Semi-Bayes and empirical Bayes adjustment methods for multiple comparisons

Epidemiological studies often involve multiple comparisons, and may therefore report many "false positive" statistically significant findings simply because of the large number of statistical tests involved. Traditional methods of adjustment for multiple comparisons, such as the Bonferroni method may induce investigators to ignore potentially important findings, because they do not take account of the fact that some variables are of greater a priori interest than others. The Bonferroni method involves "adjusting" all of the findings to take account of the number of comparisons involved, even though the a priori evidence may be very strong for some exposures, but may be much weaker (or non-existent) for the other exposures being considered Furthermore, the Bonferroni method only "adjusts" for estimates of statistical significance (p-values) and does not "adjust" the effect estimates themselves (e.g. odds ratios and 95% CI). Empirical Bayes and semi-Bayes methods can enable the avoidance of numerous false positive associations, and can produce effect estimates that are, on the average, more valid. In this paper, we report on a research in which we applied these methods to a case-control study of occupational risk factors for lung cancer and tested their performance.