Why African American Men Have Poorer Survival From Prostate Cancer (Even When They Don't)

Andrew J. Vickers, PhD


November 02, 2009

I recently read a paper reporting that, compared with white men, African American men have a statistically significant 6.2% lower chance of being alive 10 years after surgery for prostate cancer. Nonetheless, the paper concluded that "African American men do not have shorter survival from prostate cancer."

I was half expecting the authors to go on and say that Danny DeVito is no shorter than Vin Diesel. Then I remembered that some statistical analysis was involved and, as so often happens, common sense leaves the room as soon as the P values start churning out from the stats software.

The authors conducted both univariate ("single variable") and multivariable analyses. In the univariate analysis, they looked only at the effects of race on prostate cancer deaths and found the 6.2% reduction in survival mentioned above. The multivariable analysis included other variables, such as stage and grade of cancer. The idea was to see whether these other variables explained any apparent association between race and survival.

As a simple example of a multivariable analysis, imagine that we were interested in whether coffee drinking was associated with lung cancer. An increased rate of lung cancer among coffee drinkers might lead us to consider the effects of a confounding variable, such as smoking: if coffee drinkers smoked more, this might create an apparent association between coffee and cancer even if there was no relationship between the two. Multivariable analysis controls for this type of confounding by comparing cancer rates in coffee drinkers and non-drinkers separately for smokers and non-smokers, and then combining the results of the 2 analyses.

In the multivariable analysis of the study on race and prostate cancer survival, after adjusting for tumor characteristics, the authors found that African American and white men had a similar risk for death following radical prostatectomy. In other words, if you took 2 men undergoing surgery for similar tumors -- one white and one African American -- you'd have no reason to guess that the African American man would do worse. The apparent decrease in survival among African American men in the study would be explained by the fact that they presented with more advanced cancers.

The contradiction in the paper -- why African American men had poorer survival even if they didn't -- reflects the different conclusions of the univariate and multivariable analyses. However, it also reflects a fundamental misunderstanding of multivariable statistics. Being African American cannot reduce prostate cancer survival in and of itself -- it is not as if the cancer cells think, "This guy's got a dark face; let's grow as quickly as we can." There has to be some reason. A multivariable analysis helps us examine those reasons, and, in the prostate cancer study, it was trying to answer the question: "Is the reason why African American men are more likely to die from prostate cancer after radical prostatectomy that they have more advanced cancers to start with?" If the answer was "no" -- if race was still associated with survival after controlling for tumor characteristics -- then the authors could have postulated other possible causes; for example, that the biology of prostate cancer is different in African American men from that in white men.

All too many researchers think of P values as giving a simple "yes" or "no": a P value less than .05? Yes, race is associated with survival. A P value of .05 or more? No, race is not associated with survival. This is why the presence of 2 P values can lead us to silly conclusions, such as race not affecting survival even when it does. We need to stop thinking that statistics can provide "yes" and "no" answers and, instead, look at statistics as a tool for scientific understanding. Or to put it another way: we need to stop thinking of statistics in black and white.

Note: The example is given for illustrative purposes only, and I have simplified it down somewhat. As it happens, there are some known biological differences in prostate tumors associated with particular genetic polymorphisms that are more common in African American men. Differences between white and African American men in terms of access to care and choices about screening and treatment can also affect survival.

If you liked this article, you'll love Andrew Vickers' collection of stories on statistics: "What Is a P-Value Anyway?"


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.