Whole (15 page)

The most well-regarded (and therefore best-funded and most common) form of reductionist study design is prospective, meaning that information is recorded in real time, and effects are observed as they occur. In its simplest form, one group of subjects (the experimental group) is given an intervention, while the other group (the control group) is not. The gold standard of reductionist research is a form of prospective experiment known as the randomized controlled trial. The “random” part of the study refers to the way subjects are assigned to either the experimental or control group. The theory here is that random assignment eliminates the effects of potentially confounding variables by evenly distributing them across all groups. If you’re worried about whether being a heavy smoker might influence the results of an intervention, random assignment uses the power of statistics to spread this variable evenly across groups, theoretically making it irrelevant.

Randomly controlled trials often include a double-blind feature, wherein neither the researcher nor the subject knows whether the subject
is receiving the intervention being tested. In a drug trial, for instance, neither would know whether the pill the subject is taking is the actual substance or a lookalike placebo. That way, patients don’t get better just because they think they’re taking a wonder pill,
¹
and researchers don’t subconsciously treat a placebo subject differently than a subject taking the active compound.

Prospective experiments are seen as a “clean” form of study design, because they nail down the details with more precision, and because they minimize the messiness and “noise” of the real world. This allows researchers to isolate the effects of the intervention in which they’re interested. This isolation of a single variable (X) supposedly gives the researcher the right to say, “X causes Y,” where Y is an outcome that occurs after X and does not occur when X is not present.

This is most useful in cases where it makes sense to isolate a single factor, as when we need to assess the safety and effectiveness of a new drug. But even in the case of drug tests, there’s an inherent trade-off between that kind of certainty within a controlled environment and its applicability in the messy, noisy real world. The more perfectly controlled the experiment, the less it resembles reality.

While studying specific chemicals in isolation provides for pretty findings, these research methods cannot provide predictive models for complex interactions with multiple causes and effects—in other words,
life.

Another commonly used research design, regarded as less discriminating by reductionist researchers than the prospective experiment, is the case-control study. The cases—individuals who, for example, have a disease—are compared with the controls—individuals of the same sex, age group, and so forth, who do not have the disease, as researchers look for lifestyle differences between the two groups that could have influenced their different outcomes. Case-control studies typically examine influences that cannot practically or ethically be imposed on people: diets, lifestyle practices, and exposure to toxins are common examples. You wouldn’t force half of the people in your study to eat all their meals at McDonald’s, for example, but you could find people who choose this diet on their own and see what happens to them.

Case-control studies can be retrospective when researchers use previously recorded observations to explain disease outcomes. They can also be prospective, in which cohorts of subjects with different lifestyles and diets are studied to see what will happen to them. Either way, because subjects aren’t randomly assigned to these cohorts, it’s impossible to prove that the differences caused the outcomes. The problem is, people who are alike on one characteristic are probably alike on many others. It’s impossible to tell which characteristic or characteristics were the active agents leading to the varying outcomes. So researchers typically resort to a family of statistical procedures to make this problem go away, called “adjusting for confounding.”

Here’s how statistical adjustment for confounding works. Suppose you are studying the relationship between breast cancer and dietary fat. You start with two groups, one made up of women who have been diagnosed with breast cancer (the cases), and one made up of women who have not been diagnosed with breast cancer (the controls). You question them about their eating habits to figure out if the cases are eating more dietary fat than the controls. But there’s a problem: the women with breast cancer carry a higher percentage of their body weight as fat. Assuming that there is a relationship between dietary fat and body fat to begin with, what’s causing what here? Is the dietary fat causing the breast cancer? Or are the women more prone to obesity also more susceptible to breast cancer?

The more questions we allow ourselves to ask, and the more possible interactions we entertain, the further we plunge into a reductionist nightmare. Maybe these women with breast cancer and a higher percentage of body fat have a genetic predisposition both to obesity and to breast cancer, so therefore we may not have to worry about how much fat women without that same genetic predisposition consume. Maybe there’s some other variable that we haven’t even thought about; perhaps heavier women exercise less, or are more depressed because of societal prejudice, and that’s the factor that leads to breast cancer. Or maybe they’re heavier because they’re depressed, and tend to eat more and exercise less. Or maybe they’re heavier because they are less educated about healthy eating, which sometimes correlates with less access to healthcare, which correlates to low income, which correlates to less access to fresh produce, which correlates to living in neighborhoods with higher concentrations of environmental toxins.

To deal with this uncertainty, reductionists use statistics to mathematically “hold constant” all these potential sources of data pollution and make their effects magically disappear—that is, they compare, in effect, small segments of each group whose confounding variables are nearly the same. Of course, you can do this only to those confounding variables you’re able to think of and then measure in some way. No study has unlimited time or money, so there will always be potentially confounding variables that don’t get neutralized by the statistical magic wand.

But the more we scientists try to disentangle the web of influences around a specific health outcome, the less useful the “results” of a study become. Suppose, in the breast cancer example, we “adjust” for every other influence we can think of, so that the only two variables that remain are rates of breast cancer and obesity. If we then say that obese women seem to get more breast cancer, the prescription to prevent breast cancer immediately collapses into “lose weight.” Any method that purports to take off the pounds then becomes a form of breast cancer prevention. Meal-replacement shakes, low-carb regimens, lemon juice fasts, and all manner of craziness would now be tied to a healthy outcome, regardless of the actual mechanism of the relationship between obesity and breast cancer. Suppose that increased rates of breast cancer and obesity are both functions of highly processed diets with lots of animal products and not enough whole-plant products. For many women who follow this weight-loss regimen, the “get thin by any means to prevent breast cancer” message could translate into diet choices that would increase, not decrease, their cancer risk.

It’s as if you noticed that happy people tend to smile more than unhappy people, so you invented a device that stretched the human face into a smile as a cure for depression. Yes, the smile is a good marker for happiness. Yes, there’s a correlation between smiling and happiness. Yes, it’s possible that reminding yourself to smile more can affect your mood. But isolating the smile and ignoring all other factors that might contribute to happiness and depression is patently ridiculous.

Think these examples sound unbelievable? We’ll talk more in
chapter eleven
about a real-world consequence of this kind of narrowly reductionist research when we look at the hype surrounding dietary supplements. In this hype, researchers have used statistical adjustment to conclude that certain nutrients are not just markers of good health, but the cause of it,
ignoring clusters of factors surrounding those nutrients as if they didn’t matter or even exist. The result of this miscalculation isn’t merely a waste of vitamin-takers’ money; in some cases, the outcomes have been serious illness and even premature death.

The reason wholistic ways of exploring reality come under fire from many contemporary scientists is that they all smack of fuzziness, of imprecision. They don’t narrow cause and effect to the point where everything is airtight, completely repeatable, and measurable to the fifth decimal place, the way reductionist experimental design does.

Reductionism by definition seeks to eliminate all “confounding” factors: any variables that might influence the outcome in addition to the main substance under investigation. But because nutrition is a wholistic phenomenon, it simply doesn’t make any sense to study it as if it were a single variable. Studying nutrition as if it were a single-function pill disregards its complex interactions.

The whole point of wholism is that you can’t tease out one contribution and ignore the rest. Of course body fat, dietary fat, education level, depression, socioeconomic standing, and so many more characteristics are interrelated and interactive with one another and with our bodies’ systems. While statistical adjustments can pretend to wrap up reality into neat little packages, they don’t explain the underlying reality at all.

You can’t study wholistic phenomena solely through reductionist modes of inquiry without sacrificing reality and truth in the process.

At its best, epidemiology draws conclusions from many different types of study design, just as a group of blind elephant scholars pool their findings to increase their understanding of the whole beast. Sadly, however, only reductionist studies are taken seriously and funded generously, so much
so that the entire field of epidemiology is substantially biased in favor of reductionist philosophy. You wouldn’t give an electron microscope to someone studying elephants and expect them to tell you anything about the animals’ personalities or social structures. The only way to find wholistic answers is to allow for the possibility of seeing them.

Reductionist critics argue that the China Study was experimentally weak because it didn’t prove independent effects of single agents or show results applicable for individual people. As I hope I’ve shown in this chapter, this criticism is misguided. We don’t need to know the effects of single agents on health, because this is not the way that nature works. Nutrition has a wholistic effect on health; one that we consistently miss and misinterpret when we focus on isolated nutrients. Our project in China, when evaluated from a wholistic perspective as intended by the study’s design, provided unique evidence on cause-and-effect relationships between diet and disease through highly significant patterns of association between food consumption and health outcomes.

For drug trials, the most informative study is the randomized control trial. But for nutrition, the most informative study design is the wholistic study: one that allows us to see how unimaginably complex interactions can be influenced, and how radiant health can be achieved through simple dietary choices.

Explanations always go in one direction, from the complex to the simple and, in particular, toward what is less distinctly human.

—
T. H. JONES

W
e’ve just looked at how reductionist design leads to reductionist answers and excludes the true nature of biological complexity. Now it’s time to revel in that mind-boggling complexity, specifically when it comes to nutrition.

In this chapter I want to introduce you to an old friend of mine: an enzyme called mixed function oxidase (MFO), which ultimately converted me from a reductionist to a wholist.
¹
Sharing more about the function of enzymes, those amazingly complex and powerful molecules responsible for every chemical reaction that goes on in our bodies, is the best way I can think of to show you the complexity of nutrition’s effect on health—and the inadequacy of the reductionist model of scientific inquiry to address it.

As I mentioned in the book’s introduction, my first official research project as a professor at Virginia Tech back in 1965 was to analyze peanut samples for the presence of the cancer-causing chemical aflatoxin (AF).
²
A product of the mold
Aspergillus flavus
,
³
AF had recently been shown to be a very potent liver carcinogen for laboratory rats.
⁴
On the list of America’s most popular foods, peanuts rank somewhere up there with milk and T-bone steaks. They’re what help keep hands busy at cocktail parties; they’re half of that most beloved of lunchbox sandwiches, the PB&J. So the possibility of a mold-produced carcinogen in peanuts was a dreadful thought. The other troubling aspect of these findings was that the amounts of AF required for liver cancer in rats appeared to be exceptionally low, possibly making AF the most potent chemical carcinogen ever discovered, at least for rats.
⁵