|
Abstract:
|
A common design for an intervention trial is to randomly assign subjects into two groups. But in some cases, perfect randomization is not practicable, since at times the assigned intervention may not be available. For example, in a recently completed mammography peer teaching study, the objective was to compare the effectiveness of a nurse practitioner to that of a peer of the subject in explaining the importance of regular mammograms. The outcome was whether or not the subject returned for a mammogram within one year of the intervention. However, for logistical reasons, the randomly assigned intervention (nurse or peer) was not always available, and in such a case, the patient was re-assigned to the alternative intervention. In this talk we discuss using covariates, propensity scores, and instrument variables as potential tools for adjusting for the mis-randomization.
|
