|
Abstract:
|
We consider a study with a binary exposure and outcome, subject to confounding and misclassification of the outcome variable. There are several methods to adjust for misclassification of exposure; however, misclassification of outcome remains relatively unexplored. Outcomes from administrative claims data are often subject to misclassification, as diagnoses are based on coding such as ICD-10, which may not always reflect true outcomes. We use inverse probability weighting and internal validation sampling to rebalance covariates across treatment groups while mitigating misclassification bias. We discuss several validation sampling schemes and a Monte Carlo approach to approximate optimal sample size determination. A parametric bootstrap is used for variance estimation. We explore finite sample properties of the weighted estimators via simulation, with particular attention to relative efficiency of different sampling schemes for validation. We demonstrate the use of the methods through an example using administrative data.
|
