Monday, October 24, 2011

The China Study II: Animal protein, wheat, and mortality … there is something odd here!

WarpPLS and HealthCorrelator for Excel were used in the analyses below. For other China Study analyses, many using WarpPLS and HealthCorrelator for Excel, click here. For the dataset used, visit the HealthCorrelator for Excel site and check under the sample datasets area. I thank Dr. T. Colin Campbell and his collaborators at the University of Oxford for making the data publicly available for independent analyses.

The graph below shows the results of a multivariate linear WarpPLS analysis including the following variables: Wheat (wheat flour consumption in g/d), Aprot (animal protein consumption in g/d), Mor35_69 (number of deaths per 1,000 people in the 35-69 age range), and Mor70_79 (number of deaths per 1,000 people in the 70-79 age range).


Just a technical comment here, regarding the possibility of ecological fallacy. I am not going to get into this in any depth now, but let me say that the patterns in the data suggest that, with the possible exception of some variables (e.g., blood glucose, gender; the latter will get us going in the next few posts), ecological fallacy due to county aggregation is not a big problem. The threat of ecological fallacy exists, here and in many other datasets, but it is generally overstated (often by those whose previous findings are contradicted by aggregated results).

I have not included plant protein consumption in the analysis because plant protein consumption is very strongly and positively associated with wheat flour consumption. The reason is simple. Almost all of the plant protein consumed by the participants in this study was probably gluten, from wheat products. Fruits and vegetables have very small amounts of protein. Keeping that in mind, what the graph above tells us is that:

- Wheat flour consumption is significantly and negatively associated with animal protein consumption. This is probably due to those eating more wheat products tending to consume less animal protein.

- Wheat flour consumption is positively associated with mortality in the 35-69 age range. The P value (P=0.06) is just shy of the 5 percent (i.e., P=0.05) that most researchers would consider to be the threshold for statistical significance. More consumption of wheat in a county, more deaths in this age range.

- Wheat flour consumption is significantly and positively associated with mortality in the 70-79 age range. More consumption of wheat in a county, more deaths in this age range.

- Animal protein consumption is not significantly associated with mortality in the 35-69 age range.

- Animal protein consumption is significantly and negatively associated with mortality in the 70-79 age range. More consumption of animal protein in a county, fewer deaths in this age range.

Let me tell you, from my past experience analyzing health data (as well as other types of data, from different fields), that these coefficients of association do not suggest super-strong associations. Actually this is also indicated by the R-squared coefficients, which vary from 3 to 7 percent. These are the variances explained by the model on the variables above the R-squared coefficients. They are low, which means that the model has weak explanatory power.

R-squared coefficients of 20 percent and above would be more promising. I hate to disappoint hardcore carnivores and the fans of the “wheat is murder” theory, but these coefficients of association and variance explained are probably way less than what we would expect to see if animal protein was humanity's salvation and wheat its demise.

Moreover, the lack of association between animal protein consumption and mortality in the 35-69 age range is a bit strange, given that there is an association suggestive of a protective effect in the 70-79 age range.

Of course death happens for all kinds of reasons, not only what we eat. Still, let us take a look at some other graphs involving these foodstuffs to see if we can form a better picture of what is going on here. Below is a graph showing mortality at the two age ranges for different levels of animal protein consumption. The results are organized in quintiles.


As you can see, the participants in this study consumed relatively little animal protein. The lowest mortality in the 70-79 age range, arguably the range of higher vulnerability, was for the 28 to 35 g/d quintile of consumption. That was the highest consumption quintile. About a quarter to a third of 1 lb/d of beef, and less of seafood (in general), would give you that much animal protein.

Keep in mind that the unit of analysis here is the county, and that these results are based on county averages. I wish I had access to data on individual participants! Still I stand by my comment earlier on ecological fallacy. Don't worry too much about it just yet.

Clearly the above results and graphs contradict claims that animal protein consumption makes people die earlier, and go somewhat against the notion that animal protein consumption causes things that make people die earlier, such as cancer. But they do so in a messy way - that spike in mortality in the 70-79 age range for 21-28 g/d of animal protein is a bit strange.

Below is a graph showing mortality at the two age ranges (i.e., 35-69 and 70-79) for different levels of wheat flour consumption. Again, the results are shown in quintiles.


Without a doubt the participants in this study consumed a lot of wheat flour. The lowest mortality in the 70-79 age range, which is the range of higher vulnerability, was for the 300 to 450 g/d quintile of wheat flour consumption. The high end of this range is about 1 lb/d of wheat flour! How many slices of bread would this be equivalent to? I don’t know, but my guess is that it would be many.

Well, this is not exactly the smoking gun linking wheat with early death, a connection that has been reaching near mythical proportions on the Internetz lately. Overall, the linear trend seems to be one of decreased longevity associated with wheat flour consumption, as suggested by the WarpPLS results, but the relationship between these two variables is messy and somewhat weak. It is not even clearly nonlinear, at least in terms of the ubiquitous J-curve relationship.

Frankly, there is something odd about these results.

This oddity led to me to explore, using HealthCorrelator for Excel, all ordered associations between mortality in the 35-69 and 70-79 age ranges and all of the other variables in the dataset. That in turn led me to a more complex WarpPLS analysis, which I’ll talk about in my next post, which is still being written.

I can tell you right now that there will be more oddities there, which will eventually take us to what I refer to as the mysterious factor X. Ah, by the way, that factor X is not gender - but gender leads us to it.

29 comments:

  1. Check for milk as factor X

    ReplyDelete
  2. Hi Anon, thanks, but why milk? Lactose?

    ReplyDelete
  3. Paul, Aaron: Was going to get everything in one post, but realized I needed other analyses.

    ReplyDelete
  4. Funny that a re-tweeter said that animal protein “may not be so bad after all”. Must have been someone really impressed by FOK, but who nevertheless re-tweeted this!

    ReplyDelete
  5. Because of insulin Ned

    ReplyDelete
  6. Thanks Anon. So I guess you mean dairy in general, right?

    ReplyDelete
  7. As a rule of thumb the lower the average insulin level the longer the expected lifespan for any given ideological diet. So I would expect milk to be a very different animal in this respect. I would even rephrase the benefits of caloric restriction as the the benefits of insulin restriction. That's why I would check for milk as factor x.

    ReplyDelete
  8. Dairy is damned tricky. But as a rough proxy in China I would say yes. I do not expect to find long- fermented cheese down there, so no French confounder. A Finnish scenario instead.

    ReplyDelete
  9. Interesting. I came across this link from a twitter account link; I'll have to read your other posts about this as well, but just a quick question; has income disparity been taken into consideration in these county numbers? I don't know the exact case for China, but there is the impression that folks who eat a diet containing more grain and less meat are of a lower income. Folks of a lower income in many countries often die of avoidable causes because of less access to adequate healthcare.

    Perhaps you've covered this already, and I'm not disputing your claims here - I don't know if these facts hold true for these counties or not, nor if the information is available. Just curious. Regardless, all of this is quite interesting!

    ReplyDelete
  10. Hi Anon. I had to restrict my dataset on mortality to only a limited number of variables, because it takes quite a lot of effort to align the rows correctly for columns (i.e., variables) on different sheets, among other data manipulation problems (the data provided on the Oxford site is not user friendly at all).

    Still, I have just checked the table of univariate correlations with mortality that have a P value lower than 0.05 (i.e., that are statistically significant), on the monograph, and income doesn’t seem to show up there at all.

    ReplyDelete
  11. thanx for doing all the heavy lifting! But, wouldn't plant protein also include soy, which I would think would add a non-trivial amount of protein to the diet.

    ReplyDelete
  12. Hi EB. Indeed, soybean products are a source of plant protein. But when you compare the amounts of wheat flour and plant protein consumed in the dataset, it seems that virtually all of the plant protein consumed is coming from wheat. These two variables are also very highly correlated.

    ReplyDelete
  13. I would expect that there would be big differences between consuming wheat flour per se and consuming gluten directly (which, as seitan and its variants is a major food in China). But I have no idea how the sorted through that in the stats--or it it were in fact noted at all.

    Some would expect more morbidity with higher gluten intake. I would lean the other way; I think most of what is wrong with wheat is water-soluble. In other words, I think that gluten is probably healthy for people while wheat flour iself is not.

    A minority (and possibly crackpot view), I know. All I can say in its defense is that I know people who react poorly to wheat and other grains, and have zero problems with seitan.

    The Campbell data probably can't answer such questions.

    But I'm interested to hear about Factor X!

    ReplyDelete
  14. Hi David. I don’t think wheat flour was somehow confused with seitan; that is not compatible with the plant protein numbers, which suggest that the wheat flour they consumed was at most 15% protein.

    Interesting points though.

    I guess they call seitan mock meat, or mock duck, over there – or so I heard.

    ReplyDelete
  15. This comment has been removed by the author.

    ReplyDelete
  16. By the way, one of my doctoral students who is Chinese was eating the other day what he said was a very common type of candy in China. I asked what it was made of, and he said wheat.

    Apparently wheat products are more common in China than many people think.

    ReplyDelete
  17. I guess these Japanese centenarians are not all that different from the folks in this dataset:

    http://www.ncbi.nlm.nih.gov/pubmed/1407826

    Abstract: The present paper examines the relationship of nutritional status to further life expectancy and health status in the Japanese elderly based on 3 epidemiological studies. 1. Nutrient intakes in 94 Japanese centenarians investigated between 1972 and 1973 showed a higher proportion of animal protein to total proteins than in contemporary average Japanese. 2. High intakes of milk and fats and oils had favorable effects on 10-year (1976-1986) survivorship in 422 urban residents aged 69-71. The survivors revealed a longitudinal increase in intakes of animal foods such as eggs, milk, fish and meat over the 10 years. 3. Nutrient intakes were compared, based on 24-hour dietary records, between a sample from Okinawa Prefecture where life expectancies at birth and 65 were the longest in Japan, and a sample from Akita Prefecture where the life expectancies were much shorter. Intakes of Ca, Fe, vitamins A, B1, B2, C, and the proportion of energy from proteins and fats were significantly higher in the former than in the latter. Intakes of carbohydrates and NaCl were lower.

    ReplyDelete
  18. "I guess they call seitan mock meat, or mock duck, over there – or so I heard."

    That's the stuff. The pack in my refrigerator (sold in water like fresh tofu) says that per serving we have:

    Calories: 110
    Total Fat: 2 g
    Total Carb: 4 g
    Fiber: 2 g
    Protein: 20 g

    Being from grain the protein, of course, is a little short on lysine.

    Eat it as a sole food source and you're asking for "rabbit starvation" even if you get lysine from somewhere. But fried with an excess of coconut oil (and either cajun spices or black bean sauce), it's an interesting addition to a low-carb diet.

    I'll let you know about my future morbidity. But seitan never seems to have adverse effects on me.

    Bread is an entirely different matter--so I can easily believe that wheat flour poses some problems.

    ReplyDelete
  19. This comment has been removed by a blog administrator.

    ReplyDelete
  20. Thanks David. So it is mostly protein. What is the type of fat (2 g)?

    For those interested, here is a photo of mock duck:

    http://en.wikipedia.org/wiki/File:Wheat_gluten_(vegetarian_mock_duck)_2007.jpg

    ReplyDelete
  21. Hi, Ned--

    Hard to get precise details. No sat fat, no trans. The rather confused and fragmentary references I can find seem to echo the composition of wheat germ oil, which is about half omega-6, and most of the rest is mono.

    Of course, since the only way I find it appetizing is stir-fried, the total fat intake ends up looking very much like the profile of what I fry it in...

    ReplyDelete
  22. proteins solutions Protein and organic proteins probably have the same effects on health. It's the proteins package that's likely to matter.

    ReplyDelete
  23. These kind of post are always inspiring and I prefer to read quality content so I happy to find many good point here in the post, writing is simply great, thank you for the post

    ReplyDelete
  24. Hey great information! I like this post 'cause is just what I was looking for so please keep sharing things like this and keep up the good work.

    ReplyDelete
  25. As per capita income rises, so too does consumption of animal fats. Therefore consumption of wheat vs. meat may be highly dependent on income. And of course higher income people have better access to health care. So consumption of wheat = low income = poor healthcare = higher mortality

    ReplyDelete
  26. Blank, read Ned's previous comment. From the dataset, mortality was not significantly associated to wealth.

    ReplyDelete