|
Abstract:
|
Multiplicity of research hypotheses is common in epidemiologic studies investigating drug safety. However, multiple test procedures are underutilized in these studies because controlling the type I error inflation for null associations of drug with risk reduces the power to detect those non-null associations of drug with risk. Through a simulation study, we show that type I error could be seriously inflated even when there are only a small number of hypotheses to test with rare safety events in large databases. By comparing Holm's method and Benjamini-Hochberg method with methods without Type I error adjustment, we illustrate that the Holm's method and Benjamini-Hochberg method can control Type I error rate while retaining good power and sensitivity. We illustrate the application of our method on a drug safety investigation.
|
