Abstract A80: Detection bias, statin drug use, and advanced prostate cancer risk

Some previous prospective studies support that statin drugs (HMG Co-A reductase inhibitors) may protect against advanced prostate cancer. However, there is concern that these results may be due to detection bias. Men are likely to be screened for both elevated PSA and cholesterol by their internists, and thus are more likely to undergo both diagnostic workup for prostate cancer and be prescribed a statin. Therefore, men who take a statin may be less likely to be diagnosed with advanced prostate cancer, irrespective of a causal relation. Ideally, the possibility of this bias would be eliminated by studying the association between statin drugs and prostate cancer in an unscreened population, but because statins and PSA screening both became available at the same time (late 1980s), it is impossible to study this association in the pre-PSA era in the US. Thus, we performed a simulation to determine whether this source of detection bias is explanatory.

3,000 datasets with 100,000 men without a prostate cancer diagnosis were simulated for two populations, one with a high (65%) and one with a low (15%) prevalence of PSA screening. For both populations we investigated three scenarios for the true association between statin use and advanced prostate cancer: RR=1.0, 0.75, and 0.5; in each we set the association for statin use and local disease to RR=1.0. We set the prevalence of statin use to 10% and varied the percentage of statins users who underwent PSA screening from 0 - 100%. We assumed the following: an annual prostate cancer incidence rate of 1%, a weighted average of the screened and unscreened men; a risk of prostate cancer diagnosis in screened men twice that in unscreened men; and an advanced stage at diagnosis in 20% and 40% of cases in screened and unscreened men, respectively.

The observed association (RR_obs) between statin use and local and total prostate cancer varied with the correlation between PSA screening and statin use. As PSA screening and statin use became more coincident, the RR_obs between statin use and both local and total prostate cancer was biased upward from the true null association. This bias was stronger in the population with a low prevalence of PSA screening: e.g., assuming that 70% of men using statins had PSA screening, the RR_obs for total and local prostate cancer, respectively, were 1.03 and 1.04 in the population with 65%, and 1.56 and 1.88 in the population with 15% prevalence of PSA screening. However, in all the simulated scenarios, there was very little bias in the RR_obs for statin use and advanced prostate cancer: e.g., assuming that 70% of men using statins had PSA screening and that the true RR for advanced prostate cancer was 1.0, the RR_obs for advanced prostate cancer was 0.98 in the population with a 65% prevalence of PSA screening, and 0.98 in the population with a 15% prevalence of PSA screening.

Under the scenarios and assumptions we used, this simulation suggests that detection bias due to the correlation between PSA screening and statin use is unlikely to explain the inverse association observed between statin use and advanced prostate cancer in some US and European cohorts, but could account for the positive association for total prostate cancer that has been reported in some studies.

Citation Information: Cancer Prev Res 2008;1(7 Suppl):A80.