns, cakes, pastries, fruit pies and polyunsaturated margarine.
A higher score on dietary pattern 2 was associated with higher total cholesterol and higher triglycerides. 
Dietary pattern 2 showed a significant linear trend across quartiles with a higher dietary pattern score also associated with increased risk of CHD (Model 3, adjusted for age, sex and energy misreporting, ethnicity, employment grade, smoking, alcohol and physical activity, p less than 0.0001) however this relationship was no longer significant after further adjustment for BMI and blood pressure.(As far as I can see, the pattern 2 trend was never very significant and the dose-response of both patterns is all over the place. There are 6 possible statistical models for each pattern, and none in the table given reads as having anything like a 0.0001 p value).

The paper states that French dressing (21% PUFA according to wikipedia) had no independent association with CHD, and gives no information about independent associations with polyunsaturated margarine.

No comments: