A few of you lovely readers emailed me today (thanks!) about the study Low-Carbohydrate Diets and All-Cause and Cause-Specific Mortality just published in the Annals of Internal Medicine. This paper compares mortality rates for folks eating a so-called “animal-based diet” versus a so-called “vegetable-based diet,” both of them so-called “low carbohydrate.” I finally got a chance to look at it, and indeed, a glance at the abstract looks a little spooky for any low-carb omnivores out there:

A low-carbohydrate diet based on animal sources was associated with higher all-cause mortality in both men and women, whereas a vegetable-based low-carbohydrate diet was associated with lower all-cause and cardiovascular disease mortality rates.

Oh noes! This abstract sounds vaguely China-Study-esque, with the conclusion that plant-based diets are healthier than ones featuring more animal foods. Was this study really comparing hardcore meat eaters with plant noshers, like the abstract implies? Is animal protein poison after all? Is it time to ditch the steaks and bow down in phytoestrogenic reverence to the almighty tofu?

I’ve said it before and I’ll say it again: In most cases, abstracts tell you a whole ‘lotta nothing—so don’t judge a study until you’ve read the full text.

For right now, I’ll give this study the benefit of the doubt and ignore the fact that A) the researchers used a pretty lame decile-based scoring system* and B) employed the notoriously unreliable food-frequency questionnaire to collect their data.

*NOTE: The decile method divvies up dieters into ten levels of adherence—with the folks in the first decile adhering the least to a low-carb diet, and the folks in the tenth decile adhering the most. The reason it’s lame is that it uses a scoring system based on misconceptions about what a low-carb cuisine looks like, including the necessity of a high protein intake.

First, let’s take a look at what the low-carbohydrate “animal food” group and “vegetable food” group were actually eating. Click the thumbnails for a bigger pic—first one’s women, second one’s men.

Ha ha ha ha.

Oh man.

I’ll sum it up. Some of these “low carbers” were eating up to 60% of their diet as carbohydrates (first decile), which—last time I checked—is kind of not low-carb. Even the lowest low-carb eaters were still eating over 37% of their calories from carbohydrates. Whoever decided to call this study “low carbohydrate” is nuttier than a squirrel turd. That doesn’t mean it can’t offer anything useful, though, so let’s look at what else is going on in the highest decile for each group (which is the only decile the researchers really looked at):

• Folks in the Animal Group were more likely to smoke and had higher BMIs than adherents of the Vegetable Group. Along with influencing mortality outcomes, this suggests the Animal Food group, in the aggregate, may have been somewhat less health-conscious than the dieters lumped into the vegetable category. And that’s the type of thing that has repercussions for other diet and lifestyle choices that weren’t measured in the study.
• The Vegetable Group was nowhere near plant-based: They derived almost 30% of their daily calories from animal sources (animal fat and animal protein), versus about 45% for the Animal Group. If we compare the middle (fifth) decile, the Vegetable Group was eating a greater percent of total calories from animal foods than the Animal Group was. D’oh!
• The Vegetable Group ate more fruits, vegetables, and whole grains than the Animal Group—which begs the question: What kinds of carbohydrates filled this macronutrient void for the animal-food eaters? Could it’ve been refined grains and processed carbs, which the study conveniently forgot to document?
• For the Vegetable Group, cancer and cardiovascular mortality was lower in the tenth decile than the first decile, even though both deciles ate exactly the same amount of red meat and nearly the same amount of total animal foods. This suggests animal products aren’t the driving force behind differences in mortality rates.
• Similarly, at the fifth decile, the Vegetable Group had a lower cardiovascular mortality hazard ratio than the Animal Group (0.99 versus 1.21), even though the Vegetable Group was eating a slightly greater proportion of animal foods (33.3% versus 29.9% of total energy for women; 32.9% versus 31% for men).

Here are the mortality tables:

All Cause

Cardiovascular Disease

Cancer

And for the vegetable dieters, the people with the lowest cancer mortality (male) and cardiovascular mortality (both genders) were not the ones eating the most plant foods—they were the folks in the sixth and seventh deciles, respectively:

Unfortunately, the article doesn’t show us the food/macronutrient breakdowns for any deciles besides the first, fifth, and tenth, so we don’t know what the average diet looked like for these people. But since the vegetable low-carbohydrate score was based on “the percentage of energy of carbohydrate, vegetable protein, and vegetable fat” the dieters consumed, it’s pretty safe to say that the folks in the sixth and seventh deciles were eating less plant foods than the tenth-decilers (and consequently, more animal foods).

Bottom line: In this study, when you look closer at the data, differences in mortality appear to be unrelated to animal product consumption. Changes in cancer and cardiovascular risk ratios occur out of sync with changes in animal food intake.

So what is responsible for the Vegetable Group’s lower mortality hazard ratios (and the Animal Group’s higher ones)?

Here’s a clue. Every time the researchers made multivariate adjustments to the data to account for the risk factors they did document (including physical activity, BMI, alcohol consumption, hypertension, and smoking, among other things), the hazard ratio went down for the Animal Group (meaning it got better) and it went up for the Vegetable Group (meaning it got worse). That indicates pretty clearly that the Animal Group had more proclivity to disease right from the get go, regardless of meat consumption, and the Vegetable Group may have been more health-aware than most folks. (To see what I’m talking about, look at the mortality tables under the “10″ column, and compare the “Age- and energy-adjusted HR” with the “Multivariate-adjusted HR” for each group.)

In other words, it looks like what this study really measured was a Standard American Diet group (aka Animal Group) and a slightly-less Standard American Diet group (aka Vegetable Group). Both ate sucky diets, but the latter had slightly less suckage. You can bet the farm that neither was anything close to “low carb.” And if you have two farms, you can bet the other one that neither diet group was anything near plant-based, so I’m not sure the vegan crowd has much to gloat about here.

The End.

(Not only is this woefully, frustratingly, absurdly belated, but it’s also not yet finished. But I hate being a blog tease, so here’s part one!)

If you’ve been following along with the previous China Study entries (and the wild drama that ensued), you know that I’ve been promising an entry on wheat for a while now, mostly because this little snippet snagged so many eyes:

Correlation between wheat flour and coronary heart disease: 0.67

That’s a value straight from the original China Study data. Could the “Grand Prix of epidemiology” have accidentally uncovered a link between the Western world’s leading cause of death and its favorite glutenous grain? Is the “staff of life” really the staff of death? Bwah ha ha.

Damning as it seems, a single unadjusted correlation isn’t enough to make that leap. Actually, nothing in this post will be enough to make that leap, because A) it’s epidemiological data and not a controlled study, and B) correlation isn’t causation anyhow. You know the drill.

So my goal here isn’t to prove anything about wheat. Mostly, I want to see if I can find a confounder that’s creating a false association between wheat and heart disease in the China Study data. Something wheat-eating regions have in common that makes them more susceptible to ticker troubles. Because really, folks, this is serious business:

And when we pluck out the wheat variable from the 1989 China Study II questionnaire—which has more recorded data—and consider potential nonlinearity, the outcome is even creepier:

Wowza! By the way, wheat flour also correlates significantly with hypertensive heart disease and stroke, but I’m mainly going to look at coronary heart disease in this post. (And although wheat looks like it could have a nonlinear relationship with heart disease, with the highest wheat eaters having disproportionately steeper rates than non-wheat eaters, I’m going to treat it as linear for the sake of this analysis. That way, the worst that’ll happen is we’ll underestimate the potential effect of wheat, which—for now—is better than overestimating it.)

Since I’m not trying to dissect our friend Campbell’s claims anymore, I’ll be using the China Study II data (from 1989) because it recorded more descriptive variables about diet and blood samples.* And because it’s already available online. (Not that I don’t love typing thousands of numbers onto my computer by hand. Three cheers for data-entry-induced carpal tunnel!)

*Quickie note: If you want to play with the China Study numbers yourself, I recommend not just using the “all vascular diseases” variable, because it includes rheumatic heart disease—a condition spawned by rheumatic fever and generally unrelated to diet. Lumping diseases with different etiologies together dilutes the strong correlations you can find by looking at each disease independently. Try checking out stroke (M065 STROKE), ischaemic heart disease (M063 IHD), and/or hypertensive heart disease (M062 HYPTENS)  along with all vascular diseases (M059 ALLVASC).

Here’s the problem with looking at wheat and heart health. Along with correlating pretty darn strongly with heart disease, wheat-eating regions boast  a number of other factors possibly involved as well—some as protective agents and some as causative. For instance, wheat flour correlates significantly and inversely with:

• Plasma folate concentrations (and consequently, homocysteine status)
• Fish intake and DHA levels
• Yearly green vegetable consumption
• HDL cholesterol
• Vitamin C intake

And it correlates significantly and positively with:

• Height, weight, and BMI
• Blood pressure
• Latitude (as a possible marker for vitamin D status)
• Yearly milk intake
• Polyunsaturated fatty acid intake

Since all of these variables also associate (inversely or positively) with heart disease, it’s possible they could be confusing the “0.67″ figure we’ve cited for wheat. Could some other, non-grain component of the wheat eaters’ diets predispose these folks to heart disease?

On the bright side, China’s wheat eaters are less likely to drown than the wheat-shunners (r = -0.68 for the youngsters under 34). Maybe they’re all buoyant from celiac bloat.

And in case you’re wondering, here are some heart disease risk factors (the ones Conventional Nutritional Wisdom likes to toss around) that don’t positively correlate with wheat. That means we probably can’t blame ‘em for wheat’s dirty deeds. Out of curiosity, though, I’ll still include them in some of my models just to see how they behave in relation to wheat with heart disease.

• All meat intake (r = -0.35)
• Red meat intake (r = -0.30)
• Animal fat intake (r = -0.35)
• Saturated fat intake (r = -0.40)
• Total animal protein intake (r = -0.27)
• Total fat intake (r = -0.43)
• Fat as a percentage of total calories ( r = -0.41)
• Total cholesterol (r = -0.05)
• Apolipoprotein B (r = 0.02)
• Daily alcohol intake  (r = -0.37)

Mostly, what I’m looking for is a little somethin’-somethin’ that both wheat flour and heart disease have in common. A shared variable that could be slyly—and wrongfully—framing wheat as our heart-harming villain.

So how do we untangle all these variables? I’m using two methods: multiple regression analysis and stratification. Multiple regression is a handy way of looking at two or more variables and seeing how each one behaves when the others are held constant, and stratifying data can work similarly by divvying up data into groups that share or exclude a certain variable. (For the stats junkies out there, I’m using ordinary least squares for the regressions, and I’m running each model two times: once with the data as-is, and once with any non-normally-distributed dependent variables transformed (via natural log) for more reliable statistical significance testing. I’m also checking for linearity between the variables before creating each model, since a nonlinear relationship will be underestimated with linear regressions.)

And for anyone not familiar with statistics terminology, here’s a really quick rundown of what you need to know to understand the numbers in this post:

• r = the Pearson product-moment correlation coefficient  between two variables. It can range from -1 to 1. When it’s zero or close to zero, there’s pretty much no relationship between the variables. When it’s a negative number (like r = -0.50), there’s an inverse relationship between the variables, meaning one increases as the other decreases. When it’s a positive number (like r = 0.50), there’s a positive relationship between the variables, meaning they increase and decrease hand-in-hand. The closer to -1 or 1 r is, the stronger the association. R can never prove cause and effect, though—it only indicates an relationship of some sort.
• beta = the standardized coefficient for each variable in the multiple regressions I’ll  be running. This is a lot like r, in the sense that it shows how well a specific variable is predicting the outcome (eg, heart disease) and also ranges from -1 to 1. But in the case of beta, we’re also controlling for the effects of other variables, so this number tends to be more accurate than r.
• p = the probability that our results are just a fluke. P indicates how likely it is that we’d get a value of a test statistic that’s as extreme (or more extreme) as the one we have based on chance alone. Having a p-value of less than 0.05 indicates a high level of significance and means that our results are pretty sound. The lower the number, the more confident we can be that we’ve got something legit.
• r-squared = percent of variance explained. This number shows what proportion of the outcome (eg, heart disease) can be explained by the variables in a particular model (eg, wheat and HDL cholesterol). The higher the number, the more successfully the variables are predicting the outcome. (“Predicting” is a misleading way of putting it, though, since we still aren’t looking at proof of cause-and-effect—only a relationship.)

Preliminary theories

It’s no secret that I’m less-than-enamored with wheat. We parted ways long ago (he got me allergic and then ran off with some floozy—classy, eh?). Nonetheless, I don’t like pointing fingers where they shouldn’t be pointed, so I’ll entertain some alternative theories that could explain wheat’s apparent association with heart disease.

1. Folate deficiency. In northern China, about 40% of the population qualifies as folate deficient (compared to only 6% in the south)—a geographical trend that corresponds nicely with wheat consumption. Being low in folate tends to elevate homocysteine, which—you guessed it—is an independent risk factor for heart disease. So maybe it’s not the wheat itself causing mischief, but the fact that low-vegetable, wheat-centered diets in China tend to breed folate deficiency and hike up homocysteine.

On top of that, in the China Study II data, wheat flour positively correlates (r = 0.30, p<0.05) with childhood death from neural tube defects—a category of birth defects often related to folate deficiency. Although the China Study data didn’t document homocysteine levels (darnit), the 1989 data did measure plasma folate. That means we’ll be able to test whether folate levels could be obscuring the true relationship between wheat and heart disease.

2. Vitamin D deficiency. For the most part, wheat-eating regions in China are in the northern half of the country—a hotspot for vitamin D deficiency, which is strongly linked to heart disease. Given the pretty convincing correlation between latitude and heart disease mortality, it’s possible that vitamin D is playing a role in this mess. Are the wheat-eaters merely suffering from low levels of the ol’ Sunshine Vitamin due to their unfortunate geographical placement, and getting more heart disease as a result? Sure seems possible.

3. Low intake of DHA. In an earlier publication, Campbell and his crew already determined that fish and DHA intake appears protective against heart disease in the China Study data. Not too surprising, since DHA reduces blood viscosity and can lower other factors associated with heart disease (like triglyceride levels). And considering wheat-eating regions don’t consume much seafood (r = -0.43 for daily fish intake), perhaps DHA deficiency—rather than wheat consumption itself—is to blame for higher rates of heart disease.

4. Combo-abombo. Maybe a mix of low folate, vitamin D deficiency, and DHA deficiency are swirling together into a doomful vortex—some horrible, Bermuda-Triangle-esque zone of heart disease. A zone that just happens to overlay areas of wheat consumption.

5. Unexpected mystery variable. If none of the above can explain the wheat-heart disease link, we’ve still got a verdant jungle of China Study variables to plow through. So plow we shall. I’ll try running a number of common-sense models to see if I can find something that explains heart disease better than wheat alone.

Multiple regression results

Folate. Ah, theory numero uno! Like wheat, folate has a strong, statistically significant correlation with heart disease (r = -0.40, p<0.001), so what happens when we run a model using both folate and wheat as exposures? Initially, it looks like wheat clobbers folate as a predictor (beta = 0.59, p<0.001 versus beta = -0.06, p = 0.39)—which would suggest that, although China’s wheat-eaters tend to have lower folate levels, folate deficiency itself isn’t enough to explain the link with heart disease.

But I’m not ready to dismiss this one just yet. As often happens with plasma measurements and health conditions, folate may have a nonlinear relationship with heart disease—which means multiple regressions (of the linear variety) won’t show the full picture. Indeed, when I make a scatter plot for folate levels and coronary heart disease, it looks like a bit of a curve emerges, with folate being most strongly associated with heart disease when the county average dips below 10 micrograms per liter (or thereabouts). Above that, the correlation is far less dramatic.

So how do we deal with this statistical monkey wrench? For starters, I tried transforming the folate data to make it more suitable for linear regressions, but that didn’t do diddly squat to the results: The numbers were beta = 0.58, p<0.001 for wheat and beta = -0.06, p = 0.31 for folate. So then I tried stratifying the data based on “low” and “high” folate levels (10 or less micrograms/liter versus 10.1 or more micrograms/liter), but both subgroups continued showing wheat as strongly and significantly correlated with heart disease while folate was off the hook.

Just to cover my bases (and because I’m a stubborn son-of-a-gun), I kept playing  with the numbers for a while longer to see if I could excavate anything new. Nope. Bottom line: It looks like wheat is predictive of heart disease whether or not folate levels are low, whereas folate is mostly predictive of heart disease only in the presence of high levels of wheat consumption.

So, theory #1 doesn’t pan out. Bugger. But bear in mind, we’re using folate mostly as a marker for elevated homocysteine, so these results don’t mean that homocysteine itself isn’t playing a role. Other causes of high homocysteine, such as B12 deficiency, weren’t documented in the China Study data. So this is an issue that’ll have to remain annoyingly unresolved. Another bugger!

Onto the next theory: latitude. Could the folks living in northern wheat-eating regions have lower vitamin D levels, leading to more heart problems—and creating a false link between wheat and cardiovascular disease? I admit, this was my favored theory after folate, but it ain’t holdin’ water. When I run wheat and latitude together as potential contributors to heart disease, wheat remains strongly predictive (beta = 0.65, p<0.001), while latitude diminishes (beta = -0.01, p=0.96). It’s pretty clear that the raw correlation between heart disease and latitude (which is 0.43, p<0.01) is just an echo of the relationship between heart diseases and wheat-eating regions, which are typically northern.

Okay, so that’s two strikes for Denise’s heart disease theories. What about fish and DHA? Are the wheat eaters suffering due to their fishless (and low-in-DHA) diets rather than from wheat itself? Alas, it doesn’t look likely. When I run these things together as exposures for heart disease, wheat stays strongly predictive (beta = 0.68, p<0.001) while the fishies do not (beta = 0.08, p = 0.47). Likewise, DHA teeters out into statistical insignificance (beta = 0.06, p = 0.30) when used in a model with wheat.

(Wait, I know what you’re thinking! “Why does it look like fish and DHA contribute positively to heart disease?” It’s because many of the fish-eating regions are more industrialized, and—in the absence of wheat—the fish-heart disease relationship is confounded by other factors like more desk work, more smoking (especially manufactured cigarettes), less physical activity, more vegetable oil consumption, and so forth. When we add some more variables to the model that take away the “city effect” associated with fish—such as apo-B, tobacco use, or percentage of the population employed in agriculture—then both fish and DHA turn inverse again. Although wheat, it should be mentioned, stays rock-steady in its high coefficient and statistical significance.)

Other stuff

Milk. Is moo juice a cardiovascular foe obscuring the relationship between wheat and heart disease? Probably not, according to the data—which isn’t surprising, given how few counties even drink the stuff. When running daily milk intake alongside wheat intake, wheat keeps its positive correlation (beta = 0.67, p<0.001) and milk actually turns a bit inverse, though not significantly so (beta = -0.07, p=0.47). No model shows a significant association between milk and cardiovascular disease, so I’m crossing this one off the list of potential confounders.

Blood pressure, BMI, corn, millet, sorghum, rice, added animal fat, added vegetable fat, total fat, total animal food, total carbs, total protein, percent of calories from animal protein, and all the smoking/tobacco variables I tried became statistically nonsignificant (in relation to heart disease) when thrown into a model with wheat.

Income is positively associated with heart disease when wheat is held constant, but it still doesn’t put a ding in wheat’s association with heart disease.

Models with more variables

So apparently comparing wheat + one other independent variable isn’t enough to explain the Wheat Effect. Not even a little bit. But maybe, just maybe, a bigger combination of variables will do the trick. Perhaps wheat-eating regions just host a collection of heart-harming factors (low folate, low vitamin D, low EFAs, and so forth) that, together, are more powerful predictors of disease than the variable wheat.

Here are the variables I’m interested in looking at. Some could be causative and some could be preventative:

• Wheat consumption
• Corn consumption
• Millet consumption
• Rice consumption
• Total blood cholesterol
• LDL cholesterol
• HDL cholesterol
• Apolipoprotein-B
• DHA levels
• Folate levels
• Latitude
• Added vegetable oil
• Blood pressure
• Weight
• BMI
• Total fat intake
• Total monounsaturated fat intake
• Total polyunsaturated fat intake
• Total saturated fat intake
• Percent of calories as fat
• Percent of calories as carbohydrates
• Total animal protein intake
• Total plant food intake (by weight)
• Total animal food intake (by weight)
• Green vegetables (daily, not yearly)
• Vitamin C intake
• Total sodium intake
• Poultry consumption
• Egg consumption
• Red meat consumption
• All meat consumption
• Fish consumption
• Dietary cholesterol intake
• Percent of the population currently smoking
• Percent of the population who have ever smoked tobacco
• Percent of the population smoking manufactured cigarettes
• Percent of the population pipe smoking
• Percent of the population smoking cigars
• Percent of the population working in industry (typically less physical activity)
• Percent of the population working in agriculture (typically more physical activity)

I won’t bore you with the results of every single combination I tried (over 100), so here’s the gist. No matter what model I use, wheat always adds unique variance. That means wheat (or an undocumented variable associated with wheat) is contributing something to heart disease that these other variables can’t account for. No combination out of the above bumped the association between wheat and heart disease out of the “statistically significant” zone.

Incidentally, one model had the best fit out of all the others for explaining heart disease:

1. Wheat consumption (beta = 0.62, p<0.001)
2. Apolipoprotein B (beta  = 0.38, p<0.001)
3. Total cholesterol (beta = -0.22, p<0.05)

Note that the number for total cholesterol is inverse, meaning higher cholesterol was associated with less heart disease—at least in this specific model. Unless you’re an Ancel Keys groupie, this may actually be quite plausible.

Anyway, here’s the important point. No matter what variables I adjust for, I can’t make the correlation between wheat flour and heart disease go away. Sorry, wheat! Neener neener.

Cardiovascular disease: The only “Western” problem without “Western” risk factors

Here’s a mystery for ya.

In the China Study data, most Western diseases (such as breast cancer, colon cancer, lung cancer, and diabetes) are concentrated in areas that share some key characteristics: more industrial employment, less agricultural work, greater population density, and often higher levels of schooling. Folks here eat more processed starch and sugar, use more polyunsaturated vegetable oils, chug down more beer, smoke more manufactured cigarettes, and typically get less physical activity than their neighbors in pastoral communities.

In other words, the Western-disease-prone-regions are like baby Americas—slowly waddling, diapered and naive, towards the motherly lap of disease.

Most likely, these Western ailments aren’t spawned from a single food or activity, but from a tragic mix of diet choices, lifestyle habits, and environmental factors. For problems like breast cancer and colon cancer and lung cancer, it’s pretty easy to see what the matrix of risk-raisers are from looking at the data: It’s the same combination of things spurring disease in Western nations.

But oddly enough, this isn’t the case for heart conditions. The factors shared by other Western illnesses are not, in most cases, associated with heart disease in this data set. If you’ve read some of the earlier China Study posts, you might remember that I took issue with Campbell’s disease-clustering strategy because heart disease doesn’t fit cleanly with the “diseases of affluence” group, despite his insistence on sticking it there anyway. Unlike the other Western problems, heart disease isn’t associated with eating more sugar, working in industry, drinking more alcohol, using vegetable oils, having higher apo-B levels, or any of the other variables uniting the Western diseases and mirroring the traits common to industrialized countries.

What’s the only thing heart-disease-prone regions have in common with Westernized nations? That’s right: consumption of high amounts of wheat flour.

Food for thought. Kinda spooky.

Wheat eaters: fatter with fewer calories

Here’s some more weirdness. In both China Study I and II, wheat is the strongest positive predictor of body weight (r = 0.65, p<0.001) out of any diet variable. And it’s not just because wheat eaters are taller, either, because wheat consumption also strongly correlates with body mass index (r = 0.58, p<0.001):  

How odd! This aligns with a post Stephan Guyenet at Whole Health Source wrote about wheat consumption and obesity in China, speculating that wheat might wreak metabolic havoc wherever it goes—a trend that becomes apparent when comparing similar populations of wheat eaters and non-wheat eaters, such as in China. But perhaps there’s some confounding going on. What about calorie intake? Are the wheat eaters just scarfing down more food in general, leading to higher weight regardless of wheat consumption? Doesn’t look like it. Running wheat and calorie intake together as predictors with BMI as the outcome, wheat takes the weight-gaining gold:

• Wheat: beta = 0.56, p<0.001
• Calorie intake: beta = 0.13, p = 0.19

Unfortunately, we have no way of accounting for energy expenditure through physical activity—but considering wheat-eating regions tend to be pastoral and dominated by agricultural work, it seems they’d be burning through a greater wallop of calories than more sedentary regions. Indeed, independent of calorie intake, there’s a clear association between agricultural work and weight (lower) versus industry work and weight (higher), suggesting these things could be approximate measures of calorie expenditure. So once again, we’ve got a paradox: The wheat eaters are consuming lower or average levels of calories, doing more physical labor, and yet… they’re fatter.

Out of curiosity, I ran a stepwise regression on a bunch of relevant variables to see what combination would best predict BMI. (In statistics, stepwise regression is a really cool, but sometimes totally misleading method for building a statistical model. It involves adding (or winnowing away) variables one by one based on how they behave together and contribute to the outcome—BMI, in this case—until you’ve got a model where each variable offers significant variation and the highest possible percent of explanation (represented as r-squared). Unfortunately, since this process is automated and computers usually don’t understand the whole “biological plausibility” thing, you can wind up with weird models that don’t make sense in the real world. Nonetheless, it can be a worthwhile method if used with caution.)

Setting BMI as the outcome, I chose the following variables as potential exposures:

• Total calories
• Total fat
• Total carbohydrates
• Total plant food
• Total animal food
• Total plant protein
• Total animal protein
• Total monunsaturated fat intake
• Total saturated fat intake
• Total polyunsaturated fat intake
• Red meat
• All meat
• Fish
• Poultry
• Eggs
• Wheat flour
• Corn
• Millet
• Legumes
• Starchy tubers
• Green vegetables (daily, not yearly)
• Agricultural employment
• Industrial employment

(I left out milk because so few counties consumed it.)

The best-fitting model for predicting BMI (at 95% confidence)? Drum roll please. Three variables made the cut.

1. Eating more wheat flour (beta = 0.48, p<0.001)
2. Eating more polyunsaturated fat (beta = 0.44, p<0.001), and
3. Eating fewer green vegetables (beta = -0.29, p<0.01).

This model has an r-squared value of 0.53, meaning it predicts a little over half of the variation in BMI—at least in theory. That’s actually pretty high, considering we haven’t directly factored things like physical activity into the equation.

Interesting, eh? All animal foods and total dietary fat, by the way, were completely insignificant in terms of BMI.

Of course, there could be other variables involved that the China Study didn’t cover. Were the higher-BMI folks also more heavily muscled (perhaps from more physical labor), increasing their body weight but not body fat? Are the wheat eaters, some of whom are ethnic minorities in China (especially Turkic and Mongolian), genetically “bigger” than the Han Chinese? There are plenty of unknowns, and alas, no way to clarify them based on this data.

I guess we’ll leave it as a question mark for now.

Grain damage: Do other studies back it up?

But don’t those peer-reviewed, scientific studies tell us wheat is healthy? Alas, the vast majority of studies on grains—especially wheat—showcase at least one of the following problems:

• They look at the effects of whole grains versus refined grains—not whole grains versus the same diet with no grains at all.
• Study subjects increase their consumption of whole grains, and this displaces some portion of yuckfoods (processed junk, white-flour products, sugary things, and so forth). As a result, it’s hard to tell whether any health perks are due to the addition of whole grains, or from the reduction of truly-awful-for-you foods. This is particularly true in studies that scout out disease patterns in populations rather than controlled studies that measure specific changes that occur with the addition of whole grains.
• They don’t adequately account for other factors that often accompany whole-grain consumption, like a greater level of health consciousness, more exercise, other positive diet choices, and so forth.

However, a few gems are lurking in the massive slush-pile of irrelevant studies. This one’s pretty doggone interesting, and it’s from all the way back in 1959: “Comparisons of atherogenesis in rabbits fed liquid oil, hydrogenated oil, wheat germ and sucrose.” You can click on that for the full-text PDF.

As you might guess from the title, this study examines the effects of diet on the development of atherosclerosis—AKA hardening of the arteries. The researchers took cholesterol-infused rabbit food and supplemented it with liquid corn oil (yuck), hydrogenated corn oil (double yuck), wheat germ (mystery murderer?), and sucrose (sweet poison!). Sorry, I dig hyperbole. Anyway, part of the goal was to create an experiment testing the hypothesis that “the geographic differences in the incidence of coronary disease might be related to selective hydrogenation of polyunsaturated fatty acids or to degermination of cereals.”

So now, the moment of truth: Which group had the most severe atherogenesis? Perhaps the one fed the nasty hydrogenated oil, as hypothesized? Ladies and gentlemen, place your bets. From the article:

The most severe atherogenesis occurred in the animals on the wheat germ diet.

Was it a fluke? Probably not:

In an earlier study, we maintained 5 groups of 5 rabbits each for three months on 500 mg of cholesterol daily and rabbit chow supplemented with different fats or with wheat germ. Here also, the animals on the wheat germ diet showed a significantly greater degree of atheromatous lesions than the animals on rabbit chow plus 20% corn oil, cottonseed oil or hydrogenated cottonseed oil, whereas no significant difference was found between the various fats.

So what made the wheat germ contribute to atherogenesis? The researchers state that it’s “difficult to speculate” about the mechanism, which is a scientific way of saying “We dunno.” They suggest the extra dietary protein from wheat germ could be the cause, but from the literature I’ve skimmed so far, it looks like plant proteins don’t have much effect on bunnies (although animal protein does).

Of course, rabbits are truly terrible models for anything that happens in the human body. They’re hardcore herbivores. A mere billowing of the wind is practically enough to spike their cholesterol. But what explains the specific effect of wheat germ on their poor arteries? Could this have implications for humans?

My answer: It’s “difficult to speculate.”

Other studies

Prefer human studies? Me too. Here’s one that initially looks totally irrelevant but is actually pretty interesting: Flaxseed and cardiovascular risk factors: Results from a double blind, randomized, controlled clinical trial. (This also a stellar example of why it’s important to read full-text articles instead of just abstracts, which often don’t tell you diddly about the stuff you want to know.)

This particular study charted the effects of flaxseed on adults with high cholesterol. One group got food with ground flaxseed; the other group got food with added wheat bran. Other dietary elements were the same. (Low fat, low cholesterol. Fun times!)

The results? Ye Olde Flaxseed Group did pretty well: Compared to their baseline measurements, these folks had lower insulin, lower blood glucose, lower C-reactive protein (a marker for inflammation), and better insulin sensitivity (as calculated by HOMA-IR).

But poor Wheat Group was less fortunate. Since the study was about flaxseed, the results of wheat aren’t specifically discussed, but check out “Table 4″ in the link above to see the numbers for yourself. The wheat-bran eaters had a 14.9% increase in insulin resistance (calculated by HOMA-IR) and a 9.3% increase in C-reactive protein. In other words, they lost some insulin sensitivity and gained some inflammation—two risk factors for heart disease. Hmm. Was the wheat bran to blame? Some other element of the control diet? It’s impossible to say for sure based on this study, but considering the wheat group’s adverse effects were more dramatic than the flaxseed group’s benefits, it seems a little suspect.

(A rather abrupt end of part one! The next post will have some more studies and speculations on potential mechanisms for wheat as causative of heart disease.)

“The China Study”:

A Formal Analysis and Response

Denise Minger
deniseminger@gmail.com
August 2, 2010

Introduction

SECTION 1: Reiteration and Expansion of Criticisms

1. Linkage of animal protein with cancer by way of cholesterol
2. Misleading association of breast cancer with lipid intake and lipid intake with animal protein
3. Supposition that plasma cholesterol increases liver cancer risk
4. Misrepresentation of heart-protective effects of green vegetables, and the three-variable linkage between animal protein, apolipoprotein B, and cardiovascular disease
5. Biased use of unadjusted univariate correlations to confer protective benefits of plant foods but not with animal foods
6. Use of a three-variable chain to connect animal foods with “Western” diseases
7. Unexplored role of blood glucose, insulin, and disease
8. Dismissing relevant variables
9. Errors in the extrapolation of casein to all animal protein

SECTION 2: Biological Models and Cited Papers

1. Breast cancer
2. Liver cancer
3. Energy utilization
4. Affluent-poverty diseases
5. Summary

SECTION 3: Response to Points Raised by Campbell

1. Wheat: confounded variable or legitimate concern?
2. Selection of univariate correlations and confirmation bias
3. Tuoli county and erroneous data
4. Whole-food, plant-based diets versus whole-food diets with animal products

Conclusion

References

Introduction

When I first embarked on an analysis of T. Colin Campbell’s The China Study, I did not anticipate the range or magnitude of responses it would invoke—reactions that have been at times controversial, at times impassioned, and at times downright heated, but above all else intellectually provocative. It seems “The China Study” is a book that, in many cases, is either intensely revered or vehemently criticized, and its ability to generate ongoing discussion signifies a deep-seated division in the scientific community.

I would like to thank Dr. Campbell for his cordial response to my critique, as well as for the time he has taken to elucidate his philosophy of nutrition and his approach to research. While I do not agree with some of his conclusions, I honor his contributions to the field of health and nutrition, and deeply admire his courage to promote an unpopular message amidst a research sector dominated by special interests and opposing views.

I propose that Campbell’s hypothesis is not altogether wrong but, more accurately, incomplete. While he has skillfully identified the importance of whole, unprocessed foods in achieving and maintaining health, his focus on wedding animal products with disease has come at the expense of exploring—or even acknowledging—the presence of other diet-disease patterns that may be stronger, more relevant, and ultimately more imperative for public health and nutritional research.

Having lit a proverbial fuse, I feel called and compelled to make the sum of my findings available to the public so that they may add, in whatever extent or direction, to the symphony of voices engaged in this discourse. My intent with this paper is not to discredit Campbell as a scientist, nor to promote or discourage a particular diet—but rather, to present new ways of looking at the China Study data and related research while highlighting the shortcomings in Campbell’s specific conclusions. I hope this information can be valuable to readers while—above all else—encouraging the use of independent, critical thought to advance our understanding of health.

Section 1:

Reiteration and Expansion of Criticisms

Although some of the following points have been discussed previously, they were largely dismissed as “reductionist.” Given Campbell’s preference to examine nutrition from a holistic perspective with less focus on individual components, his assessment is understandable, albeit inaccurate. I cite these points not to split nutritional hairs, but to reveal a consistent pattern of bias and misrepresentation as it relates to Campbell’s hypothesis.

Here, I present my original points once more with additional information and references to highlight their relevance.

1. An attempt to link animal protein with cancer by way of cholesterol—a chain of variables that exhibits several logical and statistical shortcomings.

In citing the China Study data, Campbell states that total cholesterol is “positively associated with most cancer mortality rates” and also “positively associated with animal protein intake.”[1] However, he provides no indication that he examined or accounted for the cancer-risk-raising variables associated with cholesterol, including schistosomiasis and hepatitis B infection.[2]

Additionally, per Campbell’s own assessment, cholesterol is only one of several variables that tend to cluster alongside Western-type diseases: The others include higher blood glucose levels, increased consumption of refined carbohydrates, higher beer intake, and industrial rather than agricultural employment[3]^,[4]—with the latter bringing changes in lifestyle and increased work hazards such as benzene exposure, an extensively studied cause of lung cancer, leukemia, and other lymphatic and hematopoietic malignancies in Chinese factory workers.[5]^,[6]

This entanglement of risk-raising factors casts doubt on the usefulness of cholesterol as an indicator of animal food consumption rather than of accompanying variables—especially considering the lack of a known biological mechanism that causes cholesterol to rise from increased protein consumption.

As previously mentioned, Campbell also fails to cite direct links between animal foods themselves and cancer, relying instead on biomarkers as a liaison. Since epidemiological data can only identify trends and not cause-and-effect sequences, and because some diseases intrinsically alter blood profiles, the assumption that higher cholesterol precedes disease is also unsubstantiated. To link specific foods or a category of foods with disease requires evidence that the foods themselves—independent of confounding factors—influence disease risk; the reliance on biomarkers that only partially relate to dietary items is too indirect to yield truly meaningful conclusions.

2. The association of breast cancer with lipid intake—and lipid intake with animal protein consumption—as a means to link breast cancer with animal foods.

Although many of Campbell’s observations about breast cancer in rural China align with widely accepted risk factors, such as earlier menarche and greater exposure to hormones, Campbell relies on an intermediary variable (fat consumption) to forge a link that does not exist directly between animal foods and breast cancer. If an intermediary variable is introduced, a positive association can superficially emerge where the direct association is actually neutral or mildly negative. On page 86 of “The China Study,” Campbell suggests that in China, “the association between fat and breast cancer might really be telling us that as consumption of animal-based foods goes up, so does breast cancer”—an idea spawned from his observation that animal protein consumption correlates strongly with lipid intake. On the same page, he notes that the correlation between fat consumption and animal protein was “very high, at 70 – 84%”—with 70% expressing the linear relationship between animal protein and percentage of calories as fat, and 84% expressing the linear relationship between animal protein and total lipid intake.[7]

However, these figures—especially the persuasively high 84%—may be overestimated. Campbell has noted that the data for the county with the highest animal protein and fat intake, Tuoli, was “clearly not accurate on the 3 days that the data were being collected,” because “on those days, they were essentially eating as if it were a feast to impress the survey team.”[8] Further, Campbell states that Tuoli was “intentionally excluded from virtually all our analyses” because of its misleading values for meat intake.[9]

Tuoli County was not, however, excluded from the calculation for the association between animal protein consumption and lipid intake, which has a correlation of 0.84 or 84% only when using all 65 counties. A visual graph of the data reveals Tuoli’s strong influence in this correlation, as represented by the data point at the far right.

If the Tuoli data is ultimately unreliable and reflects a short-term “feast” of meat, its inclusion in the animal protein/lipid correlation may be erroneous. Since Campbell has provided no method for correcting the data nor indicated what more reliable values may be, omitting this county from the calculation may be warranted.

With the removal of Tuoli, the correlation between animal protein intake and total lipid intake drops from 0.84 to 0.52 (p<0.001). While still high and statistically significant, this correlation is low enough to undermine Campbell’s immediate implication of animal foods with breast cancer via lipid consumption.

More importantly, once the data set is corrected for Tuoli County, the correlation between plant oils and total fat intake rises drastically. Using all 65 counties, the variable “Oil intake other than rapeseed”—which includes soybean oil, corn oil, cottonseed oil, peanut oil, and sesame oil—has a non-significant linear correlation of 0.18,[10] and the variable “Rapeseed oil intake” correlates at 0.08.[11] When Tuoli is removed from calculation, these numbers rise to 0.42 (p<0.001) and 0.25 (p=0.514), respectively.

Using the data set with the flawed inclusion of Tuoli, Campbell cites a strong association between animal protein and lipid intake as a reason to implicate animal foods with breast cancer. Yet using the revised data set, animal foods do not contribute significantly more fat to total lipid intake than do plant oils. As a result, any association between breast cancer and dietary fat could be linked to either animal or plant-sourced foods, and there is no justification for indicting only animal products.

This may be particularly relevant because, according to one of Campbell’s publications, certain plant oils such as corn have carcinogenic properties; he notes that “Increased intake of corn oil has previously been shown to promote the development of L-azaserine-induced preneoplastic lesions in rats,”[12] a similar phenomenon to what his research demonstrated with casein.

Nonetheless, the association with lipid intake and breast cancer may not even be noteworthy. From his paper “Additional ecological evidence: lipids and breast cancer mortality among women age 55 and over in China,” Campbell et al conclude:

Although the result is consistent with a positive association between lipid intake and breast cancer risk, the observed association is weaker than the association previously observed. This finding provides only modest support for the possibility of a diet-breast cancer link.[13]

Neither the association between animal foods and lipid intake nor lipid intake and breast cancer, then, is particularly strong. This casts doubt on the accuracy of Campbell’s statement that the China Study data showed a “connection of breast cancer with dietary fat, [and] thus with animal-based foods.”[14]

Moreover, another variable may be more relevant than lipids when exploring the mechanisms behind breast cancer occurrence. Although Campbell emphasizes the importance of biological models and clinical research to corroborate epidemiological data, particularly univariate correlations, he does not examine the positive correlations between breast cancer and blood glucose or processed starch and sugar consumption in the China Study data,[15] even in light of research showing these associations may be highly relevant. Research spanning the previous decade has revealed a potential role of blood glucose levels in the development of breast cancer,[16] has linked hyperinsulinaemia with both early menarche and breast cancer,[17] and has shown that high insulin levels are a risk factor for breast cancer independent of estrogen.[18] Additional discussion of the relationship between blood glucose, insulin, and cancer is included under item 6.

The animal food/breast cancer hypothesis is directly testable in the China Study data, given the records for total animal protein consumption as well as amount and frequency of meat intake, fish intake, dairy intake, and egg intake. Yet despite his stated hypothesis that animal-based foods correspond with increased breast cancer rates, Campbell provides no indication of excavating a direct link through any of his analyses.

3. The claim that animal products and plasma cholesterol increase rates of liver cancer in high-risk populations.

Stratified data shows that high-risk areas for liver cancer have a nearly neutral association with animal food intake, and inverse associations with meat, eggs, and dairy.[19] Of all animal food variables, only fish and fish protein are strongly correlated with liver cancer[20]—an issue Campbell et al address in the publication “Fish consumption, blood docosahexaenoic acid and chronic diseases in Chinese rural populations”:

[It] is not difficult to visualise the reason for the link with liver cancer [and fish consumption]. The coastal, estuarine and lacustrine regions with the high fish and sea food intakes are also those with the highest humidities. Storage of food in regions of high humidity is known to encourage the spread and growth of hepatitis B virus and Aspergillus flavus which produces aflatoxin, both are major causes of primary carcinoma of the liver.[21]

Moreover, an identifiable relationship between cholesterol and liver cancer does not prove causality. Although Campbell implies that elevated cholesterol levels increase liver cancer risk, the inverse may be true: Hypercholesterolemia has been identified as a complication of liver cancer,[22] often in conjunction with hypoglycaemia and hypercalcaemia.[23]

4. The incomplete statement that cardiovascular disease is inversely associated with green vegetable consumption, and the three-variable linkage between animal protein intake, apolipoprotein B (apo-B), and cardiovascular disease.

In his 1998 publication “Diet, lifestyle, and the etiology of coronary artery disease: the Cornell China Study,” Campbell states that coronary artery disease mortality rates were “inversely associated with the frequency of intake of green vegetables (r = -0.43, p<0.01)”[24] in rural Chinese populations—a statement cited elsewhere as a significant finding of the China Study.[25]

Although frequency of green vegetable consumption does boast a strong inverse correlation with heart disease in the unadjusted data, the actual amount of green vegetables consumed has a weak positive correlation (r = 0.05)[26]—a paradox Campbell does not mention or seem to explore. Had Campbell examined this discrepancy closer, he would notice the strong regional patterns associated with frequency of green vegetable consumption, including humidity (r = 0.68, p<0.001), heat (r = 0.61, p<0.001), elevation (r = -0.48, p<0.001), and latitude (r = -0.60, p<0.001),[27] all of which suggest this variable serves as a geographical marker and thus is likely associated with other regional risk factors and protective factors for disease.

While Campbell has noted he prefers to view nutritional patterns in the aggregate rather than individually, the “green vegetable paradox,” as I’ve termed it, is representative of similar and repeated oversights, potentially weakening his hypothesis as a whole. When referring to the China Study data, Campbell cites misleading figures when they endorse plant food consumption—without first completing the analytical steps necessary to prove their accuracy and eliminate confounding. Likewise, he consistently omits similar correlations that indicate a neutral or protective effect between animal foods and disease, even when those trends, too, seem to form an overarching pattern.

Furthermore, Campbell cites a chain of three variables to implicate animal protein with coronary heart disease: He notes that animal protein associates with the low-density lipoprotein fraction apo-B, that apo-B associates with increased mortality from coronary artery disease, and therefore concludes that animal protein associates with heart disease.[28] Although the first two statements are correct in isolation, the leap to the latter is unsupported by logic and contradicted by the China Study data.

While Campbell found it appropriate to cite an unadjusted correlation for frequency of green vegetable intake, had he done the same for animal food variables, he would find only neutral or inverse correlations between cardiovascular disease and:

• amount of meat consumed (r = -0.28)*
• frequency of meat consumption (r = -0.15)
• amount of fish consumed (r = -0.15)
• frequency of fish consumption (r = -0.14)
• amount of eggs consumed (r = -0.13)
• frequency of egg consumption (r = -0.14)
• animal protein intake (r = 0.01)
• fish protein intake (r = -0.11)[29]

* When Tuoli county is omitted, this correlation becomes r = -0.36, p<0.05. Other listed correlations do not significantly change.

The only animal food with a positive (though still not significant) correlation with heart disease is dairy, both in amount (r = 0.06 for full data set; r = 0.12 adjusted for Tuoli) and frequency (r = 0.11 for full data set; 0.12 adjusted for Tuoli).[30] However, considering dairy is generally only consumed in three counties, the accuracy of these correlations is difficult to determine.

Non-dairy animal foods do not consistently correlate with shared geographical features,[31] vegetable consumption,[32] or plasma antioxidants,[33] thus minimizing the possibility of common protective factors masking their true effect on heart disease. In light of this—and given Campbell’s interest in finding “overarching” themes in nutrition—it seems curious he does not explore the consistent inverse relationships between most animal foods and cardiovascular disease. While it is possible that any or all of these figures require additional adjustments to account for confounding, Campbell offers no indication that he did so for patterns with vegetable consumption before embracing its inverse association with heart disease.

Likewise, while Campbell readily accepts favorable correlations with plant-food variables and disease, he does not account for the numerous correlations that run contrary to his hypothesis—particularly the association between all non-rice grains and heart disease, including: yearly ration of wheat (r = 0.51, p<0.001), yearly ration of corn (r = 0.31, p<0.05), yearly ration of sorghum (r = 0.31, p<0.05), yearly ration of millet (r = 0.37, p<0.05), wheat flour per day (r = 0.67, p<0.001), and “other cereal” intake per day, which includes corn, millet, sorghum, and barley (r = 0.39, p<0.01). Additional plant variables exhibit a positive association as well, including total fiber, per food composite (r = 0.30, p<0.05) and total plant protein, per food composite (r = 0.21), likely due to the influence of these grains.

Because these are unadjusted correlations, they are only preliminary and illustrative, not conclusive. However, if Campbell is seeking overarching patterns of nutrition and disease, it seems this is one worth inspecting.

Lastly, although Campbell also cites a correlation between apo-B and cardiovascular disease, the biomarker he typically uses to connect animal foods with health afflictions—total plasma cholesterol—is ineffective in the case of heart disease. In the paper “Erythrocyte fatty acids, plasma lipids, and cardiovascular disease in rural China,” Campbell et al conclude:

Within China neither plasma total cholesterol nor LDL cholesterol was associated with CVD [cardiovascular disease]. The results indicate that geographical differences in CVD mortality within China are caused primarily by factors other than dietary or plasma cholesterol.[34]

5. The use of unadjusted univariate correlations to link fiber with reduced colorectal cancer rates, green vegetables with reduced stomach cancer rates, and plant-food biomarkers with reduced stomach cancer rates.

Drawing from unadjusted China Study data, Campbell cites several perceived effects of plant foods and plant biomarkers on colorectal health and stomach cancer. While Campbell has stated his approach to nutrition is one of holism rather than reductionism, repeated use of unreliable correlations ultimately weakens the hypothesis they help construct.

In citing an inverse association between 14 fiber fractions and colorectal cancer,[35] Campbell provides no indication that he tested for confounding variables—a significant oversight, given that schistosomiasis infection appears to be both a major risk factor for colorectal cancer (r = 0.89, p<0.001) and less common in regions with high fiber consumption (r = -0.23 for total fiber intake). As demonstrated visually and verbally in my first response to Campbell, the protective effect of each fiber faction convincingly echoes its correlation with schistosomiasis, suggesting schistosomiasis may potentially create or accentuate the inverse relationship between fiber and colorectal cancers.[36]

Similarly, Campbell cites other unadjusted correlations related to plant foods. He notes an inverse association with stomach cancer and green vegetable intake, plasma beta-carotene, and plasma vitamin C[37]—patterns aligning with his hypothesis that plant foods are protective against disease.

The problem isn’t that the correlations are invalid; they may, after more analysis, persist. However, Campbell selects them above stronger associations that contradict his hypothesis and provides no indication of adjusting for appropriate factors. This approach allows him to build a repertoire of supportive evidence that may only be superficially congruous with his hypothesis. Scientific rigor mandates more even-handed analyses, which Campbell has not done.

6. The use of a three-variable chain to connect animal-based foods with cholesterol and cholesterol with “Western” diseases.

To form a comprehensive method for examining disease patterns, Campbell creates two dichotomous sets of diseases—one associated with affluent living conditions and one associated with poverty. While searching for underlying nutritional patterns characterizing the diseases of affluence, he observes that plasma cholesterol has a positive correlation with the collective group, and concludes that “one of the strongest predictors of Western diseases was blood cholesterol.”[38]

Given that a variety of factors—dietary and otherwise—can influence cholesterol and the cause-and-effect relationship between cholesterol and disease is not always clear, Campbell’s use of cholesterol as an intermediary between animal foods and disease is unsubstantiated. For instance, a shift from a highly active lifestyle in agriculture-dominated regions to a more sedentary one in industrialized areas may, in itself, be enough to explain higher cholesterol levels in certain areas[39]—a plausible theory, given that regions with greater industry employment in the China Study tended to have higher total cholesterol (r = 0.45, p<0.001) but exhibited no significant association with animal protein intake (r = 0.04).[40]

More importantly, a different plasma variable may be even more relevant than cholesterol in the occurrence of Western diseases: blood glucose.

Blood glucose: an overlooked clue

Although Campbell cites cholesterol as a consistent risk factor for Western diseases, blood glucose also exhibits associations with the chronic conditions most prevalent in affluent nations—including the ones less convincingly linked to cholesterol.

Notably, blood glucose tends to have a distinct nonlinear relationship with disease, and its association with cardiovascular diseases, cancer, and diabetes may therefore appear diminished when studying only linear correlations, as Campbell has generally done with cholesterol. The following graphs present the subtle “U-curve” pattern between blood glucose and diseases common to Western nations, juxtaposed with the same diseases and cholesterol.

On all left-side graphs, the horizontal axis represents blood glucose levels as mg/dL, while the vertical axis represents disease mortality per 1000 deaths. On all right-side graphs, the horizontal axis represents total cholesterol as mg/dL, while the vertical axis represents disease mortality per 1000 deaths.

Acronyms: MI/CHD = myocardial infarction/coronary heart disease; BC = breast cancer; LC = lung cancer; CC = colorectal cancer.

An examination of the Third National Health and Nutrition Examination Survey (NHANES) revealed that every 50 mg/dL increase in plasma glucose corresponded with a 22% increase in cancer mortality risk, possibly due to the proliferative and anti-apoptotic properties of glucose and insulin.[41] In another study, a cohort of more than 140,000 Austrian adults revealed that elevated blood glucose is associated with multiple cancers in both men and women, including liver cancer, breast cancer, bladder cancer, gallbladder cancer, non-Hodgkin’s lymphoma, and thyroid cancer.[42]

Although averaged blood glucose levels in China all fell within the normal range, if Campbell hypothesizes that increases in even very low cholesterol collide with greater disease risk, it would not be a stretch—by his logic—to suggest that small increases in even normal-range blood glucose could do the same.

On pages 78-79 of “The China Study,” Campbell writes:

As blood cholesterol decreased from 170 mg/dL to 90 mg/dL, cancers of the liver, rectum, colon, male lung, female lung, breast, childhood leukemia, adult leukemia, childhood brain, adult brain, stomach and esophagus (throat) decreased.

However, citing the same univariate correlations, he could have also written:

As plasma glucose decreased, cancers of the liver, rectum, colon, male lung, female lung, breast, childhood leukemia, adult leukemia, childhood brain, adult brain, stomach, bladder, and cervix, as well as childhood and adult lymphoma, decreased.

In addition to potentially spurring cancer growth, blood glucose and insulin may play a pivotal role in the development of atherosclerosis. A 2009 paper by Nunes and Silva[43] reveals that, “among several systemic parameters studied, plasma glucose was found to be correlated to coronary artery atherosclerosis lesions” and remained strongly correlated after accounting for other variables. The researchers note:

In the context of the present investigation, one may speculate that higher plasma glucose, probably in the presence of elevated plasma insulin, could be associated to a growth-stimulating effect on atherosclerotic lesions, perhaps involving magnesium as a cofactor for insulin-stimulated growth.

In addition, Nunes and Silva mention that “in the present study, we could find no evidence of an association between lipid fractions and CADB [coronary artery disease burden].”

A separate study by Brunner et al determined that glucose intolerance associates with increased mortality risk from all causes, stroke, and respiratory disease. The researchers state that the findings of their study “provide clear evidence that coronary mortality is raised among those with marginal postload hyperglycemia,”[44] and suggest several possible explanatory mechanisms:

A raised glucose level at baseline may indicate emerging insulin resistance and a downward trajectory in glycemic control, with increased risk of glucose intolerance, diabetes, and CHD in subsequent years. … Other pathways include oxidative stress and formation of advanced glycation end products that accelerate atherosclerosis when blood glucose is only slightly raised.

It is, of course, impossible to determine a cause-and-effect relationship in epidemiological data, but given that there is already corroborating research and biological plausibility for links between glucose, insulin, and disease, it seems the glucose-disease trend is one worth exploring. A combination of diet and lifestyle changes—especially ones associated with industry-dominated regions, where diseases of affluence tended to occur—could serve as precursors for metabolic abnormalities and insulin resistance, contributing to or occurring alongside heightened risk for cardiovascular disease and many cancers.[45]^,[46]

Dismissing other relevant variables

Campbell claims that “even small increases in the consumption of animal-based foods” were “associated with increased disease risk” in the China Study data,[47] but dismisses other relevant food variables because they are eaten in lower quantities in China than in most Western nations, and are thus “probably more indicative of general economic conditions and other local circumstances than of biological relationships to disease.”[48] The illogic in this statement becomes apparent when comparing the actual ranges of these variables. For instance, once the outlier Tuoli is omitted, animal protein consumption in China ranged from 0g to 23.92g per day, whereas processed sugar and starch consumption ranged from 0g to 27g and beer intake ranged from 0g to 357.1g per day. Yet even though Campbell finds the small intake animal protein intake to be relevant, he dismisses other variables with even wider ranges due to being “consumed in much lower quantities” than in the United States.[49]

How can Campbell be certain that increases of animal protein from 0g to 24g are relevant, but increases in processed starch and sugar from 0g to 27g are not—especially if those increases coincide with other shifts towards a more Western lifestyle, such as decreased physical activity?

The potential significance of other variables cannot be dismissed based on subjective assessments of their importance. As explored later in this paper, some ethnic groups who have subsisted on a constant traditional diet for centuries or millennia respond with disproportionate levels of disease to the introduction of new, Western-style foods and lifestyle habits.[50] In studies of immigrants, Asians eating Western diets appear to have excessively high rates of diabetes and insulin resistance compared to their non-Asian counterparts,[51] suggesting a “predisposition to insulin resistance and its metabolic abnormalities.”[52] In addition, studies of lean Chinese adults have demonstrated that insulin resistance often occurs independent of obesity, and even normal-weight Chinese are susceptible to impaired glucose tolerance.[53]

If there is widespread inability to handle high-glycemic foods like sugar and a predisposition to metabolic abnormalities, then what Campbell deems an insignificant intake by Western standards may, in actuality, be highly pertinent to the Chinese.

Projection of casein’s carcinogenic properties to all forms of animal protein

As I explored in my original critique and clarified again in my follow-up post, Campbell’s extrapolation of his casein research to all forms of animal protein—as well as the assumption that casein behaves the same way in whole-food form as when isolated—is supported by neither clinical evidence nor logic.

Campbell draws his animal protein/cancer hypothesis from a series of experiments conducted on aflatoxin-exposed rats, which revealed dramatic differences in the cancer growth depending on the level of protein consumed. As Campbell explains, rats fed a diet of 5% protein in the form of casein exhibited dramatically fewer lesions than rats fed a diet of 20% casein.[54] Additional experiments showed that wheat and soy protein did not promote cancer growth, even when fed at the 20% level.[55] From this, Campbell concludes that casein could be the “most relevant carcinogen ever identified.”[56]

Yet in a 1989 study, Campbell discovered that wheat protein exhibited similar carcinogenic properties as casein when lysine, its limiting amino acid, was restored.[57] This suggests that any complementary combination of amino acids will spur cancer growth under certain experimental conditions, and that carcinogenic qualities are not unique to casein nor to animal protein at large. The sole reason plant protein appeared protective in rat studies was due to a deficiency in one or more amino acids, a scenario that rarely occurs in real-world situations when a variety of foods—whether plant or animal in origin—are consumed. Campbell himself notes that eating a variety of plant foods provides a full spectrum of amino acids[58]—indicating that even a plant-only diet can yield the complete protein Campbell claims to be carcinogenic.

However, the notion that complete protein is inherently carcinogenic is contradicted by more recent literature. Although Campbell’s focus on casein is understandable, given the research is chiefly his own, he does not acknowledge the abundance of similar studies showing that whey—another milk protein—consistently boasts anti-cancer properties,[59]^,[60],[61] including when studied under the same experimental conditions that demonstrate the carcinogenic qualities of casein.[62]^,[63] This is significant, as even a single example of animal protein inhibiting rather than spurring cancer invalidates Campbell’s hypothesis that the effects of casein can be extrapolated to all animal protein.

Section 2:

Biological Models and Cited Papers

An evaluation of Campbell’s cited papers and relevance to biological models

In Campbell’s second response to my critique of “The China Study,” he notes:

The China project encouraged us not to rely on independent statistical correlations with little or no consideration of biological plausibility. In the book, I drew my conclusions from six prior models of investigation to illustrate this approach: breast cancer, liver cancer, colon cancer (minimally), energy utilization/body weight control, affluent disease-poverty disease and protein vs. body growth rates.[64]

Campbell then cites “a few representative publications” in which he applies supportive data to biological models, purportedly demonstrating their congruency.

Do these publications and the biological models they employ implicate animal products as causative of disease? To answer this question, I’ve examined several of Campbell’s cited papers, evaluating in each case whether the biological models Campbell draws from support his use of specific data to verify an animal food/disease hypothesis.

Breast cancer

Publication: Marshall JR, Qu Y, Chen J, Parpia B, and Campbell TC. “Additional ecological evidence: lipids and breast cancer mortality among women age 55 and over in China.” Eur J Cancer. 1992;28A(10):1720-7.

In the first paper he cites for breast cancer, Campbell et al use the China Study data to explore the relationship between breast cancer mortality and a variety of risk indicators— searching specifically for associations between fat consumption and higher cancer rates. They note that, although animal experiments consistently demonstrate greater breast cancer risk from increasing dietary fat, that “human data are less consistent,” and that “several well-designed and executed case-control studies apparently failing to show any association.”

Drawing from plasma samples, red blood cell samples, diet survey results, and questionnaire answers, Campbell et al study breast cancer mortality in relation to:

• total cholesterol
• high density lipoprotein (HDL) cholesterol
• low density lipoprotein (LDL) cholesterol
• triglyceride levels
• apolipoprotein A-1
• apolipoprotein B
• total lipid saturates
• total lipid polyunsaturates
• ratio of total lipid polyunsaturates to saturates
• total dietary lipid intake
• total caloric intake
• percentage of caloric intake from lipids
• physical height
• physical weight
• Quetelet index
• alcohol consumption
• age at menarche
• age at first pregnancy, and
• total live births

For the diet variables, the researchers do not distinguish between animal and plant-derived fats—an important observation, given the equivalent contribution of plant versus animal fats to total fat intake after adjusting for the outlier county Tuoli, as explained in section 1.

From this study, Campbell et al conclude that, among cholesterol fractions, the “strongest and most consistent predictor of risk is apolipoprotein A-1,” and that “higher levels are consistently associated with greater breast cancer risk” even after adjusting for other variables. Incidentally, apolipoprotein A-1 appears to be significantly associated with rapeseed oil intake, but not with animal protein.[65]

The researchers also note that “increased red blood cell saturated fat is associated with an insignificantly lower risk of breast cancer” (r = -0.16), although the plasma lipid indicators may provide unreliable measures of county status.

Although researchers do cite a weak positive relationship between lipid intake and breast cancer rates, no examination of specific animal foods such as meat, fish, dairy, eggs, or total animal protein is included in this paper’s analyses. Similarly, researchers offer no rationale for using lipid intake or blood markers as a liaison specifically between animal products and disease.

Moreover, Campbell et al state that the China Study data offers only a weak indication of a relationship between diet and breast cancer. As mentioned in section 1, they conclude from their findings that the data “provides only modest support for the possibility of a diet-breast cancer link.”

Additionally, Campbell et al acknowledge that their results “may well be confounded,” particularly because their analysis unveiled several anomalous associations—such as a relationship between higher parity (number of live-born children) and increased cancer risk, as well as higher age at first birth and lower cancer risk. These associations, the researchers note, “contradict nearly all individual-based studies.” The accuracy of this particular data set and the trends extracted from it may thus be in doubt.

At best, the associated biological model involves a possible—but “modest”—role of lipid consumption in the development of breast cancer, with no distinction made between plant or animal sources of fat. While Campbell attempts to implicate animal foods due to a univariate correlation between animal protein and total lipid intake, the estimated correlation—as addressed earlier—is likely to be steeply overestimated due to the inclusion of Tuoli county in the calculation, an outlier whose survey data Campbell deemed “unreliable.”

Publication: Key TJA, Chen J, Wang DY, Pike MC, and Boreham J. “Sex hormones in women in rural China and in Britain.” British Journal of Cancer 1990(62):631-636.

In the second paper Campbell cites, researchers examine plasma concentrations of hormones implicated with breast cancer in Chinese versus British women—namely oestradiol, testosterone, sex hormone binding globulin, and prolactin. In comparing several physical and reproductive variables between Chinese and British women, researchers highlight significant differences in height, weight, age of menarche, age of first pregnancy, and age of menopause.

Results of the analysis reveal that British women generally had higher concentrations of oestradiol and testosterone than did the Chinese group, but that variations in testosterone “may be due to the difference in body weight.”

The researchers note several shortcomings in the comparisons between these two countries, stating:

• The Chinese samples were collected at a different time of day than the British samples, and were only collected during a three-month period rather than year round—a significant observation, because hormone levels often fluctuate throughout the day as well as varying seasonally.
• Blood sample collection and processing methods differed between the two countries, possibly confounding the results.

Potential data inaccuracies aside, the researchers hypothesize that the differences in oestradiol between Chinese and British women may be due not only to the Chinese’s low fat diet, but also to their greater levels of exercise, additional dietary practices, or other lifestyle factors that differ between the two populations. As with the previous paper Campbell cited, this publication does not discuss a potential role of animal foods as either direct or indirect causative agents of breast cancer. More specifically, it offers no rationale for assuming animal foods, as a collective group, cause elevated hormone levels more than other dietary or lifestyle constituents.

Liver cancer

Publication: Campbell TC, Chen J, Liu C, Li J, Parpia B. “Nonassociation of aflatoxin with primary liver cancer in a cross-sectional ecologic survey in the People’s Republic of China.” Cancer Res 1990(50):6882-6893.

In exploring risk factors for primary liver cancer in rural China, Campbell et al conclude that liver cancer mortality was unassociated with exposure to the carcinogen aflatoxin, but was positively and significantly associated with hepatitis B infection, total cholesterol, and intake of cadmium from plant foods. The paper notes a high correlation between cholesterol and liver cancer, and posits animal foods as the cause:

Plasma cholesterol was highly significantly associated with PLC mortality rates. … In this study, even though it was very low compared to the United States, it tended to be associated with the intakes of foods of animal origin.

The referenced table cites a list of univariate correlations between cholesterol and 12 variables, as follows:

Campbell et al used only univariate correlations to link animal food variables to cholesterol, without adjusting for other cholesterol-raising factors that may cluster alongside them. Moreover, had the researchers omitted Tuoli county from these calculations—a reasonable choice, given that the erroneous data of this region strongly influences any correlations involving meat, dairy, fat intake, and animal protein intake—several of the figures would be somewhat attenuated. The corrected data set yields a correlation of 0.18 instead of 0.24 with animal protein and cholesterol, 0.03 instead of 0.21 for dairy, and 0.18 instead of 0.26 for meat. These numbers are lower than other variables correlating with cholesterol, such as daily beer consumption (r = 0.32, p<0.01), daily liquor consumption (r = 0.20), total daily alcohol consumption (r = 0.21), intake of soybean, corn, cottonseed, sesame, or peanut oil (r = 0.20) and industry employment (r = 0.45, p<0.001)—with the latter accompanying other lifestyle factors that may lead to higher cholesterol, such as reduced physical activity.

Most importantly, cholesterol may not actually be a cause of liver cancer—but rather, an effect. In a study of 792 Chinese patients with liver cancer, Hwang et al discovered that 11.4% of subjects were hypercholesterolaemic, but they exhibited a return of normal cholesterol levels following surgery and chemoembolization for their conditions:

Serum cholesterol levels fell to the normal range after treatment and rose to abnormal levels again when tumours recurred after surgery. … Serum cholesterol levels may serve as another marker in identifying tumour recurrence and the presence of a viable tumour mass in hypercholesterolaemic HCC patients.[66]

Thus, liver cancer itself may cause cholesterol to rise, independent of diet or lifestyle factors. If this is the case, the influence of animal products on blood cholesterol would be irrelevant, and a direct link between animal foods and liver cancer would be necessary to prove their association. With the exception of fish—which Campbell et al have explained associates with liver cancer due to climatic and geographical factors[67]—such a relationship is not apparent.

In addition, it should be noted that this paper has received criticism from other cancer researchers who consider its conclusions unfounded. In 1991, Christopher P. Wild and Ruggero Montesano of the International Agency for Research on Cancer submitted a letter to Cancer Research stating:

We were concerned by the conclusions drawn by Campbell et al. in their recent paper in which they reported (a) a lack of association between urinary aflatoxin metabolites and primary hepatocellular carcinoma in 48 counties in the People’s Republic of China and (b) a positive association with plasma cholesterol. We consider the conclusions unsubstantiated and misleading …[68]

Wild and Montesano proceed to outline flaws in the experimental methods used by Campbell et al—including problems associated with the urinary assay of aflatoxin, the researchers’ lack of adjustment for urine concentration, and seasonal variations in aflatoxin exposure that could yield misrepresentative data. Therefore, the validity of the paper itself may be in question.

Energy utilization

Publication: Campbell TC and Chen J. “Energy balance: interpretation of data from rural China.” Toxicol Sci. 1999 Dec;52(2 Suppl):87-94.

In this paper, Campbell and Chen synthesize information from earlier lab rat studies, other animal research, and the China Study—particularly data recorded during a three-day diet survey, which revealed that rural Chinese citizens have a high average calorie intake compared with most Americans. Based on this apparent calorie paradox, they hypothesize that a low-protein diet increases thermogenesis, and that:

some unknown but significant, and probably difficult to measure, amount [of extra calorie intake] could be due to increased energy expenditure associated with non-post-prandial basal metabolism.

The implication is that low-protein diets may be effective for maintaining a healthy body weight because they divert a “biologically meaningful but difficult to measure” amount of energy away from weight gain and into body heat. Although the researchers do not specifically describe how this mechanism would occur in humans, they draw from animal involving brown adipose tissue metabolism.

However, the researchers concede that the increased calorie intake exhibited by the Chinese may simply be due to exercise:

Undoubtedly, much of the increased energy intake but lower body weight in rural China, as measured in this survey, was attributable to their greater physical activity (i.e., it is common to see most office workers riding bicycles to work) …

Moreover, given Campbell’s earlier disclosure that one county was “essentially eating as if it were a feast to impress the survey team”[69] during the three-day survey, the validity of the recorded energy intake is in question. How certain is Campbell that other counties were not altering their eating habits to give the impression of greater wealth or food abundance, thus leading to an overestimate of the average calorie intake for the Chinese?

Given the possible overestimation of calorie intake in the China Study, the reliance on animal rather than human studies, and the inability to calculate whether the increased calorie consumption was or was not fully offset by physical activity, this hypothesis rests on the accuracy of many unknowns. While its validity is still possible, the evidence at hand is insufficient to confirm it.

Affluent-poverty diseases

Publication: Campbell TC, Junshi C, Brun T, Parpia B, Yinsheng Q, Chumming C, and Geissler C. “China: From diseases of poverty to diseases of affluence. Policy implications of the epidemiological transition.” Ecology of Food and Nutrition 1992(27):133-144.

The publication Campbell cites to explain his “diseases of poverty” and “diseases of affluence” model has already been mentioned throughout this paper, but briefly, its premise is that two disease clusters naturally emerged from the China Study data. Diseases in one group are “generally associated with impoverished conditions,” while diseases in the second group are “characteristic of more affluent societies.” The second self-clustered group includes stomach cancer, liver cancer, colon cancer, lung cancer, breast cancer, leukemia, diabetes, and coronary heart disease. By viewing each assembly of diseases in the aggregate, Campbell seeks to identify underlying nutritional patterns in their collective emergences—potentially deciphering the source of rising disease rates in affluent nations.

While this method may be useful for examining general disease patterns, Campbell’s chief errors are as follows.

1. Disregard for potentially critical variables in disease proliferation. Although Campbell acknowledges that numerous variables associate with the “diseases of affluence” cluster—including intake of processed starch and sugar, beer intake, fish consumption, egg consumption, and industry work—he dismisses all but cholesterol, citing the rest as “probably more indicative of general economic conditions and other local circumstances” than as causative of disease. In addition, other biomarkers such as plasma glucose may be of equal or greater relevance compared to cholesterol but receive no mention in his publication. Campbell’s disregard for these variables appears to be subjective, rather than a result of the thorough analysis necessary for deeming them insignificant.
2. “Reductionist” use of cholesterol as a disease indicator. In examining health and nutritional trends, Campbell takes the same reductionist approach he censures elsewhere by targeting cholesterol as the chief predictor for disease. By linking cholesterol solely to animal food consumption and disregarding the numerous other variables that may cause it to rise, Campbell overlooks the larger context of disease mechanisms as they pertain to diet and lifestyle.
3. Inaccurate representation of true diseases of affluence. Campbell’s dichotomization of diseases, while useful in some cases, does not accurately reflect disease rates in developed countries:
   • Stroke, the third leading cause of death in the United States,[70] does not fit cleanly into either the affluent or poverty disease group, so Campbell omits it entirely.
   •Heart disease correlates only weakly positively or, in three cases, inversely with the other diseases in the affluent cluster, suggesting it may not be strongly associated with the other conditions and is potentially a result of separate geographic, nutritional, or lifestyle variables.
   •Liver cancer is relatively uncommon in affluent nations, but exhibits strong correlations with the variables Campbell ascribes to diseases of affluence, most notably cholesterol. This provides further indication that cholesterol may not be an appropriate or dependable biomarker for examining true Western diseases in relation to diet.
4. Oversight of a third, potentially significant disease cluster. Myocardial infarction, hypertensive heart disease, stroke, brain and neurological diseases, and diseases of the blood and blood-forming organs share strongly statistically significant correlations with each other and with shared nutritional variables, such as non-rice grain consumption, while correlating inversely with the variables associated with diseases of affluence. Despite this, Campbell forces myocardial infarction into a disease cluster it does not naturally align with, and ignores the remaining diseases rather than attempt to explain their anomalous nonassociation with other Western conditions.

Consequently, Campbell’s use of these disease clusters to identify relationships between diet and diseases of Western nations may be unsound, especially given a myopic focus on cholesterol to the point of excluding other pertinent factors.

Summary

While biological models, as Campbell notes, are essential for developing a comprehensive understanding of nutrition and disease mechanisms, the ones he employs do not validate the claim that animal foods are unhealthful—the hypothesis that inspired my original skepticism and critique. The biological models he cites fail to support the three-variable chains he creates to implicate animal products with cancer, heart disease, and other chronic conditions, and his use of univariate correlations to impose these links remains unfounded.

Moreover, the models Campbell cites center on individual biomarkers in disease mechanisms—examples of the same reductionism Campbell claims to oppose. If disease mechanisms work in a “symphony,” as Campbell has described, and if animal products are harmful in the aggregate rather than due to single nutrients, then a direct relationship between animal food consumption and disease should be identifiable.

Section 3:

Response to Points Raised by Campbell

Although the previous sections have covered—directly and indirectly—my rationale for citing certain errors in Campbell’s work, I will use the following pages to address more specific concerns Campbell raised regarding my critique.

Wheat: confounded variable or legitimate concern?

In his second response to my critique, Campbell cites a rhetorical remark I included in my analysis—an inquiry as to why he did not appear to have explored the strong (r = 0.67, p<0.001) correlation between wheat and heart disease,[71] despite citing far weaker correlations as a means to implicate animal products with various conditions. In offering possible explanations for the wheat-heart disease link, Campbell presents several relevant correlations that are “all highly statistically significant (p<0.01 to p<0.001),” including:

Higher wheat flour consumption, for example, is correlated, as univariate correlations, with lower green vegetable consumption (many of these people live in northern, arid regions where they often consume meat based diets with little no consumption of vegetables).[72]

Although Campbell is correct in noting a highly statistically significant (r = -0.63, p<0.001) inverse correlation between wheat flour and green vegetable consumption, he uses the same variable whose speciousness I discussed earlier: frequency, rather than quantity, of green vegetable intake. Citing this correlation only shows—somewhat redundantly—that wheat is consumed in northern regions where many crops grow seasonally rather than year-round. The correlation with quantity of green vegetable consumption, however, is an attenuated -0.16, and wheat happens to positively correlate with protective plant foods such as light-green vegetables (0.10) and carrots (0.27, p<0.05).

Additionally, Campbell’s claim that “many of these people … consume meat based diets” is fairly incongruous with the China Study data itself, which reveals that only one county—the now-discredited Tuoli—consumed any significant portion of meat, even though 22 counties consumed at least 100g of wheat flour per day. And in contrast to what Campbell asserts, wheat flour correlates at -0.22 with frequency of meat consumption and at -0.09 with amount of daily meat intake (reducing further to -0.23 and -0.22, respectively, when recalculated without Tuoli). Although animal protein correlates at 0.17 with wheat flour consumption when using all 65 counties, this figure, too, reduces when Tuoli is excluded from calculation, dropping to -0.06.

Campbell is likely aware of the lack of a wheat-meat link, as a paper he coauthored in 1998, entitled “Diet, lifestyle, and the etiology of coronary artery disease: the Cornell China Study,” presents the following conclusion:

Nonetheless, the wheat-flour effect appears to be independent of meat consumption, so enhancement of coronary artery disease risk by wheat consumption may be a possibility.[73]

Despite the “possibility” that one of the most widely-consumed grains may contribute to heart disease, Campbell does not pursue this issue further through research or in his book, continuing to focus instead on animal foods.

In his response, Campbell points to another pertinent correlation with wheat: “greater body weight (higher risk of heart disease),” which correlates at 0.59 (p<0.001) with wheat flour intake.[74] This, of course, raises an important issue: Why do people in wheat-eating regions tend to have significantly higher body weight than citizens of other areas? The answer does not appear to be calories, as wheat flour only correlates at 0.07 with total caloric intake.[75] Nor is the answer lower activity associated with industrial employment, as wheat flour correlates at -0.24 with percentage of the population employed in industry.[76] And given the lack of association with meat or other animal foods, animal protein is an equally unlikely solution. Does wheat encourage body mass gain or spur growth in a way that rice, for instance, does not?

Given that wheat flour does not have an obvious relationship with other energy-dense foods Campbell ascribes to increased growth, it does seem wheat itself may be a factor. This is not a topic that can be feasibly analyzed in the span of this paper, but it may be a relevant one to explore in the future.

Another correlation Campbell notes is higher serum levels of urea, which he mentions is a biomarker of protein consumption. Although Campbell’s implication may be that animal protein is the cause—thus lending credence to his animal protein-disease theory—an examination of the data reveals otherwise. A notable feature about grain consumption in China is its dichotomization: Rice dominates southern regions while strongly inversely correlating with wheat consumption (r = -0.76, p<0.001) and other cereal grains (r = -0.68, p<0.001)[77] in the north. Given that wheat flour, depending on whether whole-grain or refined, is nearly twice as high in protein compared to white rice on a per-calorie basis,[78] it logically follows that areas where wheat is a staple have higher protein intakes—and thus higher serum urea—than areas where rice is a staple.

Indeed, wheat flour correlates at 0.34 (p<0.01) with plant protein and at 0.35 (p<0.01)[79] with total protein intake, whereas rice intake correlates at -0.20 with plant protein and -0.23 with total protein intake.[80] Coupled with the fact that wheat flour inversely associates with all forms of animal food except for milk—which is generally only consumed in three counties—a logical interpretation is that a greater intake of plant protein results in these higher urea levels. A higher intake of plant protein, perhaps, may also contribute to the higher body weights exhibited in wheat-eating regions, particularly given Campbell’s supposition about protein-restricted diets limiting weight gain and higher-protein diets fueling it.[81]

The next point Campbell raises may be significant: the relationship between wheat flour and certain serum lipid fractions. Although Campbell states that wheat flour intake is associated with “lower serum levels of monounsaturated fats,” which he notes can increase risk of heart disease, wheat also correlates with lower total lipid docosahexaenoic acid (DHA) (r = -0.34, p<0.05)[82]—an essential fatty acid linked to cardiovascular health by abundant research,[83]^,[84] including a China Study-based publication Campbell coauthored.[85]

Campbell implies that unfavorable lipid profiles may be responsible for the high rates of heart disease independent of wheat, thus creating a false correlation between wheat consumption and cardiovascular conditions. However, another possibility is that the wheat itself contributes to unfavorable blood lipid profiles, especially in the absence of more heart-protective foods such as fish—which is rarely consumed in wheat-eating regions (r = -0.37, p<0.01) but more frequently in rice-eating regions (r = 0.32, p<0.05) where heart disease is far less common (r = -0.58, p<0.001).[86] If wheat as a dietary staple is nutritionally inadequate, this would suggest that wheat-based diets may require careful planning or supplementation to reduce heart disease risk, especially in supplying certain fatty acids difficult to obtain from plant foods.

Continuing in this vein, Campbell writes:

[The] correlation of wheat flour and heart disease is interesting but I am not aware of any prior and biologically plausible and convincing evidence to support an hypothesis that wheat causes these diseases.[87]

Because my initial mention of wheat’s correlation with heart disease was intended to be speculative rather than assertive, I did not offer corroborating theories or evidence to substantiate a wheat-heart disease link. However, Joel Fuhrman—a plant-based diet advocate whom Campbell cites as one of his “physician colleagues”—has stated:

Many scientific studies show a strong association between the consumption of white flour products, such as pasta and bread, with diabetes, obesity, and heart disease. … Whole grains are the least nutrient-dense food of the seed family, and they do not show the powerful protection against disease that is apparent in the scientific studies of fresh fruit, vegetables, beans, raw nuts, or seeds.[88]

Since the China Study data provides no indication as to whether the wheat flour consumed was whole-grain or refined, the following can only be guesswork. However, one noteworthy feature of refined grains such as white flour is their connection with elevated triglyceride levels, a condition widely associated with heart disease.[89]^,[90]Another of Campbell’s colleagues, John McDougall, asserts that refined grains cause blood triglycerides to increase, and states in his October 2006 newsletter:

My experience has been that people who are having problems getting their … triglycerides under control need to stop using refined flour products and simple sugars.[91]

Indeed, wheat flour in the China Study is strongly associated with high triglyceride levels (r = 0.51, p<0.001).[92] Given that omega-3 fats have a mitigating effect on triglycerides,[93] it could be posited that rice-eating regions in China, with their frequent consumption of omega-3 rich seafood, could be more protected from heart disease than wheat-eating regions—even though white rice alone may exhibit the same effects as other refined grains.

Of course, epidemiological data cannot prove causative relationships, only highlight correlations that may or may not be meaningful. And also importantly, the above web of univariate correlations is in no way conclusive, as many of these values may change when accounting for nonlinearity and confounding. When searching for overarching themes and material for future research, however, such univariate correlations are a useful place to start, as they are often the first indication of patterns that gain magnitude once fully excavated and analyzed. Given that connections between processed grains and heart disease are already corroborated by research as well as by biological plausibility, an authentic connection between wheat and heart disease is not unfeasible.

Selection of univariate correlations and confirmation bias

My largest concern with Campbell’s conclusions, as stated in this paper and elsewhere, is that his approach to both China Study data and related research has been angled by the pursuit of a specific hypothesis—rather than an evenhanded evaluation of information and subsequent formation of a theory. In Campbell’s explanation of his approach, he writes:

I first inquired whether a collection of variables in the China survey (ranging from univariate correlations to more sophisticated analyses) could consistently and internally support each of these biologically plausible models and, second, I determined whether the findings for each of these models were consistent with the overarching hypothesis that a whole food, plant-based diet promotes health.[94]

Had Campbell approached the data from a different angle—or, better, from diverse and opposing perspectives in search of the most accurate one—he may have found multiple biologically plausible ways of incorporating China Study data trends with  known physical mechanisms. By not testing alternative hypotheses alongside his own, Campbell runs the risk of investigative tunnel vision, and cannot truly determine whether his hypothesis is more valid than another.

In his second response to my critique, Campbell also writes:

As I’ve said many times, not all the evidence in the China database supported this conclusion, although the large majority did.[95]

Considering the complexity and abundance of trends in the raw data, I would like to know what methods Campbell used to analyze and adjust the majority of the 8,000 statistically significant correlations in a way that yielded supportive results for his hypothesis. I propose that the China Study has generated enough material to bolster nearly any theory, regardless of actual validity—and for this reason, mandates an impartial and multi-perspective approach rather than a search for a predetermined outcome.

While the limitations of using univariate correlations are clear, Campbell has expressed willingness to employ them when they “consistently and internally support … biologically plausible models.”[96] Yet it appears Campbell’s chief criterion for deeming correlations valid is not just whether they’re objectively plausible, but whether they support his hypothesis. In the instances they do, as explained earlier in this paper, he cites them without performing deeper analyses; in the instances they do not, he gives them no mention nor delineates a methodology for explaining their inconsistency with his theory. Rather than evaluating discrepancies within the data, he dismisses them—a choice ultimately leading to confirmation bias, potential misrepresentation of true trends, and a missed opportunity to rework his hypothesis to account for apparent anomalies.

Tuoli county and erroneous data

In discussing my observation of the apparent good health of a Chinese county whose diet, per China Study data, was high in animal protein, Campbell clarifies:

[Tuoli county was] intentionally … excluded from virtually all our analyses on meat consumption because this county ranked very high when meat consumption was documented at survey time, but much lower when responding to the questionnaire on frequency of meat consumption. That is, these nomadic people migrate for part of the year to valleys, where they consume more vegetables and fruits.[97]

Although the information Campbell provides is useful, meat was not the dietary feature noted in my discussion of Tuoli: dairy was. Both the three-day diet survey and the frequency questionnaire reveal high intakes of dairy for Tuoli citizens, with the questionnaire indicating milk products are consumed an average of 330.3 days per year, and closer to 350 in one township.[98] In addition, despite Campbell’s comment that the Tuoli migrate seasonally and consume more vegetables and fruit for part of the year, the China Study frequency questionnaire indicates Tuoli’s vegetable intake is only twice per year and fruit intake is less than once per year on average.[99]

If Campbell believes both the three-day diet survey and frequency questionnaire were in error, I must question why Tuoli county was not excluded entirely from the data set—especially given its pronounced influence on virtually all associations involving meat, dairy, and animal protein, many of which Campbell cited as verification for his animal foods-disease hypothesis.

Efficacy of whole-food, plant-based diets versus whole-food diets with animal products

In his second response to my critique, Campbell states:

[The] results of people using a diet of whole, plant-based foods, as shown by physician colleagues (previously mentioned, McDougall, Esselstyn, Ornish, Barnard, Fuhrman, et al) as well as by many of the readers of our book are nothing less than incredible.[100]

Campbell cites several examples of physician colleagues who have successfully employed plant-based diets—often in conjunction with other lifestyle modifications—to improve patients’ health and reverse chronic conditions such as heart disease. Although these doctors unanimously advise limiting animal food consumption, their diet programs are characterized by more than just plant-based nutrition: They also drastically reduce or eliminate refined carbohydrates, processed sugar, and hydrogenated oils—foods that tend to feature prominently in Western-style cuisines alongside animal-based products.

1. John McDougall. While McDougall’s program embraces whole plant foods, he also advises against consuming refined flour, refined and sugar-coated cereals, soft drinks, vegetable oils, white rice, and other processed carbohydrates.[101]
2. Caldwell Esselstyn, Jr. The diet promoted by Esselstyn involves not only the elimination of animal products, but also the avoidance of vegetable oils and refined grains—including white rice, white flour, and products made from enriched flour such as pastas, breads, bagels, and baked goods.[102]
3. Dean Ornish. Along with eschewing meat, Ornish’s program—as outlined in the book Eat More, Weigh Less—also involves reducing “sugar and simple sugar derivatives” such corn syrup, white flour, and white rice, avoiding margarines and vegetable oils, limiting alcohol, and avoiding commercial products with more than two grams of fat per serving, which is likely to disqualify most ready-made processed foods from dieters’ menus.[103] Ornish also notes that his program involves more than just a plant-based diet: He emphasizes increased exercise[104] and other lifestyle changes to achieve better health.
4. Neal Barnard. In his book Dr. Neal Barnard’s Program for Reversing Diabetes, Barnard advises his readers to “keep vegetable oils to a minimum” and “favor foods with a low glycemic index,”[105] which ultimately eliminates refined carbohydrates, most processed foods, high fructose corn syrup, and other common sweeteners. Barnard also recommends avoiding fried foods, including fried starches such as potato chips and French fries.[106]
5. Joel Fuhrman. Along with reducing or eliminating animal products, the diet Joel Fuhrman espouses shuns refined grains, refined oils, and refined sweets; Furhman lists these foods as less healthful than all forms of animal food in terms of nutrient density,[107] and notes that “eating a diet that contains a significant quantity of sugar and refined flour … leads to an earlier death.”[108] Fuhrman also notes that “a low-fat diet can be worse than a higher-fat diet” if it centers on refined carbohydrates and contains trans fat,[109] stating specifically:

A vegetarian whose diet is mainly refined grains, cold breakfast cereals, processed health food store products, vegetarian fast foods, white rice, and pasta will be worse off than a person who eats a little turkey, chicken, fish, or eggs but consumes large volumes of fruits, vegetables, and beans.[110]

Although plant-based diets eschewing white sugar, refined grain products, trans-fatty acids, high fructose corn syrup, and other highly processed ingredients are likely to improve health compared to a standard Western diet, research comparing unprocessed plant-food diets with unprocessed omnivorous diets is sparse. The success of whole foods, plant-based diets is not itself an indication that animal foods are deleterious; to determine this would require juxtaposing the results of whole-food vegan diets with equally “clean” omnivorous eating plans and demonstrating consistent superiority of the former.

Non-Westernized omnivorous diets

As current research indicates, other dietary paradigms may offer similar benefits to the plans promoted by McDougall, Esselstyn, Ornish, et al, without the reduction of animal products. Recent studies have shown “Paleolithic” style diets—which eschew grains, dairy, legumes, processed carbohydrates, and refined fats while embracing minimally processed meat, fish, vegetables, eggs, fruit, and nuts—may reduce fasting glucose levels, improve diastolic blood pressure, promote weight loss, improve glycemic control, lower triglycerides, raise HDL or “good” cholesterol, and generally reduce risk factors for cardiovascular disease, all while allowing for liberal consumption of non-dairy animal products.[111] In a study of diabetic patients, these effects were more pronounced in a Paleolithic diet group than in a group fed a standard low-fat diet including abundant plant-based foods such as whole-grain bread, other whole-grain cereal products, vegetables, and fruit, along with low-fat dairy.[112]

Similarly, a 1999 study by Frassetto et al discovered that non-obese subjects consuming a diet “comprising lean meat, fruits, vegetables and nuts” while excluding cereal grains, legumes, and dairy led to consistent and nearly immediate improvements in blood pressure, reduction in plasma insulin, lowered total cholesterol, reduced low-density lipoproteins, and decreased triglycerides.[113] In their publication, Frassetto et al conclude:

Even short-term consumption of a paleolithic type diet improves BP [blood pressure] and glucose tolerance, decreases insulin secretion, increases insulin sensitivity and improves lipid profiles without weight loss in healthy sedentary humans.[114]

Additionally, a study conducted by Lindeberg et al showed that a grain-free diet with animal products improved glucose tolerance and reduced waist circumference more effectively than an unprocessed “Mediterranean” diet featuring whole grains, abundant plant foods, low-fat dairy, and minimal red meat.[115]

Additional clues herald from Australia. Research from O’Dea on the health and dietary patterns of Australian aborigines reveals that those eating a traditional cuisine—typically high in animal foods such as organ meats, fat deposits, and bone marrow along with tubers, vegetables, seeds, and fibrous fruits—exhibited “no evidence of the chronic diseases” common to Westerners,[116] including heart disease, diabetes, and obesity.[117]

Yet despite the apparent lack of adverse effects from their native high-meat diet, aborigines exhibit disproportionately high rates of diabetes and obesity after adopting a Western diet and lifestyle,[118] indicating genetics alone is not what protects them. If this surge in disease rates occurs with other groups shifting from a traditional omnivorous diet to a Western one, it suggests that factors other than animal food consumption may be responsible for the diseases plaguing developed countries.

In addition to the health of non-Westernized aborigines, virtually nonexistent rates of Western diseases have been reported of the Kitava, a traditional Melanesian society consuming no grains or processed carbohydrates but subsisting on a native diet of tubers, fish, coconut, and fruit.[119]^,[120] A compilation of research collectively known as the Kitava Study revealed that “stroke and ischaemic heart disease appear to be absent in this population,”[121] despite their consumption of animal products and lack of purportedly “heart-healthy” grains.

The Masai of East Africa—who consume copious amounts of meat and milk—also patently defy Campbell’s hypothesis, particularly as it relates to animal foods and heart disease. After conducting a field survey of 400 Masai in the 1960s, researchers Mann et al observed that “Despite a long continued diet of exclusively meat and milk the men have low levels of serum cholesterol and no evidence for arteriosclerotic heart disease.”[122]

In a guest editorial published in the American Journal of Clinical Nutrition, these researchers expand on their findings, describing the high animal-fat diet of the Masai and their paradoxically low serum cholesterol—a mean of 135.4 mg/dL on average, a level on par with the rural Chinese consuming plant-based cuisines:

The average daily caloric intake was estimated to be about 3,000 kcal, with 66% of the calories derived from fat. The estimated average daily cholesterol intake was from 600 to 2,000 mg per person. The serum cholesterol levels of 254 Masai of various ages were determined; a low average value of 135.4 +/- 33.5 mg/100 ml … was observed.[123]

To provide further evidence of the Masai’s noteworthy lack of heart disease, the authors note that “gross, histochemical, and chemical studies of the aortas and coronary arteries of 10 consecutive autopsies gave direct proof of the paucity of atherosclerosis in the Masai.”[124]

In Alaska, researchers have observed rising rates of cardiovascular disease coinciding with a shift away from traditional dietary patterns and towards Western-style eating and lifestyle habits—a testimony to the health risks conferred by processed foods rather than animal foods in the aggregate. In 2009, cardiovascular disease risk factors were examined in relation to differing diet patterns among Alaska Eskimos, including a native diet featuring abundant animal products, wild foods, and no sugar or other refined carbohydrates:

Participants following … the “traditional” diet consumed fish, native sea and land mammals and their fats and oils, wild greens, stew with mostly meat, stew with mostly rice or noodles, native birds, wild berries, and native berry agutuk.[125]

Compared to Eskimos following other eating patterns, including a Western-style diet and “healthy” store-bought diet, individuals consuming traditional foods had the most “desirable cardiovascular risk factor profile,” including lower blood pressure and lower homocysteine.[126]

In a separate study, Alaskan Natives who replaced processed store-bought foods with traditional Eskimo foods—including meat from sea and land creatures—resulted in reduced diastolic blood pressure, lower total and low-density lipoprotein cholesterol, lower fasting glucose, and improved glucose tolerance.[127] Additional research shows that native diets emphasizing marine mammals, fish, game animals, berries, and wild greens results in lower triglycerides, increased high-density lipoprotein cholesterol, and better cardiovascular health—even while providing levels of animal fat exceeding those of most governmental recommendations.[128]

Conclusion

If both whole-food vegan diets and non-Westernized omnivorous diets yield similar health benefits, this is a strong indication that the results achieved by McDougall, Esselstyn, Ornish, et al are not due to the avoidance of animal products but to the elimination of other health-harming items. Western diets involve far more than increased consumption of animal products, and for some groups—such as Alaskan Natives—a switch from a traditional diet to a Westernized one entails reduced animal food consumption, with the caloric void replaced by refined carbohydrates, hydrogenated oils, grains, sugar, and convenience foods. The fact that a dietary shift towards Western fare inevitably leads to proliferation of “diseases of affluence”—regardless of changes in animal food consumption—suggests that another factor, or lattice of factors, instigates this decline in health.

The success of the Chinese on plant-based diets does not invalidate the experiences of other populations who evade disease while consuming animal products. Nor does individual success on a vegan program nullify the disease reversal seen by those adhering to specific omnivorous diets. Rather than studying the dissimilarities between healthy populations, perhaps we should examine their areas of convergence—the shared lack of refined carbohydrates, the absence of refined sweeteners and hydrogenated oils, the emphasis on whole, unprocessed foods close to their natural state, and the consumption of nutritionally dense fare rather than empty calories or ingredients concocted in a lab setting. Modern foods, and the diseases they herald, have usurped the dietary seats once occupied by more wholesome fare. It is this commonality—the thread bonding healthy populations—that may offer the most meaningful insight into human health.

A theory as purportedly universal as Campbell’s should, by definition, unite the various health and disease patterns of global cultures without generating frequent anomalies. By  naming animal products as the source of Western afflictions, Campbell has created a hypothesis valid only under hand-picked circumstances—one that cannot account for other epidemiological trends or even recent case-controlled  studies. This is a symptom of a deficient theory, embodying only partial truths about broader diet-disease mechanisms.

I propose that the China Study remains a largely untapped resource for revealing potential diet-health patterns, expanded awareness of the source of disease, and inlets for future nutritional research—possibilities Campbell has not fully explored in his quest to validate a predetermined hypothesis. I invite Campbell, if he has the time and the interest, to present a more detailed account of his methodology, such as the unpublished book chapter he cited in his first response to my critique.[129] It is only through ongoing discussion and clarification that the field of nutrition can continue to evolve, progressing towards an increasingly unified understanding of health.

Lastly, I suggest that the “symphony” Campbell has heard thus far is only a partial opus. To cease listening now would be—at best—a missed opportunity for heightened health awareness, and at worst a perpetuation of the misinformation already degrading public and scientific understanding of diet and disease. I thank Dr. Campbell for both the harmonies and the dissonance his work has supplied to the field of nutrition, but implore him to continue listening. The final note has not yet sounded.

[2] Junshi C, Campbell TC, Junyao L, Peto R. Diet, Life-style and Mortality in China: A Study of the Characteristics of 65 Chinese Counties. Oxford: Oxford University Press, 1990, p. 264.

[3] Campbell TC, Junshi C, Brun T, Parpia B, Yinsheng Q, Chumming C, and Geissler C. “China: From diseases of poverty to diseases of affluence. Policy implications of the epidemiological transition.” Ecol. Food Nutr. 1992(27):133-144.

[4] Junshi C et al. Diet, Life-style and Mortality in China, p.264.

[5] Yin SN, Li GL, Tain FD, Fu ZI, Jin C, Chen YJ, Luo SJ, Ye PZ, Zhang JZ, Wang GC, Zhang XC, Wu HN, Zhong QC. “A retrospective cohort study of leukemia and other cancers in benzene workers.” Environ Health Perspect. 1989 July; 82: 207–213.

[6] Hayes RB, Yin SN, Dosemeci M, Li GL, Wacholder S, Chow WH, Rothman N, Wang YZ, Dai TR, Chao XJ, Jiang ZL, Ye PZ, Zhao HB, Kou QR, Zhang WY, Meng JF, Zho JS, Lin XF, Ding CY, Li CY, Zhang ZN, Li DG, Travis LB, Blot WJ, Linet MS. “Mortality among benzene-exposed workers in China.” Environ Health Perspect. 1996 Dec;104 Suppl 6:1349-52.

[7] Junshi C et al. Diet, Life-style and Mortality in China, p. 572.

[8] Campbell TC. “A Challenge and Response to the China Study.” Retrieved from http://tynan.net/chinastudyresponse

[9] Ibid.

[10] Junshi C et al. Diet, Life-style and Mortality in China, p. 640.

[11] Ibid, p. 638.

[12] O’Connor TP, Roebuck BD, Peterson F, and Campbell TC. “Effect of dietary intake of fish oil and fish protein on the development of L-azaserine-induced preneoplastic lesions in the rat pancreas.” Journal of the National Cancer Institute 1985 Nov;75(5):959-62.

[13] Marshall JR, Qu Y, Chen J, Parpia B, and Campbell TC. “Additional ecological evidence: lipids and breast cancer mortality among women age 55 and over in China.” European Journal of Cancer 1999(28A):1720-1727.

[14] Campbell. The China Study, p. 87.

[15] Ibid.

[16] Muti P. et al. “Fasting glucose is a risk factor for breast cancer: a prospective study.” Cancer Epidemiology, Biomarkers and Prevention 2002 Nov;11(11):1361-8.

[17] Stoll BA. “Western diet, early puberty, and breast cancer risk.” Breast Cancer Res Treat. 1998 Jun;49(3):187-93.

[18] Gunter MJ, Hoover DR, Yu H, Wassertheil-Smoller S, Rohan TE, Manson JE, Li J, Ho GY, Xue X, Anderson GL, Kaplan RC, Harris TG, Howard BV, Wylie-Rosett J, Burk RD, Strickler HD.. “Insulin, insulin-like growth factor-I, and risk of breast cancer in postmenopausal women.” Journal of the National Cancer Institute 2009 Jan;101(1):48-60.

[19] Minger D. “The China Study: Fact or Fallacy?” Raw Food SOS. Retrieved from http://rawfoodsos.com/2010/07/07/the-china-study-fact-or-fallac/

[20] Junshi C et al. Diet, Life-style and Mortality in China, p. 118.

[21] Wang Y, Crawford MA, Chen J, Li J, Ghebremeskel K, Campbell TC, Fan W, Parker R, Leyton J. “Fish consumption, blood docosahexaenoic acid and chronic diseases in Chinese rural populations.” Comparative Biochemistry and Physiology Part A: Molecular and Integrative Physiology 2003 Sep;136(1):127-40.

[22] “Hepatocellular Carcinoma (Liver Cancer): FAQs.” Johns Hopkins Gastroenterology and Hepatology. http://www.hopkins-gi.org/GDL_Disease.aspx?CurrentUDV=31&GDL_Cat_ID=488EC820-7CD8-41C1-9B23-0F343893BFDE&GDL_Disease_ID=A349F0EC-5C87-4A52-9F2E-69AFDB80C3D1

[23] Chu CW, Hwang SJ, Luo JC, Tsay SH, Li CP, Chang FY, Lee SD, Lui WY, Chiang JH. “Manifestations of hypercholesterolaemia, hypoglycaemia, erythrocytosis and hypercalcaemia in patients with hepatocellular carcinoma: report of two cases.” J Gastroenterol Hepatol. 1999 Aug;14(8):807-10.

[24] Campbell TC, Parpia B, and Chen J. “Diet, lifestyle, and the etiology of coronary artery disease: the Cornell China Study.” Am J Cardiol. 1998 Nov 26;82(10B):18T-21T.

[25] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[26] Junshi C et al. Diet, Life-style and Mortality in China, p. 618.

[27] Ibid, p. 716

[28] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Cornell Chronicle. Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[29] Junshi C et al. Diet, Life-style and Mortality in China, p. 188.

[30] Ibid.

[31] Ibid, p. 798 – 806.

[32] Ibid, p. 612 – 618 and 716 – 724.

[33] Ibid, p. 280 – 288.

[34] Fan WX, et al. “Erythrocyte fatty acids, plasma lipids, and cardiovascular disease in rural China.”

[35] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Cornell Chronicle. Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[36] Minger D. “The China Study: My Response to Campbell.” Raw Food SOS. Retrieved from http://rawfoodsos.com/2010/07/16/the-china-study-my-response-to-campbell/

[37] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Cornell Chronicle. Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[38] Campbell TC. The China Study, p. 77.

[39] O’Donovan G, Owen A, Kearney EM, Jones DW, Nevill AM, Woolf-May K, Bird SR. “Cardiovascular disease risk factors in habitual exercisers, lean sedentary men and abdominally obese sedentary men.” Int J Obes (Lond). 2005 Sep;29(9):1063-9.

[40] Junshi C, et al. Diet, Life-style and Mortality in China, p. 830.

[41] Parekh N, Lin Y, Hayes RB, Albu JB, Lu-Yao GL. “Longitudinal associations of blood markers of insulin and glucose metabolism and cancer mortality in the third National Health and Nutrition Examination Survey.” Cancer Causes Control. 2010 Apr;21(4):631-42.

[42] Rapp K, Schroeder J, Klenk J, Ulmer H, Concin H, Diem G, Oberaigner W, Weiland SK. “Fasting blood glucose and cancer risk in a cohort of more than 140,000 adults in Austria.” Diabetologia. 2006 May;49(5):945-52.

[43] Nunes JPL and Silva JC. “Systemic correlates of angiographic coronary artery disease.” PLoS ONE. 2009; 4(1): e4322.

[44] Brunner EJ, Shipley MJ, Witte DR, Fuller JH, Marmot MG. “Relation between blood glucose and coronary mortality over 33 years in the Whitehall Study.” Diabetes Care. 2006 Jan;29(1):26-31.

[45] Bruning PF, Bonfrèr JM, van Noord PA, Hart AA, de Jong-Bakker M, Nooijen WJ. “Insulin resistance and breast-cancer risk.” Int J Cancer. 1992 Oct 21;52(4):511-6.

[46] Bruce WR, Wolever TM, Giacca A. “Mechanisms linking diet and colorectal cancer: the possible role of insulin resistance.” Nutr Cancer. 2000;37(1):19-26.

[47] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Cornell Chronicle. Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[48] Campbell TC et al. “China: From diseases of poverty to diseases of affluence. Policy implications of the epidemiological transition.”

[49] Segelken R. “China Study II: Switch to Western diet may bring Western-type diseases.” Cornell Chronicle. Retrieved from http://www.news.cornell.edu/chronicle/01/6.28.01/china_study_ii.html

[50] O’Dea K. “Diabetes in Australian aborigines: impact of the western diet and life style.” J Intern Med. 1992 Aug;232(2):103-17.

[51] Knight TM, Smith Z, Whittles A, Sahota P, Lockton JA, Hogg G, Bedford A, Toop M, Kernohan EE, Baker MR. “Insulin resistance, diabetes, and risk markers for ischaemic heart disease in Asian men and non-Asian in Bradford.” Br Heart J. 1992 May;67(5):343-50.

[52] Dhawan J, Bray CL, Warburton R, Ghambhir DS, Morris J. “Insulin resistance, high prevalence of diabetes, and cardiovascular risk in immigrant Asians. Genetic or environmental effect?” Br Heart J. 1994 November; 72(5): 413–421.

[53] Wang G, Liu X, Christoffel KK, Zhang S, Wang B, Liu R, Li Z, Liu X, Brickman WJ, Zimmerman D, Xu X, Wang X. “Prediabetes is not all about obesity: association between plasma leptin and prediabetes in lean rural Chinese adults.” Eur J Endocrinol. 2010 Aug;163(2):243-249. Epub 2010 May 28.

[54] Appleton BS, Campbell TC. “Inhibition of aflatoxin-initiated preneoplastic liver lesions by low dietary protein.” Nutr Cancer. 1982;3(4):200-6.

[55] Campbell TC. The China Study, p. 60.

[56] Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[57] Schulsinger DA, Root MM, and Campbell TC. “Effect of dietary protein quality on development of aflatoxin B1-induced hepatic preneoplastic lesions.” J Natl Cancer Inst. 1989 Aug 16;81(16):1241-5.

[58] Campbell TC. The China Study, p. 30-31.

[59] Parodi PW. “A role for milk proteins and their peptides in cancer prevention.” Curr Pharm Des. 2007;13(8):813-28.

[60] Bounous G, Batist G, Gold P. “Whey proteins in cancer prevention.” Cancer Lett. 1991 May 1;57(2):91-4.

[61] Bounous G. “Whey protein concentrate (WPC) and glutathione modulation in cancer treatment.” Anticancer Res. 2000 Nov-Dec;20(6C):4785-92.

[62] Hakkak R, Korourian S, Shelnutt SR, Lensing S, Ronis MJ, Badger TM. “Diets containing whey proteins or soy protein isolate protect against 7,12-dimethylbenz(a)anthracene-induced mammary tumors in female rats.” Cancer Epidemiol Biomarkers Prev. 2000 Jan;9(1):113-7.

[63] Badger TM, Ronis MJ, Hakkak R. “Developmental effects and health aspects of soy protein isolate, casein, and whey in male and female rats.” Int J Toxicol. 2001 May-Jun;20(3):165-74.

[64] Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[65] Junshi C et al. Diet, Life-style and Mortality in China, p. 572.

[66] Hwang SJ, Lee SD, Chang CF, Wu JC, Tsay SH, Lui WY, Chiang JH, Lo KJ. “Hypercholesterolaemia in patients with hepatocellular carcinoma.” J Gastroenterol Hepatol. 1992 Sep-Oct;7(5):491-6.

[67] Wang Y et al. “Fish consumption, blood docosahexaenoic acid and chronic diseases in Chinese rural populations.”

[68] Wild CP and Montesano R. “Correspondence re: T. Colin Campbell et al. Nonassociation of aflatoxin with primary liver cancer in a cross-sectional ecological survey in the People’s Republic of China.” Cancer Res. 1991(51):3825.

[69] Campbell TC. “A Challenge and Response to the China Study.” Retrieved from http://tynan.net/chinastudyresponse

[70] “Leading Causes of Death.” Centers for Disease Control and Prevention. Retrieved from http://www.cdc.gov/nchs/fastats/lcod.htm

[71] Junshi C et al. Diet, Life-style and Mortality in China, p. 608.

[72]Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[73] Campbell TC, Parpia B, and Chen J. “Diet, lifestyle, and the etiology of coronary artery disease: the Cornell China Study.” American Journal of Cardiology 1998 Nov 26;82(10B):18T-21T.

[74] Junshi C et al. Diet, Life-style and Mortality in China, p. 608.

[75] Ibid.

[76] Ibid.

[77] Ibid.

[78] “Nutrient Data Laboratory.” United States Department of Agriculture. Retrieved from http://www.nal.usda.gov/fnic/foodcomp/search/

[79] Junshi C et al. Diet, Life-style and Mortality in China, p. 608.

[80] Ibid, p. 606.

[81] Campbell TC, Chen J. “Energy balance: interpretation of data from rural China.” Toxicol Sci. 1999 Dec;52(2 Suppl):87-94.

[82] Junshi C et al. Diet, Life-style and Mortality in China, p. 608.

[83] Pottala JV, Garg S, Cohen BE, Whooley MA, Harris WS. “Blood eicosapentaenoic and docosahexaenoic acids predict all-cause mortality in patients with stable coronary heart disease: the Heart and Soul study.” Circ Cardiovasc Qual Outcomes. 2010 Jul;3(4):406-12.

[84] Manerba A, Vizzardi E, Metra M, Dei Cas L. “n-3 PUFAs and cardiovascular disease prevention.” Future Cardiol. 2010 May;6(3):343-50.

[85] Wang Y, Crawford MA, Chen J, Li J, Ghebremeskel K, Campbell TC, Fan W, Parker R, Leyton J. “Fish consumption, blood docosahexaenoic acid and chronic diseases in Chinese rural populations.” Comparative Biochemistry and Physiology Part A: Molecular and Integrative Physiology 2003 Sep;136(1):127-40.

[86] Junshi C et al. Diet, Life-style and Mortality in China, p. 606.

[87] Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[88] Fuhrman J. “Eat for Health: Cut Back on Grains.” Disease Proof. Retrieved from http://www.diseaseproof.com/archives/hurtful-food-eat-for-health-cut-back-on-grains.html

[89] Hokanson JE, Austin MA. “Plasma triglyceride level is a risk factor for cardiovascular disease independent of high-density lipoprotein cholesterol level: a meta-analysis of population-based prospective studies.” J Cardiovasc Risk. 1996 Apr;3(2):213-9.

[90] Iso H, Naito Y, Sato S, Kitamura A, Okamura T, Sankai T, Shimamoto T, Iida M, Komachi Y. “Serum triglycerides and risk of coronary heart disease among Japanese men and women.” Am J Epidemiol. 2001 Mar 1;153(5):490-9.

[91] McDougall J. “Refined Carbohydrates for Food Addicts.” McDougall Newsletter 2006:5(10). Retrieved from http://www.drmcdougall.com/misc/2006nl/oct/sugar.htm

[92] Junshi C et al. Diet, Life-style and Mortality in China, p. 608.

[93] Harris WS, Bulchandani D. “Why do omega-3 fatty acids lower serum triglycerides?” Curr Opin Lipidol. 2006 Aug;17(4):387-93.

[94] Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[95] Ibid.

[96] Ibid.

[97] Campbell TC. “A Challenge and Response to the China Study.” Retrieved from http://tynan.net/chinastudyresponse

[98] Junshi C et al. Diet, Life-style and Mortality in China, p. 732.

[99] Ibid, p. 716.

[100] Campbell TC. “Denise Minger Reply.” Campbell Coalition for Health Change. Retrieved from http://campbellcoalition.com/?p=142

[101] McDougall J. Dr. McDougall’s Digestive Tune-Up. Summertown, TN: Healthy Living Publications, 2006: p.146.

[102] Esselstyn CB. Prevent and Reverse Heart Disease. New York: Penguin Group, 2007: p. 70.

[103] Ornish D. Eat More, Weigh Less. New York: HarperCollins Publishers, 1993: p. 43.

[104] Ibid, p. 61.

[105] Barnard N and Grogan BC. Dr. Neal Barnard’s Program for Reversing Diabetes. New York: Random House, 2007: p. 65.

[106] Ibid, p. 84.

[107] Fuhrman J. Eat to Live: The Revolutionary Formula for Fast and Sustained Weight Loss. New York: Little, Brown and Company, 2003: p. 121.

[108] Ibid, p. 33.

[109] Ibid, p. 135.

[110] Fuhrman J. “What You Need to Know About Vegetarian or Vegan Diets.” Dr. Fuhrman. http://www.drfuhrman.com/library/article5.aspx

[111] Jönsson T, Granfeldt Y, Ahrén B, Branell UC, Pålsson G, Hansson A, Söderström M, and Lindeberg S. “Beneficial effects of a Paleolithic diet on cardiovascular risk factors in type 2 diabetes: a randomized cross-over pilot study.” Cardiovascular Diabetology July 16 2009 8:35.

[112] Ibid.

[113] Frassetto LA, Schloetter M, Mietus-Synder M, Morris RC Jr, Sebastian A. “Metabolic and physiologic improvements from consuming a paleolithic, hunter-gatherer type diet.” Eur J Clin Nutr. 2009 Aug;63(8):947-55.

[114] Ibid.

[115] Lindeberg S, Jönsson T, Granfeldt Y, Borgstrand E, Soffman J, Sjöström K, Ahrén B. “A Palaeolithic diet improves glucose tolerance more than a Mediterranean-like diet in individuals with ischaemic heart disease.” Diabetologia. 2007 Sep;50(9):1795-807.

[116] O’Dea K. “Traditional diet and food preferences of Australian aboriginal hunter-gatherers.” Philos Trans R Soc Lond B Biol Sci. 1991 Nov 29;334(1270):233-40; discussion 240-1.

[117] O’Dea K. “Cardiovascular disease risk factors in Australian aborigines.” Clin Exp Pharmacol Physiol. 1991 Feb;18(2):85-8.

[118] O’Dea K. “Diabetes in Australian aborigines: impact of the western diet and life style.” J Intern Med. 1992 Aug;232(2):103-17.

[119]Lindeberg S, Berntorp E, Nilsson-Ehle P, Terént A, Vessby B. “Age relations of cardiovascular risk factors in a traditional Melanesian society: the Kitava Study.” Am J Clin Nutr. 1997 Oct;66(4):845-52.

[120] Lindeberg S, Berntorp E, Carlsson R, Eliasson M, Marckmann P. “Haemostatic variables in Pacific Islanders apparently free from stroke and ischaemic heart disease–the Kitava Study.” Thromb Haemost. 1997 Jan;77(1):94-8.

[121] Lindeberg S, Lundh B. “Apparent absence of stroke and ischaemic heart disease in a traditional Melanesian island: a clinical study in Kitava.” J Intern Med. 1993 Mar;233(3):269-75.

[122]Mann GV, Shaffer RD, Anderson RS, Sandstead HH et al. “Cardiovascular disease in the Masai.” J Atheroscler Res. 1964 Jul-Aug;4:289-312.

[123]Taylor CB, Ho KJ. “Studies on the Masai.” Am J Clin Nutr. 1971 Nov;24(11):1291-3.

[124] Ibid.

[125] Eilat-Adar S, Mete M, Nobmann ED, Xu J, Fabsitz RR, Ebbesson SO, Howard BV. “Dietary patterns are linked to cardiovascular risk factors but not to inflammatory markers in Alaska Eskimos.” J Nutr. 2009 Dec;139(12):2322-8.

[126] Ibid.

[127] Ebbesson SO, Ebbeson LO, Swenson M et al. “A successful diabetes prevention study in Eskimos: the Alaska Siberia project.” Int J Circumpolar Health 64 (2005):409–424

[128] Bersamin A, Luick BR, King IB, Stern JS, Zidenberg-Cherr S. “Westernizing diets influence fat intake, red blood cell fatty acid composition, and health in remote Alaskan Native communities in the center for Alaska Native health study.”J Am Diet Assoc. 2008 Feb;108(2):266-73.

[129] Campbell TC. “A Challenge and Response to the China Study.” Retrieved from http://tynan.net/chinastudyresponse

Here it is, folks. This should take you straight to the PDF:

“The China Study”: A Formal Analysis and Response

(Updated noon-ish PST on August 3rd with typo corrections)

If you haven’t done so yet, also read Campbell’s first response and Campbell’s second response, which this is in reply to.

I’ll see what I can do about getting this set up in blog-post form, but I really don’t have the mental capacity to work on it right now. Sorry. In the meantime, here’s the table of contents so you know what you’re getting yourself into:

SECTION 1: Reiteration and Expansion of Criticisms

1. Linkage of animal protein with cancer by way of cholesterol
2. Misleading association of breast cancer with lipid intake and lipid intake with animal protein
3. Supposition that plasma cholesterol increases liver cancer risk
4. Misrepresentation of heart-protective effects of green vegetables, and the three-variable linkage between animal protein, apolipoprotein B, and cardiovascular disease
5. Biased use of unadjusted univariate correlations to confer protective benefits of plant foods but not with animal foods
6. Use of a three-variable chain to connect animal foods with “Western” diseases
7. Unexplored role of blood glucose, insulin, and disease
8. Dismissing relevant variables
9. Errors in the extrapolation of casein to all animal protein

SECTION 2: Biological Models and Cited Papers

1. Breast cancer
2. Liver cancer
3. Energy utilization
4. Affluent-poverty diseases
5. Summary

SECTION 3: Response to Points Raised by Campbell

1. Wheat: confounded variable or legitimate concern?
2. Selection of univariate correlations and confirmation bias
3. Tuoli county and erroneous data
4. Whole-food, plant-based diets versus whole-food diets with animal products
5. Conclusion

Also, this is sort of a pre-final version, and there may be typos (please point them out!) or orphaned punctuation (ditto). If I make any changes, I’ll post the updated version with a note.

Now if you’ll excuse me, I’m going to spend a very, very long time not staring at the computer screen, catching up on a couple weeks’ worth of sleep, and hopefully regrowing the little chunks of my soul that died while writing this. Adieu!

Yeah, yeah, yeah. It’s now Saturday. Gotta love being a multiple-offense deadline breaker. (I tend to value thoroughness over timeliness–so anyone out there who was thinking of hiring me for any time-sensitive job, you’ve been duly warned.) I’m currently adding a final section on wheat to Campbell response #2, and then this puppy WILL be ready to post. Pinky swear! Thanks for bearing with me.

–(end update/start of older post)–

I’m excited to see quite a few people take interest in the China Study data (huzzah, numbers!), and even more excited that some of you are already posting the results of your analyses. To quote reader and blogger Ned Kock:

I hope more people will do their own analyses on the original data, like we have been doing. Then the discussion will move away from X or Y are saying this, to something more like “the data” is saying this.

Right on.

While I’m finishing a fairly laborious (you’ll see what I mean later) response  to Mr. Campbell, I thought I’d post some of the data I already have typed up for those of you who are gettin’ antsy. I’ll be updating this entry frequently as I upload more files, but here’s the first batch.

I’ll also use this post to link to anyone who has posted their results somewhere on the ‘net. Those will be right after the links to the data.

Also feel free to request any variable(s) you’re interested in analyzing, and I’ll type them up when I have a spare moment.

Enjoy!

Myocardial infarction/coronary heart disease:

(includes total cholesterol, HDL cholesterol, green vegetable consumption, animal protein, plant protein, dairy variables, egg variables, meat variables, and fish variables)

• Myocardial infarction variables for Excell 2007
• Myocardial infarction/CHD variables for Excell ’97 – ’03

(Note: included are the variables “amount of green vegetables consumed” and “frequency of green vegetables consumed” to illustrate the Green Veggie Paradox.)

Colorectal cancer:

(includes cholesterol, schistosomiasis, plant protein, and animal protein)

• Colorectal cancer variables for Excell 2007
• Colorectal cancer variables for Excell ’97 – ’03

(A shout out to eds. Chen Junshi, T. Colin Campbell, Li Junyao, and Richard Peto for making this stuff available in book form.)

Reader links:

So far, we have two posts from Ned Kock:

1. The China Study again: A multivariate analysis suggesting that schistosomiasis rules!
2. The China Study one more time: Are raw plant foods giving people cancer? (This one’s particularly interesting: Ned used a nonlinear regression analysis on the data with no schistosomiasis infection, and uncovered a U-curve in the relationship between cholesterol and colorectal cancer. In other words, the counties with the lowest cholesterol and highest cholesterol had higher rates of colorectal cancer than the groups with more mid-range cholesterol, who appear the most protected. Ned offers a great hypothesis for this result in his post. Additionally, while animal protein consumption correlated strongly with total cholesterol, animal protein itself correlated inversely (beta = -0.31, p<0.10) with colorectal cancer, while plant protein correlated positively (beta = 0.47, p<0.01). Remember, of course, that correlation doesn’t equal causation, and this is just a sampling of the dizzying number of variables recorded in the China Study.)

Update 7/22: Reader Ned Kock of “Health Correlator” performed a multivariate analysis on the data for colorectal cancer, animal protein, cholesterol, plant protein, and schistosomiasis from the China Study. Check his blog to read what he discovered. (Any other readers who’ve done something similar, please post and let us know what you’ve found as well.)

In other news:

If you haven’t seen it yet, Campbell has expanded his original response to my critique and posted it in two places:

1. On his website TColinCampbell.org, where it’s available for download as a Word document, and
2. On CampbellCoalition.com, where it’s in HTML format and you can contribute comments and questions.

Word has it that Campbell himself will be replying to at least some of the comments on Campbell Coalition, so this would be a wonderful opportunity for anyone with questions for him to engage in dialogue. Correction 7/22: Campbell has closed this discussion to comments with the following remark:

Based on the response received thus far, we have determined that our prior idea of a reasoned and civil discourse, with participation by Dr. Campbell, is not feasible and have decided to discontinue this discussion thread.

Bummer. Well, if you want to carry a non-reasoned and un-civil discourse, feel free to do it here. First Amendment FTW!

If you submitted comments that weren’t accepted on the Campbell Coalition website, Dave Dixon has created a special entry on his blog “Spark of Reason” where you can post them and still get your voice heard.

Campbell’s longer rebuttal has also been featured on Vegsource.com, in which the editors kindly wrote:

Previously we at VegSource had looked at some of Ms. Minger’s anti-Campbell rhetoric.  One thing we were struck by early on was the fact that Ms. Minger apparently removes comments on her blog from scientific researchers who point out the flaws in her reasoning and in her understanding of accepted research methods.

Huh. All scientific researchers who had their comments removed, please say “aye.” The one and only comment I’ve deleted thus far was one I wrote, although (as I’ve mentioned several times now) some comments do get snagged in the spam or “awaiting approval’” queue, especially if they have links–in which case they don’t show up right away.  I apologize if this has happened to you, but you’re welcome to comment here even if you disagree. Dissenting voices FTW!

Update 7/22: Looks like they edited the above to be marginally nicer but still woefully inaccurate. And, as per tradition, they took a moment to lambaste the Weston A. Price Foundation—’cause really, what China Study article would be complete without randomly evoking something completely irrelevant to the discussion? Non-sequiturs FTW!

I have (another) response to Campbell underway, so for those of you waiting for the wheat post, it just got pushed back farther in the waiting line. Many apologies. Contrary to some circulating hypotheses, I really am just one person, with limited capacity to type and crank out blog entries. When I finish rearing my army of bovine ninja babies, I’ll enslave them and outsource my research and data entry tasks, but that’s a ways off yet.

Carry on.

Instead, it went viral and racked up 20,000 page views within 24 hours.

I’m surprised, but equally thrilled. My self-marketing skills are pretty dismal, and it was only by the grace of all the bloggers who featured my critique that this page-view boom occurred. Thank you to everyone who helped spread the word. I owe y’all!

This post is going to be quite long (no shocker there) and, in places, a bit more technical than the last. I know not everyone digs science mumbo-jumbo, so I’ll try to keep that to a minimum and explain things like journal quotes in simpler terms.

First, I’d like to address a couple points I’ve seen crop up in reader comments and emails I’ve received.

One: My graphs and simple statistical explanations. The graphs I posted were not intended to stand as new hypotheses or conclusions about the data. I apologize if I didn’t make this abundantly clear. Their sole purpose was to demonstrate, to the general layperson, how raw correlations (in the instances Campbell used them) can be misleading—as well as show how dramatically a single confounder can affect a correlation and make a positive trend appear where there may not be one at all. The graphs and explanations were meant to be illustrative, not exhaustive.

Two: Bias in Campbell’s representation of the data. This is a point I feel has been overlooked by some critics who’ve myopically targeted my use of statistics.

My biggest concern is with the way data appears to be cherry-picked to create a “plant foods are good” and “animal foods are bad” dichotomy when the actual data from the China Study (as well as from Campbell’s own research) does not reflect this.

For instance, when citing the anti-disease effects of plant foods, Campbell points to inverse correlations with biomarkers for plant food consumption as well as plant food intake itself. One example is in Claim #5 when he notes stomach cancer is inversely associated with plasma concentrations of beta-carotene and vitamin C (biomarkers) as well as with green vegetable intake (a plant food). (Both of these claims are based on uncorrected correlations, by the way.)

Yet when citing the purportedly harmful effects of animal foods, Campbell relies on blood markers (usually total cholesterol or apo-B) but fails to find direct relationships between disease and the animal foods themselves. He still indicts animal foods as harmful, but comes to this conclusion by enlisting the help of intermediary variables. And as I explained in the last post and will continue explaining in this one, the link between cholesterol levels and the diseases Campbell links them to are not even as straightforward as he suggests.

To those who approach this discussion already believing animal foods are generally unhealthy, this bias is subtle and might not be obvious. But to those who approach this discussion from a place of neutrality, the bias is unmistakable.

Response to Campbell

Now, onto business.

In case you haven’t heard yet, the much-discussed T. Colin Campbell wrote a response to my critique of his book. If you haven’t already done so, hop on over and read it on Tynan.net.

Let me preface this with something important. When it comes to science, my motto is an old line from Dragnet (which, having no TV, I’ve never actually watched): “Just the facts, Ma’am.” Or sir. Science itself should be cool, neutral, and somewhat soulless. As far as I’m concerned, personal conflicts, drama, mudslinging, grudges, and other flurries of emotion should be locked out of science’s doors and banned for life.

For this reason, I want to make it clear that even though I disagree with Campbell’s interpretation of the China Study data, I have no interest in launching any personal vendetta against him. I know some readers are none too pleased with the man, but I do believe he’s trying to promote a message he deeply believes will help others. I won’t be participating in any character attacks, regardless of how I feel about his interpretations.

That said, I’m a bit disappointed Campbell didn’t offer a more revealing glimpse into his own methods of analysis. Here’s a secret: If he wants to silence his critics, all he has to do is publish the details of his process—which, apparently, he has already written up:

A more appropriate method is to search for aggregate groups of data, as in the ‘affluent’ vs. ‘poverty’ disease groups … I actually had written material for our book, elaborating some of these issues but was told that I had already exceeded what is a resonable [sic] number of pages. There simply were not enough pages to go into the lengthy discussions that would have been required–and I had to drop what I had already written.

I’ve emailed Mr. Campbell and asked him to consider publishing this material somewhere as a downloadable PDF or in another accessible form. Then the rest of us can study his methodology, look for oversights, and hopefully replicate his findings. I’ll update this when or if he responds.

UPDATE: Campbell has informed me via email:

To go back and fetch the material that I had previously written would take a lot of time that I don’t have. Also, much of it is in my peer-reviewed 300+ scientific papers.

Well, shucky darns. Although he doesn’t have time to fetch already-written material, he does have time to craft a more thorough response to my critique than the one published on Tynan.net, which he’ll be posting on his website sometime soon. Hopefully he’ll provide more details about his methods there.

Some responses to specific parts of Campbell’s letter

To clarify who’s saying what, quotes will always be italicized and indented. All bold parts of the quotations are my own emphasis and not the original author’s.

Campbell: She claims to have no biases–either for or against–but nonetheless liberally uses adjectives and cutesy expressions that leaves me wondering.

News flash: I was an English major with a creative writing emphasis. Cutesy is my thang. When the occasion calls for it, I can become Formal Monotone Academic Denise—but seeing as this is a blog and I needed to keep readers hooked for 9,000 words, I figured a more colloquial tone would be best.

Also, I wasn’t aware adjectives indicated bias, and if that’s the case, boy am I ever in trouble. You know what else? I sometimes use adverbs. That’s right. Evil adverbs. I learned them from Stalin when we worked together in the ’40s (oops, did I say that out loud?).

Campbell: As far as her substantive comments are concerned, almost all are based on her citing univariate correlations in the China project.

Actually, they’re based on the univariate correlations that Campbell cited first.

If you read my critique, you’ll see that Campbell’s claims align with the raw and uncorrected data, which—as I tried to illustrate—can be misleading due to the influence of other variables implicated with disease.

But it seems my critique wasn’t enough to convince some Campbell supporters that he did not use exhaustive analytical methods under some important circumstances, so I’ll present examples straight from his peer-reviewed papers.

First, let’s look at “Diet, Lifestyle, and the Etiology of Coronary Artery Disease: The Cornell China Study” published in the November 1998 issue of the American Journal of Cardiology. One statement from the paper, from the section discussing “Diet-Coronary Artery Disease Relations,” notes the following:

The combined coronary artery disease mortality rates for both genders in rural China were inversely associated with the frequency of intake of green vegetables (r = -0.43, p<0.01)…

Remember the “Green Veggie Paradox” from my last post, which pointed out that frequency of green vegetable consumption may be a geographical marker for southern regions where heart disease rates are lower, instead of an actual protective agent against heart disease? Well, here’s that paradox again. In a peer-reviewed article. Co-authored by Campbell. Using the raw data (-0.43). And neither him nor the rest of his team made adjustments, ran more sophisticated analyses to account for confounding variables, or even mentioned other factors that could explain the correlation between frequent green-vegetable consumption and healthier hearts.

Had Campbell tried to understand the apparent discrepancy between frequency of green vegetable consumption (which had a strong inverse association with coronary heart disease) and the amount of green vegetables consumed (which had a weak positive association with coronary heart disease), he may have realized there was more to our fibrous friends than meets the eye. For instance, geography is closely tied to heart disease in the China Study data, with lower latitudes exhibiting lower rates. And if frequency of green vegetable consumption strongly reflects geography, it seems any researcher committed to accuracy would want to tease apart these variables before citing them in a scientific paper.

In this article, Campbell also employs several other unadjusted correlations straight from the monograph:

The combined coronary artery disease mortality rates for both genders in rural China were inversely associated with … plasma erythrocyte monounsaturated fatty acids (r = 0.64, p<0.01), but positively associated with a combined index of salt intake plus urinary sodium (r = 0.42, p<0.01) and plasma apolipoprotein B (r = 0.37, p<0.01).

These numbers are all raw correlations. Campbell didn’t conduct a deeper statistical analysis on any of it to account for potential confounders, such as lifestyle habits or other dietary factors that might accompany specific biomarkers.

These apolipoproteins, in turn, are positively associated with animal protein intake (r = 0.26, p <0.05) and the frequency of meat intake (r = 0.32, p<0.01) and inversely associated with plant protein (r = 0.37, p <0.01), legume (r = 0.26, p<0.05), and light colored vegetable intake (r = 0.25,  p <0.05).

Again, we have a match with the uncorrected data. And again, Campbell and his team didn’t appear to run multiple variable regressions or any other analyses to see if the raw data was accurate. (And notice how Campbell can’t say animal protein itself associates with heart disease, but has to pull a connecting variable into the picture to make his theory fit.)

Why didn’t Campbell pay more attention to the role of confounders? Why did he accept the raw data, which showed plant foods as protective and an animal-food biomarker as harmful, without conducting deeper analyses?

This might be the answer:

The principal hypothesis of this study was that the greater the dietary proportion of a variety of good-quality plant-based foods, the lower the rate of chronic degenerative diseases.

Essentially, Campbell and his team approached the data set specifically looking for trends showing plant foods to protect against disease (and, perhaps, showing animal foods to be harmful).

If you’ll recall, the China Study has 8,000 statistically significant correlations. That’s a lot. Enough, in fact, to find pretty much anything you want if you look hard enough—especially if you use a bit of sloppy science and cite raw correlations or chains of variables when they suit your needs.

Of course, that’s not the only China-Study-based  paper showcasing analytical shortcomings. Let’s look at “Fish consumption, blood docosahexaenoic acid and chronic diseases in Chinese rural populations” published in the September 2003 issue of Comparative Biochemistry and Physiology. This paper examines the role of fish and the essential fatty acid DHA in relation to several diseases.

Campbell and his crew’s methodology for studying the variables:

Pearson’s correlation coefficient was used to explore the relationship between variables. The two-tailed test of significance was used to examine the significant differences within variables.

Alright, this is your standard high school stuff: examining the linear relationship between two variables. No multiple variable regressions. No adjustments for confounding variables. And from these rudimentary correlations, Campbell and his team cite a number of observations about the relationships between fish, other meat, total lipids, blood markers, and disease, ultimately concluding:

[T]he protective nature of DHA or aquatic foods is intrinsic and global, with implications for health world wide. The decline in sea and fresh water food consumption in many regions last century could be an adverse, contributory factor to the increasing risk of chronic diseases and the rise in mental ill health …

Researchers concluded from these raw correlations that the DHA is associated with lower risk of many chronic diseases. But might this effect become even more pronounced through different statistical models—namely ones that account for confounding variables?

It seems likely, and here’s why. In this paper, Campbell and his team noted that diabetes is positively associated with DHA in the China Study data, despite other research showing the opposite:

Diabetes showed a positive but non-significant relation with DHA in Fig. 2, which meant no clear-cut conclusion about the efficacy of lower DHA level in diabetes even though a negative association has been found between DHA and triglycerides in plasma [in previous research]. … [We] have no explanation for the positive correlation with diabetes.

No explanation, eh? I’ve got one. In this paper, Campbell and other researchers determine that fish is the most significant source of DHA in the studied counties. And we know from the China Study monograph that fish-eating regions tended to have high intakes of processed starch and sugar compared to other counties—a correlation of 0.58. Could processed sugar and starch intake be skewing the relationship between DHA and diseases like diabetes? If so, why didn’t Campbell et al run more appropriate analyses to account for this?

Still not convinced Campbell’s methods are less than perfect? Here’s some more. From “Diet and chronic degenerative diseases: perspectives from China,” published in the May of 1994 issue of the American Journal of Clinical Nutrition:

Intakes of 14 complex carbohydrate and fiber fractions were obtained in this study to determine whether particular fiber fractions were associated with particular diseases, especially cancers of the large bowel. … Based on an overview of the univariate correlations, colon and rectal cancer mortality rates were consistently inversely correlated with all fiber and complex carbohydrate fractions except for pectin, which showed no correlation.

So here we have Campbell and his team using univariate correlations to look at the relationship between fiber and colorectal (large bowel) cancer. No adjustments made for potential confounding variables. And from these correlations, he concludes:

[T]here is evidence of a weak inverse relationship between cancer of the large bowel and the intake of multiple complex carbohydrate and dietary fiber fractions.

In other words, the fiber fractions seemed to protect against colorectal cancer across the board. But is this an accurate inference?

Had Campbell looked more closely at the data (instead of assuming the raw figures were accurate, as he seems fond of doing when it supports anti-disease properties of plants), he would’ve noticed something striking. The correlations between those 14 fiber fractions and colorectal cancer seem to mirror the correlations between the fiber fractions and schistosomiasis infection.

Okay, I know what you’re thinking. “What’s Denise blathering on about this time?” Let’s back up for a minute.

Schistosomiasis (also called bilharzia) is a parasitic disease known to raise risk of colorectal cancers. If you get infected with one of these lovely worms, they’ll lay eggs that travel to your liver, intestine, or bladder, where they can cause permanent damage and inflammation. How fun!

The link with colorectal cancers isn’t something I’m just pulling out of the air, by the way. It’s pretty well established. Some references:

• “Schistosoma japonicum and colorectal cancer: an epidemiological study in the People’s Republic of China“: Prevalence of infestation with Schistosoma japonicum was highly correlated with mortality from colorectal cancer in 89 communes in four counties of Jiangsu province, China (rank correlation coefficient = 0.68) in 1973-75, and with incidence of colorectal cancer in 24 communes of Haining county, Zhejiang province in 1977-79.
• “Correlations of colon cancer mortality with dietary factors, serum markers, and schistosomiasis in China“: [P]revalence of schistosomiasis was significantly correlated with increased colon cancer mortality.
• “Schistosomiasis and its prognostic significance in patients with colorectal cancer. National Cooperative Group on Pathology and Prognosis of Colorectal Cancer“: This paper analyses 430 cases of colorectal cancer complicated with schistosomiasis. The 5 year survival rate was 45.6%, lower than that without schistosomiasis. … The infection of schistosome should be considered as one of the important factors in prognosis. (In other words: Schistosomiasis infection increases the mortality rate of colorectal cancer sufferers.)
• “A cohort study on the causes of death in an endemic area of schistosomiasis japonica in Japan“: These results suggest that schistosomiasis japonica is one of the important risk factors for cirrhosis of the liver, cancer of the liver and cancer of the colon.

With that in mind, it seems pretty obvious that Campbell would want to look at a schistosomiasis infection in relation to colon cancer occurrence, especially since 1) it’s pretty common in Asia and 2) it could be a confounding variable. In fact, schistosomiasis has a correlation of 0.89 with colorectal cancer mortality in the China Study data. (If you’re having a déjà vu moment, you’re not crazy: I wrote about this in the previous entry as well.)

So what does this have to do with fiber?

The fiber fractions Campbell cites as having a “weak inverse relationship” with “cancer of the large bowel” also have a somewhat stronger inverse relationship with schistosomiasis. In other words, fiber is already likely to be associated with less colorectal cancer simply because those who eat more of it tended to have less of another significant risk factor.

It might help to represent this visually, so here’s a graph plotting each fiber fraction’s correlation with schistosomiasis and colorectal cancer. These are the fiber fractions corresponding to the x-axis numbers:

1. Total fiber
2. Total neutral detergent fiber
3. Hemi-cellulose fiber
4. Cellulose fiber
5. Lignins remaining after cutin removed
6. Cutin
7. Starch
8. Pectin
9. Rhamnose
10. Fucose
11. Arabinose
12. Xylose
13. Mannose
14. Galactose

Bottom line: Is the inverse relationship between fiber and colorectal cancer legitimate, or is that correlation influenced by schistosomiasis rates? Given the relationship between these variables, shouldn’t Campbell have run a more thorough analysis on the data?

I sure think so. But he didn’t. Again, he seems to readily accept uncorrected correlations when they prove his theory.

So, what happens when we do adjust for confounding variables? Let’s look at another of Campbell’s peer-reviewed papers: “Erythrocyte fatty acids, plasma lipids, and cardiovascular disease in rural China” published in the December 1990 issue of the American Journal of Clinical Nutrition. Here were their statistical methods:

To adjust for the effect of other factors in the relationship between two variables, ordinary least-squares multiple-regression analysis was used. Natural logarithmic transformations of the mortality rates (the dependent variable in the models) were used to obtain a normal distribution of the outcome variable for reliable statistical significance testing of the regression coefficients.

No uncorrected correlations here. And the results:

Within China neither plasma total cholesterol nor LDL cholesterol was associated with CVD [cardiovascular disease]. The results indicate that geographical differences in CVD mortality within China are caused primarily by factors other than dietary or plasma cholesterol.

Did you catch that? After adjusting for confounding variables, researchers found that cholesterol was not associated with cardiovascular disease in the China Study data. And that includes both blood cholesterol and cholesterol from food.

Let that sink in for a moment.

Nah, this is pretty big: Give it two moments.

Or three.

Now, let’s look at Campbell’s next point, which flows quite nicely from the last:

Diseases of affluence and diseases of poverty

Campbell: A more appropriate method is to search for aggregate groups of data, as in the ‘affluent’ vs. ‘poverty’ disease groups, then examine whether there is any consistency within groups of biomarkers, as in considering various cholesterol fractions.

If you’re unfamiliar with Campbell’s disease-clustering strategy, you can read “From Diseases of Poverty to Diseases of Affluence” to get a feel for it (although be warned, the formatting is a little wonky). In essence, Campbell examined the China Study data and identified two distinct groups of diseases that were generally associated with each other—with one group representing diseases common to developing nations and the other representing “Western” afflictions.

In the article linked above, Campbell et al describe the first group:

As expected, diseases of poverty are associated more with agricultural than with industrial activity. Areas where these diseases are common are located further inland where mean elevation is higher and overall economic activity, literacy and population density are lower.

And the second group:

In contrast, diseases of affluence are found in the more densely populated rural areas nearer the seacoast where industrial activity and literacy rates are higher and more fish, eggs, soy sauce, beer and processed starch and sugar products are consumed.

More specifically, Campbell defines the “diseases of poverty” as:

• Pneumonia
• Intestinal obstructions
• Peptic ulcer
• Other digestive disorders
• Nephritis
• Pulmonary tuberculosis
• Infectious diseases (other than schistosomiasis)
• Eclampsia
• Rheumatic heart disease
• Metabolic and endocrine disease (other than diabetes)
• Diseases of pregnancy and birth (other than eclampsia)

And “diseases of affluence” include:

• Stomach cancer
• Liver cancer
• Colon cancer
• Lung cancer
• Breast cancer
• Leukemia
• Diabetes
• Coronary heart disease
• Brain cancer (ages 0-14 years)

Again, the diseases in each cluster tend to associate positively with each other but inversely with the diseases in the opposite group.

It’s not a bad strategy, really. Campbell uses this two-group method to identify general factors (such as nutritional patterns) related to each disease cluster, taking a holistic view of disease rather than examining ailments through reductionism. This approach aligns with something I very much agree with: that diseases don’t happen in isolation, but that multiple forms of chronic disease can spring from the same cause (poor nutrition, processed foods, unhealthful living, and so forth).

But while I agree with this general method, it’s not without flaws—and the way Campbell employs it to study nutrition and disease requires a few leaps of faith.

First, some problems with the groups Campbell created:

1. Not all of the “diseases of affluence” are actually common in affluent countries, raising questions about whether these disease clusters apply outside of China. For instance, the two most prevalent diseases of affluence in the China Study data are liver cancer and stomach cancer—but in the US, a decidedly affluent nation, these diseases make up less than 5% of all cancer deaths.
2. Where’s “stroke” on either list? Nowhere to be found. Campbell had to create a third group called “Other” for a few diseases that didn’t fit cleanly into the other two clusters. According to the American Heart Association, stroke is currently the third leading cause of death in America. So what explains its lack of correlation with other diseases of affluence? Campbell offers no insights.

Perhaps more importantly, Campbell makes some excellent observations about the nutritional variables correlating with diseases of affluence, but then dismisses them without any satisfying or even logical explanation. He lists the following correlations between several foods and his affluent disease cluster:

• Processed starch and sugar: 0.51
• Fish (g/day): 0.56
• Beer: 0.59
• Eggs (times per year): 0.31

Since the industrialized areas with diseases of affluence tended to be near the coast, it’s not surprising fish consumption was high. But that’s a pretty hefty correlation with processed starch and sugar, too. Could those refined carbs contribute to diseases of affluence? Eh? Eh?

Apparently not. Campbell doesn’t consider them significant in the China Study data. He states that “beer and processed starch and sugar products are also consumed in much lower quantities [than in the US],” and therefore “consumption of these foods is probably more indicative of general economic conditions and other local circumstances than of biological relationships to disease.” And that’s the last we hear about ‘em.

That’s right, folks.

Here we have evidence that areas in China with the highest rates of Western-type diseases also eat the most processed starch and sugar. Maybe not in the grotesque amounts that Americans eat them, but then again, China’s “affluent disease” rates were also lower than America’s.

But instead of examining the relationship between processed carbohydrates and poor health, Campbell zeros in on another variable associated with industrialized nations and diseases of affluence. And if you’ve been paying attention to this post and the last, that variable won’t surprise you: It’s cholesterol.

By the way, the correlation between Campbell’s affluent diseases (in the aggregate) and cholesterol is 0.48, slightly less than the correlation with processed starch and sugar. And if you’ll recall, Campbell’s own analysis showed that cholesterol levels in the China Study data didn’t associate with cardiovascular disease, a major cause of “affluent” mortality. But I guess that doesn’t matter, because Campbell says so and Campbell has lots of credentials.

But back to Campbell’s response. His statement that a more appropriate method of analysis is to “search for aggregate groups of data, as in the ‘affluent’ vs. ‘poverty’ disease groups, then examine whether there is any consistency within groups of biomarkers” is something I can at least partially agree with. Yet in examining Campbell’s own use of these disease groups, I smell another whiff of bias: He immediately implicates cholesterol (and, as a consequence, animal products) as causative of disease, when at least four other diet variables (most notably processed starch and sugar) are also heavily implicated with diseases of affluence.

Now, for something completely different:

The “Mysterious Tuoli” not so mysterious?

Campbell: [W]e discovered after the project was completed that meat consumption for one of the counties, Tuoli, was clearly not accurate on the 3 days that the data were being collected. On those days, they were essentially eating as if it were a feast to impress the survey team but on the question of frequency of consumption over the course of a year, it was very different.

I’m glad Campbell pointed this out (and I’ll be updating the Tuoli page to reflect it), but meat was not the component I found notable with the Tuoli diet: dairy was. Assuming the frequency questionnaire was more reliable than the three-day diet survey, the Tuoli still consumed dairy most days of the year and still consumed nearly no vegetables (twice per year), nearly no fruit (once per year), and ate wheat as their primary plant food. Not exactly a balanced diet—yet, compared to the rest of China, they remained in good health.

(By the way, a number of you have asked for help finding more information about the Tuoli. A Google search for “Tuoli” doesn’t reel in a whole lot of relevant hits, so you can try the alternative English spelling of “Toli,” or a search for a related group of people called “Uyghur” or “Uygur” in the Xinjiang Autonomous Region of China.)

However, Campbell’s statement about the unreliability of the diet survey for the Tuoli also calls into question the validity of the three-day diet survey as a whole—as well as the significant observations Campbell gleaned from it. For instance, on page 99 of “The China Study,” Campbell notes:

Average calorie intake, per kilogram of body weight, was 30% higher among the least active Chinese than among average Americans. Yet, body weight was 20% lower. How can it be that even the least active Chinese consume more calories yet have no overweight problems? What is their secret?

If Tuoli is any indication, there may not be a secret at all. Since Campbell drew his calorie data from the three-day diet survey, suppose multiple counties tried to impress researchers by “feasting” or otherwise altering their eating habits to reflect greater wealth, prosperity, or food abundance than they actually had. The result? Calorie intake during those three days would be higher than for the rest of the year, leading to an overestimated average calorie intake for the 65 counties studied.

Did Campbell consider this, especially given his awareness about the unreliable records for the Tuoli? Apparently not. On page 101, he states:

Chinese consume more calories both because they are more physically active and because their adoption of low-fat, low-protein diets shifts conversion of these calories away from body fat to body heat. This is true even for the least physically active Chinese.

Physical activity certain plays a role in higher calorie requirements, but eating a low fat, low protein diet may not increase thermogenesis as Campbell suggests—at least not based on the China Study data. Some counties may have simply been showing off by stuffing themselves silly, leading to high average calorie intakes. We’ve got Campbell’s assertion that at least one place did this: How do we know others didn’t as well?

Again, let me highlight what appears to be another link in a chain of bias: Campbell dismisses the low disease rates and high animal protein intake of the Tuoli because the three-day diet survey was inaccurate, yet doesn’t account for potential shortcomings in that diet survey when it helps score brownie points for plant foods.

Moving on.

Campbell: One final note: she repeatedly uses the ‘V’ words (vegan, vegetarian) in a way that disingenuously suggests that this was my main motive.

I understand—and respect—that Campbell was trying to avoid the ethical implications of the word “vegan,” since the term often conveys a complete lifestyle choice rather than just a diet. However, my intent was definitely not disingenuous, nor was I trying to peg a motive on Campbell. My own use of the term “vegan” was simply to describe a completely animal-product-free diet. I apologize if this wasn’t clear from my post.

Campbell: One further flaw, just like the Weston Price enthusiasts, is her assumption that it was the China project itself, almost standing alone, that determined my conclusions for the book (it was only one chapter!).

I guess Campbell missed the 2,135 words I dedicated to his research on casein, including the problems with extrapolating its effects to all animal protein. And the citation of his own research showing it’s a full spectrum of amino acids, not just animal protein, that apparently spurs cancer in aflatoxin-exposed rats. And the insight that a vegan diet provides all amino acids (and thus complete protein) if you eat a variety of plant foods, thereby posing similar purported risks as omnivory in terms of cancer growth. And the question about the apparent unhealthfulness of breastfeeding and exposing young, delicate-bodied children to casein. And the glaring example of bias in Campbell’s treatment of animal versus plant protein in relation to body size and disease.

Easy oversight, I guess. It was a pretty formidable post. As is this one, apparently.

By the way, if anyone had trouble following my train of thought in the casein/wheat/lysine/complete protein section of the critique, Chris Masterjohn has written a more “digestible” article (pun definitely intended) expanding on this subject and probably explaining it better than I did. Yep, that’s the same Chris who wrote a well-known critique of “The China Study” five years ago.

Next up, a very serious and momentous subject:

Does Denise work for the meat and dairy industry/is Denise a cyborg/is Denise a figment of your imagination/is Denise actually Campbell’s employee, son, dog, long-lost daughter, or alter-ego?

Campbell: I find it very puzzling that someone with virtually no training in this science can do such a lengthy and detailed analysis in their supposedly spare time.

And:

Campbell: I have no proof, of course, whether this young girl is anything other than who she says she is, but I find it very difficult to accept her statement that this was her innocent and objective reasoning, and hers alone. If she did this alone, based on her personal experiences from age 7 (as she describes it), I am more than impressed.

Then thank you for the compliment, Mr. Campbell! I’m definitely a singular person, so I’m glad to more-than-impress you.

Initially, I didn’t want to muddy this post with retorts to statements like this, but really. What’s so hard to believe about a 23-year-old Super Nerd deciding to tackle a project out of personal interest? What do I need to show to prove I’ve got a brain in this noggin? College transcripts? 4.0, three scholarships, dean’s list, top 1% of the class? I can say the alphabet backwards, too. That has to count for something.

In all seriousness, I can understand why Campbell would express skepticism that a young person would have the resources or repertoire of knowledge necessary to tackle this sort of project. And I think it may largely be a generational issue. When Campbell was a young’un, he didn’t have access to the internet or online books or PubMed or Google Scholar or any of the other self-educational tools most of us now take for granted. For him, education did necessitate sitting in a room with a teacher, pouring over textbooks, showing up to a physical classroom, and accumulating credentials to prove you’d survived the journey. These days, education can manifest in numerous other forms.

In other words, I’m more flattered than offended.

Other odds and ends

Several readers have raised an issue that probably deserves more attention than I’ve given it so far: the limitations of the China Study itself. Although I’ve focused on examining the errors and biases in Campbell’s conclusions, the fact of the matter is, this study itself is just a big ol’ epidemiological survey—and any analyses it produces, no matter how thorough, are inherently limited due to the nature of the data.

In fact, before Campbell’s “The China Study” was even released, Thomas Billings of Beyondveg.com wrote an excellent overview of the shortcomings of the study itself. I recommend reading this if you want a better understanding of what a study like the China Project can and cannot do.

A note on wheat

I know many of you are particularly interested in the correlation between wheat and heart disease. In my critique’s gargantuan cascade of words, the two little paragraphs about wheat pinged on many readers’ radar (or, perhaps, grain-dar). I’ve already seen the “correlation of 67″ statistic thrown around the ‘net as if it’s solid evidence. Holiest of molies, that spread fast!

Indeed, I feel the China Study may hold important clues—ones that research thus far has simply not explored—about the role of wheat or wheat flour in human health. However, we can’t jump the gun yet. I will be doing some more analysis of the China Study data regarding wheat and other grains, but even if this manages to paint our glutenous friends as the most malicious of dietary villains, it doesn’t prove a darn thing.

Bummer, right?

As someone who’s massively allergic to wheat, I’d love nothing more than to shove this grain in the corner with a dunce cap and revel in my victory. Karma’s a… female dog in heat. But I can’t do that. Not yet, anyway. Bottom line, this is epidemiological data we’re working with, and it can only show correlations—not causation. Not proof. Not irrefutable evidence.

What I do hope occurs—and feel free to cross your fingers with me—is that this information snags the eye of other nutritional researchers and leads to controlled experiments about the health effects of wheat.

At any rate, I have a couple more China-Study-related posts coming up (including one with the results of multiple variable regressions), and wheat will probably be the subject of the next one. Keep your eyes peeled if this is a subject that interests you.

Summary of this post

For those of you who skipped over everything above and scrolled directly to this part, well… I don’t blame you. However, there’s really only one thing you need to know about this whole ordeal, and this is it:

1. Data sets are like people. If you torture them long enough, even when they’re innocent, you’ll eventually squeeze out a false confession.

Some final thoughts, for those who haven’t clicked the “back” button on their browser yet

Although the vast majority of the feedback I’ve received (both positive and negative) has been intelligent, respectful, and ultimately constructive, I’ve received a few very fiery emails that have made me realize what a deep nerve diet debates can strike. For those whose lives have been profoundly affected—for better or for worse—by food and nutrition, diet can become a personal issue inextricably bound with identity. And as someone who’s already run through a gamut of eating styles due to allergies, ethical goals, and the pursuit of vibrant health, I know how this goes. I’ve been there. In many ways, I’m still there. For this reason, I can wholly empathize with the emotional response my critique triggered in some readers, and I understand why a backlash is apt to occur.

By the same token, I think it’s important to look at what that impassioned response signifies. Are we trying to be healthy, or are we trying to be right? Are we trying to learn, or do rigid beliefs deafen our ears to new knowledge? Have the open minds that led us to search for the truth in nutrition suddenly slammed shut, clamping tight around an ideology that may or may not truly serve us?

Critical thinking isn’t a privilege reserved for the elite; it’s a birthright. My goal is not to tell people what to think, but to show them how to think. How to sift through the vast expanse of nutritional litter and pull out the gems. How to stop blindly following the advice of so-called authorities who may not have our best interest at heart. How to think independently.

To everyone who’s taken the time to plod through this post and the last, to read, to write, to comment, to think, or to reconsider any limiting beliefs you hold about diet, I extend my deepest gratitude and wish you nothing but health and happiness.

Thank you for reading.

When I first started analyzing the original China Study data, I had no intention of writing up an actual critique of Campbell’s much-lauded book. I’m a data junkie. Numbers, along with strawberries and Audrey Hepburn films, make me a very happy girl. I mainly wanted to see for myself how closely Campbell’s claims aligned with the data he drew from—if only to satisfy my own curiosity.

But after spending a solid month and a half reading, graphing, sticky-noting, and passing out at 3 AM from studious exhaustion upon my copy of the raw China Study data, I’ve decided it’s time to voice all my criticisms. And there are many.

First, let me put out some fires before they have a chance to ignite:

1. I don’t work for the meat or dairy industry. Nor do I have a fat-walleted roommate, best friend, parent, child, love interest, or highly prodigious cat who works for the meat or dairy industry who paid me off to debunk Campbell.
2. Due to food sensitivities, I don’t consume dairy myself, nor do I have any personal reason to promote it as a health food.
3. I was a vegetarian/vegan for over a decade and have nothing but respect for those who choose a plant-based diet, even though I am no longer vegan. My goal, with the “The China Study” analysis and elsewhere, is to figure out the truth about nutrition and health without the interference of biases and dogma. I have no agenda to promote.

As I mentioned, I’m airing my criticisms here; this won’t be a China Study love fest, or even a typical balanced review with pros and cons. Campbell actually raises a  number of points I wholeheartedly agree with—particularly in the “Why Haven’t You Heard This?” section of his book, where he exposes the reality behind Big Pharma and the science industry at large. I admire Campbell’s philosophy towards nutritional research and echo his sentiments about the dangers of scientific reductionism. However, the internet is already flooded with rave reviews of this book, and I’m not interested in adding redundant praise. My intent is to highlight the weaknesses of “The China Study” and the potential errors in Campbell’s interpretation of the original data.

(IMPORTANT NOTE: My response to Campbell’s reply, as well as to some common reader questions, can be found in the following post: My Response to Campbell. Please read this for clarification regarding univariate correlations and flaws in Campbell’s analytical methods.)

(If this is your first time here, feel free to browse the earlier posts in the China Study category to get up to speed.)

On the Cornell University website (the institution that—along with Oxford University—spawned the China Project), I came across an excellent summary of Campbell’s conclusions from the data. Although this article was published a few years before “The China Study,” it distills some of the book’s points in a concise, down-n’-dirty way. In this post, I’ll be looking at these statements along with other overriding claims in “The China Study” and seeing whether they hold up under scrutiny—including an in-depth look at Campbell’s discoveries with casein.

(Disclaimer: This post is long. Very long. If either your time or your attention span is short, you can scroll down to the bottom, where I summarize the 9,000 words that follow in a less formidable manner.)

(Disclaimer 2: All correlations here are presented as the original value multiplied by 100 in order to avoid dealing with excessive decimals. Asterisked correlations indicate statistical significance, with * = p<0.05, ** = p<0.01, and *** = p<0.001. In other words, the more stars you see, the more confident we are that the trend is legit. If you’re rusty on stats, visit the meat and disease in the China Study page for a basic refresher on some math terms.)

(Disclaimer 3: The China Study files on the University of Oxford website include the results of the China Study II, which was conducted after the first China Study. It includes Taiwan and a number of additional counties on top of the original 65–and thus, more data points. The numbers I use in this critique come solely from the first China Study, as recorded in the book “Diet, Life-style and Mortality in China,” and may be different than the numbers on the website.)

From Cornell University’s article:

“Even small increases in the consumption of animal-based foods was associated with increased disease risk,” Campbell told a symposium at the epidemiology congress, pointing to several statistically significant correlations from the China studies.

Alright, Mr. Campbell—I’ll hear ya out. Let’s take a look at these correlations.

Campbell Claim #1

Plasma cholesterol in the 90-170 milligrams per deciliter range is positively associated with most cancer mortality rates. Plasma cholesterol is positively associated with animal protein intake and inversely associated with plant protein intake.

No falsification here. Indeed, cholesterol in the China Project has statistically significant associations with several cancers (though not with heart disease). And indeed, plasma cholesterol correlates positively with animal protein consumption and negatively with plant protein consumption.

But there’s more to the story than that.

Notice Campbell cites a chain of three variables: Cancer associates with cholesterol, cholesterol associates with animal protein, and therefore we infer that animal protein associates with cancer. Or from another angle: Cancer associates with cholesterol, cholesterol negatively associates with plant protein, and therefore we infer plant protein protects against cancer.

But when we actually track down the direct correlation between animal protein and cancer, there is no statistically significant positive trend. None. Looking directly at animal protein intake, we have the following correlations with cancers:

Lymphoma: -18
Penis cancer: -16
Rectal cancer: -12
Bladder cancer: -9
Colorectal cancer: -8
Leukemia: -5
Nasopharyngeal: -4
Cervix cancer: -4
Colon cancer: -3
Liver cancer: -3
Oesophageal cancer: +2
Brain cancer: +5
Breast cancer: +12

Most are negative, but none even reach statistical significance. In other words, the only way Campbell could indict animal protein is by throwing a third variable—cholesterol—into the mix. If animal protein were the real cause of these diseases, Campbell should be able to cite a direct correlation between cancer and animal protein consumption, which would show that people eating more animal protein did in fact get more cancer.

But what about plant protein? Since plant protein correlates negatively with plasma cholesterol, does that mean plant protein correlates with lower cancer risk? Let’s take a look at the cancer correlations with “plant protein intake”:

Nasopharyngeal cancer: -40**
Brain cancer: -15
Liver cancer: -14
Penis cancer: -4
Lymphoma: -4
Bladder cancer: -3
Breast cancer: +1
Stomach cancer: +10
Rectal cancer: +12
Cervix cancer: +12
Colon cancer: +13
Leukemia: +15
Oesophageal cancer +18
Colorectal cancer: +19

We have one statistically significant correlation with a rare cancer not linked to diet (nasopharyngeal cancer), but we also have more positive correlations than we saw with animal protein.

In fact, when we look solely at the variable “death from all cancers,” the association with plant protein is +12. With animal protein, it’s only +3. So why is Campbell linking animal protein to cancer, yet implying plant protein is protective against it?

In addition, Campbell’s statement about cholesterol and cancer leaves out a few significant points. What he doesn’t mention is that plasma cholesterol is also associated with several non-nutritional variables known to raise cancer risk—namely schistosomiasis infection (correlation of +34*) and hepatitis B infection (correlation of +30*).

Not coincidentally, cholesterol’s strongest cancer links are with liver cancer, rectal cancer, colon cancer, and the sum of all colorectal cancers. As we saw in the posts on meat consumption and fish consumption, schistosomiasis and hepatitis B are the two biggest factors in the occurrence of these diseases. So is it higher cholesterol (by way of animal products) that causes these cancers, or is it a misleading association because areas with high cholesterol are riddled with other cancer risk factors? We can’t know for sure, but it does seem odd that Campbell never points out the latter scenario as a possibility.

Campbell Claim #2

Breast cancer is associated with dietary fat (which is associated with animal protein intake) and inversely with age at menarche (women who reach puberty at younger ages have a greater risk of breast cancer).

Campbell is correct that breast cancer negatively relates to the age of first menstruation—a correlation of -20. Not statistically significant, but given what we know about hormone exposure and breast cancer, it certainly makes sense. And there is a correlation between fat intake and breast cancer—a non-statistically-significant +18 for fat as a percentage of total calories and +22 for total lipid intake. But are there any dietary or lifestyle factors with a similar or stronger association than this? Let’s look at the correlation between breast cancer and a few other variables. Asterisked items are statistically significant:

Blood glucose level: +36**
Wine intake: +33*
Alcohol intake: +31*
Yearly fruit consumption: +25
Percentage of population working in industry: +24
Hexachlorocyclohexane in food: +24
Processed starch and sugar intake: +20
Corn intake: +20
Daily beer intake: +19
Legume intake: +17

Looks to me like breast cancer may have links with sugar and alcohol, and perhaps also with hexachlorocyclohexane and occupational hazards associated with industry work. Again, why is Campbell singling out fat from animal products when other—stronger—correlations are present?

Certainly, consuming dairy and meat from hormone-injected livestock may logically raise breast cancer risk due to increased exposure to hormones, but this isn’t grounds for generalizing all animal products as causative for this disease. Nor is a correlation of +18 for fat calories grounds for indicting fat as a breast cancer risk factor, when alcohol, processed sugar, and starch correlate even more strongly. (Animal protein itself, for the record, correlates with breast cancer at +12—which is lower than breast cancer’s correlation with light-colored vegetables, legume intake, fruit, and a number of other purportedly healthy plant foods.)

Campbell Claim #3

For those at risk for liver cancer (for example, because of chronic infection with hepatitis B virus) increasing intakes of animal-based foods and/or increasing concentrations of plasma cholesterol are associated with a higher disease risk.

Ah, here’s one that may be interesting! Even if animal products don’t cause cancer, do they spur its occurrence when other risk factors are present? That would certainly be in line with Campbell’s research on aflatoxin and rats, where the milk protein casein dramatically increased cancer rates.

So, let’s look only at the counties with the highest rates of hepatitis B infection and see what animal food consumption does there. In the China Study, one documented variable is the percentage of each county’s population testing positive for the hepatitis B surface antigen. Population averages ranged from 1% to 29%, with a mean of 13% and median of 14%. If we take only the counties that have, say, 18% or more testing positive, that leaves us with a solid pool of high-risk data points to look at.

Animal product consumption in these places ranges from a meager 6.9 grams per day to a heftier 148.1 grams per day—a wide enough range to give us a good variety of data points. Liver cancer mortality ranges from 5.51 to 59.63 people per thousand.

Let’s crunch these numbers, shall we? Here’s a chart of the data I’m using.

When we map out liver cancer mortality and animal product consumption only in areas with high rates of hepatitis B infection (18% and higher), we should see cancer rates rise as animal product consumption increases—at least, according to Campbell. That would indicate animal-based foods do encourage cancer growth. But here’s what we really get.

In these high-risk areas for liver cancer, total animal food intake has a correlation with liver cancer of… dun dun dun… +1.

That’s it. One. We rarely get a perfect statistical zero in the real world, but this is pretty doggone close to neutral. Broken up into different types of animal food rather than total consumption, we have the following correlations:

• Meat correlates at -7 with liver cancer in high-risk counties
• Fish correlates at +11
• Eggs correlate at -29
• Dairy correlates at -19

In other words, it looks like animal foods have virtually no effect—whether positive or negative—on the occurrence of liver cancer in hepatitis-B infected areas.

Campbell mentioned plasma cholesterol also associates with liver cancer, which is correct: The raw correlation is a statistically significant +37. If it’s true blood cholesterol is somehow an instigator for liver cancer in hepatitis-B-riddled populations, we’d expect to see this correlation preserved or heightened among our highest-risk counties. So let’s take a look at the same previous 19 counties with high hepatitis B occurrence, and graph their total cholesterol alongside their liver cancer rates.

In the high-risk groups, the correlation between total cholesterol and liver cancer drops from +37 to +8. Still slightly positive, but not exactly damning.

If I were Campbell, I’d look at not only animal protein and cholesterol in relation to liver cancer, but also at the many other variables that correlate positively with the disease. For instance, daily liquor intake correlates at +33*, total alcohol intake correlates at +28*, cigarette use correlates at +27*, intake of the heavy metal cadmium correlates at +38**, rapeseed oil intake correlates at +25*—so on and so forth. All are statistically significant. Why doesn’t Campbell mention these factors as possible causes of increased liver cancer in high-risk areas? And, more importantly, why doesn’t Campbell account for the fact that many of these variables occur alongside increased cholesterol and animal product consumption, making it unclear what’s causing what?

Campbell Claim #4

Cardiovascular diseases are associated with lower intakes of green vegetables and higher concentrations of apo-B (a form of so-called bad blood cholesterol) which is associated with increasing intakes of animal protein and decreasing intakes of plant protein.

Alright, we’ve got a multi-parter here. First, let’s see what the actual correlations are between cardiovascular diseases and green vegetables—an interesting connection, if it holds true. The China Study accounted for this variable in two ways: one through a diet survey that measured how many grams of green vegetables each county averaged per day, and one through a questionnaire that recorded how many times per year citizens ate green vegetables.

From the diet survey, green vegetable intake (average grams per day) has the following correlations:

Myocardial infarction and coronary heart disease: +5
Hypertensive heart disease: -4
Stroke: -8

From the questionnaire, green vegetable intake (times eaten per year) has the following correlations:

Myocardial infarction and coronary heart disease: -43**
Hypertensive heart disease: -36*
Stroke: -35*

A little odd, oui? When we look at total quantity of green vegetables consumed (in terms of weight), we’ve got only weak negative associations for two cardiovascular conditions, and a slightly positive association for heart attacks (myocardial infarction) and coronary heart disease. Nothing to write home about. But when we look at the number of times per year green vegetables are consumed, we have much stronger inverse associations with all cardiovascular diseases. Why the huge difference? Why would frequency be more protective than quantity? What accounts for this mystery?

It could be that the China Study diet survey did a poor job of tracking and estimating greens intake on a long-term basis (indeed, it was only a three-day survey, although when repeated at a later date yielded similar results for each county). But the explanation could also boil down to one word: geography.

Let me explain.

The counties in China that eat greens year-round live in a particular climate and latitude—namely, humid regions to the south.  The “Green vegetable intake, times per year” variable has a correlation of -68*** with aridity (indicating a humid climate) and a correlation of -60*** with latitude (indicating southerly placement on the ol’ map). Folks living in these regions might not eat the most green vegetables quantity-wise, but they do eat them frequently, since their growing season is nearly year-round.

In contrast, the variable “Green vegetable intake, grams per day” has a correlation of only -16 with aridity and +5 with latitude, indicating much looser associations with southern geography. The folks who eat lots of green veggies don’t necessarily live in climates with a year-round growing season, but when green vegetables are available, they eat a lot of them. That bumps up the average intake per day, even if they endure some periods where greens aren’t on the menu at all.

If green vegetables themselves were protective of heart disease, as Campbell seems to be implying, we would expect their anti-heart-disease effects to be present in both quantity of consumption and frequency of consumption. Yet the counties eating the most greens quantity-wise didn’t have any less cardiovascular disease than average. This tells us there’s probably another variable unique to the southern, humid regions in China that confers heart disease protection—but green veggies aren’t it.

Some of the hallmark variables of humid southern regions include high fish intake, low use of salt, high rice consumption (and low consumption of all other grains, especially wheat), higher meat consumption, and smaller body size (shorter height and lower weight). And as you’ll see in an upcoming post on heart disease, these southerly regions also had more intense sunlight exposure and thus more vitamin D—an important player in heart disease prevention.

(And for the record, as a green-veggie lover myself, I’m not trying to negate their health benefits—promise! I just want to offer equal skepticism to all claims, even the ones I’d prefer to be true.)

Basically, Campbell’s implication that green vegetables are associated with less cardiovascular disease is misleading. More accurately, certain geographical regions have strong correlations with cardiovascular disease (or lack thereof), and year-round green vegetable consumption is simply an indicator of geography. Since only frequency and not actual quantity of greens seems protective of heart disease and stroke, it’s safe to say that greens probably aren’t the true protective factor.

So that about covers it for greens. What about the next variable in Campbell’s claim: a “bad” form of cholesterol called apo-B?

Campbell is justified in noting the link between apolipoprotein B (apo-B) and cardiovascular disease in the China Study data, a connection widely recognized by the medical community today. These are its correlations with cardiovascular disease:

Myocardial infarction and coronary heart disease: +37**
Hypertensive heart disease: +35*
Stroke: +35*

And he’s also right about the negative association between apo-B and plant protein, which is -37*, as well as the positive association between apo-B and animal protein, which is +25* for non-fish protein and +16 for fish protein. So from a technical standpoint, Campbell’s statement (aside from the green veggie issue) is legit.

However, it’s the implications of this claim that are misleading. From what Campbell asserts, it would seem that animal products are ultimately linked to cardiovascular diseases and plant protein is ultimately protective of those diseases, and apo-B is merely a secondary indicator of this reality. But does that claim hold water? Here’s the raw data.

Correlations between animal protein and cardiovascular disease:

Myocardial infarction and coronary heart disease: +1
Hypertensive heart disease: +25
Stroke: +5

Correlations between fish protein and cardiovascular disease:

Myocardial infarction and coronary heart disease: -11
Hypertensive heart disease: -9
Stroke: -11

Correlations between plant protein and cardiovascular disease (from the China Study’s “diet survey”):

Myocardial infarction and coronary heart disease: +25
Hypertensive heart disease: -10
Stroke: -3

Correlations between plant protein and cardiovascular disease (from the China Study’s “food composite analysis”):

Myocardial infarction and coronary heart disease: +21
Hypertensive heart disease: 0
Stroke: +12

Check that out! Fish protein looks weakly protective all-around; non-fish animal protein is neutral for coronary heart disease/heart attacks and stroke but associates positively with hypertensive heart disease (related to high blood pressure); and plant protein actually correlates fairly strongly with heart attacks and coronary heart disease. (The China Study documented two variables related to plant protein: one from a lab analysis of foods eaten in each county, and one from a diet survey given to county citizens.) Surely, there is no wide division here between the protective or disease-causing effects of animal-based protein versus plant protein. If anything, fish protein looks the most protective of the bunch. No wonder Campbell had to cite a third variable in order to vilify animal products and praise plant protein: Examined directly, they’re nearly neck-and-neck.

If you’re wondering about the connection between animal protein and hypertensive heart disease, by the way, it’s actually hiked up solely by the dairy variable. Here are the individual correlations between specific animal foods and hypertensive heart disease:

Milk and dairy products intake: +30**
Egg intake: -28
Meat intake: -4
Fish intake: -14

You can read more about the connection between dairy and hypertensive heart disease in the entry on dairy in the China Study.

At any rate, Campbell accurately points out that apo-B correlates positively with cardiovascular diseases. But to imply animal protein is causative of these diseases—and green vegetables or plant protein protective of them—is dubious at best. What factors cause both apo-B and cardiovascular disease risk to increase hand-in-hand? This is the question we should be asking.

Campbell Claim #5

Colorectal cancers are consistently inversely associated with intakes of 14 different dietary fiber fractions (although only one is statistically significant). Stomach cancer is inversely associated with green vegetable intake and plasma concentrations of beta-carotene and vitamin C obtained only from plant-based foods.

This is congruous with conventional beliefs about fiber being helpful for colon health. And as a plant-nosher myself, I hope it’s true—but that’s no reason to omit this claim from critical examination. Here are all of the China Study’s fiber variables as they correlate to colorectal cancer:

Total fiber intake: -3
Total neutral detergent fiber intake: -13
Hemi-cellulose fiber intake: -10
Cellulose fiber intake: -13
Intake of lignins remaining after cutin removed: -9
Cutin intake: -14
Starch intake: -1
Pectin intake: +3
Rhamnose intake: -26*
Fucose intake: +2
Arabinose intake: -18
Xylose intake: -15
Mannose intake: -13
Galactose intake: -24

Surprise, surprise: I agree with Campbell on this one! All but two of the fiber variables have inverse associations with colorectal cancers. The first part of Campbell Claim #5 passes Denise’s BS-o-Meter test. Let us celebrate!

…But before we get too jiggy with it, I do have a nit to pick. Fiber intake also negatively correlates with schistosomiasis infection, a type of parasite. Try Googling “schistosomiasis and colorectal cancer” and you’ll get more relevant hits than you’ll ever have time to read. I’ll elaborate on this in a few paragraphs, so hang tight—but for now, I’ll just point out two things:

1. Schistosomiasis infection is a very strong predictor for colon and rectal cancers, more so than any of the other hundreds of variables studied in the China Project (it has a correlation of +89 with colorectal cancer).
2. The only fiber factions that don’t appear protective of colorectal cancer (pectin and fucose) also have the most neutral associations with schistosomiasis infection (+1 and -5, respectively—whereas other fiber factions had correlations ranging from -9 to -27 with schistosomiasis). In all cases, the correlation between each fiber faction and colorectal cancer parallels its correlation with schistosomiasis.

In other words: Is it the fiber itself that’s protective against colorectal cancer, or is it the fact that the counties eating the most fiber happened to also have the lowest rates of schistosomiasis? It would, I think, be wise to prune these variables apart before declaring fiber itself as protective based on the China Study data.

There is research conducted outside of the China Project suggesting fiber benefits colon health, but often that association dissolves when researchers adjust for other dietary risk factors, such as with the this pooled analysis of colorectal cancer studies published in the Journal of the American Medical Association. Bottom line: It’s never a good idea to go looking for a specific trend just because we believe it should be there. Chains of confirmation bias are often what cause nutritional myths to emerge and persist. Fiber may be beneficial, but we shouldn’t approach the data already expecting to find this—lest we overlook other important influences.

Moving on. Now, what about the second part of this claim: Stomach cancer is inversely associated with green vegetable intake and plasma concentrations of beta-carotene and vitamin C obtained only from plant-based foods.

Is this a fair assessment? Let’s find out. Here are the correlations between stomach cancer and each of these variables.

Green vegetables, daily intake: +5
Green vegetables, times eaten per year: -35**
Plasma beta-carotene: -14
Plasma vitamin C: -13

Ah, looks like we’re facing the Green Veggie Paradox once again. The folks with year-round access to green vegetables get less stomach cancer, but the the folks who eat more green vegetables overall aren’t protected. Once again, I’ll suggest that a geographic variable specific to veggie-growing regions could be at play here.

As for beta-carotene and vitamin C concentrations in the blood, Campbell is correct in noting an inverse association with stomach cancer. However, the correlations aren’t statistically significant, nor are they very high: -14 and -13, respectively.

Campbell Claim #6

Western-type diseases, in the aggregate, are highly significantly correlated with increasing concentrations of plasma cholesterol, which are associated in turn with increasing intakes of animal-based foods.

From his book, we know Campbell defines Western-type diseases as including heart disease, diabetes, colorectal cancers, breast cancer, stomach cancer, leukemia, and liver cancer. And indeed, the variable “total cholesterol” correlates positively with many of these diseases:

Myocardial infarction and coronary heart disease: +4
Diabetes: +8
Colon cancer: +44**
Rectal cancer: +30*
Colorectal cancer: +33**
Breast cancer: +19
Stomach cancer: +17
Leukemia: +26*
Liver cancer: +37*

Perhaps surprisingly, total cholesterol has only weak associations with heart disease and diabetes—weaker, in fact, than the correlation between these conditions and plant protein intake (+25 and +12, respectively). But we’ll put that last point aside for the time being. For now, let’s focus on the diseases with statistical significance, which include all forms of colorectal cancer, leukemia, and liver cancer. (Despite classifying stomach cancer as a “Western disease,” by the way, China actually has far higher rates of this disease than any Western nation. In fact, half the people who die each year from stomach cancer live in China.)

First, let’s dive into the dark, murky chambers of the digestive tract and start with colorectal cancers. Off we go!

What Campbell overlooks about colorectal cancers and cholesterol

As I mentioned earlier, a little somethin’ called “schistosomiasis” is a profoundly strong risk factor for developing colon cancer and rectal cancer. In the China Study data, schistosomiasis correlates at +89*** with colorectal cancer mortality. Yes, 89—higher than any of the other 367 variables recorded.

This, ladies and gentlemen, is what we call a positive correlation.

It just so happens that total cholesterol also correlates with schistosomiasis infection, at a statistically significant rate of +34*:

Basically, this means that areas with higher cholesterol levels also had—for whatever reason—a higher incidence of schistosomiasis infection. It’s hard to say for sure why this is, but it’s likely that the high-cholesterol and high-schistosomiasis groups had a third variable in common, such as infected drinking water or other source of schistosomiasis exposure.

From this alone, it shouldn’t be too shocking that higher cholesterol also correlates with higher rates of colorectal cancer (+33*):

Clearly, we have three tangled-up variables to sort through: total cholesterol, colorectal cancer rates, and schistosomiasis infection. Is it really higher cholesterol that increases the risk of developing colon and rectal cancers, or is the influence of schistosomiasis deceiving us?

To figure this out, let’s look at what cholesterol and colorectal cancer rates look like only in regions with zero schistosomiasis infection. If cholesterol is a causative factor for colorectal cancers, then cancer rates should still increase as total cholesterol rises.

The above graph showcases a correlation of +13. Still positive, but not statistically significant, and a major downgrade from the original correlation of +33*. It does seem schistosomiasis inflates the correlation between cholesterol and colorectal cancers—something Campbell never takes into account. Is blood cholesterol still a risk factor? It’s possible, but we would need more data to know whether the +13 correlation persists or whether there are additional confounding variables at work. For instance, beer intake is another factor correlating significantly with both total cholesterol (+32*) and colon cancer (+40**).  If we remove the three counties that drink the most beer from of the data set above, the correlation between cholesterol and and colorectal cancer drops to -9.

See how tricky the interplay of variables can be?

What Campbell overlooks about leukemia and cholesterol

Next in our lineup of “Western diseases” is leukemia, which has a statistically significant correlation of +26* with total cholesterol. (Although the implication here is that animal product consumption raises leukemia risk, it should be noted that animal protein intake itself has a correlation of -5 with leukemia, whereas plant protein actually has a correlation of +15 with this disease. But let’s humor this claim anyway by looking solely at the role of blood cholesterol.)

If you’ll recall from the post on fish and disease in the China Study, leukemia correlates very strongly with working in industry (+53**) and inversely with working in agriculture (-53**). Although it’s possible the cause is nutritional, it’s also quite likely that an occupational hazard is to blame—such as benzene exposure, which is a major and well-known cause of leukemia in Chinese factory and refinery workers.

Lo and behold, cholesterol also correlates strongly with working in industry (+45**) and inversely with working in agriculture (-46**). If an industry-related risk factor raises leukemia rates, it could very well appear as a false correlation with cholesterol. How can we tell if this is the case?

Let’s try looking at the correlation between leukemia and cholesterol only in counties where few members of the population were employed in industry. If cholesterol itself heightens leukemia risk, our positive trend should still be in place. In the China Study data set, the range for percent of the population working in industry is 1.1% to  41.3%, so let’s try looking at the counties where the value is under 10%:

For the low-industry counties, the correlation between leukemia and total cholesterol is close to neutral—a mere +4. As you can see, this is hardly a damning trend. And in case you’re wondering if higher cholesterol could possibly spur the rates of leukemia in folks who are already at risk, this isn’t the case either: Using only counties that had 20% or more of the population working in industry, presumably the folks who had the most exposure to chemicals like benzene, the correlation between cholesterol and leukemia is a slightly protective -3.

What Campbell overlooks about liver cancer and cholesterol

I may not be vegan, but that doesn’t mean I like beating dead horses. Instead of rehashing the earlier analysis of liver cancer under Campbell Claim #3, I’ll just repeat that cholesterol does not have a significant correlation with liver cancer when you divide the data set into separate groups: areas with high hepatitis B rates an areas with low hepatitis B rates.

From page 104 of his book:

Liver cancer rates are very high in rural China, exceptionally high in some areas. Why was this? The primary culprit seemed to be chronic infection with hepatitis B virus (HBV). …

… But there’s more. In addition to the [hepatitis B] virus being a cause of liver cancer in China, it seems that diet also plays a key role. How do we know? The blood cholesterol levels provided the main clue. Liver cancer is strongly associated with increasing blood cholesterol, and we already know that animal-based foods are responsible for increases in cholesterol.

Campbell connects some of the dots, but misses a very important one. Indeed, hepatitis B associates strongly with liver cancer. Indeed, cholesterol associates with liver cancer. But what he doesn’t mention is that cholesterol also associates with hepatitis B infection. In other words, the groups with higher cholesterol are already at greater risk of liver cancer than groups with lower cholesterol—but it’s not because of diet.

In addition to greater rates of hepatitis B infection, higher-cholesterol areas had additional risk factors for liver cancer, such beer consumption, which also inflated the trend. Despite Campbell’s claims, cholesterol itself does not appear to significantly heighten cancer rates in at-risk populations.

Given Campbell’s casein research and earlier observations about the animal-protein consuming children in the Philippines getting more liver cancer, I wonder if Campbell approached the China Study already expecting a particular outcome. In a massive data set with 8,000 statistically significant correlations, even a smidgen of confirmation bias can cause someone to find a trend that isn’t truly there.

An example of bias in “The China Study”

Body weight, associated with animal protein intake, was associated with more cancer and more coronary heart disease. It seems that being bigger, and presumably better, comes with very high costs. (Page 102)

Consuming more protein was associated with greater body size. … However, this effect was primarily attributed to plant protein, because it makes up 90% of the total Chinese protein intake. (Page 103)

Let’s read between the lines. Here we have Campbell claiming two things, a few paragraphs apart: One, that body weight is associated with more cancer and heart disease, and two, that body size in China is linked not only with a greater intake of animal protein, but also with a greater intake of plant protein. In fact, the link between body size is stronger with plant protein than with animal protein.

Yet notice how Campbell only implicates animal protein in the association between body weight, cancer, and heart disease. If he were to describe the data without bias, Campbell’s first statement would be this:

Body weight, associated with animal protein intake and plant protein intake, was associated with more cancer and more coronary heart disease.

Maybe his editor just overlooked that omission, eh? Right afterward, Campbell notes:

But the good news is this: Greater plant protein intake was closely linked to greater height and body weight. Body growth is linked to protein in general and both animal and plant proteins are effective! (Page 102)

Wait a minute. This is good news? Didn’t Campbell just say being bigger “comes with very high costs” and that it’s associated with “more cancer and coronary heart disease?” Why is body size a bad thing when it’s associated with animal protein, but a good thing when it’s associated with plant protein?

Does less animal foods equal better health?

People who ate the most animal-based foods got the most chronic disease. Even relatively small intakes of animal-based food were associated with adverse effects. People who ate the most plant-based foods were the healthiest and tended to avoid chronic disease.

This oft-repeated quote from “The China Study” is compelling, but is it true? Based on the data above, it seems like an unlikely conclusion—but let’s try once more to see if it could be valid.

As an illustrative experiment, let’s look at the top five Chinese counties with the lowest animal protein consumption and compare them against the top five counties with the highest animal protein consumption. A data set of 10 won’t yield any confident conclusions, of course, and I won’t treat this as representative of the collective body of China Study data. But since animal protein consumption among the studied counties ranged from 0 grams* to almost 135 grams per day, we should see a stark contrast between the nearly-vegan regions and the ones eating considerably more animal foods. That is, assuming it’s true that “even relatively small intakes of animal-based food” yield disease.

*The county averaging zero grams per day wasn’t completely vegan, but the yearly consumption of animal foods was low enough so that the daily average appeared less than 0.01 grams.

Here are the counties I’ll be using. The first five are our near-vegans; the second five are our highest animal product consumers. From both groups, I had to exclude a top-five county due to missing data for most mortality variables (illegible documentation, according to the authors of “Diet, Life-style and Mortality in China”) and replaced it with a sixth county where animal protein consumption matched within a few hundredths of a gram.

Below are the names of each county, as well as values for their daily animal protein intake, the percentage of their total caloric intake coming from fat, and their daily intake of fiber (in case the latter two variables are also of interest).

To give you a visual idea of these quantities, 135 grams of animal protein is the equivalent of 22 medium eggs per day, 24 grams of animal protein is the equivalent of four medium eggs per day, 12 grams is two eggs, and 9 grams is one and a half eggs. Obviously, that’s quite a wide range even among the top consumers of animal foods, so the highest animal-food-eating counties (Tuoli and XIanghuang qi) may be the most important to study in contrast with the near-vegan counties.

Animal protein intake by county:

For reference, some other diet variables:

And now, mortality rates for important variables (as per 1000 people). I’ll save you my commentary and just show you the graphs, which should speak for themselves. Remember, the five left-most bars (Jiexiu through Songxian) on each graph are the near-vegan counties, and the five right-most bars (Tuoli through Wenjiang) are the highest consumers of animal products.

As you can see, the mortality rates for both groups (near-vegan and higher-animal-foods) are quite similar, with the animal food group coming out more favorably in some cases (death from all cancers, myocardial infarction, brain and neurological diseases, lymphoma, cervix cancer). This little comparison might not carry a lot of scientific clout due to its small sample size, but it does blatantly undermine Campbell’s assessment:

People who ate the most animal-based foods got the most chronic disease … People who ate the most plant-based foods were the healthiest and tended to avoid chronic disease.

Sins of omission

Perhaps more troubling than the distorted facts in “The China Study” are the details Campbell leaves out.

Why does Campbell indict animal foods in cardiovascular disease (correlation of +1 for animal protein and -11 for fish protein), yet fail to mention that wheat flour has a correlation of +67 with heart attacks and coronary heart disease, and plant protein correlates at +25 with these conditions?

Speaking of wheat, why doesn’t Campbell also note the astronomical correlations wheat flour has with various diseases: +46 with cervix cancer, +54 with hypertensive heart disease, +47 with stroke, +41 with diseases of the blood and blood-forming organs, and the aforementioned +67 with myocardial infarction and coronary heart disease? (None of these correlations appear to be tangled with any risk-heightening variables, either.)

Why does Campbell overlook the unique Tuoli peoples documented in the China Study, who eat twice as much animal protein as the average American (including two pounds of casein-filled dairy per day)—yet don’t exhibit higher rates of any diseases Campbell ascribes to animal foods?

Why does Campbell point out the relationship between cholesterol and colorectal cancer (+33) but not mention the much higher relationship between sea vegetables and colorectal cancer (+76)? (For any researcher, this alone should be a red flag to look for an underlying variable creating misleading correlations, which—in this case—happens to be schistosomiasis infection.)

Why does Campbell fail to mention that plant protein intake correlates positively with many of the “Western diseases” he blames cholesterol for—including +19 for colorectal cancers, +12 for cervix cancer, +15 for leukemia, +25 for myocardial infarction and coronary heart disease, +12 for diabetes, +1 for breast cancer, and +10 for stomach cancer?

Of course, these questions are largely rhetorical. Only a small segment of “The China Study” even discusses the China Study, and Campbell set out to write a publicly accessible book—not an exhaustive discussion of every correlation his research team uncovered. However, it does seem Campbell overlooked or ignored significant points when discerning the overriding nutritional themes in the China Project data.

What about casein?

Along with trends gleaned from the China Project, Campbell recounts the startling connection he found between casein (a milk protein) and cancer in his research with lab rats. In his own words, casein “proved to be so powerful in its effect that we could turn on and turn off cancer growth simply by changing the level consumed” (page 5 of “The China Study”). Protein from wheat and soy did not have this effect.

This finding is no doubt fascinating. If nothing else, it suggests a strong need for more research regarding the safety of casein supplementation in humans, especially among bodybuilders, athletes, and others who use isolated casein for muscle recovery. Unfortunately, Campbell extrapolates this research beyond its logical scope: He concludes that all forms of animal protein have similar cancer-promoting properties in humans, and we’re therefore better off as vegans. This claim rests on several unproven assumptions:

1. The casein-cancer mechanism behaves the same way in humans as in lab rats.
2. Casein promotes cancer not just when isolated, but also when occurring in its natural food form (in a matrix of other milk substances like whey, bioactive peptides, conjugated linoleic acid, minerals, and vitamins, some of which appear to have anti-cancer properties).
3. There are no differences between casein and other types of animal protein that could impose different effects on cancer growth/tumorigenesis.

Campbell offers no convincing evidence that any of the above are true. We do share some metabolic similarities with rats, so for the sake of being able to entertain the possibility that #2 and #3 are valid, let’s assume that the effect of casein on rats translates cleanly to humans.

How does Campbell justify generalizing the effects of casein to all forms of animal protein? Much of it is based on a study he helped conduct: “Effect of dietary protein quality on development of aflatoxin B[1]-induced hepatic preneoplastic lesions,” published in the August 1989 edition of the Journal of the National Cancer Institute. In this study, he and his research crew discovered that aflatoxin-exposed rats fed wheat gluten exhibited less cancer growth than rats fed the same amount of casein. But get this: When lysine (the limiting amino acid in wheat) was restored to make the gluten a complete protein, the rats had just as much cancer occurrence as the casein group. Jeepers!

Campbell thus deduced that it’s the amino acid profile itself responsible for spurring cancer growth. Because most forms of plant protein are low in one or more amino acids (called “limiting amino acids”) and animal protein is complete, Campbell concluded that animal protein, but not plant protein, must encourage cancer growth. Time to whip out the veggie burgers!

Of course, this conclusion has some gaping logical holes when applied to real life. Unless you consume nothing but animal products, you’ll be ingesting a mixed ratio of amino acids by default, since animal foods combined with plant foods still yield limiting amino acids. The rats in Campbell’s research consumed casein as their only protein source, the equivalent of someone eating zero plant protein for life. An unlikely scenario, to be sure.

Moreover, certain combinations of vegan foods (like grains and legumes) have complementary amino acid profiles, restoring each other’s limiting amino acid and resulting in protein that’s complete or nearly so. Would these food combinations also spur cancer growth? How about folks who pop a daily lysine supplement after eating wheat bread? If Campbell’s conclusions are correct, it would seem vegans could also be subject to the cancer-promoting effects of complete protein, even when eschewing all animal foods.

Also, it seems Campbell never mentions an obvious implication of a casein-cancer connection in humans: breast milk, which contains high levels of casein. Should women stop breastfeeding to reduce their children’s exposure to casein? Did nature really muck it up that much? Are children who are weaned later in life at increased risk for cancer, due to a longer exposure time the casein in their mothers’ milk? It does seem strange that casein, a substance universally consumed by young mammals, is so hazardous for health—especially since it’s designed for a time in life when the immune system is still fragile and developing.

At any rate, Campbell’s theories about plant versus animal protein and cancer are essentially speculation. Despite a single experiment with restoring lysine to wheat gluten, he hasn’t actually offered evidence that all animal protein behaves the same way as casein.

But check this out. After delineating his discovery of the link between casein and cancer, Campbell writes:

We initiated more studies using several different nutrients, including fish protein, dietary fats and the antioxidants known as cartenoids. A couple of excellent graduate students of mine, Tom O’Conner and Youping He, measured the ability of these nutrients to affect liver and pancreatic cancer. (Page 66)

So he did experiment with an animal protein besides casein! Unfortunately, Campbell never mentions what the specific results of this research were. In describing the studies he conducted with his grad students, Campbell says only that a “pattern was beginning to emerge: nutrients from animal-based foods increased tumor development while nutrients from plant-based foods decreased tumor development.” (Page 66)

I don’t know about you, but I’d sure like to see the actual data for some of this.

After a little searching, I found one of the aforementioned experiments conducted by Campbell, his grad student Tom, and two other researchers. It was published in the November 1985 issue of the Journal of the National Cancer Institute: “Effect of dietary intake of fish oil and fish protein on the development of L-azaserine-induced preneoplastic lesions in the rat pancreas.”

(A preneoplastic lesion, by the way, is a fancy term for the growth that occurs before a tumor.)

In this study, Campbell and his team studied three groups of carcinogen-exposed rats: One fed casein plus corn oil, one fed fish protein plus corn oil, and one fed fish protein plus fish oil (from a type of high omega-3 fish called menhaden). All groups received a diet of about 20% protein and 20% fat and ate the same amount of calories.

Providing background for the study, the authors note that previous research has showed fish protein to have anti-cancer properties (emphasis mine):

Gridley et al. [n15,n16] reported on two studies in which intake of fish protein resulted in a reduced tumor yield when compared to other protein sources. Spontaneous mammary tumor development in C3H/HeJ mice was reduced. The incidence of herpes virus type 2-transformed cell-induced tumors in mice was also reduced in animals fed a fish protein diet.

Perhaps this should’ve tipped Campbell off that not all sources of animal protein spur cancer growth like casein does. For reference, the cited studies are “Modification of herpes 2-transformed cell-induced tumors in mice fed different sources of protein, fat and carbohydrate” published in the November-December 1982 issue of Cancer Letters, and “Modification of spontaneous mammary tumors in mice fed different sources of protein, fat and carbohydrate” published in the June 1983 issue of Cancer Letters.

So what were the results of Campbell’s experiment? According to the study, both the casein/corn oil and fish protein/corn oil groups had significant preneoplastic lesions. We don’t know whether to blame this on the protein or the corn oil, since—according to the researchers—“intake of corn oil has previously been shown to promote the development of L-azaserine-induced preneoplastic lesions in rats.” However, the group that ate fish protein plus fish oil exhibited something radically different:

It is immediately apparent that menhaden oil had a dramatic effect both on the development in the number and size of preneoplastic lesions. The number of AACN per cubic centimeter and the mean diameter and mean volume were significantly smaller in the F/F [fish protein and fish oil] group compared to the F/C [fish protein and corn oil] group. Furthermore, no carcinomas in situ were observed in the F/F group, whereas the F/C group had an incidence of 3 per 16 with 6 total carcinomas.

There’s some significant stuff here, so let’s break this down point by point.

One: The cancer-inducing properties of fish protein, if there are any to begin with, were neutralized by the presence of fish oil. This means that even if all animal protein behaves like casein under certain circumstances, its effect on cancer depends on what other substances accompany it. Animal protein is therefore not a universal cancer promoter; only a situational one, at best.

Two: What does “fish protein” plus “fish fat” start to resemble? Whole fish. Campbell just demonstrated that animal protein may, indeed, operate differently when consumed with its natural synergistic components.

Since there wasn’t a rat group eating casein plus fish oil, we don’t know what the effect of a dairy protein plus fish fat would have been. However, it would be interesting to have more studies looking at cancer growth in mice fed diets of casein plus milk fat. If casein loses its cancer-promoting abilities under that circumstance, as fish protein did with fish oil, then we’d have good reason to think the various factions of whole animal products might reduce any cancer-promoting properties a single component has in isolation.

And Campbell and his team conclude:

[A] 20% menhaden oil diet, rich in omega 3 fatty acids, produced a significant decrease in the development of both the size and number of preneoplastic lesions when compared to a 20% corn oil diet rich in omega 6 fatty acids. This study provides evidence that fish oils, rich in omega 3 fatty acids, may have potential as inhibitory agents in cancer development.

Remember how Campbell said, summarizing this research, that “nutrients from animal-based foods increased tumor development while nutrients from plant-based foods decreased tumor development”? Last I checked, fish oil ain’t no plant food.

Why does Campbell avoid mentioning anything potentially positive about animal products in “The China Study,” including  evidence unearthed by his own research? For someone who has openly censured the nutritional bias rampant in the scientific community, this seems a tad hypocritical.

But back to casein and milk for a moment. It’s interesting that the only dairy protein Campbell experimented with was casein, since whey—the other major protein in milk products—repeatedly shows cancer-protective and immunity-boosting effects, including when tested side-by-side with casein. Just a sampling of the literature:

• Diets containing whey proteins or soy protein isolate protect against 7,12-dimethylbenz(a)anthracene-induced mammary tumors in female rats. ” When 100% of the casein-fed rats had at least one tumor, soy-fed rats had a lower tumor incidence (77%) in experiment B (P < 0.002), but not in experiment A (P < 0.12), and there were no differences in tumor multiplicity. Whey-fed rats had lower mammary tumor incidence (54-62%; P < 0.002) and multiplicity (P < 0.007) than casein-fed rats in both experiments. … Furthermore, whey appears to be at least twice as effective as soy in reducing both tumor incidence and multiplicity.” (So much for plant protein being more protective against cancer!)
• Developmental effects and health aspects of soy protein isolate, casein, and whey in male and female rats. We found that SPI [soy protein isolate] accelerated puberty in female rats (p < .05) and WPH [whey protein hydrolysate] delayed puberty in males and females, as compared with CAS (p < .05). … Female rats fed SPI or WHP or treated with genistein had reduced incidence of chemically induced mammary cancers (p < .05) compared to CAS controls, with WHP reducing tumor incidence by as much as 50%, findings that replicate previous results from our laboratory.
• Tp53-associated growth arrest and DNA damage repair gene expression is attenuated in mammary epithelial cells of rats fed whey proteins. “Results indicate that mammary glands of rats fed a WPH [whey protein hydrolysate] diet are more protected from endogenous DNA damage than are those of CAS [casein]-fed rats.”
• A role for milk proteins and their peptides in cancer prevention. “Animal models, usually for colon and mammary tumorigenesis, nearly always show that whey protein is superior to other dietary proteins for suppression of tumour development.”
• A bovine whey protein extract stimulates human neutrophils to generate bioactive IL-1Ra through a NF-kappaB- and MAPK-dependent mechanism. “Our data suggest that WPE [whey protein extract] … has immunomodulatory properties and the potential to increase host defenses.”
• Whey proteins in cancer prevention.
• Whey protein concentrate (WPC) and glutathione modulation in cancer treatment.

Given all this, it seems unlikely that casein’s effects on cancer apply to other forms of milk protein—much less all animal protein at large. Isn’t it possible (maybe even probable) that casein has deleterious effects when isolated, but doesn’t exhibit cancer-spurring qualities when consumed with the other components in milk? Could casein and whey work synergistically, with the anti-cancer properties of whey neutralizing the pro-cancer properties of casein?

I’ll let you be the judge.

In summary and conclusion…

Apart from his cherry-picked references for other studies (some of which don’t back up the claims he cites them for), Campbell’s strongest arguments against animal foods hinge heavily on:

1. Associations between cholesterol and disease, and
2. His discoveries regarding casein and cancer.

For #1, it seems Campbell never took the critical step of accounting for other disease-causing variables that tend to cluster with higher-cholesterol counties in the China Study—variables like schistosomiasis infection, industrial work hazards, increased hepatitis B infection, and other non-nutritional factors spurring chronic conditions. Areas with lower cholesterol, by contrast, tended to have fewer non-dietary risk factors, giving them an automatic advantage for preventing most cancers and heart disease. (The health threats in the lower-cholesterol areas were more related to poor living conditions, leading to greater rates of tuberculosis, pneumonia, intestinal obstruction, and so forth.)

Even if the correlations with cholesterol did remain after adjusting for these risk factors, it takes a profound leap in logic to link animal products with disease by way of blood cholesterol when the animal products themselves don’t correlate with those diseases. If all three of these variables rose in unison, then hypotheses about animal foods raising disease risk via cholesterol could be justified. Yet the China Study data speaks for itself: Animal protein doesn’t correspond with more disease, even in the highest animal food-eating counties—such as Tuoli, whose citizens chow down on 134 grams of animal protein per day.

Nor is the link between animal food consumption and cholesterol levels always as strong as Campbell implies. For instance, despite eating such massive amounts of animal foods, Tuoli county had the same average cholesterol level as the near-vegan Shanyang county, and a had a slightly lower cholesterol than another near-vegan county called Taixing. (Both Shanyang and Taixing consumed less than 1 gram of animal protein per day, on average.) Clearly, the relationship between animal food consumption and blood cholesterol isn’t always linear, and other factors play a role in raising or lowering levels.

For #2, Campbell’s discoveries with casein and cancer, his work is no doubt revelatory. I give him props for dedicating so much of his life to a field of disease research that wasn’t always well-received by the scientific community, and for pursuing so ardently the link between nutrition and health. Unfortunately, Campbell projects the results of his casein-cancer research onto all animal protein—a leap he does not justify with evidence or even sound logic.

As ample literature indicates, other forms of animal protein—particularly whey, another component of milk—may have strong anti-cancer properties. Some studies have examined the effect of whey and casein, side-by-side, on tumor growth and cancer, showing in nearly all cases that these two proteins have dramatically different effects on tumorigenesis (with whey being protective). A study Campbell helped conduct with one of his grad students in the 1980s showed that the cancer-promoting abilities of fish protein depended on what type of fat is consumed alongside it. The relationship between animal protein and cancer is obviously complex, situationally dependent, and bound with other substances found in animal foods—making it impossible extrapolate anything universal from a link between isolated casein and cancer.

On page 106 of his book, Campbell makes a statement I wholeheartedly agree with:

Everything in food works together to create health or disease. The more we think that a single chemical characterizes a whole food, the more we stray into idiocy.

It seems ironic that Campbell censures reductionism in nutritional science, yet uses that very reductionism to condemn an entire class of foods (animal products) based on the behavior of one substance in isolation (casein).

In sum, “The China Study” is a compelling collection of carefully chosen data. Unfortunately for both health seekers and the scientific community, Campbell appears to exclude relevant information when it indicts plant foods as causative of disease, or when it shows potential benefits for animal products. This presents readers with a strongly misleading interpretation of the original China Study data, as well as a slanted perspective of nutritional research from other arenas (including some that Campbell himself conducted).

In rebuttals to previous criticism on “The China Study,” Campbell seems to use his curriculum vitae as reason his word should be trusted above that of his critics. His education and experience is no doubt impressive, but the “Trust me, I’m a scientist” argument is a profoundly weak one. It doesn’t require a PhD to be a critical thinker, nor does a laundry list of credentials prevent a person from falling victim to biased thinking. Ultimately, I believe Campbell was influenced by his own expectations about animal protein and disease, leading him to seek out specific correlations in the China Study data (and elsewhere) to confirm his predictions.

It’s no surprise “The China Study” has been so widely embraced within the vegan and vegetarian community: It says point-blank what any vegan wants to hear—that there’s scientific rationale for avoiding all animal foods. That even small amounts of animal protein are harmful. That an ethical ideal can be completely wed with health. These are exciting things to hear for anyone trying to justify a plant-only diet, and it’s for this reason I believe “The China Study” has not received as much critical analysis as it deserves, especially from some of the great thinkers in the vegetarian world. Hopefully this critique has shed some light on the book’s problems and will lead others to examine the data for themselves.

They live here:

Which looks like this:

Where they eat a lot of this:

But not a lot of this:

The Tuoli diet is so abnormal for China, in fact, that T. Colin Campbell et al omitted this county from analysis in several China Study papers—such as “Vitamin A and cartenoid status in rural China,” published in the British Journal of Nutrition:

One county (Tuoli County in Xinjiang Autonomous Region), composed primarily of an ethnic minority population of herdspeople, had disproportionately high values for retinol, lipid and protein intake due to an exceptionally high intake of animal foods. This ‘outlier’ was not included in the analysis, to characterize more accurately the average intakes of the rural Chinese population and to avoid the undue influence of one data point on the results.

Given the prevailing beliefs about nutrition and health—such as saturated fat and cholesterol as a cause of heart disease, the necessity of fiber for colon health, the immunity-boosting properties of fruits and vegetables, and the dangers of a diet high in animal fat—it would seem the Tuoli should showcase the health woes that come from breaking every rule in the diet book.

But is that the case?

Tuoli diet

First, let’s take a closer look at what he China Project data has to say about these Tuoli folks.

In terms of macronutrients, the Tuoli consumed an average of 185.6 grams of fat, 172.5 grams of protein, and 322 grams of carbohydrates per day. Average energy intake was a whoppin’ 3704 calories, and average fiber intake was 17.9 grams per day—only slightly more than your run-of-the-mill American.

The average diet of all counties studied in the China Project is clearly carb-based, low in fat and protein (as a percent of total calories):

In contrast, the Tuoli diet is nearly half fat:

Main items on the Tuoli menu included:

• Dairy: 856.5 grams per day (almost two pounds)
• Wheat flour: 371.6 grams per day (0.82 pounds)
• Meat: 121 grams per day (a bit over a quarter of a pound)

Sparse and non-existent items included:

• Potatoes: five to six times per year
• Green vegetables: twice per year
• Fruit: less than once per year
• Legumes: never
• Sea vegetables: never
• Nuts: never
• Eggs: never
• Fish: never
• Plant oils (rapeseed, soybean, sesame, corn): never
• Soy sauce: never

Basically, these folks live on dairy, meat, and wheat, and refuse to eat their vegetables. Sounds like some Americans I know.

Tuoli blood markers and diseases

If the Tuoli’s meat-and-dairy-heavy diet is the source of disease, we’d expect to see these folks facing more chronic conditions than the regions eating plant-based diets. To test whether this is the case, let’s compare Tuoli with the 13 counties in the China Project that consumed less than 1 gram of animal protein per day—the closest thing we have to Chinese vegans.

I’ll be putting these all in a bar graphs, but to prevent an uber-cluttered x-axis, I’ll just use numbers corresponding to each county:

1. Cixian
2. Jingxing
3. Huguan
4. Jiangxian
5. Jiexiu
6. Linxian
7. Songxian
8. Jianhu
9. Taixing
10. Qingzhen
11. Cangxi
12. Shanyang
13. Longxian
14. Tuoli

The first 13 counties will always be blue bars; Tuoli will always be red.

Before getting to the mortality statistics, let’s look at some basic blood markers for heart disease. Here we have total cholesterol of the above counties, lined up side-by-side for comparison.

As you might expect, Tuolians* have higher total cholesterol than most of the near-vegan counties, although it’s still a healthy number by American standards. However, the difference between a couple of those counties isn’t all that profound: Tuoli’s cholesterol is tied with that of Shanyang and lags a bit behind Taixing, both of which consume only trivial amounts of animal products. Curious, indeed. Obviously, something other than animal product consumption affects blood cholesterol.

*”Tuolian” may or may not be an actual term.

Next, let’s peek at triglycerides—a type of blood fat that, in high amounts, can raise your risk of heart disease.

It seems Tuoli is pretty much neck-and-neck with the plant-eating counties. By American standards, triglyceride levels between 150 and 199 are considered borderline high, and lower numbers are considered normal—so only one county, a near-vegan one, had values outside a healthy range.

Disease rates

First up: Death from all causes (per 1000 people under the age of 65). Remember, Tuoli county is the red bar; the blue bars represent the near-vegan counties in the China Project that consumed less than 1 gram of animal protein per day on average.

Okay, so the Tuoli don’t have a higher death rate than the near-vegans. In fact, Tuoli’s total mortality rate is lower than 11 of the other counties and higher than only two.

But what about cancer? Let’s look at mortality from all cancers for Tuoli and the plant-lovin’ regions. Again, Tuoli is the red bar, and the near-vegan counties are the blue ones.

How ’bout them apples? Tuoli doesn’t appear to have higher cancer rates than the near-vegan areas. Eight counties have higher rates and only five have lower ones, leaving Tuoli hovering near the lower-middle end of the spectrum.

Next we have mortality from myocardial infarction (heart attacks) and coronary heart disease, per 1000 people. Tuoli is red, near-vegan counties are blue… you know the drill.

Surprised? Despite a massive intake of cholesterol, saturated fat, calories, animal protein, and all those other horrors ascribed to declining heart health, the Tuoli have relatively low levels of coronary heart disease and heart attacks. Seven near-vegan counties have higher rates than Tuoli, and six have lower rates.

And now for stroke mortality (per 1000 people).

Again, no significantly higher stroke rates for the Tuolians. Seven near-vegan counties have more incidences of stroke, and six have fewer incidences of stroke.

And since lack of fiber is supposed to harm colon health, here is a comparison of colon cancer and rectal cancer mortality (per 1000 people) between the plant-noshing counties and the vegetable-phobic Tuolians.

Looks like they’re doing pretty dandy without much fiber, right?

But what about leukemia? Let’s check it out:

As you can see, Tuoli isn’t significantly worse off than the near-vegan counties in terms of chronic disease. Total mortality rate is lower, cancer rates are lower or similar, heart attacks aren’t more common than usual, stroke rates are average. From this data alone, we’d have no basis for claiming that eating two pounds of dairy per day (and minimal vegetation, aside from wheat flour) is less healthful than consuming a mostly vegetarian diet. For sure, this data fails to support Campbell’s claim that chronic disease rates climb when animal protein intake rises.

(And as you’ll see in an upcoming post, it’s pretty surprising that the Tuoli had low rates of cardiovascular disease while eating high levels of wheat—but we’ll get to that later.)

Why aren’t these people sick and diseased?

We have plenty of evidence showing hormone-pumped dairy, grain-fed meat, pasteurized and homogenized milk, processed lunch meats, and other monstrosities are bad for the human body. No debate there. But we do have a woeful lack of research on the effects of “clean” animal products—meat from wild or pastured animals fed good diets, milk that hasn’t been heat-zapped, antibiotic-free cheeses and yogurts, and so forth. Perhaps the best data we have is from observational studies of isolated or primitive peoples (such as those studied by Weston A. Price), but those lack detailed documentation about mortality rates and don’t usually meet standards of scientific rigor.

In other words, this is one area where nutritional research is pretty deficient.

Is it possible the diseases we ascribe to animal products aren’t caused by animal products themselves, but by the chemicals, hormones, and treatment processes we expose them to? If the Tuoli are any indication, this may be the case. Hopefully future research will shed more light on the matter.

Mongolian yaks: A source of Chinese dairy.

I’ll admit it: Out of all the variables in the China Project, dairy is the one I’ve been most eager to analyze. Not because I’m a dairy lover myself (I haven’t touched it in years) or because I’m secretly a billionaire milk tycoon with my own thousand-acre Holstein farm (au contraire; I’m strangely phobic of cows). In his book, T. Colin Campbell makes such a compelling case about casein (a milk protein) as a cancer-promoting agent that I’m left wondering: Does the China Study data shows an equally convincing link between dairy and disease?

After all, the counties studied in the China Project weren’t eating the hormone-laden, antibiotic-stuffed, factory-farmed dairy we find in most stores. Their dairy was from pastured animals—typically sheep, goats, or yaks along with cattle—raised on natural diets in rural areas. As best I can deduce, milk products were neither pasteurized nor homogenized. This means that any connections we find between dairy and mortality variables are probably from dairy itself—not the nastiness that accompanies the dairy Westerners are more familiar with. This could be one of our best opportunities for studying dairy consumption in its raw, natural state. Yeehaw!

(Note: If this is your first visit to my site and you’re on a quest for China Study information, you may want to start with the earlier posts in this series:)

• What the China Study says about meat and disease
• What the China Study says about fish and disease
• What the China Study says about eggs and disease

Dairy consumption in rural China

First, the bad news—from a scientific standpoint, at least. Only three out of the 65 counties in the China Project consumed any noteworthy amount of dairy at all. The rest were completely dairy-free or consumed dairy only a handful of times per year. That means our sample size of hardcore dairy eaters is tiny, and drawing any conclusions about dairy consumption is trickier than if we had lots of dairy-eating regions to examine.

But there’s some good news, too. Two of the counties that did eat dairy ate a lot of it—856.5 grams per day for Tuoli (just shy of 2 pounds!) and 147 grams per day for Xianghuang qi (about a third of a pound). In terms of diet, both of these places are proverbial black sheep in China, making them quite interesting as case studies. And because these regions consumed so many milk products in contrast to everyone else, any correlations genuinely tied to dairy should be pretty dramatic.

Baoqing county, the third highest consumer of dairy, ate about 19.1 grams of milk products per day. Not a whole lot, for sure, but we should keep an eye on this county as well when looking for links with disease.

A tad more info on these places for anyone who’s curious:

Tuoli county. Located at the tippy-top of northwest China in the Uyghur region—a straggler county on the map, much farther west than any other area studied in the China Project. The Tuoli get over half of their daily calories from dairy products (holy cow, literally; this is even more than Americans eat) and are quite fond of yogurt. They consume very few vegetables, fruits, or nuts, but do eat a significant amount of wheat flour.

Xianghuang qi county. Located in inner Mongolia. Local cuisine includes mutton and dairy products—especially yogurt, fermented milk, Mongolian milk tea, butter, and cheese—as well as oats and buckwheat. Vegetables haven’t been a big part of their menu until quite recently. You can read more about the diet of this region at China.com.

Baoqing county. Located at the northernmost and easternmost spot out of all the counties studied. These folks are one of our highest egg eaters, eat moderate amounts of meat, and consume wheat as their primary grain.

Dairy consumption in rural China

In China, dairy intake ranged from zero grams per day to 856.5 grams per day, as mentioned earlier. Since so few counties consumed milk products, our correlations are strongly swayed by the habits of the two dairy-eating regions. Those include snuff use (correlation of +98), meat eating (+52) and wheat intake (+35). Dairy-eating people also tended to be heavier and taller than other Chinese citizens (+34 and +23, respectively). Although the additional fat and protein from dairy foods could be responsible for a bigger body size, inhabitants of these regions tend to be ethnic minorities in China, and it’s possible they have larger builds genetically.

Dairy intake correlates positively with HDL or “good” cholesterol (+29), but not with LDL or “bad” cholesterol (-15). And in terms of general diet composition, dairy-eating regions had a low intake of fiber (-28), soluble carbohydrates (-23), and plant protein (-25) but a high intake of total protein (+80), animal protein (+99), total fat (+78), and total calories (+62).

So does all this animal protein and fat equate to more chronic diseases, especially cancer? Let’s take a look-see. Surprisingly, there’s only one statistically significant correlation, so I’ll list even the weak correlations to give the full picture.

NEGATIVE CORRELATIONS (more dairy = fewer of these diseases)

Liver cirrhosis: -21
Peptic ulcer: -19
Lymphoma: -19
Other metabolic diseases: -17
Rectal cancer: -14
Penis cancer: -13
Diabetes: -10
Colorectal cancer: -9
Death from all causes: -9
Other heart disease: -9
Bladder cancer: -8
Diseases of the blood and blood-forming organs: -7
Colon cancer: -6
Leukemia: -6
Liver cancer: -4
Neurological diseases: -2

POSITIVE CORRELATIONS (more dairy = more of these diseases)

Hypertensive heart disease: +30*
Stomach cancer: +10
Breast cancer: +9
Stroke: +9
Oesophageal cancer: +8
Death from all cancers: +7
Myocardial infarction: +6
Brain cancer: +4
Cervix cancer: +2
Rheumatic heart disease: +1

Remember that, for both positive and negative correlations, smaller numbers are essentially insignificant. Zero is perfect neutrality, but we rarely get a statistical zero in the real world—especially when we have a maximum of 65 data points to work with. With the exception of hypertensive heart disease, none of the positive correlations appear meaningful, and perhaps only lymphoma and “other metabolic diseases” warrant further study among the negative correlations. (Liver cirrhosis and peptic ulcers, the other marginally strong inverse trends, most likely have non-nutritional causes.)

Hypertensive heart disease

Unlike atherosclerosis, hypertensive heart disease is not caused by plaque building up on arterial walls. With this condition, chronic high blood pressure (AKA hypertension) forces the heart to work harder, leading to a thickening of the muscle. Does something about dairy consumption raise blood pressure and lead to a big heart (the diseased kind, not the generous kind)?

Looking solely at the “hypertensive heart disease” variable, we can see that a few other factors outshine the dairy correlation. Daily salt intake correlates at +37, salt intake plus urine salt correlates at +50, weight correlates at +33, and wheat flour correlates at +54. The “other foods” category—which includes vinegar, MSG, baking powder, tea, and melon seeds—has a correlation of +51. Negative associations include rice (-45), yearly green vegetable intake (-36), steamed bread and pancakes (-57), and daily alcohol consumption (-27).

Since we really only have three counties that consumed significant levels of dairy, it won’t be easy—and maybe not even possible—to pinpoint the role of dairy itself in hypertensive heart disease. Here’s our untampered graph, charting every county that reported hypertensive heart disease  mortality.

Kinda sparse, huh?

There certainly is an upward trend amongst dairy-eating counties, but let’s face it: We’re basing that on two measly dots. Dots that, more importantly, represent complete dietary rebels in this data set. Along with consuming milk products, our two dairy-eating regions have some other nutritional divergences compared to the rest of China. For instance:

• Both significantly fewer vegetables than any of the other 65 counties in the China Project: Tuoli only consumes green vegetables an average of two times per year (!) and Xianghuang qi consumes them 24 times per year, about twice per month. The average intake for all counties is closer to 200 times per year. Could lack of green vegetables, which we already know correlates with hypertensive heart disease, be raising these county’s rates?
• Tuoli consumes a whoppin’ average of 172.5 grams of protein per day, 134.55 of which come from animal sources. Even a gym-rat bodybuilder might consider that excessive. No other county comes close to that intake; the next highest is 90.8 grams per day. Could an uber-high protein intake contribute to hypertension?
• Tuoli also consumes far more fat than any other county: 185.8 grams per day. Wowza. The average for all of China is 44.2. Is a high fat intake playing a role?
• As you’d expect from a very high-fat diet, the Tuoli eat more calories per day than any other county studied in the China Project: 3704 on average. No, that’s not a typo. They’re also the heaviest county among the 65. Do these folks have higher rates of obesity, which surely is a risk factor for hypertensive heart disease?
• Both dairy-eating counties had a higher levels of dietary and urine sodium than average. This one’s a no-brainer; excess sodium is a well-known cause of high blood pressure and, consequently, hypertensive heart disease.
• Both dairy-eating counties eat significant quantities of wheat flour, which—as you shall soon see—has some really mind-bogglingly crazy associations with cardiovascular disease. Seriously, it’ll knock your socks off. But that’s a few more posts away.

And despite eating almost two pounds of dairy a day, Tuoli’s hypertensive heart disease mortality (far right dot) is still surpassed by two other counties—both of which eat no dairy at all. And the disease rate for Xianghuang qi county is right smack dab in the middle of the data set range, ranking behind six dairy-free counties. Even if dairy is a factor in developing hypertensive heart disease, which we can’t say for certain, it’s clearly not the only contributor.

At any rate, I’m not interested in vilifying nor vindicating dairy here; I’m just exploring alternative possibilities for the correlation between milk products and hypertensive heart disease.

Bottom line:

• China’s dairy eaters don’t have significantly more cancers, myocardial infarction, stroke, and so forth than the dairy-free regions.
• Dairy’s only significant mortality correlation, hypertensive heart disease, may be related to any number of variables we don’t have enough data to tweeze apart. (Lack of vegetables, excess sodium, high body weight, and high caloric intake, to name a few.)
• Despite T. Colin Campbell’s findings with the milk protein casein spurring cancer in lab rats, there does not seem to be a correlation between high dairy consumption and cancer in the China Study data.

Are you as fascinated as I am with the Tuoli’s incredibly high-fat, high-protein, vegetableless diet? Are you wondering what their disease rates are, how long they live, and whether they’re any healthier or sicker than the Chinese eating plant-based diets? I sure am. Up next will be a mini-post about the Tuoli specifically, including answers to the aforementioned questions. This should be pretty interesting!