mammograms are caused by technical faults, a third are radiologically occult, and a third are fast-growing tumours not yet detectable at screening,.10 By following a number of strict rules, we are attempting to reduce the number of falsenegative cases to the theoretical minimum.1' I We thank S.

No abstract

The costs and effects of different invitation schedules of breast cancer screening are compared. The effect estimates are based on trials from the USA, Sweden and the Netherlands. The cost estimates use registration data, file studies and organization charts. The calculations were performed with the MISCAN computer simulation package, which is developed especially for the evaluation of mass screening programmes. Screening women of 50-70 years at 2-yearly intervals is a relatively cost-effective schedule. In a real population, it will reduce breast cancer mortality by 12%. Screening of women under 50 is probably far less cost-effective. Screening induces a considerable shift towards breast-conserving therapy. Although a 12% mortality reduction may seem low, in absolute numbers this represents more than the total mortality from, e.g., cervical cancer. Moreover, cost per death prevented or per life-year saved is much lower than for most other medical interventions for which cost-effectiveness ratios are known, screening for cervical cancer included.

Study objective-The aim was to demonstrate the benefits of breast cancer screening on mortality.Design-The study was an evaluation of a breast cancer screening programme by means of different approaches: (1) a casecontrol study, breast cancer deaths being the cases; (2) comparing the numbers of breast cancer deaths in screened and unscreened women; (3) comparing breast cancer mortality before and after start of the programme; (4) comparing breast cancer mortality in different large cities; (5) comparing screening activity with mortality reduction.Setting-The setting was a breast cancer screening programme in the city ofUtrecht, the DOM project, for women aged 50-64 years old at intake, birth cohort 1911-1925 The first effects on mortality from breast cancer of a mass screening project in the city of Utrecht using both physical examination and (xero)mammography were published in 1984.' The project was called the DOM project, which in Dutch signifies "Diagnostisch Onderzoek (= investigation) Mammacarcinoom". It has a double meaning, because Utrecht is sometimes called the DOM city for its medieval cathedral. As the DOM project used a non-randomised design, we measured the effects on mortality from breast cancer by means of a case controlled study.In the case-control study, a case was defined as a breast cancer death in a woman born between 1911 and 1925, diagnosis and death occurring after the start of the project. For each case, three controls were selected at random from the local population registries, matched for year of birth. For both the case and the corresponding controls, the screening history was ascertained from the records of the DOM project. The screening history was taken for the time up to and including the date of diagnosis of the case, ie, a "restricted period" compared with the period of the total screening programme. A protective effect of screening against dying from breast cancer was found (odds ratio 0 30, 95% confidence interval 0 13-070).No biasing factors were found: in particular the incidence rate of breast cancer in the unscreened group was the same as in the population before the start of the project and the stage of the tumours did not much differ either.2 There were no changes in therapy in that period.Although this was not the first time that a screening project had been evaluated by means of a case-control approach,3 and at the same time a screening programme in Nijmegen was evaluated in the same way4 and others followed as well,5 some criticisms have been raised. Therefore we felt it desirable, in addition to updating the case-control study and searching further for confounding variables, to use alternative approaches to the evaluation of the screening effect. This paper includes: (1) an update of the case-control study (odds ratio) and further search for confounding factors; (2)

In medical usage, relative operating characteristics (ROC) refer to the symptoms, signs, and laboratory values used for making a diagnosis. Diagnosis analysis by ROC measures diagnosis performance independently of the diagnostic criteria used, thus revealing the arbitrary character of diagnosis. In order to carry out ROC analysis, the diagnosis has to be graded in levels of certainty of the presence of disease (or suspicion of a lesion). When both the graded diagnosis decisions and the ultimate proof of presence or absence of disease are available, ROC analysis can easily be carried out. Analysis of 50,102 successive examinations of 19,400 women in the Utrecht breast cancer screening projects presents a remarkably good fit of the observed data with the calculated values. In the medical context even small deviations of the actual sensitivity and specificity values from the ROC curve values must be considered as signs of differences in the diagnosis procedure and should lead to further analysis of the diagnostic criteria and the examined population. Consequently the separate effect of diagnosis impairment by the mammographic risk patterns (Wolfe) on diagnostic accuracy could be isolated. The effect of age on the detection capability of breast cancer by mammography appeared to operate in narrow age groups.