TCS Daily

Nonsense for the Diet Season

By Sandy Szwarc - January 26, 2005 12:00 AM

After ringing in the New Year, the party's over. The diet season has begun and so do our resolutions to diet and exercise, quit smoking and begin a new life. "Yup, it'll be All Bran all the time in 2005," said food writer Gwyneth Doland.

Just in time to accompany the weight loss advertisements that beset us every year at this time, a new report claims to provide the evidence that dieting and attaining a trim figure are imperative. Women must be "toned and trim," we're told, because no amount of exercise can offset the risks of dying prematurely when overweight. The bearers of those gloomy tidings are researchers from the Harvard School of Public Health, in a report published in the New England Journal Medicine. They claim to have studied 116,564 women for 24 years and found that even exercising 3 1/2 hours a week, fat women have a 91% greater risk of premature death as compared to lean women.

This veritable death threat left women feeling angry and hopeless that their fat was going to kill them regardless of what they did. The latest comprehensive review of the nation's top ten diets by University of Pennsylvania researchers confirmed, yet again, that there is no diet that offers "more than glossy ads and dramatic testimonials when it comes to promising long-term results."

But despite this Harvard study's claim we must shed body fat or else, closer examination reveals that it deserves nothing more than a good belly laugh and to be tossed out with all the fad diets.

Brought to you by the data dredge experts

It's easy for researchers to conclude whatever they set out to find - even results that counter the body of clinical and epidemiological evidence - when they don't actually do a clinical study. Like increasingly more researchers, these Harvard authors instead did a data dredge: mining a database of questionnaires and manipulating selected data looking for possible correlations. They went back to the largest quarry of questionnaires: the Nurses Health Study, headed by co-author JoAnne Manson of Brigham and Women's Hospital in Boston. It's been their source for more than 500 such "studies" to date -- studies which have "found" coincidental correlations that often contradict each other or are altogether preposterous.

In this case, the soundest clinical and epidemiological studies have consistently shown the opposite of what this Harvard study claims: physical activity abrogates any excessive mortality associated with increased body weight and weight itself is irrelevant to long-term health, heart disease, diabetes or premature death from all causes. "The impression ... that everyone who is overweight faces an elevated risk for mortality... is simply not true," said Steven Blair of the Cooper Institute for Aerobic Research in Dallas.

Your first clue that this Harvard study is just another spurious population study is its very first sentence: "We examined the association..."

To arrive at the results they wanted, they selectively picked and chose their data. They didn't use a cumulative measure of weights that had been gathered throughout the study period. Instead, they only used a single weight reported in 1976 and claimed from it to be able to predict the womens' mortality decades later.

Why does this matter -- besides the fact that we have no idea what the women weighed for 24 years? Because, as two of the co-authors of this study reported in a 1999 Annals of Internal Medicine, 54.9% of the women in the Nurses Study database were yo-yo dieters. That means for years they've been dutifully dieting, with the inevitable weight regain before trying another diet, again and again. The result is weight cycling, with most getting heavier with each diet.

Yet in their statistical analysis, the researchers didn't consider the possible adverse effects of dieting or weight cycling, even though they themselves have reported on them. In fact, clinical studies have confirmed for decades that weight cycling of just 5 to 10 pounds increases rates of high blood pressure by up to 400%; the chances of insulin resistance (type 2 diabetes) and heart disease by 200%; appears to harm the immune system, increasing risks for infections and cancer, such as a nearly 400% increase risk for kidney cancer in women; decreases bone density and increases osteoporosis, and increases the build-up of visceral fat around organs and weight gain especially in the upper body. Their own 1999 study found it also increases gall bladder disease by 138%. And, as the January issue of the Journal of Human Hypertension reports, "obese" women are most vulnerable to the harmful effects of even short term weight loss and weight cycling, with their risks of high blood pressure increasing 300% and 574% respectively.

Even though the researchers knew that most of these women were dieters, they also didn't factor in dieting methods, including the use of weight loss drugs and other risky measures, which have been shown to impart serious dangers for cardiovascular disease, stroke and cancer.

Statistical Hocus Pocus

Your second clue that this study is another untenable data dredge: they said they employed various risk models and "calculated the attributable risk [of fatness and sedentary behavior] to estimate the percentage of premature deaths in our cohort that, theoretically, would not have occurred if all women [had been thin exercisers], assuming a causal relationship between risk factors and mortality."

All of this is a fancy way of saying that they guessed, while they made the most bogus assumption of all: that correlation proves causation.

Consumers, and all too many healthcare professionals, are regularly misled as to the meaning of "risk factors." When researchers use that term, all it means is that they've found a correlation between two variables. For example, baldness is a "risk factor" for heart disease and wearing a bra is a "risk factor" for breast cancer. But a risk factor is not the same as a cause, and a risk factor doesn't actually increase the risk for a disease. Aging men (who normally have hair loss) are more likely to have heart disease and women (who typically wear bras) are more likely to get breast cancer, but hair transplants and going braless doesn't prevent either disease. So, while data dredgers look for more health problems to link to obesity, it doesn't mean that fatness causes any of them, no matter how hard they try to convince us of that.

Like all data dredges, to make their correlations sound more dramatic, they state them as "relative risks" rather than give us actual numbers. Relative risks are fractions that compare two groups, say their incidences of a certain disease. So a relative risk of 1 means there's no difference between the groups. If in a group of people there's one case of some disease, and in another group there's two cases, that's a relative risk of 2 for the second group -- an impressive-sounding 100% increase -- but in real life reflects only one additional case.

Because lots of correlations are due to chance, statistical bias, random errors and other factors not accounted for, in these types of studies relative risks aren't even considered credible or worth noticing unless they're at least 2 or preferably 3. Even then, no matter how significant the relationship between two variables, such as women wearing bras, it doesn't ever mean it's meaningful or useful. In looking at the study's actual data, not the spin, it found death from all causes were 69% higher (a relative risk of 1.69) among those who were sedentary and "obese," as compared to those with "normal" weights exercising the most. This is an untenable figure.

The real findings

A significant finding that, not surprisingly, didn't make the news was that age-adjusted death rates were higher among those with BMIs under 21 (what the authors claim is ideal), than those with heavier BMIs and remained so all the way up to 27 (well into the "overweight" category). Risks for deaths were then infinitesimally different among the weights and didn't even reach a relative risk of 2 until BMIs were greater than 40, the upper extreme of obesity, which comprises a tiny 4% of the general population and accounted for only 170 deaths in this study.

That's your Trojan number: The real number of women that this study's ominous conclusions were based upon was 170, not the 116,564 supposedly studied.

Like all data dredges, the unearthed correlations are often nonsensical and defy clinical evidence. If improved health outcomes can be attributed to something such as exercise, in a sound study we would see a dose effect: the more exercise the greater the benefit, to a point. But this study, while claiming exercise was still important, didn't find that. Cancer deaths, for instance, were 13% lower among the obese women doing less exercise (1 to 3.4 hours a week) than those doing the most (3.5 hours a week). Go figure.

But here's the juiciest example of why this study is just more statistical shenanigans and should be taken with a grain of salt. Of course this finding didn't make the news, either. Their own data "found" that women who smoke or once smoked, have significantly lower chances of dying prematurely than those who've never smoked! Yet, the researchers certainly aren't out there trying to terrify nonsmoking women that they have as much as a two-times greater risk of dying, even though that's what their study found!

I pray no one believes that smoking will help them live longer, because that's just as ridiculous as this study's claim that you can't be fit at whatever weight you might be.



TCS Daily Archives