|
Abstract:
|
Measures of receipt of adjuvant therapies (chemo- and radiation therapy) are important variables in studies of patterns of cancer care, but no single source reports these therapies with complete accuracy. The Cancer Care Outcomes Research and Surveillance (CanCORS) study collected these variables from multiple sources including a patient survey, medical records abstraction, cancer registry databases, and Medicare claims data. The sample coverages of these sources are different, and the corresponding reporting patterns vary with different accuracy. We propose to impute the true therapy status using Bayesian hierarchical models, synthesizing data from all sources while correcting the misreporting and incorporating the correlations among them. Valid analysis of the therapy variables can then be based on multiply imputed data. We present results using the CanCORS data.
|
