|
Abstract:
|
In clinical studies, the estimation of treatment/exposure effects can be biased when an important confounding variable is omitted from a statistical model evaluating the effects of treatment/exposure. It is necessary to evaluate the magnitude of bias due to such omitted variable and if needed, to adjust the estimation for this bias. When information about this covariate is missing in the current clinical study, one could possibly utilize the external/historical information about this covariate, e.g., from literature or from historical studies, into the analysis of the current clinical study data to control for such bias. In this talk, we present findings through a local sensitivity analysis of the potential impact from an unobserved confounder on the treatment/exposure effect estimation. Specifically, we derived simple formula to adjust the treatment effect estimates ignoring the omitted variables for such unobserved confounding. We demonstrate the use and effectiveness of these simple adjustment formula using simulations and datasets with/without external information on confounding variables.
|
