Optimal screening mammography reading volumes; evidence from real life in the East Midlands region of the NHS Breast Screening Programme

Abstract

AIM: To assess real-life reader performance as a function of both volume of mammograms read and reading experience in a multicentre cohort. MATERIALS AND METHODS: Thirty-seven film readers from the East Midlands Breast Screening Programme had 3 years of consecutive screen reading results available for comparison. Markers of screening performance as the first film reader [cancer detection rates, small cancer detection rates, recall rates, positive predictive value of recall (PPV), and missed cancers] were compared with both volume of films read and years of film reading experience. For reading volume, readers were categorized according to film reading volume over the 3 year period: < 15,000 (i.e., on average less than the recommended 5000/year); 15-< 20,000: 20-< 25,000; and >= 25,000. For years of experience, readers were categorized into the following groups: < 5 years, 5-< 10 years, 10-< 15 years, and 15-< 20 years. RESULTS: There was no evidence to suggest a relationship between years of film reading experience and film-reading performance. For reading volume, there was a significant difference in the distribution of cancer-detection rate between the four groups (p = 0.01); however, there was no difference in small cancer-detection rates, missed cancers or PPV. The median cancer detection rate in the high-volume group (>= 25,000 mammograms/3 years) was significantly lower than the other groups combined (6.9 per 1000 women screened versus 7.9 per 1000 women screened). The lowest median recall rate was also in the high-volume group, whilst those readers not meeting the NHSBSP minimum requirement had the highest median recall rate; however, there was borderline evidence to suggest a difference in the distribution of recall rates between the four groups. CONCLUSION: The data from the East Midlands do not provide any evidence for reducing the threshold volume of 5000 cases /year. However, there appears to be an upper limit above which reader performance deteriorates in terms of lower cancer-detection rates. With the imminent expansion of the programme, this has implications for service quality. These higher-volume readers should be encouraged to recall more borderline cases to assessment. Analysis of national data is recommended to further evaluate these findings. (c) 2010 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.