June 27, 2002
revised June 28, 2002
Vegans say no to extreme "not milk" spin
I have no idea what you're talking about.
On January 29, 2002, the dairy industry’s most belligerent critic became the subject of a scathing public attack.
Three months earlier, the self-styled "Notmilkman" Robert Cohen, author of MILK: The Deadly Poison and a popular stump speaker on the evils of dairy, had denounced a headline-grabbing study that suggested milk prevents breast cancer as "the fraudulent study of the century." Now the New Jersey founder and executive director of the Dairy Education Board was reaping the backlash. But it wasn’t the study’s authors or the evil minions of the milk industry who were rushing to discredit Cohen.
It was his fellow vegans.
"There is no need," wrote Stephen Walsh, Ph.D., in the open letter to Robert Cohen that spearheaded the attack, "to exaggerate or invent in order to show that the dairy industry, which causes much animal suffering and environmental damage, is at best unnecessary for human health and at worst harmful. Distorting to overstate this case only serves to weaken the credibility of all vegan advocates, and ultimately to weaken veganism."
Walsh is Vice Chair of the UK Vegan Society, the oldest vegan organization in the world. It actually coined the term "vegan" at its inception in 1944. A systems engineer and part-time lecturer at Imperial College of Science, Technology and Medicine, Walsh has been praised by leading vegan nutritional experts like Brenda Davis, R.D, and Vesanto Melina, M.Sc., R.D. (co-authors of Becoming Vegetarian and Becoming Vegan), and Jack Norris, R.D. (President of Vegan Outreach), for his brilliance as a medical scholar. As the Vegan Society's official Spokesperson on Nutrition and Health, Walsh has helped unite the vegan community (with some exceptions, like Robert Cohen) in underscoring the need to obtain vitamin B12 from fortified foods or supplements to ensure long-term health. He has dispassionately weighed the pros and cons of dairy in academic-grade "briefing papers." Recently Walsh wrote a formal letter on behalf of the Vegan Society and the International Vegetarian Union to the World Health Organization (WHO) and the United Nations Food and Agriculture Organization (FAO) scientifically arguing the case that the WHO/FAO's important draft report on "Diet, nutrition and the prevention of chronic diseases" should encourage increased consumption of omega-3 fatty acids from plant sources moreso than from fish.
"Today," Cohen wrote, "I present you with an amazing story of scientific fraud and deceit. This lie made every major American newspaper and television news program." As far as Stephen Walsh was concerned, it was Cohen who was doing the lying.
In contrast, Cohen's writings are compromised by a systematically distorted, if not deceitful presentation of the evidence (scientific and otherwise) about the enemy he has built a career out of bashing. (For several examples, see "Selected Articles about Robert Cohen" on this page.) Lately, from the bully pulpit of his "notmilk" mailing list, and behind the scenes, Cohen has turned with increasingly McCarthyesque venom and imaginativeness to smearing leading vegans like Walsh, Jeff Nelson, John Robbins, and Jack Norris who disagree with him, while evading the substance of their criticisms.
Unfortunately, most people who are drawn to Robert Cohen's "message" lack the scientific background to recognize his misinformation. Couched in an aura of scientific rigor, his anti-dairy, anti-corporate revelations sound "right" – no matter if they often happen to be factually wrong and, in some cases, dangerous (as in his aggressive rejection of vitamin B12 food fortification or supplementation for vegans).
Cohen's pseudo-exposé of "the fraud of the century" was, therefore, nothing new. But it was to become a glaring exposé of Cohen himself – of his untruthfulness and, less forgivably, of his refusal to admit and correct his errors. A stickler for honest, evidence-based advocacy, Stephen Walsh was Cohen's reality check – the "rigorous peer reviewer" (to use Cohen's own rhetoric) – that had finally came home to roost. But instead of dealing with Walsh's peer review, Cohen ignored him. And then he smeared him.
Conflict between Walsh and Cohen had, in fact, been brewing and flaring for months when the Norwegian cancer study was published in the September 15, 2001 issue of the International Journal of Cancer. (For an account of that conflict by Jeff Nelson, see "Cohen and Dr. Walsh" on this page. For Robert Cohen’s version of events, see "True Evil Is Exposed" on this page.) Conducted by Anette Hjartåker and her associates of the Universities of Oslo and Tromso, the subjects of the study were 48,844 premenopausal women randomly invited to enter the mammoth Norwegian Women and Cancer Study (NOWAC). Over an average follow-up period of 6.2 years, the study found, women with a childhood and recent adulthood history of drinking lots of milk were about half as likely to develop breast cancer as those who drank little or none. Moderate milk drinkers had a roughly 20 percent reduced risk.
The dairy industry (predictably) milked the study. And the media, for its part, happily lapped up the good news about a popular, but beleaguered beverage. Little or no mention was made in press releases, articles, and soundbites of the fact that previous research on milk and breast cancer was so contradictory the Norwegian researchers themselves thought it likely "that any association [pro or con] between milk consumption and breast cancer is not a strong one."
But for Robert Cohen, whose public image has been built on the motto that milk is "a deadly poison" and that any study that suggests otherwise must either be the product of incompetence or fraud,it was war.
On October 29, in a column titled "Fooling Most of the People Most of the Time," published by Cohen on his website and mailing list, Cohen lit into the study, dubbing it, in all caps, "THE FRAUDULENT STUDY OF THE CENTURY."
"Today," Cohen wrote to an audience of thousands (his mailing list alone boasted a membership of roughly 4500 at the time), "I present you with an amazing story of scientific fraud and deceit. This lie made every major American newspaper and television news program."
As far as Stephen Walsh was concerned, it was Cohen who was doing the lying. On December 21, 2001, he emailed Cohen, exposing point-by-point the "glaring misstatements" in his column. It had been five weeks since Walsh had sent Cohen an accurate review of the Norwegian study (subsequently published here and here), but the diatribe on Cohen's website was still unchanged, and Walsh was now giving the Notmilkman an ultimatum:
If you do not correct the nonsense in the tomfoolery article by January 21, 2002, and apologise to the authors of the study you unjustifiably accused of fraud, this letter will become an open letter to the vegan community.
If you are prepared to try to work within the constraints of truth I would be happy to help you produce material based on sound science.
I do not like your tone, or your agenda.
I hope that nothing is lost in the translation when I tell you to go and have carnal relations with yourself.
In Jeff Nelson, Walsh found a sympathetic ear. As Nelson would detail a few weeks later in the first of several VegSource articles debunking Robert Cohen’s reputation as the Ralph Nader of vegan activism, Cohen had been privately alienating person after person, most recently Nelson’s friend John Robbins, author of the seminal classic of contemporary veganism, Diet for a New America.
Nelson agreed to publish Walsh’s open letter. But first he emailed Cohen.
"I'd like to get a response from you, if you'd like, to publish with this when I run it," Nelson wrote. "I don't want to take sides here, I'm not an expert. But I know that I, and I think a lot of our audience, would enjoy seeing a dialog on this issue. If Walsh is all wet, now is a good time to point out exactly where. On the other hand, if he raises some valid points, on consideration you might want to respond and make some fixes on your article."
Cohen flatly refused, accusing Nelson and Walsh of having an anti-Robert Cohen "agenda."
Weeks later, I also sought to engage Robert Cohen in a forthright dialogue on the charges levelled in "Fooling Most of the People Most of the Time," but to no avail. As I discuss in "Encountering Cohen," the Notmilkman was as evasive and, ultimately, hostile as he had been with Stephen Walsh.
Putting Cohen’s Critique to the Test
In his open letter to Robert Cohen, Stephen Walsh concisely explained the reasons for his rejection of the former’s "melodramatic" (as Walsh understatedly put it) charges against Anette Hjartåker and her associates. Walsh presented his case in a manner that should have been perfectly understandable to Cohen, who styles himself as a savvy former research scientist and probing, Columbo-like critic of biomedical research. Cohen could have accepted Walsh’s criticisms, or taken issue with them. He did neither. Instead, in "True Evil Is Exposed" (archived on this page), he exalted himself as a Gandhi-like martyr while smearing Walsh as an industry-sponsored "infiltrator" and would-be "destroyer" of the vegan movement, with Nelson his unwitting dupe.
For some of Robert Cohen’s followers and others unacquainted with the details of the breast cancer study, Walsh’s succinct open letter may leave too much to the imagination. It may even seem like one man’s "interpretation" against another’s. As a vegan who shares Walsh’s frustration that someone who claims to speak for our cause speaks with a forked tongue, I would like to try and pick up where Walsh left off and fill in the blanks, so there can be no room for doubt.
The following discussion will, therefore, be quite technical in places. I welcome feedback from readers who may find parts difficult to understand, so I can improve the article in any future revisions.
Now to the study.
Based on what Cohen claimed to be a truly critical analysis of the full published paper by Anette Hjartåker and her associates (which, as Cohen correctly pointed out, reporters seldom read), Cohen charged the authors with four counts of bias, gross incompetence, and even fraud.
He began by insinuating important data had been swept under the rug:
1) 317 of the 48,844 women in the study got breast cancer (six tenths of one percent), but the study actually began with 57,664 women. Why were the data from 8820 women eliminated? It turns out that 986 of those women had cancer too (11%). What does that indicate regarding the entire study?Cohen left his rhetorical question dangling in the air, and proceeded to his next charge. But for anyone familiar with epidemiological research (the major source of scientific information about diet and human disease), the unspoken answer was a no-brainer, and nothing to get the least bit suspicious about.
The study is what’s known in epidemiology as a prospective cohort study. By design, such studies follow a sample, or "cohort," of people randomly selected from a given "population" (premenopausal Norwegian women, in this case) over an extended period of time ("prospective") to see who gets sick and why. "Sick" is typically (for practical and economic reasons) defined at the outset of the study by one or, at most, a few criteria: heart attack, stroke, breast cancer . . . And the "why" is also circumscribed by the range of data the researchers have collected about the cohort at the "baseline" of the study (and sometimes one or more times thereafter).
You’ve undoubtedly heard of many prospective studies without necessarilly knowing that’s what they are – studies like the famous Framingham Heart Study (now in its 55th year) which has contributed so much to our knowledge about the causes of cardiovascular disease, or the Harvard-based Nurses' Health Study which has cast an unusually wide net over its 26 years (including the Nurses' Health Study II, begun in 1989), correlating baseline dietary and other characteristics of nearly 250,000 nurses to everything from bone fractures and cancer to heart attacks and strokes.
The Norwegian Women and Cancer Study, or "NOWAC," was exactly this kind of investigation. Its subjects (the "cohort") were a randomly selected sample of women who according to the National Central Person Register of Norway had been born between 1943 and 1957. In 1991 and 1992, 100,000 such women had been mailed a questionnaire and a letter by Statistics Norway inviting them to enter the study in return for allowing the researchers to link to their records in Norway’s mandatory cancer registry. 57,664 women returned the questionnaire with informed consent.
Because NOWAC was "prospecting" for all kinds of cancer, women whose linked records showed they already had cancer were automatically disqualified. Such is the nature of prospective research. As Hjartåker herself explained to me in an email: "Cohort studies always only include subjects that are free from the disease under investigation. In our situation, free from cancer."
As you’ll recall, Cohen had asked: "Why were the data from 8820 women eliminated? It turns out that 986 of those women had cancer too (11%). What does that indicate regarding the entire study?"
It indicates (as Walsh had futilely explained to Cohen that fall) that the study was properly conducted. The 986 women who already had cancer were ineligible for the study. Had they been included, the researchers would have been incompetent.
But what of the rest of the 8820 women whose "data" was "eliminated"? Again, in the very same six-page paper by Hjartåker et al. that Cohen claimed to have "read and analyzed . . . on a recent plane trip to Chicago" diligently enough to denounce it as "an amazing story of scientific fraud and deceit," the authors gave a full accounting.
Because they had set out to study breast cancer in premenopausal women, Hjartåker et al. eliminated 3851 women who were either postmenopausal when they returned the questionnaire or who turned fifty during the follow-up period, "a dividing line for menopausal status . . . based on data from an older sub-cohort of NOWAC."
A few hundred more women were eliminated for routine reasons, like dying before the study began or being lost to follow-up due to emigration. Finally, 101 of the women simply didn’t answer any of the milk questions and another 3,694 had been given no milk questions to answer in the first place because 6000 of the 100,000 questionnaires had includedno dietary questions at all. There was no skullduggery here: it was part of a sub-experiment to see how the length of the questionnaire would affect response rate.
So in answer to Cohen’s nudge nudge, wink wink question, "What does that indicate regarding the entire study?" the reply must be: "It was conducted properly."
Cohen continued to build suspicion with his second criticism:
"How much milk did you drink as a child each day?"
Even the authors recognize how poorly they designed this so-called study. In the discussion section (page 891), they write:
"Our questionaire [sic – Cohen’s transcription error] included only a single question on childhood milk consumption... we do not know how well the question reveals real differences...although no significant association between childhood milk consumption and breast cancer incidence was found in our study, one may speculate on a negative association."
IS THAT REAL SCIENCE OR REAL BIAS?
Unlike Cohen, I find it reasonable that the designers of the questionnaire (evidently not Hjartåker et al.) didn’t expect the women, now in their thirties and forties, to accurately recall their childhood milk consumption in finer detail. Hjartåker and associates do note that almost all the milk consumed in Norway during the 1950s, when most of the women were growing up, was whole. They candidly question whether the multiple choice categories were set too high and wonder, as researchers always do, how much stock they can put in this vicarious test of long-term recall. But when they venture to "speculate on a negative association" (i.e., that drinking more milk means less cancer), it is in fact "real science," not "real bias."
What Cohen doesn’t say is that there was a moderate trend for women who later developed breast cancer to have initially recalled drinking less milk as children. For example, breast cancer was about 25% less common among the women who drank 1 to 3 glasses a day than among those who drank none and about 40% less common in the highest intake group. But this trend was far too close to being a chance difference to be considered statistically significant, i.e., unlikely to be due to coincidence. However, among the youngest women in the study (34-39 years old at baseline), the trend was highly significant, greater than a thousand to one against chance. Had the researchers not "speculate[d] on a negative association," they would have shown bias against milk.
The baseline questionnaire asked in much greater detail about the women’s recent (past year) levels of milk consumption. It enquired separately about whole milk, low fat, and skim, and in each case, "nine different answering categories were given, ranging from ‘almost never’ to ‘6-10 glasses per day.’"
Here, the researchers again found a trend in favour of milk. There were as many as 44% fewer breast cancer cases in the highest intake group. The odds against this being a chance effect were about 8 to 1. Those may sound like good gambling odds. But to a scientist, the odds against chance must equal or exceed 20 to 1 to be considered statistically significant. There are good reasons for such a conservative approach. For example, in a study like NOWAC where data on 28 foods were available, one would mathematically expect one or two foods to appear to be significantly related (at 20 to 1 odds) to breast cancer by chance alone.
The researchers next decided to see what kind
of results they would get if all the data they had on the women’s childhood
and adulthood milk consumption were combined. To do this, they had
to create new milk intake categories into which to pour the separate childhood
and adulthood answers. They used the following formula:
But as scientists love to say, correlation does not equal causation. Correlative data like these linking milk consumption to reduced breast cancer risk are always shadowed by doubts about whether even a highly statistically significant correlation (or "association," to use a common synonym) is due to the identified cause (in this case, milk) or to some related one lurking in the shadows. For example, what if girls and women who drink more milk tend to eat better, drink less alcohol, and exercise more, and that’s why they get less breast cancer? A good study anticipates such questions and collects the data to answer them. NOWAC was no exception. Indeed, Hjartåker et al. had already performed statistical analyses to "adjust" for possible "confounding" by these and several other well-established risk factors for breast cancer. The trends reported above are what emerged after these adjustments had been made.
But at least one potential confounder wasn't included: smoking.
A recent meta-analysis (a pooling and averaging of data from many similarly designed studies) of studies of smoking and breast cancer found a positive association between the two (more smoking = more breast cancer). This was strongest for premenopausal breast cancer (as in the NOWAC study), for which smoking increased the risk by 21%. Similarly, in the Nurses’ Health Study, women who began smoking before age 17 had a 20% increased risk of breast cancer. Recently, another U.S. study found that white teenage girls who smoke were 26% less likely to drink milk, and that the more a girl smoked the less milk she tended to drink. In one conflicting survey of Swiss women, smokers tended to drink more milk. Taking these studies as a whole, it’s reasonable to speculate that milk’s protective effect in the NOWAC study might at least partially have been the result of females who drink lots of milk being less likely to smoke. In the last sentence of their paper, Hjartåker et al. do indeed acknowledge that "unmeasured lifestyle factors related to milk consumption" could be responsible for the milk effect.
There was another reason to question the milk effect. In order to make it statistically significant, the researchers had to repackage their data.
Scientists tend to regard the results of such post hoc measures warily as the (potentially) ill-gotten gains of "fishing expeditions" for statistical significance. Nonetheless, combining the study’s only two data pools into one was hardly a case of fishing far from the shore – it was an obvious next step.
But by far the most important caveat to the NOWAC study were all the other epidemiologic studies on milk and breast cancer. Cohen doesn’t mention these. But Hjartåker et al. – as researchers must – certainly do.
Just a few years earlier, another large prospective study from Norway had yielded results diametrically opposite to NOWAC: breast cancer was nearly three times as common in women who drank three glasses or more of whole milk per day compared to those who drank five ounces or less. In five other "case-control" studies (a scientifically weaker design), milk again seemed to promote breast cancer.
At this point, some anti-dairy activists would quit while they’re ahead. Unfortunately, still other studies have "sided" with milk. Like NOWAC, another prospective study – this time from Finland – found 50% fewer cases of breast cancer in the highest vs the lowest "tertile" (third) of milk consumption. A protective effect for milk also turned up in four case-control studies. Finally, in another four prospective and five case-control studies, no significant association at all was found between milk and breast cancer. Despite at least 21 epidemiologic studies, we’re still none the wiser. (Lest this make you feel cynical about the value of such research, compare this to prostate cancer where 9 of 14 prospective and case-control studies have implicated milk as a cause, and none have suggested it's protective.)
Given the remarkable inconsistency of the research, Hjartåker and her colleagues could not help but conclude that "the contradicting results may indicate that any association between milk consumption and breast cancer is not a strong one."
Some day, we may figure out whether milk – skim, low-fat, and/or whole – is indeed a breast cancer threat to women, perhaps depending on other dietary, lifestyle, environmental, or even genetic factors. And we may also find that milk is protective for some women. But for now, if you want firm answers, look to milk marketting boards, psychic hotlines, and Notmilkmen.
Evidently feeling the need to counter a shrill
pro-dairy soundbite with a 200-point headline, Cohen next loped clumsily
at the study's jugular with a rusty butcher knife:
Based upon population statistics supplied by the authors, the expectation of breast cancers for low milk consuming females was 156 cases out of 311. The actual number of cases was only 42.
The expected number of cases of breast cancer for the moderate and high milk consumption group was 155 cases. The actual number of cases of breast cancer for the milk drinkers was 269.
In other words, the authors mis-read their own data.
Women who drank a lot
of milk as children developed more cases of breast cancer than notmilk
users. How much more? A factor of 640%!
The data Cohen describes can be confusing for a nontechnical reader, and even a scientist would be at a loss to interpret it without actually seeing the original paper. I have, as have Stephen Walsh and Stephen Kaufman, M.D., medical director of the Christian Vegetarian Association and an assistant clinical professor at Case Western Reserve University School of Medicine and Northeast Ohio Universities College of Medicine. Based on our own careful reading of the original paper, on our correspondence with Anette Hjartåker, and on our abortive attempts to squeeze simple answers from Robert Cohen, we all agree that Cohen’s figures are not "based upon population statistics supplied by the authors." They are pulled from thin air. Hjartåker and her associates did not "mis-read [sic] their own data." Robert Cohen mis-read their data – or made it up.
Let's examine Cohen's claims in detail. He
Had Robert Cohen made these calculations he would have known that no more than 16.1% of the 48,844 women in the study could have been in the "low" group. Consequently, the "expected" number of women with breast cancer in this group would also be no more than 16.1% of the 311 cases (of breast cancer) who were included in the age-adjusted statistical model in Table IV upon which Cohen based his figures. This works out to 50 expected cases of cancer. Yet Cohen claimed 156 expected cases, while the actual number of cases reported by Hjartåker et al. was just 42.
When I queried Hjartåker, she agreed with my assessment that "Cohen must have made a gross and arbitrary exaggeration of the proportion of women in the low milk intake group." But she also noted that she and her associates had used a somewhat more sophisticated calculation formula called Cox regression analysis. "In these analyses," Hjartåker explained, "you don't use number of persons in the different categories, but rather person-years, and calculate RATE RATIOS." Stephen Walsh had done his estimate without the unpublished person-year data, but Hjartåker wrote that his calculations would still be "close enough." It was only after Hjartåker provided the missing data that I could see how remarkably close Walsh had been.
For the "age-adjusted model" of cancer risk in Table IV, cited by Cohen (there had been another model, adjusted for all the other potential confounding variables, with very similar results), the person years provided by Hjartåker for the three milk consumption groups were:
First you divide the number of cases by the number of person years in each group to obtain crude IRRs:
Moderate: 254/242,209 = 0.1049.
High: 15/22,271 = 0.0674.
High IRR: 0.0674/0.13 = 0.518
Earlier I mentioned that Hjartåker’s person-year data proved how close Walsh had been in his calculations. Walsh had deduced that the percentage of women in the groups had been 11%, 82%, and 7%, respectively. When I calculated the percentages using person years, the results were almost identical: 10.9%, 81.6%, and 7.5%. In contrast, Cohen’s figures – based on the "population statistics" he has persistently refused to disclose – were 50% for low and 50% for moderate and high combined.
There is only one way that Cohen could be right and the rest of us wrong: Hjartåker and her associates fabricated the data in their paper and in their correspondence with us, and somehow Cohen has figured this out and knows what their real data is.
Not only is there no conceivable way this could be true without Cohen having triumphantly revealed it by now, but Cohen himself claims he based his figures on the researchers’ own data which they "mis-read" – that is, on the data published in their paper. If Robert Cohen had any rational explanation for how he arrived at these figures – which have caused him to sink deeper and deeper into disrepute – he would have provided it long ago, if not to Stephen Walsh then surely to myself or to Stephen Kaufman, M.D., with whom he had no bad blood at the time we approached him. Nevertheless, I continue to invite Cohen to do so.
Cohen had one more aspersion to cast upon the authors of the NOWAC study before he was done "sham[ing]" them "for their deceit":
discussion by writing (page 892):
"Calcium intake, however,
That conclusion was
based upon another
I wrote about that in great detail:
It’s pitiable that Robert Cohen has such a fanatical "anti-dairy agenda" that even the health benefits of essential nutrients unlucky enough to be guilty by association with his mortal dietary enemy are a threat to be neutralized with charges of "fraud" and "murder."
Stephen Walsh said it well in his open letter to Cohen this January:
If you are prepared to try to work within the constraints of facts, I would be happy to help you produce material based on sound science.
See also these sidebars to "Spinning Out of Control":
and this appendix:
For more on the controversy over milk see "Milk: What is the Deal" at www.aquarianonline.com/Wellness/Milk.html
Copyright © 2002 by Syd