Objective: To assess double reading effectiveness in mammography screening.
Design: Retrospective study of 177,631 consecutive mammograms double read during 1998–2003.
Setting: The Florence screening programme, involving 11 trained radiologists. Abnormalities reported by at least one reader prompted assessment.
Results: The referral rate was 2.89% for the first reader, 3.15% for the second reader, and 3.59% for either reader. Of 713 total cancers detected, 43 were suspected only by the second reader (6.4% relative, 0.024% increase in absolute detection rate) and had a lower stage compared to the first reader (pTis–pT1b = 65.7 versus 52.0%): 41 were reviewed and classified (error type) as 'minimal sign' in six, and 'screening error' in 35 cases, or as BI-RADS 3 in one, 4a in 20, 4b in 13, and 4c in three cases. The second reading cost was €2.70 per woman examined, or €11,168 per additional cancer detected (versus €11,585 at a single reading).
Discussion: Second reading is effective in detecting a limited number of additional cancer cases. Tumour stage (one-third over 1 cm in diameter) and review findings (high rate of 'screening errors' and BI-RADS R4b-c categories) suggest that second reading detects small 'difficult cases' as well as larger cancers missed due to fatigue or loss of attention. Second reading reduces screening specificity to a minor extent, and since cancer detection at second reading seems cost-effective the procedure is recommendable in routine practice.
