Thursday, February 19, 2015

The Harcombe Meta-Analysis

The Meta-Analysis

A new meta-analysis regarding fat and coronary heart disease (CHD) was recently published by Harcombe, et al, titled: ‘Evidence from randomised controlled trials did not support the introduction of dietary fat guidelines in 1977 and 1983: a systematic review and meta-analysis’.  The title pretty much says it all.  This meta-analysis pooled the results from clinical trials that investigated the effects of reduced and/or modified fat* consumption on CHD, but which were also available at the time when the first US and UK dietary guidelines were introduced (1977 and 1983 respectively).  They concluded “…that dietary advice not merely needs review; it should not have been introduced”

The results for total mortality (RR=0.996) and CHD mortality (RR=0.989) can be seen in figure 2 and figure 3 respectively.  For those familiar with the trials/other meta-analyses (or who just read my previous blog post), the results of this meta-analysis shouldn’t be much of a surprise as previous meta-analyses have consistently reported no significant benefit for CHD mortality and total mortality*.  The meta-analysis also raises many good points regarding the limited evidence at the time of the US and UK dietary guidelines: only 2467 men, no women, no pure primary prevention studies and not testing the proposed guidelines.

* The meta-analysis included 5 fat modification trials (RCOT, LAVAT, MRCT, ODHS, SDHS) and only 1 low fat trial (Ball, et al).  They excluded the Finnish Mental Hospital Study (FMHS) and the Anti-Coronary Club for not being randomised)

** Except Mozaffarian, et al for CHD mortality due to their inclusion of FMHS and Skeaff & Miller for total mortality due to their incorrect interpretation of FMHS.

The Responses

As expected many people raised objections to the meta-analysis, most of which trivialised the results of the RCTs and took the position of arguing from the ‘totality of the evidence’, turning their attention to specific observational studies and the lipid hypothesis to support their position on SFA.

Unfortunately, many of the experts discussed the relationship between SFA, cholesterol and CHD as if anything that raises/lowers cholesterol will increase/decrease CHD.  While this assumption is correct some of the time, there are certainly exceptions:

  • CETP increases the total-C:HDL-C ratio, but torcetrapib, a CETP inhibitor, significantly increased CHD events and total mortality in clinical trials [8]
  • Replacing SFA with MUFA reduces the total-C:HDL-C ratio, but the Jakobsen meta-analysis of cohort studies found that doing so is associated with an increased risk of CHD events [5] (you can use the Jakobsen meta-analysis to support PUFA consumption, but not to vilify SFA)
  • With the exception of DART, in all of the main diet heart trials the experimental group lowered their cholesterol, but this didn’t translate into reduced CHD in the better controlled trials: near significant (P = 0.05-0.10) increase in major CHD events in RCOT; no difference in major CHD events and CHD mortality in MRCT; significant increase in CHD mortality in SDHS; and a 31% increase in CHD events among women in MCS 

Most of the evidence from observational studies they provided comes from ecological studies (both geographical and temporal), which is a poorer quality type of observational study that’s more susceptible to confounding variables.  However, the evidence from recent meta-analyses of observational studies (usually cohort studies) find that: (1) SFA intake is not associated with CHD [1] [2] [3] [4]; (2) PUFA may or may [5] [6] not be [1] [3] [4] inversely associated with CHD; and (3) meta-analyses that find PUFA is inversely associated with CHD do so regardless of whether SFA, carbohydrate, and probably also MUFA, is replaced with PUFA [5] [6].  The observational evidence of non-specific benefit for PUFA may be due to nuts (or at least what nuts replace) rather than vegetable oils [7] and is not that relevant to SFA consumption, particularly because Jakobsen found a minor increased risk in CHD when SFA was replaced with MUFA or carbohydrate [5].

Some of the objections related to dietary adherence, arguing essentially that (1) we don’t know what the participants were eating and (2) that it’s difficult to get people to eat differently.  The first point equally applies to observational studies, which are notorious for their inaccuracies.  The second point is true of most trials, but not these ones where compliance was really good (see table below), suggesting the ‘experts’ and other people commenting haven’t actually read the original papers.

Average Reported Intake
80g of olive/corn oil
58g olive oil and 64g corn oil
40% fat with 2/3 from vegetable oils
Adherence was approximately 80%
85g soybean oil
80g soybean oil
High soybean oil
28% of calories from soybean oil
SFA intake 10%, PUFA intake ≥15% of total calories
SFA intake 8.9%, PUFA intake 15.1%
Low Fat (Ball, et al)
40g fat
45g fat (control group was 110-130g)


  1. "Unfortunately, many of the experts discussed the relationship between SFA, cholesterol and CHD as if anything that raises/lowers cholesterol will increase/decrease CHD. While this assumption is correct some of the time, there are certainly exceptions"

    Since virtually all cholesterol lowering drugs (and statins in many populations) have failed to lower CV/CHD death and total mortality, they would often latch on to softer events to claim cholesterol lowering works.

    In addition to CETP inhibitors you mentioned, hormone treatment also has no effect on events despite reducing LDL and increasing HDL (HERS,WHI), and varespladib increases heart attacks despite reducing LDL (VISTA-16).
    ARISE was interesting where succinobucol increased LDL but a secondary endpoint was decreased driven by non-fatal MI and stroke, and the AleCardio trial also where LDL increased but non-fatal events did not increase. Then there's aspirin and fibrates which can reduce non-fatal MI to a similar extent as statins despite little to no effect on LDL.

    Collectively, this to me seriously undermines the idea that LDL per se reduces non-fatal events.

    1. Thanks for the comment Zahc. I figured there would be other examples, but wasn't familiar with those studies.

      Is this your slideshare account?

  2. "Is this your slideshare account?"

    Yep, that's my account.

    1. Thanks! It was my first time creating slides so the presentations are very simple and not as fancy or easy on the eyes as I would like it. I'll probably redo them when I get the motivation.

  3. This comment has been removed by the author.

  4. An excellent and concise defense of Harcombe et al.
    The new line seems to be "we have tested the Med diet over and over because that's what we've been spending our money and time doing, rather than testing other alternatives, and it's better than the SAD and the old guidelines, so now we want everyone to eat our PC lite travesty of the Mediterranean diet, because it has some evidence behind it".
    Which is the man looking for his car keys under the lamppost on a darkened street, because "that's where the light is", all over again, and because it allows them to cling to their wretched lipid hypothesis for a few years longer, hopefully till they retire.

    1. Thanks

      I agree. I'm planning to discuss 'eating patterns' like the Med diet in the next post. Some of the responses to Harcombe, were essentially 'food not nutrients', but then this transitioned to discussing eating patterns, which can be useful, but also often both dogmatic and susceptible to self-fulfilling prophecies in research