Objective
Using a combination of performance measures, we updated previously proposed criteria for identifying physicians whose performance interpreting screening mammograms may indicate suboptimal interpretation skills.
Materials and Methods
In this Institutional Review Board-approved, HIPAA-compliant study, six expert breast imagers used a method based on the Angoff approach to update criteria for acceptable mammography performance on the basis of combined performance measures: (Group 1) sensitivity and specificity, for facilities with complete capture of false-negative cancers; and (Group 2) cancer detection rate (CDR), recall rate, and positive predictive value of a recall (PPV1), for facilities that cannot capture false negatives, but have reliable cancer follow-up information for positive mammograms. Decisions were informed by normative data from the Breast Cancer Surveillance Consortium (BCSC).
Results
Updated, combined ranges for acceptable sensitivity and specificity of screening mammography are: (1) sensitivity ≥80% and specificity ≥85% or (2) sensitivity 75–79% and specificity 88–97%. Updated ranges for CDR, recall rate, and PPV1 are: (1) CDR ≥6/1000, recall rate 3–20%, and any PPV1; (2) CDR 4–6/1000, recall rate 3–15%, and PPV1 ≥3%; or (3) CDR 2.5–4/1000, recall rate 5–12%, and PPV1 3–8%. Using the original criteria, 51% of BCSC radiologists had acceptable sensitivity and specificity; 40% had acceptable CDR, recall rate, and PPV1. Using the combined criteria, 69% had acceptable sensitivity and specificity and 62% had acceptable CDR, recall rate, and PPV1.
Conclusion
The combined criteria improve previous criteria by considering the inter-relationships of multiple performance measures and broaden the acceptable performance ranges compared to previous criteria based on individual measures.