|
Abstract:
|
We use a Bayesian approach to estimate casual effects, as defined by Rubin, in studies comparing folic acid dosing regimes with regard to blood folate level and the occurrence of neural tube defects. In these studies, noncompliance is associated with only a subset of Rubin's classifications, compliers, always takers, never takers and defiers, in that non or partial compliance implies a lowering of true folate levels in all study arms. In addition to estimating causal effects, we propose updating self-reported compliance values using observed folate levels and additional covariates. Potential association between compliance, outcome and missingness is assessed using a sensitivity analysis.
|
