the campbellcoalition.org site is no longer existent due to restrategizing efforts. some of you never got around to seeing some of the excellent items that were on that site, so nelson campbell has kindly offered to make these available for posting within our group.
Dr. T. Colin Campbell Blog 5 Are All Proteins Created Equal?
I recently received a question that is one of the most commonly asked questions of me. Because this particular question seems to be of such interest, I thought it might make sense to turn my answer to this recurring question into this week’s blog. Here is the question: isn’t protein from any source, whether from animals or plants, similar in its biological effects?
The short answer to this question is “No.” When a dozen or so animal proteins were compared in animal studies with a dozen or so plant proteins, ALL animal proteins generated higher levels of serum cholesterol than did ALL plant proteins. Similarly, these two forms of protein do not have the same 'biological value', a concept that speaks to their abilities to support body growth and their efficiency of utilization.
And if you read “The China Study,” you may already know of our experiments showing how animal protein can stimulate cancer growth. Animal protein promotes cancer growth when fed at levels in excess of protein needs, but plant proteins when fed in isolation and at the same levels do not. We studied this protein effect extensively, it was peer-reviewed multiple times, it was supported by NIH funding and it was published in the very best science journals.
Interestingly, the animal protein we used in these experiments was casein (from dairy). Based on the data from these studies and on traditional carcinogen testing criteria, casein is the most relevant chemical carcinogen we ever tested, with a high probability that all animal-based proteins would do the same thing, similar to their behavior with other responses.
I realize that one could make the argument that plant-based proteins could be carcinogenic and act like animal-based proteins if they are fed in isolation and at a high enough level. But in reality, we get our nutrients from whole foods, or at least we should, and for this reason it is impossible to consume plant protein at such a high level. In addition, any theoretical ill effects would be limited by the consumption of multiple anti-disease components of plants.
In the case of plant and animal proteins, they are not created equal. As components of food, one can cause illness; the other becomes an essential element of a healthy diet.
Dr. T. Colin Campbell Blog 6 Filling a Hole in Government
(FROM AN EDITORIAL RECENTLY PUBLISHED IN THE HUFFINGTON POST)
Finally, a long overdue scientific correction is happening. The human genome project is failing to advance the cause of human health as was promised. There is a fundamental reason why this is happening.
Working out the details of the human genome was worth doing, on several accounts--monitoring environmental pollutants, evaluating evolutionary lineages, identifying criminal suspects. But initially promising great advances in human health was not and should not have been one of these promises. Hypothesizing that knowledge of the associations of specific genes with serious diseases like cancer, heart disease and related diseases would lead to great health advances (through drug development) was a superficial and costly oversimplification of disease causation.
Although genes and/or their mutated forms are fundamental to the initiation of all disease events, it is not their mere presence or absence that determines disease outcomes. Genes may start the job but they do not finish it. The far more important question we should ask is: what controls the expression of genes (to produce products, mostly enzymes) that lead to health and disease events? Experimental and extensively published research from my laboratory over several decades has long convinced me that nutrition primarily provides this control.
We have failed to acknowledge this question or sought its answer for far too long because we have failed to understand the scientific fundamentals of nutrition. Not one medical school in the nation adequately teaches this science, although a few give it lip service.
Still worse is the failure of federal funding agencies to recognize nutrition as a legitimate medical science. The National Institutes of Health, the most prominent biomedical research institution in the world, has kept the promise of nutrition well hidden. Not one of its 27 institutes and related centers is dedicated to nutrition! Some NIH administrators say nutrition is embedded in other programs, but do not be fooled. First, dedicated nutrition funding is meager (less than five percent of the heart and cancer institutes). Second, this small amount has been used primarily to study single nutrient effects in randomized clinical trials, a seriously flawed hypothesis. Nutrition should not to be defined by the effects of isolated nutrients. That’s pharmacology, a strategy now known not to work, in spite of the $25 billion or so that we annually spend on nutrient supplements.
Unequivocal evidence now exists to show that nutrition, when provided through whole, plant-based foods, can control the expression of our mischievous genes that otherwise would lead to serious ailments such as heart disease, diabetes, certain autoimmune diseases and many lesser ailments. For many years, experimental findings from my laboratory have shown that genetic initiation of cancer (by a powerful chemical carcinogen) can be stalled, even reversed by a modest nutritional modification that is consistent with this same whole, plant-based food effect. Numerous physician colleagues of mine have now published peer-reviewed findings showing this kind of nutrition not only to prevent serious diseases such as heart disease, diabetes and related ailments but to treat them into remission.
There is no other strategy in contemporary health science or medical practice that comes close to the breadth and depth of health benefits achievable by nutrition. We must begin to understand, communicate and apply this knowledge if we ever hope to reduce health care costs. We will never do this by depending on outmoded notions of what single (or even a few) genes, single nutrients or single chemicals (i.e., drugs) will do to create health. That thinking generates wealth for a few at the expense of health for the many. It is time to recognize the natural and harmonious biological complexity of health processes, and choose the lifestyle strategy that best maintains and restores that harmony. Nature has had eons of time to work this out. It’s also time to develop a professional science of nutrition that serves the biological health of the population, not the economic health of commerce.
As for health professionals who claim they cannot convince patients to change their dietary practices, this is not surprising when the professionals themselves are not educated in this field and are vested in a strategy that is the antithesis of good nutrition. It is time we recognize what nutrition can do and a good place to start is to establish an NIH Institute of Nutrition dedicated for this purpose.
The concepts of correlation and causation are often misunderstood and misused. I’ve been asked to express my own views on this matter but, to do so, I believe I should begin with a few more general comments on the use of statistics.
I find that statistics is an enigma for many people. Opinions range from those who seriously disparage statistics (‘statistics lie’) to those who believe that statistical proof is an absolute criterion for making decisions. Then there are the many in the middle who are not quite sure what to believe. I suggest—as do many others—that the correct answer is that statistics is only a tool to guide us, as we move from observation to observation, experiment to experiment and ultimately to seek the elusive truths. One thing about statistics is quite clear. Many people use and praise statistics much more when the analyses support their favorite hypotheses and biases!
Statistical analysis really does contribute value to our research investigations, but to appreciate this value, we should know its limitations. Although the math of statistical analysis may be a limitation for some, I suggest that the main limitation is the way statistically evaluated results are interpreted. To illustrate, I will consider one of the more troubling interpretations, the relationship between correlation and causation.
For example, let’s consider the observation that fat consumption is associated with breast cancer, based on data obtained for a group of countries. We graph our findings and statistically show that they track together more or less in unison. As one increases, the other increases. We say, therefore, that they are ‘correlated’. If these two variables track in the same direction, as they do, we say they are directly correlated (notated by a plus sign). If one variable increases while the other decreases, as observed with dietary fiber and breast cancer, for example, we say that they are inversely correlated (notated by a minus sign). Now let’s examine the analytical quality of a correlation like this. But before doing so, I should briefly describe a couple descriptors of statistics used to describe correlations.
From the graphical perspective, a perfect correlation exists if the x,y points (each point on the graph has an x value for one of the variables and a y value for the other variable) line up perfectly on a straight line that deviates away from the horizontal. The strength of a correlation is described by a coefficient r, in this case having a value of ±1.00 for the perfect correlation. Less than perfect correlations (involving almost all correlations in real life) are indicated by some fraction of ±1.00, that is, r values ranging from 0.00 to +1.00 (direct correlation) or 0.00 to -1.00 (inverse correlation). An r value of 0.00, for example, tells us there is no correlation, that is, data points on the graph are randomly scattered with no discernible relationship. The higher is the (absolute) r value, the closer are the data points to the line and the stronger is the correlation. The following graphs should illustrate.
Statistics can be used to assess the likelihood (statistical significance), of whether the observed correlation (i.e., deviation of r from 0.00) is real or merely the play of chance. This assessment becomes more sensitive for detecting a true correlation—when there are more data points. The degree of statistical significance is conventionally (and arbitrarily) indicated by cut-off points. If, for example, we find that there is a probability of, at most, one in twenty (or less, i.e., <5%), we describe it as a probability p of less than 5% (p<0.05). Similarly, one chance in 100 (or less, p<0.01) and one chance in 1000 (or less, p<0.001) are the more significant probabilities. Conventionally we describe a finding as “statistically significant” when p is less than 0.05 (p<0.05) and as “highly statistically significant” when p is less than 0.01 (p<0.01). If it is p<0.001, we are likely to be impressed if the observed effect supports our initial hypothesis! Sometimes, the degree of statistical significance may be expressed more precisely as in p=0.06 or p=0.13. I like this latter approach because it offers the observer more leeway in deciding how ‘significant’ the results may be, thereby encouraging consideration of other factors likely to be playing a role.
In our data collected for our China database, we had 367 variables (disease mortality rates, nutrient intakes, blood plasma biomarkers, etc.), each of which was correlated with every other variable. This provided about 100,000 correlations, about 8000 of which were statistically significant. We published correlation coefficients, r values, , either as +r (direct correlation) or -r (inverse correlation) for each of these correlations. Also, for convenience, we added *, **, and *** to each correlation coefficient to represent r p<0.05, p<0.01 and p<0.001, respectively. And finally, we added several other parameters to help us evaluate the quality of these correlations.
Now that we have acquired some basic terminology, I will discuss what I consider to be an equally, if not more relevant part of this story. It is the question of how to interpret and apply statistical estimates of correlations obtained in a single survey or experiment, in order to extract information that suggests causation.
First, it is worth repeating that for any random sample survey of a larger population, we are assuming that the correlations represent the larger population. Said differently, if we select a second sample of the same population, we would like to have some confidence that we would get very similar results. A repeat sample will not give us exactly the same results but they should be similar if both samples were truly random and representative. Sample correlations, strictly speaking, apply only to that sample, but nonetheless reflect what is true for the larger population.
In human population surveys like the China project, before we proceed with interpretation of the data, we must first, if possible, screen for any methodological limitations affecting the reliability of the variables that comprise the correlations. Different laboratories, different methodologies and different analysts will produce varied results for the same data. Because of these sources of variation, it is important to examine these variables for any methodological shortcomings that can be appropriately eliminated or adjusted. Correlations are only going to be as reliable as the quality of their variables.
In the China database, for example, dietary and lifestyle measurements were collected in 1983 (a few in early 1984) whereas mortality rates were for the years of 1973-1975. Thus, there is a legitimate concern that the hypothetical causal factors of disease were measured 8-10 years after deaths occurred when, ideally, they should have been measured prior to the deaths. Thus it is important to have some idea that the data collected in 1983-1984 were representative of the past. We had evidence, published elsewhere, to provide some of this confidence. We also had to consider the quality of the disease mortality rates and, again, we obtained information that these rates likely were, in most cases, surprisingly reliable. And finally, we had to know for each variable whether the distribution of values across the range of counties was reasonably bell-shaped, and not skewed to concentrate at one end of their range.
Every observational, epidemiological study has its own set of experimental conditions to consider when validating the data for statistical evaluation. After data are validated to the best of one’s ability, correlations and other statistical parameters can be computed. At this stage of the investigation, however, correlations can still be quite superficial and misleading because they may be substantially confounded by the effects of other variables. That is, each correlation represents a one-on-one relationship between two variables. These variables and their correlations may be acting only as indicators of some other factor or even some influence not recorded in the study. This brings us to the well-accepted proclamation that correlations of two variables, without further consideration and analysis, cannot alone be used to make conclusions. They should be statistically evaluated and adjusted for confounding factors.
A few real-life examples of misusing correlations to infer causality may illustrate. During the 1970s and 1980s, a very impressive correlation of dietary fat with breast cancer was observed when different countries were compared. For many people, this inappropriately suggested that fat specifically causes breast cancer, a misinterpretation that had major consequences. National policies and marketplace practices followed suit and gave further weight to this misunderstanding of fat causing breast cancer, especially during the 1970s and 1980s, a misunderstanding that still remains with us even today. Fortunately, some researchers correctly used this correlation to develop an hypothesis for further investigation. These follow-up studies, which were founded on this hypothesis, conclusively demonstrated (and surprisingly so for many people) that decreasing fat in foods and the overall diet did not explain the initially observed correlation. The initial association of fat with breast cancer proved to be far more complex, with dietary fat only indicating the real relationship. I further elaborate on this issue in chapter 14 of The China Study. The initial misinterpretation was profoundly important because it had far-reaching public consequences. Untangling the initial correlation required considerable investigation on what is biologically plausible and what is not.
On a second example, beta carotene intakes were shown in a couple of dozen studies about 30-40 years ago to be inversely correlated with lung cancer occurrence—higher beta carotene, lower lung cancer (actually published as a meta-analysis of about 20 studies). Because of this focus on beta carotene, a promising market for supplements of this nutrient quickly emerged to exploit this inverse correlation, based on the hypothesis that this nutrient could prevent cancer. But about 10 years later, two major studies unequivocally showed the opposite relationship when supplements of beta carotene were tested. Higher beta carotene intake increased by about 6-fold lung cancer deaths. Untangling the initial correlation required considerable investigation on what is biologically plausible and what is not.
As a third example, polyunsaturated fat, typically found in plant oils, was found to be inversely correlated with heart disease risk. This finding led the marketplace to substitute polyunsaturated fat-rich cooking oils and oleomargarine for butter and lard. Major food industries adopted this message for promoting their food products. Yet, eventually, it was shown that this was an inappropriate interpretation of the original ‘correlation’. Polyunsaturated fats were shown to have the potential to increase the risk of certain cancers (as in high fat diets) probably because of their induction of free radical production—which is not the case for a whole foods, plant based diet naturally low in fat. Again, the effect of polyunsaturated fats are now known to depend on other dietary factors. Untangling the initial correlation required considerable investigation on what is biologically plausible and what is not.
And as a fourth example, many studies have shown a direct correlation of calorie consumption with obesity and other adverse health effects. These results led many to conclude that decreased calorie intake is all that is needed to control these diseases. Although there was some merit to this hypothesis, it still ignored a very important role that diet composition had played. Untangling the initial correlation required considerable investigation on what is biologically plausible and what is not.
There are hundreds of similar stories where inappropriate conclusions have been drawn from correlations or their equivalent, whether or not they are statistically significant. Aside from the use of poor quality data and/or weak scientific studies that fail to record possible confounding factors (no research investigation measures more than a mere handful of such factors), the most common problem associated with the misinterpretation of correlations may be the direct use of their ‘raw’, unadjusted forms for making conclusions about causation. First, the variables need to be validated; second, the correlations need to be adjusted, if possible, for confounding factors; and third, biological plausibility must be taken into account.
So how is causation determined? As a researcher primarily experienced in biology, as opposed to statistics, I have mostly relied on statistician colleagues to do the statistics while I, in turn, have relied more on my understanding of the biological explanations (i.e., plausibility) of diet and health relationships. It is well accepted among researchers that herein lies a problem. Statisticians know statistics best, while biologists know biology best and each sometimes comes up short in properly using the information and technology of their colleagues. This difficulty of specialization has been widely acknowledged in the research community, and for most serious research groups, a statistician is an equal member of the research team. Even so, however, I still find that discussions of causation too often emphasize, to a fault, one kind of interpretation over the other.
For me, assessing correlation goes as follows—and I will use our experience with our database in China to illustrate. First, and foremost I accept that correlations are primarily intended for generating interesting hypotheses, meaning that more discriminating research needs to be done. Most researchers are adamant that this precaution be observed and will refuse to consider any other possible use for these correlations. I don’t fully support this view because, under some circumstances, I believe that it is reasonable to use them, if done properly.
By “properly”, I need to explain my concern about a very basic problem with the way we think about biological phenomena. In research, whether we do surveys, as in our China project, or use controlled experiments, we focus our attention on finding specific factors related to disease, either those that cause or those that prevent disease (by ‘focus’, I mean either testing isolated nutrients in controlled trials or searching for independent effects through adjustment for confounding, as in human epidemiological studies). This strategy is highly reductionist and was discussed in The China Study. Even though this kind of research is essential for much of what we learn about basic biological events, it also has a serious shortcoming when investigating nutrition. My interpretation of nutrition refers to the manner in which countless food components (many of which we call nutrients), operating through countless but highly integrated mechanisms, produce a symphonic-like response involving a large variety of health and disease outcomes.
Unfortunately, most researchers do not make this assumption. Instead, they adhere to the reductionist view of biological science. Such a preference for biological reductionism then becomes their rationale for adamantly rejecting the use of correlations for drawing inferences about causation in studies like ours. They are looking for the independent effects of specific agents, whether they are causal or preventive. Were I a strong supporter of reductionist research, I too would agree with their views. But, concerning the effects of nutrition on disease etiology, I don’t support this philosophy and have a somewhat different view of the use of correlations to support inferences about causality.
I rely on wholistic hypotheses to develop models about nutrition that rely, in turn, on biologically plausible explanatory models. These models are composed of, or are dependent on, virtually countless causative factors, countless mechanisms of action and countless outcomes. I then feel comfortable scanning a database for correlations, adjusted when possible, that support or deny the internal consistency of these models. With a large collection of correlations, as in our comprehensive China database, we had the advantage of searching from diverse perspectives for evidence for or against these models.
In our book, The China Study, I presented six such explanatory models and then discussed evidence found in the China database that consistently corroborated for each model the plant-based diet hypothesis, even though I was initially skeptical that this would be possible for this cohort of people. That is, their dietary practices were mostly confined (in reference to Western dietary experience) to the use of a plant-based diet (about 80 to 100% of total protein from plants), thus leaving a restricted range of associations and a less sensitive means of detecting significant associations. The fact that these models all indicated a beneficial nutritional effect provided by a plant-based diet (i.e., diseases are prevented) and, further, that the supporting evidence in the China database was produced within such analytically restrictive conditions made our conclusions favoring a plant-based diet all the more remarkable. In effect, I began with multi-factor models of cause-effect relationships that were demonstrated, beforehand, to be biologically plausible. Then I merely inquired whether the evidence from the China database (including correlations from the simple to the complex) was supportive.
I am well aware that there may be those researchers who would not do this, for fear that they might be leading from a personally biased position. I agree with this assessment if the investigator is being influenced by personal hunches and preferences. However, it is different when one begins with an hypothesized model of causation based on documented biological plausibility, as we did with our China database. We had six models and each model contained multiple factors that could be compared with correlations (appropriately qualified) that reflect the influence of plant-based nutrition. In this kind of analysis, the first question is whether the nutrition-based variables and their respective correlations within each model were consistent with the health advantages of a plant-based diet. The second question probed whether there was consistent support for the plant-based hypothesis among all six models (actually there were more). We were convinced that supportive evidence did exist, made more remarkable by the restrictive range of dietary experience mentioned above. Further, the variables that were considered included data on food intakes, clinical biomarkers, and disease mortality rates, thus providing an unusual opportunity to evaluate the hypothesis on the health benefits of a plant-based diet from multiple biological perspectives.
To statistically evaluate such a biologically complex pathway (as in the hypothesis on plant-based diets), one can simply ask what the probability is of demonstrating that all six models were not only internally consistent within each model but also that the China survey data showed that all six models were externally consistent—all six favored the same hypothesis.
In summary, I agree that using univariate correlations of population databases should not be used to infer causality, when one adheres to the reductionist philosophy of nutritional biology and/or when one ignores or does not have prior evidence of biological plausibility beforehand. In this case, these correlations can only be used to generate hypotheses for further investigation, that is, to establish biological plausibility. If in contrast, we start with explanatory models that represent the inherent complexity of nutrition and is accompanied by biological plausibility, then it is fair to look for supportive evidence among a collection of correlations, especially when we examine these correlations from multiple biological perspectives. The difference between these two approaches is a matter of order. If we go from simple correlations (without biological plausibility) to inference of causality, this should lead to the generation of hypotheses. If we already have biologically plausible models (of sufficient complexity), then we can search for supportive correlations that are appropriately qualified.
One final note: in the case of The China Study book, the theme and the conclusions expressed therein arise from substantially more than the correlations available in the China project database—only one of eighteen chapters in the book! The book’s conclusions were obtained by sequentially developing biologically plausible models, then testing for possible affirmation in a human database and finally concluding with exploration of key research done by others. The final validity of any hypothesis on diet and health, however it may be scientifically explored and interpreted, is its ability to predict outcomes. On this point (now worth an entire new book itself), the reports of the health benefits of the nutrition provided by a whole foods, plant-based diet are astounding.
Prad you are beyond wonderful:) Thank you for doing this! Is the site ever going to be back up? do you know what the restrategizing efforts are? I hope they will make this material widely available elsewhere if not on the old site.
and you are very kind jenna! thank you for bringing this to our attention.
from what nelson said, i do not think that the site will return - it only showed up in of june, 2010 as a trial initiative to get plant-based diets promoted along the lines of campbell's discoveries. (some of the 'cheerleaders' on minger's blog like to think that it was developed as a response to her efforts, but i think that fantasy should be left to those who have enough time on their hands to reshape the universe in their own minds :D ).
having acquired some preliminary 'data', my guess is that they have a better idea now as to the possible directions they feel are efficient ways in which to proceed. several weeks ago colin told us that there are some very interesting opportunities that they are exploring. i do not know any details, but he was quite enthused by these. so it is quite likely that they have shifted their energies to prepare for these rather than spend it on developing the site (which was after all little more than a collection of blog entries).
i completely agree with you that the material should become widely available. to some extent, that is going to be an undertaking for our group. colin can put great stuff on his site: http://tcolincampbell.org/
but it's people like those in this group who need to spread the word of its existence.
one (of several) ways this will be done is through the delightful blog bandits. looks like fun when they get the go ahead!
btw, you and everyone else here should know that both nelson and colin are most appreciative of the work that is being done in this group by everyone. they have said so on several occasions.
My family is very grateful for Colin and Nelson Campbell's works. The China Study (by Dr. C Campbell) and The Food Revolution (by John Robbins) were the frist nutrition books I read 7 years ago. They have changed our life for more harmonious and healthier way of living.
Peace & Love, Chang-yu