Another cruel breast cancer scare falls flat

There was a press release about a new study... sent out to media before the study was published in the medical journal... more than 500 media outlets reported on the study on the same day and all saying the same thing... Stop me if you’ve heard this before.

This week, people around the world were told about a new study of 65,000 women said to have found another threat for girls and young women who fail to exercise regularly: a higher risk for breast cancer, “which kills 40,000 American women every year.” From the press release, media reported that parents should get their daughters off the couch because women ages 12-35 who exercise regularly can cut their risk for breast cancer by 23%. This study was said to be one more reason to require girls to take PE to stay in shape, said a Mercury News editorial.

Science isn’t marketed through public relations firms, nor does it try to manipulate or scare you. If the media’s lockstep coverage, verbatim from a single press release, wasn’t enough of a clue that this was marketing and “science by press release,” then your next clue was the size of the study. No, the bigger the study doesn’t necessarily mean the findings are more important. In fact, it rarely does. The bigger the number of people studied, the less likely it is to have been an actual clinical trial, the only kind of study that can test a hypothesis and credibly suggest a cause or effective intervention. That’s because good clinical trials are expensive to conduct, while those done in a computer, using numbers and data rather than people and verified clinical evidence, aren’t.

Since we know all studies are not created equal, let’s look at this one. Was it a randomized clinical trial that followed tens of thousands of women from age 12 and found that vigorous daily exercise resulted in fewer cases of breast cancer? No. The study, published in the Journal of the National Cancer Institute, was authored by researchers at Brigham and Women’s Hospital and Harvard Medical School in Boston, which houses the largest and longest-running Rorschach test of epidemiology: The Nurses Health Study. This is a huge quarry of questionnaires gathered since 1976 from more than 120,000 nurses and has been used by its researchers to pick out characteristics in unlimited combinations to find all sorts of correlations and conclude just about anything they set out to find. Well over 500 such computer studies have been published from this database and many of the correlations reported even contradict each other.

Definitions

Study population. For this paper, they first culled down the original 116,608 self-reported questionnaires on lifestyle factors in the Nurses Health Study II cohort (began in 1989). They took only those questionnaires turned in since 1997 among women age 33-51 who had also answered questions about their current exercise levels, as well as retrospectively estimated their activity decades earlier as young girls. This left them with data on 64,777 women. This is our Trojan Number reported in the news.

Cancers. For this paper, they only considered invasive premenopausal breast cancers that had been reported as occurring between 1997 and 2004, for which they also had physical activity data. They further eliminated all in-situ and unconfirmed cases (another 189 cases), leaving 550 cases for their analysis. There were too few (129) cases of invasive postmenopausal breast cancers to include in their analysis. So, this study was actually only looking at 550 cases of premenopausal breast cancer.

Physical activity. For this paper, they defined physical activity only as leisure-time activity and used the average hours per week at each age the women reported and recalled having done, in each of three specific activity categories: strenuous (running, aerobics, lap swimming), moderate (hiking, power walking, bicycling and yard work) and walking to and from school or work. Not only is leisure-time activity a marker for things like socioeconomic status, but retrospective recall data is the weakest type of data to use in research. Beyond the fact that few people can accurately remember what they were doing decades earlier, it suffers from recall bias. For instance, if you believe that your fatness is caused by being too sedentary, you are more likely to believe you must have been too inactive and to report it thusly. The further back the recall data, the less reliable it is. The authors then took this self-reported data and estimated a metabolic equivalent (MET) value based on the intensity, and labeled the activity by METs per hour per week. They then assigned MET values of 7.0, 4.5 and 3.0 to strenuous, moderate and walking categories, respectively.

Despite claims that young women are sloths, all of the women had been active during their teen years and 20s and activity declined in their mid-30s. The women who engaged in the most leisure-time activity were also slightly more likely to not have babies, be taller, use oral contraceptives and to smoke. Their weights and diets and menstrual cycles didn’t significantly differ from the less active women, however.

Computer model. They tossed all of this into their handy dandy computer and tried every possible permutation of activity at each age, and at various cut-offs of age groups, and at every type of activity and intensity, trying to derive correlations between activity and premenopausal breast cancer.

Findings

None of the relative risks were tenable and beyond random chance or statistical error for this type of study. This study was unable to find any credible relationship between exercise and premenopausal breast cancer. All of the relative risks hugged null (RR=1).

We could stop right there, but, if we want to split hairs, there are a few observations that can help reinforce the lack of a credible correlation. Looking at their adjusted relative risks (which adjusted for age, contraceptive use, history of benign breast disease, mother or sister with breast cancer, parity, alcohol, and height — but not socioeconomic status, smoking, dieting history, etc.), there was no consistent dose-relationship. In other words, more exercise didn’t mean lower risks for premenopausal breast cancer.

Among the women engaging in strenuous-type activities, for example, those doing 3 hours a week had lower risks than those doing 4 or more hours; among the walkers, those walking a mere 1 hour a week had lower risks than those doing up to 2.5 hours a week.

This same quirk was seen in most of the different age groups. Among women 23-34 years of age, those doing the lowest levels of activity (<15 MET-h/week) had lower risks than all of the higher activity counts but one. Among women 35 years of age and older, those doing the most exercise had the same risks as those doing the least.

Only when they created an age group of 12-22 years, was any kind of linear correlation produced, although none of the relative risks were tenable and all hugged null. But the women who reported that as teens they’d done the highest levels of exercise (calculated as 72 MET-h/wk) had a mere 7% difference in relative risks from the women who as girls had done about half that (36-47.9 MET-h/wk) — but the difference in actual numbers of cases was only 0.13% versus 0.15%, respectively, not a credible correlation, let alone clinically meaningful. Especially given the questionable quality of the data going back decades.

Nevertheless, the authors said in their conclusion:

In conclusion, these results suggest that consistent physical activity during a woman’s lifetime is associated with decreased breast cancer risk. Unlike many risk factors for breast cancer, physical activity is an exposure that can be modified ed. This association, if found to be causal, has public health implications for prevention.

And if you just read the Abstract, it appeared to suggest even stronger findings, concluding:

High levels of physical activity during ages 12 – 22 years contributed most strongly to the association. Leisure-time physical activity was associated with a reduced risk for premenopausal breast cancer in this cohort. Premenopausal women regularly engaging in high amounts of physical activity during both adolescence and adulthood may derive the most benefit.

Flashback: Correlations are not causations

The correlations found in this study weren’t tenable at all. But even if they’d been huge, they’d still have been just correlations. Remember, correlations derived from data dredges and looking at characteristics among groups of people (epidemiology), no matter how strong or intuitively correct they might seem, cannot provide evidence of a cause, and most certainly not for any treatment we should act on.

Most observational, epidemiological studies are actually overturned and found to be bogus when their hypotheses are tested in randomized, controlled clinical trials. Growing numbers of research experts are so concerned that epidemiological studies are so often wrong, they say they’re next to worthless. As Dr. Stan Young, a statistician at the National Institute of Statistical Sciences in Research Triangle Park, N.C., said: “We spend a lot of money and we could make claims just as valid as a random number generator.”

We all need to be regularly reminded of why this happens. The more correlations you look for in a lump of data, said Dr. Young, the more likely you’ll be able to pull out something statistically significant — just by chance, luck, nothing more. It’s the classic parlor trick of cold reading and getting an audience to believe you can read their minds or are able to communicate with their dead relatives. Ask dozens of questions and eventually you’re bound to come up with something positive.

But our minds also trick us. Few of us realize that coincidences, even huge ones, are not that unusual. Nor do they usually mean anything. Our understanding of probabilities and statistics isn’t so great, though, and we’re easily taken in by coincidences. Instinctively, we want to believe that there must be a reason for all links, rather than the truth being that they’re just random chance.

We also don't understand the laws of dealing with numbers, especially big ones. The law of truly large numbers means that when enough data or people are involved (as in epidemiological studies), "unusual" occurrences become highly usual.

Epidemiology can dredge up all sorts of correlations and it’s inordinately successful at making us worried sick about our health. Nowadays, epidemiology is primarily being misused to link diets and lifestyle behaviors to cancer, diseases, obesity, and death; then twist them around into causation, and make us believe that our health is a matter of personal responsibility. As Dr. Paul R. Marantz of Albert Einstein College of Medicine, New York, wrote in the American Journal of Public Health: “The misleading message that an individual will prevent a particular disease by altering a particular behavior or exposure (and its converse, than an individual will develop a particular disease if such behavior is not changed) has unfortunately been widely conveyed.”

The sad fact of the matter is that beliefs and fears surrounding risk factors defy the fundamental laws of human biology, and bodies that are naturally robust and impervious to small changes in the external world, as Dr. James LeFanu, M.D., wrote in “The Fall of Medicine.” How else could the human species have survived to this point? But epidemiological scares have undermined our confidence in our bodies and have left so many perfectly normal people feeling weak and vulnerable, at risk of falling ill at the slightest exposure, he said. They’ve affected our very concept of what it means to be healthy. Rather than realize that most of us are healthy most of the time and only occasionally get sick and then get better again; it’s become widely believed that healthy young people need regular medical attention and constant diligence to stay healthy.

One perceptive blogger wrote a great comeback about this week’s study, that I’ll refer you to because it was such fun. The scientist author of Big Liberty wrote about our cultural programming of sloth and fat = cancer:

You know what also equal cancer risk? Being tall. Yup, says so right here. They found that the risk of experiencing breast cancer increased by 7% with each 5cm increase in height for post-menopausal women, with a marginally increased risk among pre-menopausal women...

But wait, there’s more from this fun little article: “Paradoxically, however, being obese appears to have a protective effect in women before the menopause.” Since they’re scared of stating this stat (scared of losing their grant money, that is), we don’t know how *much* of a protective effect being obese before menopause has. It’s not good marketing to use comparative statistics on the non-lucrative stats, you know... So it seems like there isn’t some magical prescription to make sure you don’t get cancer...

How does this week’s study fit with what’s been shown to date?

If you hadn’t tossed this study for the bird cage upon reading the headlines — recalling that recent news story trying to scare women that having a single drink a day could give them breast cancer — you might be curious about how its findings fit into the scheme of things.

First of all, despite the scary death claims in the news, tens of thousands of young women are not dying of premenopausal breast cancer each year. Breast cancers, like most cancers, are most related to advancing age. Age is the single greatest risk factor for post-menopausal breast cancer and breast cancer deaths, occurring 34-fold more often among women over age 50 (with most over age 65).

According to the American Cancer Society’s Breast Cancer Facts and Figures, 2005-2006, breast cancer death rates are also dropping and have decreased 2.2% every year since 1990. There is no evidence to support a sudden need for new public health interventions or for panic. Nor is there any evidence that young women need to fear that by not following a certain lifestyle, they could get breast cancer.

In a 2005 issue of the Nurses Health Study Annual Newsletter, Walter Willett, MD, MPH, the principal investigator of the Nurses Health Study II, which is the cohort used for this week’s study, had a very different report on what the Nurses Health Study data had shown: “Being physically active doesn’t seem to offer women much protection against premenopausal breast cancer — and being lean actually seems to increase risk,” he wrote.

His message on the exact same data calmly defused worries that premenopausal breast cancer is a woman’s fault. We’ll give him the last word:

Despite most efforts [to identify a cause or risk factor], premenopausal breast cancer remains something of a mystery, and women understandably feel frustrated when they’ve ‘done everything right’ and still develop this disease. Clearly, the usual advice to exercise and eat well is not enough to prevent breast cancer.

© 2008 Sandy Szwarc

* The principle investigator of the first Nurses Study is Dr. Graham A. Colditz, who originated the “economic costs of obesity” claim, saying obesity cost $39.3 billion in 1986, and has continued to be the lead proponent of this claim. The numbers have grown, as have claims of and death counts attributed to an obesity epidemic. Most “costs of obesity” figures are extremely flawed elaborate creations of computer models.