Sunday, September 17, 2017

The PURE Study

The Prospective Urban Rural Epidemiology study (PURE study) has lately been making the rounds in the media following 3 publications that I will discuss below

The PURE study is a very large prospective cohort study with 135335 participants from 18 countries, that were followed up for a median of 7.4 years.  The PURE study is quite unique as the vast majority of prospective cohort studies for long term diet and disease relationships use North American and European cohorts (with the occasional Japanese cohort), whereas the PURE study uses 18 countries, includes countries in other parts of the world, and those middle and lower income levels.  They categorised the countries into the following regions:

“Regions included China, south Asia (Bangladesh, India, and Pakistan), North America, Europe (Canada, Poland, and Sweden), South America (Argentina, Brazil, Chile, and Colombia), Middle East (Iran, occupied Palestinian territory, Turkey, and United Arab Emirates), southeast Asia (Malaysia), and Africa (South Africa and Zimbabwe)”

In this study, fruit, vegetable and legume consumption was combined for the main analyses because legumes are sometimes categorised as vegetables.  In addition potatoes and other tubers weren’t counted as vegetables, and fruit and vegetable juices weren’t counted as fruit and vegetables

Combined fruit, vegetable and legume consumption was associated with more education and physical activity, a higher intake of calories (2869 kcal for ≥ 8 serves vs 1442 kcal for < 1 serve), red meat, and white meat (both absolute intake and relative to energy intake), less smoking and slightly lower starch intake

The study found that fruit, vegetable and legume consumption was associated with lower mortality up to 3-4 serves per day (375-500g), which was due to both CVD and non-CVD mortality, and there was no additional benefit with higher consumption.  Fruit, vegetable and legume consumption on their own were associated with lower mortality, with no additional benefit beyond 2 serves per day for fruit and vegetables each, and 1 serving per month to less than 1 serving per week for legumes.  Also, raw vegetable intake was more strongly inversely associated with mortality than cooked vegetable intake

The results of this study are pretty much what would expect if you’ve read the Mensink meta-analysis.  There are some surprises such as MUFA and PUFA intake being associated with higher LDL-C, PUFA intake being more strongly associated with higher LDL-C than MUFA, and carbohydrate intake being only weakly associated with triglycerides (~0.08 mmol/l from 20% points of carbohydrate)

The study emphasises the ApoB:ApoA1 ratio (the ratio of LDL particles to HDL particles).  This measure is better than LDL-C and HDL-C [9] [10] [11], because the cholesterol in those particles is in some ways irrelevant, it’s about LDL and HDL function and particle number at least partly implies net function

In addition, protein didn’t have much relationship with any measure of blood lipids, and dietary cholesterol had no significant relationship with on ApoB despite being associated with higher LDL-C

This paper attracted the most media attention by far.  For the purposes of this blog post I’m going to focus on total mortality as it tells you more than just looking at one cause of death and it avoids the bias of cherry picking subgroups to fit a particular narrative

Carbohydrate intake (by % of energy intake) was associated with an increased total mortality (Q5 vs Q1: RR = 1.28), which was due to an increase risk of death from non-CVD causes (Q5 vs Q1: RR = 1.36).  Whereas fat and protein intake were both inversely associated with lower total mortality (Q5 vs Q1: RR = 0.77 and 0.88), which in both cases was due to an reduced risk of non-CVD mortality (Q5 vs Q1: RR = 0.70 and 0.85)*

When protein intake was separated by animal and plant protein, they found that “Animal protein intake was associated with lower risk of total mortality and no significant association was observed between plant protein and risk of total mortality”

* “Among noncardiovascular disease mortality, in all regions except Africa, the most common cause of mortality was cancer followed by respiratory diseases. In Africa, infectious disease was the first and respiratory disease was the second most common cause of non-cardiovascular disease mortality.”

Fat intake was then broken down to SFA, MUFA and PUFA.  Intake of all three classes of fats was inversely associated with total mortality (Q5 vs Q1: RR = 0.86, 0.81, 0.80).  which in all cases was due to an reduced risk of non-CVD mortality (Q5 vs Q1: RR = 0.86, 0.79, 0.75).

* SFA was the only class of fatty acids to be inversely associated with stroke mortality.  This isn’t the first hint that such a relationship may exist, as the Siri-Tarino meta-analysis also found an inverse relationship between SFA intake and stroke (events not deaths) but one that didn’t reach significance (RR = 0.81, CI = 0.62-1.05, P = 0.11).  This may be a bit of surprise, particularly to diet heart true believers, but then the lipid and diet heart hypotheses were always based on coronary heart disease, not stroke or other manifestations of CVD.

They then separated the data by Asian and non-Asian regions, which found the relationships between total mortality with carbohydrate, total fat, SFA and MUFA intake are stronger in non-Asian countries, and that PUFA intake is associated with lower mortality in Asian countries, but not non-Asian ones

So what to make of these results?

If we take these results on face value it challenges a few dietary dogmas in conventional dietary advice (CDA):

  • CDA: low fat (20-35%), high carbohydrate (45-65%) diets decrease the risk of chronic disease [1].  PURE: reducing fat intake below 35% and increasing carbohydrate intake above 45% of total energy intake is associated with higher total mortality*
  • CDA: SFA increases the risk of CHD and/or PUFA decreases the risk of CHD [2].  PURE: SFA and PUFA were not associated with CHD and the RR for SFA and PUFA were near identical (1.17 and 1.12)
  • CDA: plant protein is healthy, animal protein is unhealthy [3].  PURE: animal protein is healthy, plant protein is neutral
  • CDA: recommends 5 servings of fruit and vegetables per day.  PURE: no additional benefit beyond 3-4 serves of combined fruits, vegetables, and legumes per day 

In the discussion the authors cite previous observational studies that did not find a relationship between total fat and carbohydrate intake with mortality, but that previous observational studies in North American and European cohort have a narrower and lower range of carbohydrate intakes compared to the PURE study (~35-56% vs 46-77%).  Regarding fat and carbohydrate intakes, one could make an argument based on these results that the relationship between health/mortality and intake of fat and carbohydrate generally exists as an inverse U-shaped curve, just like protein and micronutrients.  Where, within the range of intakes in the PURE study, higher fat and lower carbohydrate intake sits higher on the inverse U-shaped curve.  But if we were to just look at the PURE study, the exact shape of the inverse U-shaped curve is unknown because the PURE study can’t make any good conclusions about fat intakes > 35% and carbohydrate intakes < 45% of total energy intake

One of the most common argument from mainstream sources is that the study simply shows that refined carbohydrates are bad.  A strong line of evidence to back this up is that combined fruit, vegetable and legume intake was associated with lower mortality in another of the recent publications from the PURE study [4], not to mention the many other observational studies that find fruit & vegetables = good, refined carbohydrates = bad.  But this argument cuts both ways if you’re looking to have a discussion on the health effects of fats vs carbohydrates.  Just as fruit, vegetables, whole grains and legumes don’t have the same health effect as refined carbohydrates; full fat dairy, eggs, avocado, nuts and seeds don’t have the same health effects as refined oils.  Even though these days many conventional and alternative experts will proudly declare (particularly when the latest study doesn’t go their way), that they are beyond nutrients and instead focus on foods or eating patterns (however they define that), most will still object or have some problem with a non-extreme (≤ 10% or ≥ 70%) intake of either fat or carbohydrate.  They still care

Another explanation for the results is that there was insufficient adjustment for confounding variables.  In the contexts of standard western observational studies health consciousness is a major confounding variable.  Those who better follow conventional dietary advice do so mostly because they are more health conscious, and because they are more health conscious they are also more likely to exercise regularly and less likely to smoke, etc.  Therefore, when extracting the health effects of a given nutrient or food, observational studies are also extracting the health effects of eating patterns and lifestyle.  I have serious doubts that multivariate analyses always sufficiently adjust for such confounding variables.  For example, a very large observational study was published early this year on meat and mortality.  Basically it found red meat increased the risk total mortality and white meat decreased it.  Is red meat and white meat so different, or are these divergent outcomes a product of who eats red meat vs who eats white meat?  The answer probably lies more with the latter.  The strongest associations for red meat and white meat were respiratory mortality and liver disease, probably just simply due to differences smoking and drinking, and insufficient adjustments for these confounding variables [5].  This is a similar result with another observational study that found saturated fat was associated with respiratory mortality and it being one of the stronger associations, not CHD or CVD by the way, and also that SFA intake was associated with smoking [6].  The PURE study has a unique issue in that it’s also got the country’s income as a confounding variable.  For example, if country A eats a higher carb diet and has a lower life expectancy than country B, then an observational study that either combines data from both countries (like the PURE study) or compares two countries (an ecological study) could easily find that that carbohydrate intake is associated with an increased risk of death and in a more complicated data set may have difficulty in adjusting for these confounding variables.  So ideally the data would be separated by each country or at least income level but that may leave the study quite underpowered.  Even though the researchers adjusted for confounders there were some strange associations and ones that are quite unusual to see

For example, despite a being inversely associated with total mortality intakes of total fat, SFA, MUFA and PUFA all had RRs above 1.00 for myocardial infarction (1.12, 1.17, 1.12 and 1.12), but in each case were nowhere near significance (P = 0.40 for all of them).  Conversely, the RR for carbohydrate intake and myocardial infarction was 0.90, despite carbohydrate intake being associated with higher total mortality.  This is an unexpected result as in another of the papers, all classes of fats were associated with a lower ApoB:ApoA1 ratio, total-C:HDL-C ratio and triglyceride:HDL-C ratio, whereas carbohydrate intake was associated with a higher ratios of all three.  An explanation that may partly explain these results is that richer countries eat more fats and less carbohydrate, live longer, and get more heart disease, and this wasn’t able to be sufficiently adjusted for.  Another example is from the fruit, vegetable, and legume study where an intake of 1 serving of legumes per month to less than 1 serving per week, in other words probably like 2 servings of legumes per month, was associated with 20% lower mortality vs having none at all.  I doubt that ~2 serves of legumes per month would have such an impact

I initially found the fat versus carbohydrate debate that this study reignited to be quite old.  Despite the popular version of the Mediterranean diet and the standard western diet having very similar fat and carbohydrate intakes, it’s clear that they have very divergent health effects.  This difference in health outcomes is far more likely due to differences in diet quality rather than a few percentage points difference in fat and carbohydrate intake.  It can be easy to live in a bubble where this knowledge is taken for granted.  But this bubble bursts on occasion and then it becomes apparent that many in the general population (including nutrition students!!!) place a large emphasis on reducing fat intake, believing fat increase the risk of weight gain and chronic disease, even though the such beliefs should be at least a decade or two out of date.

“For prevention of chronic diseases, nutrient-based recommendations are more difficult to translate to the public. Few individuals can accurately estimate their daily calorie consumption, much less their intake of total fat or specific fatty acids [78, 79]. Interestingly, while 67% of consumers are trying to limit their fat intake, few are aware of how much fat they should actually be eating [76]. Only 22% of consumers correctly identified the recommended range of calories from fat [9]. Sixty percent of consumers believed that fat intake should be less than 14% of daily calories [9].” [7]

These beliefs are not harmless either as people are enticed by health halo around low fat junk food and reject the possibility of low carb diets being especially therapeutic for some diseases such as type 2 diabetes and not that difficult to maintain

The PURE study is one prospective cohort study among many, but a very large and unique one.  To draw a robust conclusions about diet and disease relationships you should look at the other observational studies rather than just cherry pick the one that supports your narrative (assuming RCTs aren’t available).  At the very least, the PURE study will teach some people about the importance of confounding variables, although I suspect the many of the same people complaining about PURE will forget about confounding variables and the difficulties in adjusting for them when championing the next ‘red meat is bad for you’ study

* The authors recommended a 35% fat and 50-55% carbohydrate intake [8], but this doesn’t line up with their results (maybe being politically correct?) and also only leaves a protein intake of 10-15%.  Whereas the results of the PURE study alone would suggest protein:fat:carbohydrate should be 20:35:45, which is almost what western countries eat anyway (minus a few % protein + a few % alcohol)


  1. The weird associations between fats and MI can possibly be explained by the inability to measure trans fats.
    Trans fats impair DHA metabolism which influences clotting and inflammation, and seem to have no effect or even a beneficial effect on purely metabolic diseases like diabetes or obesity, so MI is exactly where we'd expect to see them.
    Trans fats in PHVO and shortening are made by the conversion of PUFA to trans MUFA and SFA respectively. So any of these three fat types depending on the food could be most strongly correlated with the missing TFA. Of course PUFA may be associated with MI because it's sourced from cheap or overheated oils rather than nuts or other foods (there's a note in PURE that the study gives no evidence about oils, as distinct from other food sources of PUFA).
    Why wasn't trans fat measured? I'm not sure, but maybe because PHVO was being removed from the food supply in some countries, notably India and Canada, during the study period.

    As for that respiratory mortality effect in NHS/HPFS, the news from NHS2 is that disinfectant use is associated with COPD (OR around 1.25). This wasn't known about or controlled for in NHS/HPFS. Airborne pollutant exposure other than smoking never is, but people in Western populations who work with or live near pollutants are those most reliant on meat and eating more saturated fat.
    You can bet the vegan and vegetarian are not going to apply for the job in the oil refinery or tyre plant. Similarly the nurses working with bleach most will be tend to be the underpaid blue collar individuals cleaning up in rest homes, not the lucky ones working in upscale surgeries.

    1. And, of course, a hidden TFA content of MUFA and PUFA would also explain the correlation with LDL - without that necessarily being the link to MIs.

    2. Thanks for the comment George, that makes some sense. Are you saying that artificial TFA (e.g. elaidic acid) may have a beneficial effect on metabolic health, or just ruminant TFA (vaccenic acid and CLA) which seem to have some strong inverse associations with type 2 diabetes?

      The disinfectant > COPD link sounds like it could be part of the story. There's probably several other similar factors as well, and therein lies the problem with relying so much on observational studies - you can only adjust what you measure and you will only measure what you think is relevant at the time

    3. In the de Souza meta-analysis industrial TFA is associated with CHD mortality -
      Industrial, but not ruminant, trans fats were associated with CHD mortality (1.18 (1.04 to 1.33) v 1.01 (0.71 to 1.43)) and CHD (1.42 (1.05 to 1.92) v 0.93 (0.73 to 1.18)
      but not diabetes (less evidence but probably a small non-significant association).
      In DeLany et al elaidic acid behaves much like oleic acid - there's no sign of it accumulating
      In rats it's healthier than linoleic acid

      So I think that ruminant TFA (CLA is actually a cis, trans fat so TFA is a misnomer in that case) is beneficial - in its dietary context, not necessarily as a supplement. Elaidic acid and (perhaps more so) the whole raft of rarer, novel TFAs formed in PHVO etc are bad for the heart, maybe through their effects on DHA and clotting. But attempts to say TFA (or at least elaidic acid) is behind cancer, obesity, diabetes etc have less evidence - the food it's in, probably, but a distinct metabolic difference between elaidic acid and other fats is less obvious.

      Thinking about disinfectant and COPD, it could damage the lungs directly, especially bleach; there could well be a hygiene hypothesis interaction with the lung microbiome - but the more prosaic explanation would be that the nurses who need to use bleach most often are exposed to the most virulent airborne pathogens, are dealing with the sickest people and greatest discharge of infectious fluids.