Thursday, December 2, 2010

Diet-Heart Controlled Trials: a New Literature Review

Many controlled studies have measured the cardiovascular effects of replacing animal ("saturated") fats with seed oils (predominantly the omega-6 polyunsaturated fat linoleic acid) in humans. A number of these studies recorded heart attacks and total mortality during the following 1-8 years. Several investigators have done meta-analyses (literature reviews) to try to tease out overall conclusions from these studies.

I'm pleased to point out a new meta-analysis of these controlled trials by Dr. Christopher Ramsden and colleagues (1). This paper finally cleans up the mess that previous meta-analyses have made of the diet-heart literature. One recent paper in particular by Dr. Dariush Mozaffarian and colleagues concluded that overall, the controlled trials show that replacing animal fat with linoleic acid (LA)-rich seed oils reduces heart attack risk (2). I disagreed strongly with their conclusion because I felt their methods were faulty (3).

Dr. Ramsden and colleagues pointed out several fundamental flaws in the review paper by Dr. Mozaffarian and colleagues, as well as in the prevailing interpretation of these studies in the scientific literature in general. These overlap with the concerns that I voiced in my post (4):
  1. Omission of unfavorable studies, including the Rose corn oil trial and the Sydney diet-heart trial.
  2. Inclusion of weak trials with major confounding variables, such as the Finnish mental hospital trial.
  3. Failure to distinguish between omega-6 and omega-3 fatty acids.
  4. Failure to acknowledge that seed oils often replaced large quantities of industrial trans fats in addition to animal fat in these trials.
Dr. Ramsden and colleagues accounted for all of these factors in their analysis, which has never been done before. They chose inclusion criteria* that made sense, and stuck with them. In addition, they did an impressive amount of historical work, digging up old unpublished data from these trials to determine the exact composition of the control and experimental diets. The paper is published in the British Journal of Nutrition, an excellent journal, and overall is written in a scientific and professional manner.

What did they find?
  • Interventions that replaced animal and trans fat with seed oils that were rich in LA but low in omega-3 caused a non-significant trend toward increased heart attacks (13% increase) and overall mortality.
  • Interventions that replaced animal and trans fat with a combination of LA and omega-3 fats significantly reduced heart attacks (by 22%). The numbers for total mortality followed a similar trend.
In other words, LA-rich seed oils do not prevent heart attacks (and may actually promote them), but correcting an omega-3 deficiency and reducing industrial trans fat intake may be protective. This is similar to what I've been saying for a while now, based on my own interpretation of the same studies and others. However, Dr. Ramsden and colleagues have taken the idea to a new level by their thorough and sophisticated detective work and analysis. For example, I didn't realize that in virtually all of these controlled trials, the intervention group reduced its trans fat intake substantially in addition to reducing animal fat. From the paper:
...experimental diets replaced common ‘hard’ margarines, industrial shortenings and other sources of [trans fat] in all seven of the [controlled trials] included in the meta-analysis by Mozaffarian et al. The mean estimated [trans fat] content of the seven control diets was 3·0 [% of calories] (range 1·5–9·6 [%]).
...the displacement of [trans fat], rather than the substitution of mixed n-3/n-6 [polyunsaturated fat] for [saturated fat], may account for some or all of the 22% reduction in non-fatal [heart attacks and heart attack] death in our meta-analysis. By contrast, the increased [heart attack] risks from n-6 specific [polyunsaturated fat] diets in our meta-analysis may be underestimated as n-6 [polyunsaturated fat] also replaced substantial quantities of [trans fat] (Table 3). The consistent trends towards increased [heart attack] risk of n-6 specific [polyunsaturated fat] diets may have become significant if the n-6 [polyunsaturated fat] replaced only [saturated fat], instead of a combination of [saturated fat] and [trans fat].
In other words, it looks like trans fat is probably the issue, not animal fat, but these trials replaced both simultaneously so we can't know for sure. I will note here that trans fat does not generally promote atherosclerosis (thickening and hardening of arteries) in animal models, so if it does truly increase heart attack risk as many studies suggest, it's probably through a mechanism that is independent of atherosclerosis.

The article also contains an excellent discussion of the Finnish mental hospital trial (5, 6) and why it was excluded from the meta-analysis, in which Dr. Ramsden and colleagues point out major design flaws, some of which I was not aware of. For example, trans fat intake was on average 13 times higher in the control groups than in the experimental groups. In addition, one of the control groups received more than twice as much of the antipsychotic drug thioridazine, which is known to be highly toxic to the heart, as any other group. Ouch. I'm glad to see this study finally discussed in an open and honest manner. I discussed my own problems with the Finnish trial in an earlier post (7).

I was also glad to see an open discussion of the Oslo Diet-heart study (8), in which diet changes led to a reduction in heart attack risk over five years. Dr. Mozaffarian and colleagues included it in their analysis as if it were a controlled trial in which animal fat was replaced by seed oils only. In reality, the investigators changed many variables at once, which I had also pointed out in my critique of Dr. Mozaffarian's meta-analysis (9). Here's what Dr. Ramsden and colleagues had to say about it:
First, experimental dieters were instructed to substitute fish, shellfish and ‘whale beef’ for meats and eggs, and were actually supplied with ‘considerable quantities of Norwegian sardines canned in cod liver oil, which proved to be popular as a bread spread’(32)... Second, the experimental group consumed massive amounts of soybean oil, which provided large quantities of both LA (15·6 en %) and ALA (2·7 en %). ALA consumption was about 4·5 times average US intake(42), or about twelve typical flax oil pills (1 g pill ¼ 560 mg ALA) per d. In addition, the fish and cod liver oil consumption provided Oslo (598N latitude) dieters with 610 IU (15·25 mg) of daily vitamin D3, recently linked to lower blood pressure, plaque stabilisation, and reduced [heart attack risk] (64). Furthermore, experimental dieters were encouraged to eat more nuts, fruits, and vegetables; to limit animal fats; and to restrict their intake of refined grains and sugar.
trans fat intake was also reduced substantially by excluding margarine in the experimental group. Other review papers have used this trial as a justification to replace animal fat with seed oils. Hmm... The only reason they get away with this is because the trial was published in 1966 and almost no one today has actually read it.

One criticism I have of Dr. Ramsden's paper is that they used the Oslo trial in their analysis, despite the major limitation described above. However, they were extremely open about it and discussed the problem in detail. Furthermore, the overall result would have been essentially the same even if they had excluded the Oslo trial from the analysis.

Overall, the paper is an excellent addition to the literature, and I hope it will bring a new level of sophistication to the dialogue on dietary prevention of cardiovascular disease. In the meantime, brace yourselves for an avalanche of criticism from the seed oil brigade.


* Guidelines that determine which studies to include in the analysis. For example, you want to exclude any study that wasn't randomized, because it will not be interpretable from a statistical standpoint. You also want to exclude trials where major variables differ between groups besides the specific variable you're trying to test. The Finnish mental hospital trial fails by both criteria.