Overdiagnosis in the Mammography Breast Cancer Screening Programme in Norway

There is concern over the potential of breast screening to lead to overdiagnosis. Overdiagnosis means the diagnosis of cancer as a result of screening, which would never have been diagnosed in the patient's lifetime if screening had not taken place. Overd iagnosis is often estimated by comparing number of cancers diagnosed after screening starts with the numbers expected on the basis of trends in cancer occurrence before screening. This is complicated by the fact that other factors than screening may be af fecting the occurrence of cancer, and by the fact that screening brings forward the time of diagnosis by a few years. This means that some excess cancers are not overdiagnosed, but simply brought forward in time. We have been analysing breast cancer in cidence before and after the introduction of screening and found as expected an excess of cancers in the screening period. However, the excess of cases is larger than the number of cancers detected by screening. Thus the excess contains cases which cannot be overdiagnosed, and there must be other influences on breast cancer occurrence which are independent of screening (possibly use of hormone replacement therapy) and are changing over time. We are now working on estimation of the excess for screened and not screened women separately, and on better adjustment for indpendent trends in breast cancer incidence.

We plan to retreive data from Norwegian agencies on breast cancer incidence by region, age group and calendar period, along with data on screening exposure by the same factors. In addition, we plan to obtain individual screening data, including numbers sc reened, cancers detected, interval cancers, and tumours diaognised in non-attenders. We shall estimate the effect of screening on long-term cumulative incidence of disease, including at least ten years after the upper age limit for screening. An excess in cumulative incidence still observed years after stopping screening can reasonably be attributed to overdiagnosis. We shall also estimate overdiagnosis explicitly from multistate disease progression models. For this we shall use both conventional statis tical estimation and inference, and microsimulation techniques.