Sunday, March 30, 2014

Low Fat and CHD

I originally didn’t think about writing on low fat and CHD because I didn’t see the point.  After all, throughout the literature and the dietary guidelines there was no suggestion that low fat diets would reduce the risk of CHD.  But then I don’t subscribe to pop science, where calories make you fat, protein damages your kidneys, fat clogs your arteries, carbs give you diabetes and everything causes cancer.  For most people fat (of any kind) still remains the villain in CHD. 

Remember that the three main lines of evidence against saturated fat were: 

1.      Observational studies
2.      Clinical trials measuring blood lipids
3.      Clinical trials measuring end-points (events and mortality) 

I’m going to briefly go through each of these as they relate to low fat and CHD 

Observational Studies 

The Australian National Heart Foundation’s ‘Summary of evidence. Dietary fats and dietary cholesterol for cardiovascular health’ (2009) cites a review by Mozaffarian as evidence for “there is no direct relationship between total fat intake and the incidence of CHD”. 

The Australian Heart Foundation later use a meta-analysis of cohort studies by Jakobson, et al as evidence against saturated fat.  The Jakobson, et al meta-analysis found that replacing SFA with PUFA was associated with less CHD events and death, but replacing SFA with MUFA or carbohydrate had no effect* [1]. 

Another meta-analysis of cohort studies by Skeaff & Miller found that intake of total fat wasn’t significantly associated with CHD events or CHD mortality [2]. 

And finally, there’s also the well-known meta-analysis of cohort studies by Siri-Tarino, et al, which found “no significant evidence for concluding that dietary saturated fat is associated with an increased risk of CHD or CVD” [3] 

* So if you’re going to debate the National Heart Foundation I don’t think you should spend much time arguing against low fat diets, because they will simply say that they don’t recommend a low fat diet, which is exactly what they did on Catalyst.  I would rather argue that (1) it’s unwise to make recommendations from observational studies and clinical trials measuring blood lipids; and (2) the better controlled fat modification trials don’t support the DHH and suggest some harm.  If you want to argue that they are recommending a low fat diet (because it seems most people challenging mainstream advice are pro-low carb), please define what ‘low fat’ means (I’ll be providing my take in a future post), otherwise it’s going to be one of those pointless ‘yes you are, no I’m not’ type of debate 

Clinical Trials Measuring Blood Lipids 

Two meta-analyses of clinical trials found that replacing fat with carbohydrate increases the total cholesterol to HDL-C ratio* [4] [5], which is perhaps the best CHD risk factor based on blood lipids [6]
* Only palmitic acid and artificial TFA increased the ratio, but the fats in food are comprised of many different fatty acids so even fat sources high in palmitic acid and/or artificial TFA like butter, shortening and palm oil reduce the total cholesterol to HDL-C ratio relative to carbohydrate 

** They also found replacing fat with carbohydrate increased triglycerides [4] [5] 

Clinical Trials Measuring End-Points (Events and Mortality) 

But we shouldn’t dismiss low fat diets based on findings of observational studies and their effect on blood lipids.  Just as it’s wrong to use observational studies to ‘prove’ there is an effect, it’s also wrong to use them to ‘prove’ there isn’t one either 

To my knowledge, there are two main meta-analyses looking at low fat (fat reduction) and CHD: the same Hooper, et al meta-analysis and the meta-analysis by Skeaff & Miller (who also did one for the fat modification trials*).  The results of the meta-analyses are in the table below: 

CHD Events
CHD Mortality
Total Mortality
Hooper, et al
0.97 (0.87, 1.08)
0.96 (0.82, 1.13)
0.97 (0.90, 1.04)
Skeaff & Miller
0.93 (0.84, 1.04)
1.00 (0.80, 1.24)
0.98 (0.90, 1.06)

Skeaff & Miller only included a trial by Ball, et al and the Women’s Health Initiative (WHI), while Hooper, et al included several others that were either very small (only a few events or deaths in each group), for cancer (mostly breast cancer and one bowel cancer) and one (the DO IT trial) that looks more like a fish oil trial.  So I’m only going to do Ball, et al and the WHI. 

This won’t limit me (or Skeaff & Miller) that much.  The WHI was a huge trial, lasting 8.1 years with 46,558 participants.  Consequently the WHI has well over 50% of the weighting in statistical analyses, which is probably a main reason why the results of the two meta-analyses are so similar, even though Hooper, et al included several other trials 

* They included Rose (but only compared olive oil to corn oil), LA Vets, MRC, Oslo, Finnish, DART and STARS and found for: CHD events RR = 0.83 (0.69, 1.00), p = 0.073; CHD mortality RR = 0.84 (0.62, 1.12), p = 0.335; and total mortality RR = 0.88 (0.76, 1.02), p = 0.005.  The somewhat positive result isn’t surprising considering their poor trial selection (pretty much as bad as Mozaffarian, et al) 

** My main objection to the fat modification trials was that the trials changed a lot more than SFA and PUFA intakes.  Some accidental changes would be expected when you make a dietary change.  For example: you would expect the control group (higher SFA) to have a higher vitamin A and cholesterol intake and the experimental group (higher PUFA) to have a higher vitamin E intake.  Other changes, like the amount of hydrogenated oil and junk food, is not expected and makes for a poorly controlled trial.  In the fat reduction trials you would expect the low fat diet would have more accidental changes, like increasing intake of vegetables, fruit and grains, as a function of increasing carbohydrate intake 


Despite the dogma that ‘fat will clog your arteries’, low fat diets wouldn’t be expected to reduce CHD based on both observational studies and their effect on blood lipids.  Also, low fat diets haven’t been found to reduce CHD events, CHD mortality or total mortality in clinical trials.

Tuesday, March 25, 2014

Some Criticisms of the Paleolithic Diet Trials

There are a number of potential criticisms of the Paleo diet trials that I want to address 

Define Paleo 

Paleo is poorly defined 

Paleo critics are quite hypocritical on this point.  On the one hand they want a unified definition, but criticise it for being inflexible and say stuff like: ‘you can’t promote The Paleo diet because the diets of hunter-gatherers varied by geography and season and sometimes by a great deal’ and “chucking the many different hunter–gather diets into a blender to come up with some kind of quintessential smoothie is a little ridiculous”. 

Never mind that the Paleo community acknowledges the diversity of hunter-gatherer diets.  I also don’t understand the last objection.  If we draw a made up Venn diagram (below) to symbolise the differences between HG diets, then there are going to be many areas of common ground.  What’s so bad about finding value in the common ground?

What Paleo Diet? 

The Paleo diet used in the trials is very different to the Paleo diet that is being promoted 

I agree, but I don’t think it’s so black or white.  If people could only promote the exact diet people adhered to in clinical trials we would have a lot of half-assed diets and a lot of recommendations that lack evidence to support them, such as SFA and salt restriction in the Paleo trials. 

There are some people who describe the Paleo diets in the trials as something they’re not, and then use them as evidence to support their position.  That’s clearly dishonest and wrong.  But I can’t see the harm in using these studies to say: 

‘The Paleo diet (which contained A, but not/restricted/reduced B, with emphasis on C) used in these trials was more successful at doing X than the Mediterranean/Diabetes (whatever) diet (which contained D, but not/restricted/reduced E, with emphasis on F).  Since there were many differences between the two diets, it’s very difficult to know for sure which particular difference(s) was/were responsible for the difference(s) between the groups.  The Paleo diet used in the trials is similar to, but also different from the diet that I recommend.  Here are my reasons for how the diet I recommend differs from the Paleo diet in the trials… 

Obviously that won’t fit in a tweet 

For example: one of the main points of this objection is that the Paleo diet in the trials is low in SFA, but as it’s more commonly promoted, the Paleo diet doesn’t have restrictions on SFA.  So it would be dishonest and wrong to use these trials as evidence that SFA is ok (obviously), but there shouldn’t be a problem with using other sources of evidence (the fat modification trials) as support for SFA being ok.  As long as they acknowledge that cholesterol levels in the Paleo group would have been slightly higher if the SFA content was higher 

* Just in case you’re not aware here are some of the differences.  The Paleo diet used in the trials is a lot more consistent than the various incarnations of Paleo as it’s more commonly promoted, so the last column depends a lot on who you’re talking to (especially regarding macronutrients, dairy, fruit and starch).  Many popular versions of Paleo also include some variation of things like cured meats, fermented foods, some dark chocolate, other parts of the animal and ‘Paleo’ junk food (which should be an oxymoron) 

The Paleo Diet in the Trials
The Paleo Diet as it’s more Commonly Promoted
High protein, moderate fat, moderate carb.  Low SFA, high MUFA and PUFA
Moderate-high protein, high fat, low carb.  High SFA and MUFA, low PUFA
Ad Libitum Foods
Lean meats, fruit, vegetables
Fatty meat, non-starchy vegetables, eggs, fats (mostly animal fats, olive oil and coconut oil)
Moderate Amounts Of
Eggs, potatoes, nuts, olive/canola oil
Dairy, fruit, starchy vegetables, salt
Grains, legumes, dairy, refined sugar, salt
Grains (except white rice), legumes, refined sugar

The Trials Aren’t Well Controlled 

There are many differences between the two diets in the Paleo diet trials, so we don’t know what effect each variable has 

I completely agree.  My main complaint against the fat modification trials was that the experimental groups did more than just replace SFA with PUFA.  They were all advised to restrict TFA, many of them were advised to eat less junk food (baked goods, ice cream) and some were also advised to eat more whole foods (fruits, vegetables, etc).  So how could you say the differences in outcomes were due to fat modification rather than something else that was altered? 

Same thing is true in the Paleo trials where the Paleo diet tends to be: 

·         Higher in fruit, vegetables and meat, but lower in grains and dairy
·         Higher in protein, but lower in calories, carbohydrate and saturated fat 

If there’s a difference in outcomes between the groups we don’t know which of these factors was responsible.  Sometimes studies try to answer this by doing correlations, but I don’t think they are that useful.  Often they will really just be a marker of compliance, not only to grains/dairy/etc, but also to things like junk food and perhaps even exercise and other lifestyle factors. 

The bottom line is that the trials aren’t designed to say whether grains are bad.  Hopefully there will be a trial with a similar design to DART (multiple groups, each with a different combination of diet advice) to better test individual components and perhaps even examine whether there are any interactions 


Compliance was poor in the most recent trial, so you can’t really compare the diets 

I agree to some extent.  Metabolic wards and the more community based trials have their advantages and disadvantages.  Metabolic wards are better for measuring the actual diet (construct validity), while free living trials are better at measuring what will happen in real life (external validity).  You may have come up with the best diet, but in practice will people stick to it*?  If people find it too difficult to stick to the diet, then chances are that’s a problem of the diet. 

The recent Paleo trial had poor compliance in both groups.  I’m not sure if the level of compliance was better/worse than in other trials, but it was pretty bad and would have affected the results quite a bit.  Just to focus on the Paleo diet, the diet used is fairly difficult to adhere to (no grains and dairy is the main thing, but also lean meats and restricting salt**).  Perhaps the adherence, and then consequently the result would have been more favourable if some of the restrictions (like lean meats, low salt, low SFA and no cream and butter) were lifted.  Particularly those without much evidence to support them like low salt and low SFA (so lean meats). 

I imagine one of the main difficulties for dieticians is giving sufficient unpalatable advice to be effective, but not too much to be sustainable, and all of which depends on the person and their needs. 

* This is the issue with calorie restriction for weight loss.  Of course it works, but it’s very difficult to stick to long term. 

** I’m not counting legumes because people don’t eat much of them and don’t have the same desire for them as bread, pasta, baked goods, milk, cheese, butter, ice cream, etc. 

*** This isn’t only an issue for Paleo, but rather an issue for all dietary advice, whether it’s a particular diet, national dietary guidelines, etc.
Finally, there's also the issue of the trials only measuring stuff like weight and biomarkers, but not disease endpoints, which is a function of them being small and short

Monday, March 17, 2014

Paleolithic Diet Trials: Part 3

Participants and Diets 

70 reasonably healthy, non-smoking, post-menopausal women with a BMI ≥ 27 were randomised to one of two diets: a Paleolithic diet (PD) or a diet that followed the Nordic Nutrition Recommendations (NNR).  The groups were similar at baseline, except for higher HDL-C in the PD group 

The macronutrient recommendations were: 

Protein (%)
Fat (%)
Carbohydrate (%)

In addition, the PD group was advised to have a high intake of MUFA and PUFA, base the diet on “lean meat, fish, eggs, vegetables, fruits, berries and nuts”, with additional fat sources from avocado, rapeseed oil and olive oil and avoid “dairy products, cereals, added salt and refined fats and sugar”.  The advice to the NNR group also emphasised on “low fat dairy products and high-fibre products”.  “Both diets were consumed ad libitum” 


The only dietary information we are given is the macronutrients: 

Energy (kcal)
Protein (%)
Fat (%)
Carb (%)
6 Months
24 Months
6 Months
24 Months

Other dietary data besides P:F:C include: 
  • Both groups (especially the PD group) increased MUFA and omega 3
  • The PD group decreased SFA and increased dietary cholesterol
  • Both groups (especially the PD group) increased omega 6 at 6 months, but both groups roughly halved their omega 6 at 24 months
  • There wasn’t any difference in fibre or sugar intake 

Compliance at least to macronutrients was quite poor.  In particular the PD group increased protein at 6 months, but not at 24 months and only less than 50% of the protein target (by %)*.  Also the NNR group didn’t decrease fat or increase carbohydrate and fibre, and actually reduced the absolute amount of carbohydrate at both 6 and 24 months 

At 6 months body weight, fat mass, waist circumference and sagittal diameter (of abdomen) (depth) decreased significantly more in the PD group than the NNR group.  However, at 24 months the differences weren’t significant as between 6-24 months there was some regain in the PD group and some further weight loss in the NNR group. 

Several biomarkers were measured (glucose, insulin, SBP, DBP, HR, hsCRP, tPA, PAI-1, cholesterol, HDL-C, LDL-C and triglycerides).  There were no significant differences in the changes in these biomarkers at 24 months except triglycerides, which decreased significantly more in the PD group.  Generally though the PD group had non-significantly greater improvements, but I hesitate to say that because those p values were nowhere near 0.05. 

In conclusion, for weight loss the PD group did slightly better at 6 months, but there was no significant difference at 24 months.  Adherence to both diet plans was quite a problem and was probably the cause of the modest regain in the PD group 

I can understand why many in the Paleo community are keen to promote these RCTs, particularly with the (mostly) unfounded criticisms over the last year, although I don't think this trial is exciting enough to be considered ‘long term scientific verification of the Paleo diet’.  If verification means 'is safe/healthy long term' you shouldn't need a study to see whether a diet based on meat, fruit and vegetables is safe and healthy; or if verification means 'better than', then this trial doesn't deliver.  Also, while the advice to the NNR group was a low fat, high carb diet, they didn’t actually change their macronutrient intake, so the Paleo diet wasn’t really being compared to a low fat diet (although that depends on how you define low fat I guess)*** 

* Urinary nitrogen wasn’t significantly different at 6 months or at 24 months, even though both absolute and relative protein intake was higher at both time points.  Perhaps the women in the PD group over reported their protein intake.  (As an aside, I hear how getting women to eat more protein or meat is often challenging.  They say things like ‘I couldn’t eat meat for lunch, it’s so filling’, but isn’t that the point if you’re trying to lose weight?) 

** In table 2 the baseline energy intake is lower than total energy expenditure in both groups.  Suggesting either that they were on a weight loss diet before the trial or that this is another example of the classic under reporting of food and/or over reporting of exercise that you hear so much about 

*** The Nordic Nutrition Recommendations have since been updated (which isn’t low fat, high carb), but would have occurred after the trial started 

**** The study is pay per view but I hear you can request access here

Monday, March 3, 2014

The Diet Heart Hypothesis: Part 4

People who support the diet heart hypothesis (DHH)* may use the Hooper, et al meta-analysis or the Mozaffarian, et al meta-analysis as evidence to replace SFA with PUFA, even though both meta-analyses found no benefit for total mortality.  After looking at the trials included in the meta-analyses, I found that those that were better controlled didn’t support the DHH**, and in fact suggested some harm, particularly for women.  Whereas the trials that supported the DHH were quite multifactorial in favour of the experimental group
The meta-analyses were only able to find some benefit (CHD events in Hooper, et al and CHD events and CHD mortality in Mozaffarian, et al) by including trials that were poorly controlled – trials that not only replaced SFA with PUFA, but also either altered the high PUFA diet in other ways and/or had other confounding variables that favoured the experimental group.  The Mozaffarian, et al was particularly guilty of this as they included three favourable trials that were multifactorial and also excluded two unfavourable trials for poor reasons
* Replace SFA with PUFA to lower LDL-C to reduce risk of CVD
** The only benefit was in a non-blinded trial that found a decrease in possible + probable angina (not very persuasive)
Some Possible Objections
The vegetable oils and margarine in the 60s-90s were often high in trans fats, therefore the trials were unbalanced in favour of the control group
The investigators generally took a lot of care to reduce the amount of trans fats in the experimental group by restricting hydrogenated oils and/or junk food, and by using soft margarines rather than common margarines.  But let’s say for the sake of argument that the experimental groups had some trans fats in their seed oils/margarine.  Then that might just balance out the restriction of hydrogenated oils and/or junk food that was present in all trials (except MCS because I don’t know), which leaves us with the same four trials that were better controlled: Rose, MRC, Sydney and MCS
Clinical trials are difficult to design and execute so we shouldn’t rely on them for answers.  Instead we should rely on observational studies, which prove/suggest replacing SFA with PUFA reduces CHD.
Yes, trials are hard, but if you don’t do them properly, then why do them at all?  What is the point of trials like the ACC, Oslo, FMHS and STARS?  You can’t even say which variables worked, which were unnecessarily and which were harmful.  All you can say is that the results are interesting and the variables need to be tested individually in further trials.  One of the earliest lessons in high school science is how important it is to test one variable at a time, so why didn't the researchers do that.
It’s really quite bad, if people want to know whether they should eat butter or margarine and whether to cook with animal fats or seed oils a trial should have just looked at that (one group butter and animal fats, the other group margarine and seed oils), nothing else.
As for observational studies or trials with blood lipids as the end-point, you shouldn’t rely on them for dietary guidelines or for establishing causality.  There are instances where things that appear favourable in observational studies and things that lower cholesterol or improve the total:HDL-C ratio have been found to have no effect or an adverse effect in RCTs (see these links)
The precautionary principle suggests that because LDL-C is associated with CVD and SFA increases LDL-C while PUFA decreases LDL-C we should replace SFA with PUFA
If anything the precautionary principle should be used against replacing SFA with PUFA.  After all the consumption of refined vegetable (seed) oils is very recent (~ 100 years), which is incidentally when deaths from CVD started climbing.  There is also a lack of long term data on the safety of high (~ 10%) PUFA diets (what Mozaffarian, et al and others suggest), and based on the studies Chris Masterjohn discusses here: as linoleic acid builds up in adipose tissue, vegetable oils seem to increase vitamin E depletion and consequently impair health*
* This may also counter the potential argument that the trials weren’t able to show benefit because they didn’t last long enough.  Besides that, some of the multifactorial trials found benefit and some of the better controlled found harm quite quickly (within ~ 5 years)