Sunday, February 23, 2014

The Diet Heart Hypothesis: Part 3

The Inclusion/Exclusion of Trials 

How is it that that three different meta-analyses arrive at three different conclusions?  Simple, they include/exclude different trials.  Let’s have a look at the meta-analyses and the reasons for including/excluding certain trials (I’ll just discuss the trials where there’s disagreement). 

Hooper
Mozaffarian
Ramsden*
Mine**
Rose Corn Oil Trial
X
X (6)
X
Anti-Coronary Club
Los Angeles Veterans
X
X
X (6+3)
Medical Research Council Trial
X
X
X (6+3)
X
Oslo Diet-Heart Study
X
X
X (6+3)
St Vincent’s Hospital Study
Sydney Diet Heart Study
X
X (6)
X
Finnish Mental Hospital
X
Minnesota Coronary Survey
X
X
X (6)
X
Diet and Reinfarction Trial
X
X
St Thomas Atherosclerosis Regression Study
X
X
X (6+3)
* ‘6’ Refers to a trial where SFA was replaced by omega 6, ‘6+3’ refers to a trial where SFA was replaced by both omega 6 and omega 3
** Hypothetically 

The Rose Corn Oil Trial 

Rose was both randomised and controlled, but Mozaffarian, et al didn’t include it because of “multiple interventions”.  I agree, it did have multiple interventions as the oil groups were encouraged to eat less junk food (fried food, pastry, ice-cream and cakes), but as I pointed out last post if we were to be that strict then unfortunately none of the trials would be eligible.  But they probably didn’t mean that because Mozaffarian, et al included trials that were obviously and deliberately multifactorial (like Oslo and STARS).  They were probably referring to the two oil groups, which isn’t really that relevant because the olive oil group served as a semi-control for corn oil supplementation.  While not exactly the same, Mozaffarian, et al included DART and STARS, which also had ‘multiple interventions’ 

The Los Angeles Veterans Administration Trial 

LA Vets was randomised and well controlled except for the reheating of butter (and consequently vitamin E deficiency) and increase in TFA in the control group, which is why I wouldn’t include it.  However, it was one of the better trials being double blinded and the longest (8 years) and all the meta-analyses included it 

The Oslo Diet Heart Study 

Oslo was randomised, but not well controlled as there were many other dietary differences in favour of the experimental group that were most likely deliberate (more whole foods, LCO3, CLO and vitamin D and less hydrogenated fish oil, refined grains and sugar).  Hooper, et al seems aware of this and the information I got about the diet in Oslo came from Ramsden, et al (so they’re aware of it as well), but they both included it anyway and Ramsden, et al classified it as an omega 3 + omega 6 trial, even though there are many more differences besides the omega 3. 

The Sydney Diet Heart Study 

Sydney was both randomised and controlled.  It was excluded by Mozaffarian, et al because “non-CHD endpoint”, but this wasn’t an issue for Hooper, et al or Ramsden, et al.  The original data did only mention total mortality, but 60 of the 67 (89.6%) deaths were due to CHD.  Easy solution: include it in your analysis for total mortality but not CHD events/mortality 

The Finnish Mental Hospital Study 

FMHS wasn’t randomised or well controlled.  It was excluded by Hooper, et al and because it wasn’t randomised and excluded by Ramsden, et al because it wasn’t randomised and the difference in cardiotoxic medication use between the groups.  Mozaffarian, et al included it and said it was “cluster-randomised”.  Regardless of whether that counts (see Stephan Guyunet’s take), I wouldn’t include it because of all the confounders that on balance were in favour of the experimental group (see the table at the end of this post) 

The Diet and Reinfarction Trial 

DART was randomised and fairly well controlled except for some differences in diet that didn’t seem deliberate (the experimental group increased fruits, vegetables and fish and decreased biscuits and cakes).  Hooper, et al included it, but as a low fat + fat modification trial, rather than a fat modification only trial because the ‘fat advice’ group was also told to reduce their total fat consumption, although the difference between the groups was only 3.2% of total calories.  Ramsden, et al didn’t include it because we don’t know the omega 3 and omega 6 intake in DART, which is understandable since the aim of their meta-analysis was to compare SFA vs. omega 6 trials with SFA vs. omega 6+3 trials 

The St Thomas Atherosclerosis Regression Study 

STARS was randomised, but not well controlled as there were many other dietary differences in favour of the experimental group that were deliberate (weight loss, less trans fat, more LCO3, less baked goods and potentially more whole foods).  Also, the increase in omega 6 in STARS was quite low (difference of 1.6% of total calories).  Like DART, Hooper, et al included it as a low fat + fat modification trial and this time there was a significant difference in fat intake between the groups (27% vs. 37%).  As there were more differences than that, Hooper, et al they flagged it as a trial that wasn’t ‘free from systemic bias in care’ and ‘free from dietary differences other than fat.  Ramsden, et al included it as a omega 6+3 trial even though they acknowledged the small difference in omega 6 intake and were critical of Mozaffarian, et al for including it. 

The Meta-Analyses 

Hooper, et al 

I mostly agree with the study inclusion/exclusion in the Hooper, et al meta-analysis.  I disagree with their inclusion of Oslo because it was so multifactorial (which they were aware of) and deliberately so.  I understand the inclusion of LA Vets, since the trial was high quality (double-blind and long) and differences (vitamin E and TFA) weren’t so obvious.  While DART was intended to be a low fat + fat modification trial (which is why Hooper, et al classified it that way) there was only a slight difference in fat consumption (3.2% of total calories) and actually many of the other trials had an accidental change in fat consumption that was similar in magnitude, but not a big deal 

* I’m also grateful to Hooper, et al as their meta-analysis was where I got all (except one) of the studies for my ‘studies associated with the trial’ section 

Mozaffarian, et al 

I have a few disagreements with the Mozaffarian, et al meta-analysis.  In addition to including Oslo, they also included two other poorly controlled trials that supported the DHH (FMHS and STARS) and FMHS wasn’t randomised, while excluding two unfavourable trials (Rose and Sydney) for poor reasons.  The combination of including three poorly controlled favourable trials while excluding two better controlled unfavourable trials for poor reasons brings up the possibility of bias.  Despite their selection, they still weren’t able to find a reduction in total mortality. 

* The three authors are all epidemiologists at the Harvard School for Public Health.  Are you surprised? 

Ramsden, et al 

The trials included in Ramsden, et al are almost identical to those in Hooper, et al, except the inclusion.  Like Hooper, et al, I can understand why Ramsden, et al didn’t include DART but can’t understand why they included Oslo, since they were aware of all the differences (and informed me of them).  They also included STARS as an omega 6+3 trial even though there are many other differences between the groups.  Ramsden, et al has been a very informative meta-analysis, but I think it misses the mark slightly by dividing the trials up by SFA vs. omega 6 trials and SFA vs. omega 6+3 trials, as many of the ‘SFA vs. omega 6+3 trials’ are very multifactorial and should be considered as such

No comments:

Post a Comment