Table 1.
Comparison of breast tissue interpretation from 115 pathologists compared to the reference diagnosis
| Participating Pathologists’ Interpretation | Rate of Over- and Under- Interpretation Compared to the Reference Diagnosis | Overall Agreement Rate Compared to the Reference Diagnosis% | |||||||
|---|---|---|---|---|---|---|---|---|---|
| n (row %) | % (95% CI) a | (95% CI) a | |||||||
| Reference Diagnosis of Ductal Carcinoma in situ (DCIS) | No. of Interpretations | Benign w/o Atypia | Atypia | Low-grade DCIS | High-grade DCIS | Invasive | Over-interpretation | Under- interpretation | Agreement |
| Low-grade DCIS | 512 | 82 (16%) | 73 (14%) | 238 (46%) | 112 (22%) | 7 (1%) | 23 (19, 28) | 30 (26, 34) | 46 (42, 51) |
| High-grade DCIS | 1585 | 51 (3%) | 73 (5%) | 93 (6%) | 1321 (83%) | 47 (3%) | 3 (2, 4) | 14 (12, 16) | 83 (81, 86) |
a
Generalized estimating equations (GEE) with an independent correlation structure and robust standard errors were used in a logistic regression model to account for correlated response data.
Estimates and standard errors of LS-means were computed and transformed back to the original response scale via the inverse-link function to provide misclassification and agreement rates and their 95% confidence intervals.