Interobserver agreement of a probabilistic approach to reporting breast fine-needle aspirations on ThinPrep

Abstract

We have previously demonstrated the accuracy and reproducibility of a probabilistic/categorical approach for reporting breast fine-needle aspiration (FNA). However, the interobserver agreement in the application of this approach has not been assessed. Twenty breast FNA cases (each on one ThinPrep slide) were pulled from the cytology files of Beth Israel Deaconess Medical Center. The cases included benign epithelial proliferative lesions (6), DCIS (4), and infiltrating carcinoma (10), as shown by subsequent histology. Six pathologists with 14-25 yr of experience in interpreting breast FNA and 0-8 yr of experience with ThinPrep preparations rendered diagnoses according to the probabilistic approach. The kappa statistic for the unremarkable/proliferative, atypical, suspicious, and positive categories were 0.64, 0.08, 0.43, and 0.75, respectively (P < 0.001 for all except for the atypical category [P = 0.09]). Spearman's rho correlating the individual pathologist's diagnosis and the histologic diagnosis ranged from 0.51 (P = 0.02) to 0.78 (P < 0.0001). This was not correlated with the pathologists' years of experience interpreting breast FNA (P = 1.0) or with their years using ThinPrep preparations for breast FNA (P = 0.96). In conclusion, the interobserver agreement was excellent for the positive category in the probabilistic approach, poor for the atypical category, and fair to good for the other categories. The specific level of experience interpreting breast FNA or using ThinPrep among experienced pathologists did not seem to influence their accuracy in reporting the cases in our study.