Interval breast cancers in an Australian mammographic screening program
Mary T Rickard, Richard J Taylor, Mohamed A Fazli and Nadima El Hassan
Objective: To determine the incidence of interval
cancers which occurred in the first 12 months after mammographic
screening at a mammographic screening service.|
Design: Retrospective analysis of data obtained by crossmatching the screening Service and the New South Wales Central Cancer Registry databases.
Setting: The Central & Eastern Sydney Service of BreastScreen NSW.
Participants: Women aged 40-69 years at first screen, who attended for their first or second screen between 1 March 1988 and 31 December 1992.
Main outcome measures: Interval-cancer rates per 10 000 screens and as a proportion of the underlying incidence of breast cancer (as estimated by the underlying rate in the total NSW population).
Results: The 12-month interval-cancer incidence per 10 000 screens was 4.17 for the 40-49 years age group (95% confidence interval [CI], 1.35-9.73) and 4.64 for the 50-69 years age group (95% CI, 2.47-7.94). Proportional incidence rates were 30.1% for the 40-49 years age group (95% CI, 9.8-70.3) and 22% for the 50-69 years age group (95% CI, 11.7-37.7). There was no significant difference between the proportional incidence rate for the 50-69 years age group for the Central & Eastern Sydney Service and those of major successful overseas screening trials.
Conclusion: Screening quality was acceptable and should result in a significant mortality reduction in the screened population. Given the small number of cancers involved, comparison of interval-cancer statistics of mammographic screening programs with trials requires age-specific or age-adjusted data, and consideration of confidence intervals of both program and trial data.
The principal aim of a mammographic screening service is to reduce
mortality from breast cancer. As this outcome cannot be measured
effectively for many years after the establishment of screening, and
as it applies to spatially defined populations, a range of interim
measures are used to evaluate screening services and to determine the
likelihood of mortality reduction. Interim measures include
screen-detected cancer rates and the rates of prognostic indicators
for these cancers, such as size, nodal status and grade.1 Another interim measure of
screening performance is the interval-cancer rate.2,3 As survival among women with
interval cancers is similar to that among unscreened women, then as
the incidence of interval cancers approaches that of the underlying
incidence of breast cancer the benefit of screening
Interval-cancer rates can be compared with background or underlying
rates to better assess the effectiveness of screening. Further,
comparing the rate of interval cancers with the underlying rate
allows comparison between different screening programs in
populations with different incidences of breast cancer.
Our aim was to evaluate first-year interval breast cancers after screening at the Central & Eastern Sydney Service of BreastScreen NSW (a mammographic screening and assessment service). We report the incidence and proportional incidence of interval breast cancers during the 12 months after both the first and second rounds of screening, over five years.
The Central & Eastern Sydney Service of BreastScreen NSW is located in
inner Sydney, and started screening in March 1988 as one of the first
pilot mammography screening programs in Australia. Any woman aged 40
years or more was eligible to attend, although during the study period
only 50-69-year-old residents in the Central Sydney Area Health
Service were actively recruited for two-yearly screening. Since the
progressive introduction of systematic, government-funded,
population-based screening from 1991, the Service has become a
regional component of BreastScreen NSW and BreastScreen
The study population comprised women who attended the Service for mammographic screening between 1 March 1988 and 31 December 1992. These women were all screened by two-view mammography. The technical quality of the Service's screening mammograms met all requirements for accreditation by BreastScreen Australia.5 All films were read independently by two radiologists and were reported as "normal" for routine rescreen or as "suspicious" and requiring assessment. The design of the screening and assessment Service has been previously described.6
|The definition of primary breast cancer used for this study includes invasive cancer and ductal carcinoma-in-situ (DCIS), but excludes lobular carcinoma-in-situ.5 All cases of primary breast cancer diagnosed by the screening and assessment Service in women attending for the first time were classified as prevalent (first-round) screen-detected cancers. Cancers in women attending for their first routine rescreen after two years were classified as second-round screen-detected cancers. Data were analysed according to age at first screen, by 10-year age groups (40-49, 50-59 and 60-69 years).|
We reviewed the data for interval cancers, defined as primary cancer
of the breast diagnosed up to 12 months after a first-round or
second-round screening mammogram in women aged 40-69 years at the
time of screening. The date of diagnosis of these interval cancers was
taken as the date of histological confirmation of cancer. Cases in
which clinical and imaging findings were consistent with breast
cancer but no surgery had been performed were included in the study and
the date of diagnosis was taken as the date of cytological
confirmation of cancer. Data were analysed by women's age at their
Some interval cancers were reported to the screening Service and others were identified by linking the Service's screening database with the NSW Central Cancer Registry (NSW CCR), which has operated since 1972 as a population-based registry. Notification of malignant neoplasms has been a statutory requirement for all NSW public and private hospitals, radiotherapy departments and nursing homes since 1972, and for pathology and outpatient departments since 1985. The date of diagnosis used by the NSW CCR was defined as "date of first definitive treatment for cancer" during the period covered by this study, although incidence was designated mostly on the basis of the first notification (usually the biopsy report).
We matched records of the screening database with the NSW CCR in late 1994 with the aid of probabilistic linkage7,8 using Automatch.9 Equivocal matches were investigated by individual examination of all details available, and by active follow-up by the screening Service. Positive linkages between the screening database and the cancer registry were investigated by the screening Service to ensure that all data on those interval cancers diagnosed in the first year after a negative screen were available for inclusion in the study.
The age-specific incidence of interval cancers was determined by
dividing the number of interval cancers found in women screened
between 1988 and 1992 by the age-specific (age at first screen) number
of women screened over the same period. Age groups were 40-49, 50-59
and 60-69 years, although the latter two age groups are aggregated for
reporting purposes. The underlying incidence of breast cancer from
1988 to 1992 was obtained by dividing NSW reported cases (as reported
by the NSW CCR.10,11) by the
State female population for the same three 10-year age groups. To
express the interval-cancer incidence as a proportion of the
underlying breast cancer incidence rate, an indirectly
age-standardised incidence ratio12 was calculated using the State
age-specific incidences as the standard. Because of small numbers of
cases, 95% confidence intervals were derived by the Poisson
We compared our findings with those of studies from Sweden,3 the Netherlands14 and the United Kingdom15 using the 12-month interval-cancer data from the first two screening rounds for all studies except the UK study, for which only the first-round data were available. Confidence limits for interval-cancer rates from comparison populations were calculated from the published data using the same Poisson method as above.
During the study period, the Central & Eastern Sydney Service carried
out 39 988 first-round and second-round screens on women aged 40-69
years, and 226 screen-detected cancers were diagnosed. In the
prevalent (first-round) screens, the screen-detected cancer rates
per 1000 screens were 3.0 for women aged 40-49 years, 6.2 for
50-59-year-olds and 9.6 for 60-69-year-olds. Second-round rates
were 3.5, 4.2 and 4.6, respectively. Eighteen interval cancers
occurred in the first 12 months after screening. All of these interval
cancers were invasive. The 12-month interval-cancer incidences are
shown in Box 1.
The underlying annual breast cancer incidences were 13.8 per 10 000 for the 40-49 years age group and 21.2 per 10 000 for the 50-69 years age group (19.4 for 50-59 years and 23.2 for 60-69 years). The interval-cancer incidences as a proportion of these underlying incidences are shown in Box 2, in which they are compared with those of the Swedish Two-County Study.3
The proportional interval-cancer incidence for women aged 50-69 years in the Central & Eastern Sydney Service (22.0%; 95% CI, 11.7%-37.7%) is compared with that for similar age groups from international studies3,14,15 in Box 3. The proportional interval-cancer incidence point estimate for the Central & Eastern Sydney Service was higher and had wider 95% confidence intervals than that of the Swedish Two-County Study (13.2%; 95% CI, 8.4-19.9), but was lower than that of the UK (31.4%; 95% CI, 24.9-39.1) and Dutch (39.5%; 95% CI, 20.4-69.0) studies.
The screen-detected cancer rates achieved by the Central & Eastern
Sydney Service exceed the National Accreditation Requirements and
compare favourably with those reported from the Swedish Two-County
Study and the other Australian mammographic screening
Our findings indicate that the first 12-month interval-cancer rate (as a proportion of underlying incidence) for women attending the Central & Eastern Sydney Service is higher than that of the Swedish Two-County Study,3 but lower than the rates reported from Nijmegen (the Netherlands)14 and North West Region UK.15 However, there is considerable overlap of 95% confidence intervals, and the conclusion must be that the Central & Eastern Sydney proportional first-year interval-cancer rate is not significantly different from that in these other studies. As a proportion of interval cancers occur because an abnormality has not been detected at the time of screening, the first-year interval-cancer rates reflect the proportion of false negative screens in the screening episode. The Central & Eastern Sydney results therefore indicate acceptable screening sensitivity and, as they are not significantly different from the results of the Two-County Study, they would be expected to indicate a similar future mortality reduction.
The interval-cancer data presented in this study are not affected by significant ascertainment bias. Restricting our analysis to interval cancers occurring within the first year after a screen eliminates the difficulties of accurately differentiating interval cancers and screen-detected cancers diagnosed around the 21-27-month rescreen interval. We directed considerable attention to the linkage with the NSW CCR to ensure that no interval cancers were missed.
The underlying rate of breast cancer is reliable. This was determined in a population that included the women who were screened. However, during the study period, the Central & Eastern Sydney BreastScreen Service was one of only two pilot mammographic screening and assessment endeavours in New South Wales, and screened a small fraction of the State's female population. Its clientele was not limited to the immediate geographic area. For the busiest years in this period (1991 and 1992) less than 5% of breast cancers in NSW were detected through these two mammographic screening services.18
Completeness of enumeration is difficult to determine precisely for cancer registries, but the standard indicators suggest reasonably good completeness for the NSW CCR,10,19 and its data are accepted for inclusion in Cancer incidence in five continents.20 Comparisons of the Central & Eastern Sydney Service and another pilot mammographic screening service with the NSW CCR for the period 1988-1992 showed 100% enumeration of invasive breast cancer by the NSW CCR (R T, NSW Cancer Council, unpublished data).
The design of the Australian mammographic screening program and the comparison studies in this paper differ. The Swedish Two-County Study employed single-view mammography and double reading. The Dutch and UK studies used single-view mammography and single reading. Double-view mammography and double reading, as used in Australia, would be expected to produce better results and fewer interval cancers in the first year than the comparison studies.
The confidence intervals of the proportional interval-cancer rates from the Swedish Two-County Study and those reported for North West Region UK do not overlap, and direct statistical testing has shown a significant difference between these rates.15 The authors of the UK article15 and those of the accompanying editorial21 expressed concern that the screening sensitivity in the British National Health Service program may not be sufficient to achieve mortality reduction targets. They noted the increased sensitivity resulting from two-view mammography, good film quality and two independent film readings. These features are part of BreastScreen Australia and of the Central & Eastern Sydney Service.
Even when data are collected over reasonable lengths of time, the numbers of interval cancers in many screening services are often small because of the size of the base populations. In the 50-69 years age group, there were 23 interval cancers in the Swedish Two-County Study, compared with 12 in the Nijmegen study and 13 in our study. However, performance can still be evaluated through calculation of interval-cancer rates, provided that interpretation is qualified by consideration of confidence intervals based on exact or Poisson methods. Australian mammography screening services are evaluated by comparison with National Accreditation Requirements standards.5 The point estimate of the Swedish Two-County Study has come to be regarded as a standard, as this study achieved significant mortality reduction. However the Two-County Study did not involve large numbers, and its interval-cancer rates must be interpreted in relation to their statistical confidence intervals. That is, standards cannot be derived from point estimates alone as they emanate from real studies in real populations. Further, comparison of the performance of screening services with such standards also requires that the confidence intervals of the screening service data be taken into account. Evaluations should not be based on point estimates of rates which derive from small numbers with considerable stochastic variation. The standard for Australian National Accreditation Requirements is less than six interval cancers per 10 000 screens occurring in the first 12 months. However, no age range or standardisation is specified, and confidence intervals are not considered.
Another consideration in setting performance standards for mammographic screening services is the difference in results from experimental and operational studies. Randomised trials and other specially constructed studies usually attract significant financial resources and interested investigators. It may be more reasonable to derive standards from studies of operational mammographic screening services whose data, when sufficient studies have been reported, could be assessed to determine performance standards which could reasonably be expected. However, until operational mammography screening services have been shown to produce reduced breast cancer mortality in the screened populations, the results of successful trials must set the gold standard.
|The Central & Eastern Sydney Service of BreastScreen NSW and BreastScreen Australia is jointly funded by the Commonwealth and the NSW State Governments. The NSW Central Cancer Registry, which is administered by the NSW Cancer Council and funded by the NSW Health Department, provided valuable assistance with data linkage.|
(Received 7 Aug 1997, accepted 3 Apr 1998)
Authors' detailsBreastScreen NSW, Central & Eastern Sydney, Sydney, NSW.
Mary T Rickard, FRACR, MPH, Director and Radiologist;
Mohamed A Fazli, BScEng, ME, Database Manager;
Nadima El Hassan, BEc(Hons), Computer Consultant.
Department of Public Health and Community Medicine, Faculty of
Medicine, University of Sydney, Sydney, NSW.
Richard J Taylor, FAFPHM, FRCP, Associate Professor in Public Health.
Reprints will not be available from the authors.
Correspondence: Dr M T Rickard, BreastScreen NSW, Central & Eastern Sydney, PO Box 1535, Strawberry Hills, NSW 2012.
Readers may print a single copy for personal use. No further reproduction or distribution of the articles should proceed without the permission of the publisher. For permission, contact the Australasian Medical Publishing Company
Journalists are welcome to write news stories based on what they read here, but should acknowledge their source as "an article published on the Internet by The Medical Journal of Australia <http://www.mja.com.au>".
Received 19 February 2019, accepted 19 February 2019
Publication of your online response is subject to the Medical Journal of Australia's editorial discretion. You will be notified by email within five working days should your response be accepted.