RESPONSE: Re: Zinc Supplement Use and Risk of Prostate Cancer

Extract

We respond briefly to the points raised by Costello et al. concerning our paper on zinc supplement use and the risk of prostate cancer (1). First, Costello et al. suggest that we should have provided P values for the pair-wise statistical comparisons of each zinc-user group with the nonuser group and used the Bonferroni correction for multiple comparisons. However, the standard method of providing estimates of disease risk according to distinct levels of an exposure in epidemiologic studies is the relative risk with its corresponding 95% confidence interval, and not the P value (2). We recognize that there is a difference of opinion regarding the adjustment for multiple comparisons: We do not believe that a uniform adjustment of the critical level of statistical significance for the P value is the best approach in this circumstance (2) because our study does not represent a multiple-inference situation.

Second, Costello et al. express concern about our small sample size at the highest level of zinc intake. We acknowledge that our risk estimate for men who consumed more than 100 mg/day of supplemental zinc (relative risk = 2.29) is somewhat imprecise as indicated by its wide confidence interval (95% confidence interval = 1.06 to 4.95). However, Costello et al. do not dispute the highly statistically significant positive tests for trend for level of supplemental zinc intake (P_trend = .003) and duration of supplemental zinc use (P_trend<.001) that remained statistically significant even after we excluded nonusers of zinc supplements from the analysis. The test for trend used in our study is a robust overall test for statistical significance that takes into account the entire range of supplemental zinc intake. Moreover, the Cox proportional hazards regression model used in our analysis (3) is not characterized by large sample-approximation requirements, as is the contingency table approach proposed by Costello et al.