Broccoli, cancer, and evaluation of risk

Oct 25 2010 Published by under Medicine, Uncategorized

We have some learning to do today, thanks friend of the blog, becca.  The other day, I took issue with a press release published on another website.  It was titled, Discovery may help scientists boost broccoli's cancer-fighting power, which I found to by hyperbolic and deceptive.  The actual study being reported regarded the ability of certain compounds found in cruciferous vegetables such as broccoli to be absorbed from the cecums of rats.  I dismissed the entire piece as being unsupportive of its ambitious headline.

Becca took me to task for being too dismissive:

This is a paragraph from a review article (Keck and Finley, 2004) the manuscript cites:

“Epidemiologic studies have demonstrated inverse associations between crucifer intake and the incidence of lung, pancreas, bladder, prostate, thyroid, skin, stomach, and colon cancer.3 Prospective dietary assessment of 628 men diagnosed with prostate cancer found that increasing crucifer intake from 1 to 3 or more servings per week resulted in a 41% decreased apparent risk.7 A 10-year cohort study of 47,909 men reported that increased crucifer intake, but not fruits and other vegetables, was associated with decreased risk for bladder cancer (relative risk = 0.49, 95% confidence interval = 0.32-0.75, P = .008).6″

Those numbers are simply quite solid evidence, in the context of epidemiology. Is sulphoraphane the only compound in crucifers that is important? Of course not. But this epidemiology *combined* with the cell studies you so blithely write off strongly suggest that the long term goal of the scientist you take issue with “to increase bioavailability of sulphoraphane” is, in fact, a valid pursuit.

Reading and understanding the medical literature is not an art but a skill, one that must be learned.  This learning never ends.  When I run into studies I'm not sure I understand, I can run questions by my colleagues both online and in real life.  For me, it's always work, and I'm happy to be told when I'm wrong. Medical literature can be very different from other scientific literature, as it often focuses on risk, and reported measurements of risk can be quite deceiving (you may have to copy and paste the link address into a search engine).  We also have to look at studies in the context of other studies evaluating similar questions.  Because the results of medical studies often drive changes in practice that affect millions of people, we have to pay close attention to what risk and risk reduction really mean.

For this exercise, we'll focus on the two main assertions quoted by becca (the assertions are from a review article published in a somewhat questionable journal, so separating theses assertions from folklore is particularly important).

Cruciferous Vegetables and Prostate Cancer (Odds ratios are confusing)

Study design is important.  The type of study helps determine how association between two variables can be expressed.

The cited study is a retropective case-control study.  This means that a group of patients with prostate cancer were compared to a group of similar men who did not have known prostate cancer, and they were asked to look back in time and report their intake of cruciferous vegetables over the last five years.  This sort of study is vulnerable to recall bias, in which respondents' memories may not accurately reflect the truth.

Looking at the numbers from Cohen study, comparing the  "most cruciferous eaters" and the "least cruciferous eaters" there is an (adjusted) odds ratio for prostate cancer of  0.59.  You could say that the "broccoli" group had a 41%  decreased odds of having prostate cancer compared to the broccoli-avoiders.  But odds ratios are a tricky statistic and aren't intuitive. For rare diseases, odds ratios are comparable to "relative risk", a more intuitive number.  But for common diseases (and prostate cancer is relatively common), an odds ratio can be deceptive.  That's one of the many reasons a prospective cohort study is more useful in this case, and such a study has been done.

The study cited below by Giovannucci took a sample of tens of thousands of males who were keeping track of their eating habits and at the end of the study period compared the intake records of those who did or did not have prostate cancer.  This significantly stronger study found no significant association between cruciferous vegetable consumption and the risk of developing prostate cancer (although some of the subgroup analyses were tended toward interesting).

Bladder cancer (relative risk and number needed to treat)

In the Michaud study, comparing men who ate the most cruciferous veggies to those who ate the fewest, there was, as stated, a "relative risk" of 0.49.  What does this mean?  It means an absolute difference in risk  for bladder cancer between the two groups of 0.038%.  It also means that to prevent one cancer (number needed to treat) would require 2622 person-days of high-cruciferous diet.  The initial 49% relative risk sounds big, but in real cases, it's not a terribly significant number.

Compared to the prostate data, though, there is evidence from this and other prospective studies that consuming large amounts of cruciferous vegetables may have a small protective effect against the development of bladder cancers.

These subtleties are difficult, and definitely not sexy.  But they are closer to reality.  While it would be reasonable for me to tell patients that the sum of available data indicate that a diet higher in fruits and vegetables is probably healthier than a high-calorie, meat-based diet, there are not sufficient data for me to "prescribe" a high-cauliflower diet to prevent bladder or prostate cancer.  They certainly don't allow us to assume that "broccoli has cancer-fighting power" for us to "boost".  None of the studies looked at the specific use of any compound, just the use of vegetables.

It takes a long time for basic science to move into the clinic---for good reason.


Keck, A., & Finley, J. (2004). Cruciferous Vegetables: Cancer Protective Mechanisms of Glucosinolate Hydrolysis Products and Selenium Integrative Cancer Therapies, 3 (1), 5-12 DOI: 10.1177/1534735403261831

Cohen, J. (2000). Fruit and Vegetable Intakes and Prostate Cancer Risk Journal of the National Cancer Institute, 92 (1), 61-68 DOI: 10.1093/jnci/92.1.61

Giovannucci E, Rimm EB, Liu Y, Stampfer MJ, & Willett WC (2003). A prospective study of cruciferous vegetables and prostate cancer. Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology, 12 (12), 1403-9 PMID: 14693729

Michaud, D., Spiegelman, D., Clinton, S., Rimm, E., Willett, W., & Giovannucci, E. (1999). Fruit and Vegetable Intake and Incidence of Bladder Cancer in a Male Prospective Cohort JNCI Journal of the National Cancer Institute, 91 (7), 605-613 DOI: 10.1093/jnci/91.7.605

Zeegers MP, Goldbohm RA, & van den Brandt PA (2001). Consumption of vegetables and fruits and urothelial cancer incidence: a prospective study. Cancer epidemiology, biomarkers & prevention : a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology, 10 (11), 1121-8 PMID: 11700259

5 responses so far

  • ecologist says:

    Interesting post. The link to which one is sent from

    "and reported measurements of risk can be quite deceiving"

    doesn't really work (it goes to a log-in page for Pharmacy Practice News, but doesn't show the title or reference of the article). I'm interested to learn more; could you give the citation in question?

    In fact, do you have a suggestion of a good text or review of the use and interpretation of risk measurements in clinical studies? I ask because similar issues arise in ecology, but are not well understood.


  • Mark Liberman over at language log has a number of posts talking about using odds ratios in non-technical writing (e.g., the link in my name).

    It might pay to trawl through the archives - Professor Liberman has written quite a number of posts about how to report on group differences when intra-group variation is large.

    As large intra-group variation is common in ecology, it might be a useful perspective - especially if your work has management implications or you plan to do a lot of outreach, so you need to understand not just how to do things properly, but how to talk about them in English so your audience will draw the right conclusions.

    • PalMD says:

      Odds ratios can be very deceptive, depending especially on the frequency of the variable in question. There is a way to "guess" at number needed to treat/harm from an OR using PEER, but I don't know how reliable that is.

      • ecologist says:

        "Number needed to treat" is an interesting concept. I suspect that there's an contrast of population-level and individual-level analysis involved. But maybe not.

    • ecologist says:

      Thanks; that's a helpful link.