Fish oil and prostate cancer

A week or so ago the media was telling us about a new study linking omega-3 and prostate cancer. The local news folks were suggesting that we might all want to do without those fish oil capsules. The story on CCN was headlined: Hold the salmon: omega-3 fatty acids linked to higher risk of cancer. The reports were based on the paper “Plasma Phospholipid Fatty Acids and Prostate Cancer Risk in the SELECT Trial published in the Journal of the National Cancer Institute.

Of course the health food industry did not like the study. The first criticism I heard was that it was a correlation study and thus in no way could show cause and effect. A full list of criticism from many sources is given in piece at Nutraingredients.com.

I managed to get a copy of the paper and after reading it over the last week have a number of issues with the work that was done. My bottom line is that I would ignore the claims that are being made. The study has a number of major shortcomings and problems that go well beyond the issue of being just a correlation study that cannot show cause and effect.

As a start the study did not ask or attempt the find out if those in the sample used fish oil. They measured omega-3 in the blood. I do not know the way the body digest and maintains omega-3 in the blood, but can envision that there are a number of reasons why the level would vary independent of how often or how much one used fish oil capsules or by varying their diet.

The people entered the study between 2001 and 2004 and the occurrence of prostate cancer was ascertained up to 2009. There was only one measurement of the level of omega-3 in the blood. This then implies a very short term effect of the omega-3 in causing an increased risk for prostate cancer. To claim that such an increased risk can be measured based on the level of omega-3 at one point in time is to say the least a bit dubious. Perhaps the fellow ate salmon the night before the test.

The problems do not stop there. The paper provides a detailed table of the demographics with several biomarkers for the sample. It is clear from the table the prostate cancer risk is associated with the body mass index, with race, with baseline PSA to name just three. Yet the study does not claim to take these factors into the model.

The authors created a control group for their tests. But because the cost of the tests were high they performed a subsample of those without prostate cancer. They did this by “matching men randomly selected for the subcohort from the set of men with blood samples available within the same age-race stratum. A ratio of 1:3 was used for black men and 1:1.5 for men of other races.” In simple terms they sampled the non black at twice the rate used for black men in creating the control group. A a result nonblacks are overrepresented in the control group relative to there presence in the prostate cancer group. No attempt is provided in the paper describing any efforts made to account for this imbalance.

Table 2 of the paper provides mean levels of omega-3 for those without cancer and those with cancer. For those without cancer the value is 4.48 with a confidence interval(95%) from 4.41 to 4.55. For those with cancer the mean is 4.66 with a confidence interval(95%) from 4.56 to 4.75). The means are clearly different. But that is the wrong viewpont. The proper perspective here is the overlap in the distributions of the value of the omega-3 level between the two groups. With an n=1364 in the no cancer group and n=834 in the cancer group there is clearly considerable spread in the distributions. The overlap in values between the two groups is substantial. The lesson here is look at both the variance and the mean not just the mean. In simple terms I would like to know what is the probability that someone with the cancer has a higher omega-3 value than did someone without the cancer.

Then there is table 3 witch shows relative risks of prostate cancer by level of omega-3. For levels under 3.68 they give a reference risk of 1.00. For levels between 3.68 and 4.41 the risk is 1.15(0.87 to 1.51). for levels between 4.42 and 5.31 the risk is 1.28(0.97 to 1.69). And for levels over 5.31 the risk is 1.43(1.09 to 1.88). The numbers in brackets give the 96% confidence intervals. At the bottom of the table section is noted P_trend with a value of 0.007. No indication of how this value was calculated is provided. Given the size of the confidence intervals it is very hard to envision how a statistically significant trend line could have been obtained. But even if there is a trend all this shows is a correlation. It does not show a cause and effect.

I will ignore the study.

I am beginning to think it would be worthwhile creating a table of all the things that various papers and reports link to increased risk the things like cancer, diabetes, heart disease.