Incidence and interval breast cancers in retrospective assessment

Abstract

The aims of this thesis were to investigate how different review methods used for radiological classification of breast cancers would affect the proportion judged to be false negative, to investigate how computer aided detection (CAD) would affect screening sensitivity, to study the balance between false positive and false negative results and to investigate patient outcome in relation to review result.

In a mass screening programme in Stockholm, commencing in 1989 and performed at five independent screening units, interval cancers (presenting in the time interval between two consecutive screening examinations) from two of the units, 103 women, and incidence cancers (diagnosed at subsequent screening examination) from one of the units, 117 women, diagnosed up until July 1993 were identified. Within a retrospective study set-up the screening mammograms, preceding breast cancer presentation, were reviewed mixed with images of healthy women at a ratio of 1:8 and non-mixed, including only cancer patients. Reviewers, both from the two units responsible for the screening mammograms and from two other units, participated in the study.

When using the mixed method, resembling the screen reading situation, significantly fewer cases were correctly selected (including very subtle findings) by the reviewers compared to when using the non-mixed method. Furthermore the number of reviewers (one, a majority or all) deemed necessary to classify a case to be detectable/false negative also influenced the rate that all in all varied between 7% and 34% for interval cancers and between 9% and 53% for incidence cancers. If cases correctly selected by a majority using either method were considering to be false negatives, the rate was found to be 22% for interval and 24% for incidence cases. Whether the reviewer worked at the unit responsible for the preceding examinations or not made no difference to the results. Trying to decrease false negative cases by increasing the number of women selected for recalls was in this retrospective study found likely to worry too many healthy women without providing a corresponding gain in cancer detection.

Preceeding screening images of 58 women with interval breast cancer were reviewed mixed with other screening mammograms without and with the aid of Computer Aided Detection (CAD) and non-mixed by three radiologists. Despite sufficient sensitivity for CAD alone, equal to that of the reviewers, radiologists were not inclined to revise their original interpretations according to CAD and no increase in radiologist sensitivity, or decrease in specificity, was noted when using CAD. Specificity for CAD alone was low, 38%. Large scale prospective studies would be required in order to demonstrate the usefulness of radiologists using CAD in every day mass mammography screening.

No significant differences regarding tumour prognostic factors or survival could be demonstrated between different review categories, i.e. whether or not an abnormality was detected on preceding images. Reviewing of previous mammograms of women later diagnosed with breast cancer is an important tool in learning and teaching for both radiologists and radiographers. However there are better methods available for quality control of screening than the number of tumours retrospectively judged to be false negative.