Use of Bayesian Modeling to Estimate the Sensitivity of Stereotactic Directional Vacuum-Assisted Breast Biopsy When the Gold Standard is Incomplete

Abstract

This study is conducted to estimate the sensitivity of stereotactic directional vacuum-assisted breast biopsy (ST DVAB) using Bayesian modeling and to predict how many more cancers can be inferred from those lesions without surgical correlation. We retrospectively reviewed the 103 lesions from 84 women who underwent ST DVAB. The study was approved by the Institutional Review Board of our hospital. We estimated the sensitivity and prevalence of the study population for ST DVAB by two types of approaches: for the type I approach, the gold standards were surgical correlation or postbiopsy mammographic follow-up. For the type II approach using Bayesian modeling by a beta-binomial model, the only gold standard was surgical correlation and the predicted number of cancerous lesions in those patients without surgical correlation was estimated. For the type I approach, the sensitivity was 92.3%, and the prevalence 12.6%. For the type II approach, the mean sensitivity of ST DVAB was 89%, and the mean prevalence was 15%. We predicted that an average of 1.7 cancerous lesions occurred among those lesions without surgical correlation by the Bayesian estimation. The mean sensitivity of ST DVAB using the Bayesian (type II) approach was lower than that using the type I approach, because we regarded the surgery as the only gold standard in Bayesian modeling and the nonoperated lesions were thought to be with unknown true disease status. The Bayesian approach is thus more appropriate to use than the type I approach when the gold standard is incomplete.