Elwood Jm

Impact of independent double reading of mammograms from the inception of a population-based breast cancer screening programme

Summary The Otago-Southland Breast Screening programme was set up to find out how acceptable and effective breast screening would be for New Zealand women aged between 50 and 64 years. This report examines the performance of the radiologists in the first 18 months of the pilot programme. The majority, 5659 (80.0%), of the two view mammograms from 7074 women, were read independently by two radiologists. Women, who both radiologists believed had an abnormality, were automatically referred for further assessment. The mammograms of those thought to have an abnormality by only one were reconsidered before a referral by consensus was made. The majority of cancers (58) were detected in mammograms read by radiolo- gists A and B. The positive predictive value for those read as abnormal by both was 12.4%. If only one had read the mammograms the number of cancers detected would have been 56 or 55 depending on the radiologist. Overall 73 women were found to have cancer, 67 of whom were identified by two radiologists. The positive predictive value for all the radiologists was 10.5% (95% CI 7.7–12.4). The inter-observer agreement for radiologists A and B about whether or not a woman should be referred for further assessment measured by kappa was 0.65. Because of the debate about whether or not kappa is the most appropriate measure of agreement other measures are also presented. Alpha which is based upon the idea that mammograms fall into two classes, those that can be consistently classified by two observers according to some well defined rules and those, that because they are more difficult, are classified by chance. In this study alpha was 0.82 for radiologists A and B suggesting that they were in accord, taking account of chance, for 82% of the mammograms. The intraclass correlation, a measure of agreement among all radiologists was 0.64. These measures show higher levels of agreement than have been reported in other larger studies.