Monday, April 25, 2016

Ramsden et al Recovers Data from the Minnesota Coronary Survey

Ramsden et al have previously written a meta-analysis [1] and a review [2] critical of the mainstream view that replacing SFA with mainly n-6 PUFA would reduce the risk of coronary heart disease (as well as other research on n-6 PUFA).  They later recovered some missing data from the Sydney Diet Heart Study (SDHS) [3] and more recently did the same for the Minnesota Coronary Survey (MCS) (they call it the Minnesota Coronary Experiment (MCE)) [4].  SDHS and MCS are both unfavourable trials for the diet heart hypothesis [5].  This paper was widely publicised in the media because those guys love controversy

Ramsden et al described MCS as “perhaps the most rigorously executed dietary trial of cholesterol lowering by replacement of saturated fat with vegetable oil rich in linoleic acid. The MCE is the only such randomized controlled trial to complete postmortem assessment of coronary, aortic, and cerebrovascular atherosclerosis grade and infarct status and the only one to test the clinical effects of increasing linoleic acid in large prespecified subgroups of women and older adults.” [4]

It’s this autopsy and subgroup data that Ramsden et al’s paper has recovered and is largely publishing here (see table 1)

It’s previously been reported that compared to the control group, total mortality was equal in men (RR = 0.99) and slightly higher in women (RR = 1.16) (average RR = 1.08).  The newly recovered data show this slight increase in mortality to be due to people aged at least 65.  What’s interesting about the mortality figures is that the separation in the ≥ 65 age group starts to begin at around 600 days (figure 5)

This is similar to observation Chris Masterjohn made about the Los Angeles Veterans Administration Trial (LAVAT), that cancer mortality began to be higher in the experimental group at 2 years and non-CVD mortality at 4-7 years [6].  Forget about the defence against this paper that the trial was too short to demonstrate the benefits of n-6 PUFA – cholesterol levels change within a week or two and so if that’s the primary mechanism then the benefits should start right away.  On the other hand, if the harms of excess n-6 PUFA are somewhat mediated by changes in tissue lipid composition, then you would expect the harms of excess n-6 PUFA to take a while to develop

This is followed by looking at the relationship between changes in cholesterol during the trial with total mortality (figure 6 and table 4).  They found that a reduction in cholesterol during the trial was associated with higher mortality, in both groups.  This is really an observational study like result within an RCT and can’t be used to establish causality as you can’t determine whether diseases and their treatment lowers cholesterol or whether lowering cholesterol makes one more susceptible to other diseases.  It is consistent with reviews of observational studies that found the relationship between total-C and total mortality to be a U-shaped curve [7], and another that found among older people this U-shaped curve was shifted to the right (lowest mortality at higher total-C) or an inverse relationship [8]

The autopsies data found that: “41% (31/76) of participants in the intervention group had at least one myocardial infarct, whereas only 22% (16/73) of participants in the control group did (incidence rate ratio 1.90, 95% confidence interval 1.01 to 3.72; P=0.035). Also, participants in the intervention group did not have less coronary atherosclerosis or aortic atherosclerosis (table 5).”  However, as Ramsden, et al point out “These findings should be interpreted with caution because of partial recovery of autopsy files. There was no association between serum cholesterol and myocardial infarcts, coronary atherosclerosis, or aortic atherosclerosis in covariate adjusted models (table G in appendix)”.

Ramsden et al’s paper isn’t earthshattering but is certainly nice to have.  Reading this study brought to the front of mind a few things I’m aware of but haven’t emphasised much yet:

·         The diets used in MCS were based on diets in the National Diet Heart Study.  The experimental diet in MCS was based on diet BC and the control diet In MCS was based on diet D.  NDHS aimed for most of the fat to come from specially formulated food products (filled meats, dairy, etc).  The added fat in the filled foods came from vegetable oils for the experimental diets such as BC and “either animal fat or hydrogenated shortening” for diet D.  There’s more detail on the diets in [4].  When you consider this, even a neutral result in the trial would be unfavourable for the diet heart hypothesis

·         MCS finished in 1973 but the manuscript was published in 1989 – 16 years later.  There are three abstracts in a supplement of Circulation in 1975 as the trial was also discussed in an American Heart Association conference around that time*.  But still this could be argued to be borderline publication bias**.  I also doubt that the publication of MCS would have changed much in the first dietary guidelines.  Even after MCS and DART (a neutral trial) were published people who support the diet heart hypothesis still find ways to do so and to rationalise away unfavourable results (you only need to look at the rapid responses and the media for this)

·         On a somewhat related note, the favourable diet heart trials were the ones with the most publications (ODHS (4 + monograph), LAVAT (12 + monograph), FMHS (6), HDAT (4) and STARS (9)), while the unfavourable trials had the least (RCOT (1), SDHS (2) and MCS (1))

* The 3 abstracts are difficult/impossible to find online but I was able to find them at Melbourne University’s library.  They give very little additional information unless you’re totally obsessive, because they report a different number of participants (Abstracts 9449, Manuscript 9057, Ramsden et al 9570) and the number of participants who had a major CVD event (CVD M,P) whereas the manuscript reported the number of major CHD events (CHD M,E)

** Funnel plots are used in meta-analyses to look for the probability of publication bias, where an asymmetric funnel plot suggests publication bias (here are some really bad examples [9] [10]).  I noticed some asymmetry in my funnel plots of the diet heart trials for major CHD events (top left) and total CHD events (top right), but not CHD mortality (bottom left) or total mortality (bottom right).  This is limited by the small number of studies (N = 11) and may simply reflect methodological differences rather than publication bias.  But it makes you wonder


I would also recommend reading George Henderson's rapid responses and Chris Masterjohn's post

3 comments:

  1. Thanks Steve,

    This is a very interesting paper, despite being inconclusive and same-old same-old to us, for a number of reasons.
    One is the mention in the paper of Ramsden et al's RCT of reduced-linoleic acid diets for headache.

    http://www.ncbi.nlm.nih.gov/pubmed/23886520

    and the later QOL sub-group analysis

    http://www.ncbi.nlm.nih.gov/pubmed/25958314

    Most people who went in to bat for or against the diet-heart hypothesis probably won't have got this far, but consider this result in terms of the "vegetarian diet will kill you" study, which found that the FADS2 polymorphism that lowered tolerance to high-omega 6 diets was distributed thus:
    incidence at present is 70% in South Asians, 53% in Africans, 29% in East Asians, and 17% in Europeans.

    http://mbe.oxfordjournals.org/content/early/2016/03/09/molbev.msw049.abstract

    Then consider the results of this study led by Daisy Zamora, the lead epidemiologist with Ramsden et al -
    greater compliance with US dietary guidelines is associated with more diabetes and higher BMI in US blacks but merely has no benefit in whites.

    "Among blacks, higher DQI was associated with higher baseline BMI, but the opposite relation was seen in whites (Supplementary Table 1). We found race (but not sex) to be an effect modifier of the association between DQI score and diabetes risk (Table 1). In Cox models adjusted for lifestyle and sociodemographic characteristics, there was no significant association between DQI score and diabetes risk in whites. However, blacks in the third (vs. lowest) DQI quartile had 49% higher risk of developing diabetes. This association was no longer statistically significant after further adjusting for baseline BMI."
    "Participants in the highest (vs. lowest) DQI quartile had significantly less increase in blood pressure (systolic and diastolic) and greater increase in HDL cholesterol (Supplementary Table 2). Among blacks, higher DQI scores were associated with greater increase in insulin resistance, even after adjusting models for initial BMI (P for trend <0.01)."

    http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3114488/

    This, and also the high susceptibility of Asian and Polynesian (derived from SE Asia) populations to these adverse effects, point to the FADS2 polymorphism being a potential mediator of these effects.
    If this, or anything like it, is true, that kind of scuttles the idea of one-size-fits-all recommendations around fat type in diets.

    ReplyDelete
    Replies
    1. Thanks George, those are some very interesting connections. Could be worthwhile to follow up on that Zamora study to try to tease apart which factors in the diet score are more likely to be responsible

      Ethnicity was rarely/never reported in the diet heart trials, though I think we can reasonably assume the vast majority were white (given history and locations (US, UK, Europe, Australia)). I wonder what would happen if someone decided to repeat these trials with Asians, Africans, etc as well...may not end so well

      Delete
    2. Yes, I think it's likely that adverse events would be seen more quickly.
      The connection with the restricted-omega 6 RCT I didn't make clearly - that research shows that some people are affected by higher n-6 in a way easy to show in a shorter-term trial.
      Then the other research gives a clue about who these people might be, and their distribution in various populations.

      Delete