Tuesday, April 11, 2017

The USPSTF takes a step backward on PSA screening

I don't agree with all of the statistics cited in this infographic, particularly the optimistic estimate that 1-2 men out of every 1000 screened with the PSA test avoid death from prostate cancer. I believe that the USPSTF's 2012 estimate of 0-1 men remains more accurate, but even if the new figure is true, I don't think that changing the PSA recommendation from a "D" (benefits no greater than harms, don't do) to a "C" (small net benefit, do selectively) is warranted, given the collateral damage to men's health that screening produces. In an editorial in JAMA, the USPSTF Chair and Vice-Chairs have invited the public to comment on the draft recommendations, which are more consistent with those from the American Cancer Society and American Urological Association than the American Academy of Family Physicians and the American College of Preventive Medicine (which both recommend against PSA-based screening). They will hear from me, and I hope that they will hear from others in primary care about the physical, psychological, and opportunity costs of taking a step backward on PSA screening.


  1. I agree, Kenny. Very disappointing. Some day I'd like to see definitive evidence that treatment of prostate cancer makes a difference, as opposed to "if you have the bad kind, you die despite treatment, and if you have the good kind you live die from something else, treatment or not" (although of course treatment has significant morbidity).

  2. Your Medscape article points out that ERSPC is a combination of 7 trials that various bodies, including the current USPSTF panel, claim satisfy a homogeneity test for risk ratio of prostate cancer mortality between screening and control arms across all 7 trials.

    Whatever statistical test they have used to make this claim, it is not a simple Chi-square test of homogeneity because of the heterogeneity between screening and non-screening arms. I am very dubious that they have done a good test because the Swedish risk ratio is easily inconsistent with the average risk ratio of all 7 trials.

    It is absolutely vital that homogeneity of risk ratio is established otherwise the confidence interval given for that risk ratio has no meaning and the statistical significance of a general prostate cancer mortality reduction does not exist.

  3. "I don't agree with all of the statistics cited in this infographic, particularly the optimistic estimate that 1-2 men out of every 1000 "

    Indeed that 1-2 range isn't an accurate reflection from a formally defined confidence interval. The average figure is 1.25 if I recall correctly but the lower confidence interval bound is a long way below 1. If they are going to use an informal approximation it would be far more honest to say 0-2.