Bayesian analysis of surveillance data.
EPICOH 16th International Conference on Epidemiology in Occupational Health, Barcelona, Spain, October 10-14, 2002. Rome, Italy: International Commission on Occupational Health, 2002 Sep; :12
Traditional methods of adjustment for multiple comparisons (eg., the Bonferroni adjustment) have fallen into disuse. It has been argues that such multiple comparison adjustments are unnecessary and in fact ill-advised, because they assume a global null hypothesis which is neither plausible nor of interest, and because they are too conservative and may lead investigators to ignore unexpected but important findings (Rothman, 1990). When faced with a large number of comparisons, many epidemiologists currently do no adjustment at all, but instead we use whatever a priori knowledge exists, as well as common sense and biological plausibility, to evaluate what findings are important in their data. However, Greenland and Robins (1991) and Greenland and Poole (1994) have argued that in some circumstances empirical or semi-Bayes adjustments can be useful as an alternative to traditional multiple comparison adjustments. These circumstances are that 1) a large number of comparisons are made, 2) the comparisons can be grouped into sets within which all comparisons can be considered similar or "exchangeable", 3) random error is present and presumably accounts for much of the observed variation in the parameters estimated to evaluate the comparisons (eg., relative risks, rate ratios, regression coefficients), and 4) investigators must choose which comparisons to investigate further, and there is a significant cost to such further investigations.
Case-studies; Mortality-rates; Mortality-data; Mortality-surveys; Morbidity-rates; Epidemiology; Exposure-assessment;; Statistical-analysis; Cancer; Cancer-rates; Carcinogens; Carcinomas; Breast-cancer; Risk-analysis; Risk-factors
EPICOH 16th International Conference on Epidemiology in Occupational Health, Barcelona, Spain, September 14, 2002