- Email: [email protected]
Breast cancer screening with mammography: overview of Swedish randomised trials
Saturday
17
April
No 8851
1993
ARTICLES
Breast cancer screening with mammography: overview of Swedish randomised trials
Despite encouraging results from screening trials the efficacy of mammography in reducing mortality remains somewhat controversial. Five studies have been done in Sweden. This overview, based on 282 777 women followed for 5-13 years in in randomised trials Malmö, Kopparberg, Östergötland, Stockholm, and Gothenburg, reveals a 24% (95% confidence interval 13-34%) significant reduction of breast cancer mortality among those invited to mammography screening compared with those not invited. To avoid the potential risk of differential misclassification causes of death were assessed by an independent end-point committee after a blinded review of all fatal breast cancer cases. The mortality reduction was similar, irrespective of the end-point used for evaluation ("breast cancer as underlying cause of death" or "breast cancer present at death"). There was a consistent risk reduction associated with screening in all studies, although the point estimate of the relative risk for all ages varied non-significantly between 0·68 and 0·84. The cumulative breast mortality by time since randomisation was estimated at 1·3 per 1000 within 6 years in the invited group compared with 1·6 in the control group. The corresponding figures after 9 years are 2·6 and 3·3 and after 12 years 3·9 and 5·1. The largest reduction of breast cancer mortality cancer
observed among women aged 50-69 at randomisation. Among women 40-49 there was a non-significant 13% reduction. In this younger age group cumulative breast cancer mortality was similar in the invited and control group during the first 8 years of follow-up. After 8 years there was a difference in favour of the invited women. There was no evidence of any detrimental effect of screening in terms of breast cancer mortality in any age group. Among women aged 70-74 years screening seems to have had only a marginal impact.
(29%)
was
Introduction The Health Insurance Plan of Greater New York (HIP) trial, which started in 1964, provided the first scientific evidence indicating that early diagnosis and treatment through breast screening can reduce breast cancer mortality. ADDRESSES: Department of Epidemiology and Public Health, Umeå University, S-901-85 Umeå, Sweden (L. Nyström, PhD, S. Wall, MD), Oncologic Centre, Karolinska Hospital, Stockholm (L E. Rutqvist, MD); Academic Hospital, Uppsala (A. Lindgren, MD); Karolinska Hospital (M Lindqvist, MD); Angelholm Hospital (S. Rydén, MD); Malmö Hospital (J. Andersson, MD); Sahlgrenska
Hospital, Gothenburg (N. Bjurstam, MD); University Hospital, Linköping (G. Fagerberg, MD); South Hospital, Stockholm (J Frisell, MD); Falun Hospital (L. Tabár, MD), and Oncologic Centre, Umeå University Hospital (L.-G. Larsson, MD). Correspondence to Dr Lennarth Nyström
974
TABLE I-BASIC CHARACTERISTICS OF FIVE RANDOMISED TRIALS
*Women with breast cancer diagnosed before date of randomisation excluded tFrom round 3 single or two views according to parenchymal pattern. tAverage for age group 40-49 and 50-74, respectively
The HIP trial was a randomised study of 62 000 women aged 40-64. The study group was invited to four annual screening rounds, including clinical examination and sometimes supplemented by mammography, thermography. There were significantly fewer breast cancer deaths in the study group than in the control group after only a few years.! That study stimulated worldwide interest in breast screening, especially by mammography.2 However, the benefit obtained in the HIP trial was only partly due to mammography, which identified only about one-third of all detected cancers. This was probably related to the technical quality of mammography during the 1960s. Furthermore, there was concern over the high radiation doses.3 In the early 1970s Lundgren and collaborators showed that efficient breast cancer screening could be done by a single oblique view and without concomitant clinical examination.4,s The radiation dose at mammography was further reduced by the introduction of intensifying screens in the mid-1970s. The pilot studies of Lundgren and colleagues prompted the National Board of Health and Welfare to initiate a large controlled study in two Swedish provinces, Kopparberg (W) and Ostergotland (E), in 1977. A similar but smaller trial had already started in the city of Malmo. Studies of similar design were later also initiated in the municipalities of Stockholm (1981) and Gothenburg (1982). These Swedish randomised trials were all population based-that is, they included all women belonging to a selected age group within a defined geographical area and compared mortality from breast cancer in the groups invited and not invited (control) to screening with mammography. The first mortality results from the Kopparberg/ Ostgotland (WE) trial were published in 1985 and showed a significant reduction of breast cancer mortality for women in the age group 40-74.6 A few years later, mortality data from the trials in Malmy and Stockholm8 indicated a reduction in breast cancer mortality which, however, was not statistically significant. Mortality data from the Gothenburg trial have not yet been published. Despite the encouraging results from some randomised trials breast screening remains somewhat controversial. The confidence intervals (CI) of the relative risks (RR) in individual studies have been wide making the true magnitude of any reduction in breast cancer mortality uncertain. No trial has so far recorded convincing evidence of a mortality reduction in women aged 40-49.
In 1987 the Swedish Cancer Society, in collaboration with the project leaders for the Swedish trials, initiated an overview of all five randomised trials. It was agreed that the overview should include extensive checks of the quality of the follow-up information using Swedish cancer and cause of death registries and extramural "blind" review of all breast cancer deaths by an independent "end-point committee" (EPC). The objective was to assess uniformly the efficacy of mammography screening by a blind review of causes of death among the breast cancer cases in all randomised Swedish trials.
Material and methods Invited women The characteristics of the five trials are summarised in table i. The designs were ahnost identical except for type of randomisation. However, no bias was introduced by the cluster method used in the WE trial.9 The screening centres sent magnetic tapes of information on their cohorts to the administration center of the overview at the department of epidemiology and public health in Umea. The cohorts were merged and linked to the six regional cancer registers to identify cases with breast cancer diagnosed between 1958 and 1989 and to the Swedish Cause of Death Register to identify women who died between 1951 and 1989 and the cause of death according to Statistics Sweden. The Swedish Cancer Registry was established in 1958. Later the data collection, coding, and computerisation of the cancer reports were relocated to six regional registers, and our record linkage was to these, to obtain the most recent data. Completeness of breast cancer registration in the Swedish Cancer Register has been estimated at 98%. This figure does not include breast cancer identified through death certificate diagnosis alone, since such cases are routinely not included in the Swedish Cancer Register. Collection of cause-of-death data has a long tradition in Sweden. Information has been continuously reviewed, coded, and computerised since 1951.
Exclusion criteria All analyses were based on exact age at randomisation, despite the fact that most trials, for practical reasons, used year-of-birth cohorts. This explains slight discrepancies between the numbers of women reported here and in previous reports. Since invitation to breast screening does not influence the clinical course of the disease among those who already have breast cancer, women with breast cancer diagnosed before the date of randomisation according to the Swedish Cancer Register, were excluded. The final combined cohort consisted of 282 777 women (invited group = 156 911, control group = 125 866) aged 40-74.
975
TABLE i)—PERSON-YEARS OF FOLLOW-UP AND END-POINTS WITH RELATIVE RISKS, ON FOLLOW-UP MODEL
to be discussed by consensus and in 89 of these the disagreement was whether breast cancer was underlying cause or
had
Reviewing The EPC consisted of a pathologist (A. L.), a diagnostic radiologist (M. L.), a surgeon (S. R.), and an oncologist (L. E. R.), none of whom had been involved in the screening trials. From the 27 582 deaths during follow-up to Dec 31, 1989, the review was restricted to the 1301 women who were reported to the Swedish Cancer Register with breast cancer after the date for randomisation and who died before Jan 1,1990 according to the Swedish Cause of Death Register. Women whose breast cancer was not reported to the Swedish Cancer Register, but who had breast cancer (ICD 174) as underlying or contributory cause of death were also reviewed if the death happened after randomisation but before Jan 1, 1990.
Retrieval of information A list of cases was sent to each screening centre asking for copies of medical records including necropsy reports, death certificates, and histopathology reports. Medical records and/or necropsy protocols with clinical summary were available for 99% of the breast cancer deaths, medical records for 86%, necropsy protocols for 39%, death certificates for 99%, and histopathology reports for 90%. Information relevant for the assessment of cause of death was blinded in respect of identity and mammography screening and the material on every case was checked by an oncologist (L.-G. L.) before submission to the EPC. All four members of the EPC then individually and independently of each other reviewed every case to determine the underlying cause of death and whether breast cancer was present at death. The rationale for having these two definitions of "breast cancer death" was that the second one had been used in previous analyses of the WE and Stockholm trials, whereas the former definition was used in the HIP and Mahnö trials. Histopathological slides or X-ray images were re-examined if requested by any of the committee members. Further clinical or other information could also be requested. If there was disagreement between the committee members the case was discussed at closed meetings until a consensus was reached. For the 1301 cases agreement was immediate for 1170 (89-9%). 131 cases
Fig
. Breast
cancer
0 Breast
cancer as
not.
Models In four of the five screening centres the controls were later also invited to screening. It can be argued that deaths in women whose cancer was diagnosed after completion of the screening of the control group cannot contribute useful information about the effects of screening. For this reason two models were used-a "follow-up" and an "evaluation" model (fig 1). The follow-up model included all breast cancer deaths among women with a primary diagnosis after the date of randomisation, and before a common fixed study end-point at Dec 31, 1989. The evaluation model ignored breast cancer deaths among women whose primary tumour, according to the cancer register, was diagnosed after completion of the screening of the controls. Any dilution of screening effects originating from cancers detected after the completion of screening of the control group was thus minimised. This design implies that both analyses will have the same study base
(person-year distribution). Statistical methods We used the QUEST software program." RRs were calculated by the density method, with total person-time experience of the cohort by time interval of follow-up as a basis for estimating the mortality rates. Weighted RRs and CIs were calculated by Mantel-Haenszel procedures.
Results
Study end-points On the follow-up model the results for the two study end-points are shown in table II. In the study group (1429 880 person-years) and control group (1138 850
accordmg to the Cancer Register underlying cause of death accordmg
to EPC
1-lnclusion and exclusion criteria in follow-up and evaluation models.
976
TABLE III-PERSON-YEARS OF FOLLOW-UP AND END-POINTS WITH RELATIVE RISKS, ON BOTH MODELS I
I
I
person-years) there were 418 and 425 deaths, respectively, with "breast cancer as the underlying cause of death", according to the EPC. The corresponding figures for
age groups.
"breast cancer present at death" were 440 and 442. The RR estimates were, however, similar (0-77 and 0-78) and the large numbers yielded narrow 95% CIs of 0 67-0-88 and 0’68-0’89, respectively. RRs for the two study end-points were similar in individual studies too. In what follows the only outcome measure will be "breast cancer as an underlying cause of death".
all centres and for all age groups. The largest reduction of breast cancer mortality (29%) was among women aged 50-69 at randomisation. Reductions among women aged 40-49 and 70-74 were lower, although a statistical test of heterogeneity between age groups was not significant. Among women aged 40-49 when invited to screening there was a non-significant 13% reduction in breast cancer mortality. Point estimates and 95% CI of RRs by screening centre and by age group are summarised in fig 2.
"Follow-up" and "evaluation"models Table III presents the results for screening centres separately, on both models and the age-specific results based on all centres. The weighted RR (and 95% CI) for all centres combined differed only marginally between the two models and the pattern was similar for specific centres and
Fig 2-Point estimates and 95% CI of RR, by screening
centre
and by age.
Although the weighted result was highly significant, individually only the Kopparberg data were significant (p < 0-05). However, RRs were below unity for
Cumulative mortality Cumulative mortality per 1000 person-years by time since randomisation was estimated on the follow-up model (fig 3). About 13 breast cancer deaths per 1000 occurred within 6 years in the invited group compared with 1-6 in the controls. The corresponding figures after 9 years were 26 and 3-3 and after 12 years they were 3-9 and 5-1. Fig 4 shows cumulative mortality by time since randomisation for each centre. Below the figure for each centre the cumulative proportion of women invited to screening of the control group is shown to illustrate its successive "contamination". Screening outside the protocols is not included in these figures. In the Maimo study the control group was not invited to screening during the study period. The cumulative mortality curves for the invited and control groups showed a similar pattern. However, the curves for the Gothenburg trial diverged earlier than in other trials. Cumulative mortality according to age is shown in fig 5. Curves for age groups 50-69 and 60-69 follow a similar pattern; the curves for the age group 70-74 years intersect several times; and those for the age group 40-49 seem to diverge after about 8 years.
Discussion Overview design
Fig 3-Cumulative breast cancer mortality per 1000 in invited group (-)and control group (- - - -) for breast cancer as underlying cause of death according to EPC follow-up model.
To avoid the potential risk of differential misclassification of breast cancer death we set up an extramural review by an independent committee whose members were unaware of whether a woman had been randomised to screening or not. Record linkage with the Swedish cancer and cause-of-death registries will have ensured almost complete follow-up of all women randomised and a near-complete identification of incident breast cancers. The potential risk of bias resulting from any differences in follow-up routines between the
977
Fig 4-Cumulative breast
proportions screened
cancer mortality per 1000 in invited groups (-) and control groups in controls by year since randomisation for individual screening centres.
(- - - -)
and cumulative
invited and control groups should thus have been minimised. The mortality reduction was similar irrespective of whether the end-point used was "breast cancer" as the underlying cause of death or "present at death". The estimated RR for total mortality associated with screening was 1-00. This result is as expected from the relative importance of breast cancer as a cause of death in the female population and the observed effect of screening.
"Vo//ow-1//?" a/?
"evaluation"models
In four centres control women were invited to screening 3-6 years after the trials began (fig 4) and with time this would tend to have obscured any difference in breast cancer mortality between the groups. This is why we used two
different models-namely, "follow-up", which included all breast cancer deaths before the common end-date for follow-up, and "evaluation", which excluded breast cancer deaths in women whose breast cancer was detected after the completion of the screening of the controls. However, the RR associated with screening was much the same with either model, probably because benefit did not emerge until about 4 years after screening was initiated. Since only 2-7 years (table I) had elapsed between the invitation to control groups and the common end-date for follow-up most control women will not yet have gained from screening. On the evaluation model the reduction of breast cancer mortality was estimated at 24%. This is a conservative estimate of the effect of mammography because the calculation includes those who chose not to attend. Moreover, some controls had mammography outside the trials. However, 24% represents an unbiased estimate of the effect of invitation of screening in comparison with no invitation. Years after randomisation
Fig 5-Cumulative breast cancer mortality per 1000 in invited group (-) and control group (- - - -) by year since randomisation and by age at randomisation.
Benefit by centre and by age There was a risk reduction associated with invitation to screening in all five studies though the RR for all ages varied,
978
the evaluation model, between 0-68 and 084. The individual studies differed in setting, attendance, penetrance of the intervention into the control group, and length of follow-up and we cannot tell whether there was any true variation in the efficacy between centres. A statistical test of heterogeneity of RRs for the five centres was not significant, implying that the variation observed may have been due to chance. The effect of screening by age has received considerable attention. For instance, it has been suggested that the more dense breast parenchyma in young women may contribute to a lower screening sensitivity and hence a smaller benefit than in older women. In a previous analysis of the Maimo trial there was even a slight numerical increase of breast cancer mortality among young women invited to screening. Our overview points to considerable age-related variation in estimated screening benefit even though a statistical test of heterogeneity between the age groups (40-49, 50-59, 60-69, 70-74) was not significant. The largest benefit in terms of the proportionate breast cancer mortality reduction was observed among the women aged 50-69 years at randomisation. Among women aged 40-49 the benefit was small and delayed (cumulative mortality curves did not diverge until after about 8 years). In the pooled material there was no evidence of an early increase of breast cancer mortality in this age group among those invited to screening. Among women aged 70-74 the cumulative breast cancer mortality was similar among the invited and control group during the entire period of observation. Only the WE trial included women aged 70 or more but the study was closed after only two screening rounds. This closure and the limited number of women (less than 18 000) make it difficult to draw any conclusions about the benefit of screening in women aged 70 years and above. on
Screening interval A previous analysis of the cost of breast cancer screening in Sweden suggested that the screening interval was a more important determinant of programme costs than the age groups selected." The screening interval in these studies varied from 18 to 33 months and no study included a randomised comparison for different intervals. However, the distribution of interval cancers in the WE study indicates that the incidence of such cancers approaches that expected without screening during the third year after the initial screening round-leading to the suggestion that the screening interval should not exceed two years. However, this conclusion should be considered tentative because interval cancers are only a proxy measure of the efficacy of screening in terms of reduced breast cancer mortality.
Conclusion
Many factors other than breast cancer mortality should be considered in a comprehensive evaluation of screening-for instance, the benefit of being able to use breast-conserving surgery in patients with small tumours, a decreased need of systemic treatment, a reduced fear of breast cancer among women with a negative screening result, the risk of anxiety caused by screening, the risk of diagnosing biologically benign lesions as cancer, the cost, inconvenience, and harm of negative biopsies, and the cost of the screening programme. This overview has been focused on breast cancer mortality. The greatest benefit was among women aged 50-69, who showed a proportionate breast cancer mortality reduction of 29%. However, longer follow-up
than the 7-12 years so far is important because the analysis suggests that a putative benefit for younger women may not emerge until after 8-10 years of observation. We thank Prof Jan Ponten, one of the initiators of this project and chairman of the Research Board of the Swedish Cancer Society which generously supported this project; Anna-Lena Johansson (department of epidemiology and public health, Umea University) who handled the medical records, necropsy reports and other documents for blind reviewing by the End Point Committee; and Vivi-Maj Miren and Gunilla Andersson in Malino, Barbro Ahlqvist in Kopparberg, Gunnel Nygren in Ostergotland, Birgit Angelholm in Stockholm, and Agneta Eliasson in Gothenburg who collected the information required.
REFERENCES
Shapiro S, Strax P, Venet L. Periodic breast screening in reducing mortality from breast cancer. JAMA 1971; 215: 1777-85. 2. Rutqvist LE, Miller AB, Andersson I, Hakama M, Hakulinen T, Sigfrisson BF, Tabar L. Reduced breast cancer mortality with mammography screening: an assessment of currently available data. Int J Cancer 1990; 5 (suppl): 76-84. 3. Bailar JC III. Mammography: a contrary view. Ann Intern Med 1976; 84: 1.
77-84,2783. 4.
Jacobsson S, Lundgren B, Melander O, Noren T. Mass-screening of a female population for detection of early carcinoma of the breast. Acta
5.
Lundgren B, Jacobsson S. Single view mammography. Cancer 1976; 38:
Radiol 1975; 14: 424-32. 1124-29. 6. Tabár L, Fagerberg CJG, Gad A, Baldetorp L, Holmberg LH, Gröntoft O, et al. Reduction in mortality from breast cancer after mass screening with mammography. Lancet 1985; i: 829-32. 7. Andersson I, Aspegren K, Janzon L, Landberg T, Lindholm K, Linell F, et al. Mammographic screening and mortality from breast cancer: the Malmö mammographic screening trial. BMJ 1988; 297: 943-48. 8. Frisell J, Eklund G, Hellström L, Lidbrink E, Rutqvist LE, Somell A. Randomized study of mammography screening: preliminary report on mortality in the Stockholm trial. Breat Cancer Res Treat 1991; 18: 49-56. 9. Duffy S, Tabár L, Krusemo UB. Randomization by cluster in the Swedish two-county trial: results from Kopparberg and implications for interpretation. Presentation at Nordic Cancer Union Symposium
(Stockholm, Aug 17-19, 1989). L. QUEST: a program system for statistical and epidemiological data analysis. Umeå: Umeå University, 1990. 11. Anon. Mammographic screening for early detection of breast cancer. Stockholm: National Board of Health and Welfare, 1990. 10. Gustafsson
From The Lancet A
screw too
few
The mishap which overtook the Umbria in mid-Atlantic and occasioned widespread alarm both here and in America, though now happily passed, has emphasised the serious peril which is involved in trusting to the capabilities of a single propeller however excellently constructed and skilfully worked. A break-down must occasionally happen whatever measures may have been taken to prevent it and then if no provision has been made for efficiently navigating the disabled ship an emergency arises which is most painful to contemplate. In some of the best ships afloat the propeller consists of twin screws, and if one should become incapacitated the other can still be relied upon with no worse result than some small delay. We believe that we are correct in stating that upon one of the voyages which the City of New York made last year one of her screws became disabled, but the ship, instead of becoming unmanageable, was propelled by the uninjured screw alone at a rate of seventeen knots an hour. If this be correct, surely no sea-going steamer should be built without this most necessary equipment of
duplicated propelling machinery. (Jan 7, 1893)
17
April
No 8851
1993
ARTICLES
Breast cancer screening with mammography: overview of Swedish randomised trials
Despite encouraging results from screening trials the efficacy of mammography in reducing mortality remains somewhat controversial. Five studies have been done in Sweden. This overview, based on 282 777 women followed for 5-13 years in in randomised trials Malmö, Kopparberg, Östergötland, Stockholm, and Gothenburg, reveals a 24% (95% confidence interval 13-34%) significant reduction of breast cancer mortality among those invited to mammography screening compared with those not invited. To avoid the potential risk of differential misclassification causes of death were assessed by an independent end-point committee after a blinded review of all fatal breast cancer cases. The mortality reduction was similar, irrespective of the end-point used for evaluation ("breast cancer as underlying cause of death" or "breast cancer present at death"). There was a consistent risk reduction associated with screening in all studies, although the point estimate of the relative risk for all ages varied non-significantly between 0·68 and 0·84. The cumulative breast mortality by time since randomisation was estimated at 1·3 per 1000 within 6 years in the invited group compared with 1·6 in the control group. The corresponding figures after 9 years are 2·6 and 3·3 and after 12 years 3·9 and 5·1. The largest reduction of breast cancer mortality cancer
observed among women aged 50-69 at randomisation. Among women 40-49 there was a non-significant 13% reduction. In this younger age group cumulative breast cancer mortality was similar in the invited and control group during the first 8 years of follow-up. After 8 years there was a difference in favour of the invited women. There was no evidence of any detrimental effect of screening in terms of breast cancer mortality in any age group. Among women aged 70-74 years screening seems to have had only a marginal impact.
(29%)
was
Introduction The Health Insurance Plan of Greater New York (HIP) trial, which started in 1964, provided the first scientific evidence indicating that early diagnosis and treatment through breast screening can reduce breast cancer mortality. ADDRESSES: Department of Epidemiology and Public Health, Umeå University, S-901-85 Umeå, Sweden (L. Nyström, PhD, S. Wall, MD), Oncologic Centre, Karolinska Hospital, Stockholm (L E. Rutqvist, MD); Academic Hospital, Uppsala (A. Lindgren, MD); Karolinska Hospital (M Lindqvist, MD); Angelholm Hospital (S. Rydén, MD); Malmö Hospital (J. Andersson, MD); Sahlgrenska
Hospital, Gothenburg (N. Bjurstam, MD); University Hospital, Linköping (G. Fagerberg, MD); South Hospital, Stockholm (J Frisell, MD); Falun Hospital (L. Tabár, MD), and Oncologic Centre, Umeå University Hospital (L.-G. Larsson, MD). Correspondence to Dr Lennarth Nyström
974
TABLE I-BASIC CHARACTERISTICS OF FIVE RANDOMISED TRIALS
*Women with breast cancer diagnosed before date of randomisation excluded tFrom round 3 single or two views according to parenchymal pattern. tAverage for age group 40-49 and 50-74, respectively
The HIP trial was a randomised study of 62 000 women aged 40-64. The study group was invited to four annual screening rounds, including clinical examination and sometimes supplemented by mammography, thermography. There were significantly fewer breast cancer deaths in the study group than in the control group after only a few years.! That study stimulated worldwide interest in breast screening, especially by mammography.2 However, the benefit obtained in the HIP trial was only partly due to mammography, which identified only about one-third of all detected cancers. This was probably related to the technical quality of mammography during the 1960s. Furthermore, there was concern over the high radiation doses.3 In the early 1970s Lundgren and collaborators showed that efficient breast cancer screening could be done by a single oblique view and without concomitant clinical examination.4,s The radiation dose at mammography was further reduced by the introduction of intensifying screens in the mid-1970s. The pilot studies of Lundgren and colleagues prompted the National Board of Health and Welfare to initiate a large controlled study in two Swedish provinces, Kopparberg (W) and Ostergotland (E), in 1977. A similar but smaller trial had already started in the city of Malmo. Studies of similar design were later also initiated in the municipalities of Stockholm (1981) and Gothenburg (1982). These Swedish randomised trials were all population based-that is, they included all women belonging to a selected age group within a defined geographical area and compared mortality from breast cancer in the groups invited and not invited (control) to screening with mammography. The first mortality results from the Kopparberg/ Ostgotland (WE) trial were published in 1985 and showed a significant reduction of breast cancer mortality for women in the age group 40-74.6 A few years later, mortality data from the trials in Malmy and Stockholm8 indicated a reduction in breast cancer mortality which, however, was not statistically significant. Mortality data from the Gothenburg trial have not yet been published. Despite the encouraging results from some randomised trials breast screening remains somewhat controversial. The confidence intervals (CI) of the relative risks (RR) in individual studies have been wide making the true magnitude of any reduction in breast cancer mortality uncertain. No trial has so far recorded convincing evidence of a mortality reduction in women aged 40-49.
In 1987 the Swedish Cancer Society, in collaboration with the project leaders for the Swedish trials, initiated an overview of all five randomised trials. It was agreed that the overview should include extensive checks of the quality of the follow-up information using Swedish cancer and cause of death registries and extramural "blind" review of all breast cancer deaths by an independent "end-point committee" (EPC). The objective was to assess uniformly the efficacy of mammography screening by a blind review of causes of death among the breast cancer cases in all randomised Swedish trials.
Material and methods Invited women The characteristics of the five trials are summarised in table i. The designs were ahnost identical except for type of randomisation. However, no bias was introduced by the cluster method used in the WE trial.9 The screening centres sent magnetic tapes of information on their cohorts to the administration center of the overview at the department of epidemiology and public health in Umea. The cohorts were merged and linked to the six regional cancer registers to identify cases with breast cancer diagnosed between 1958 and 1989 and to the Swedish Cause of Death Register to identify women who died between 1951 and 1989 and the cause of death according to Statistics Sweden. The Swedish Cancer Registry was established in 1958. Later the data collection, coding, and computerisation of the cancer reports were relocated to six regional registers, and our record linkage was to these, to obtain the most recent data. Completeness of breast cancer registration in the Swedish Cancer Register has been estimated at 98%. This figure does not include breast cancer identified through death certificate diagnosis alone, since such cases are routinely not included in the Swedish Cancer Register. Collection of cause-of-death data has a long tradition in Sweden. Information has been continuously reviewed, coded, and computerised since 1951.
Exclusion criteria All analyses were based on exact age at randomisation, despite the fact that most trials, for practical reasons, used year-of-birth cohorts. This explains slight discrepancies between the numbers of women reported here and in previous reports. Since invitation to breast screening does not influence the clinical course of the disease among those who already have breast cancer, women with breast cancer diagnosed before the date of randomisation according to the Swedish Cancer Register, were excluded. The final combined cohort consisted of 282 777 women (invited group = 156 911, control group = 125 866) aged 40-74.
975
TABLE i)—PERSON-YEARS OF FOLLOW-UP AND END-POINTS WITH RELATIVE RISKS, ON FOLLOW-UP MODEL
to be discussed by consensus and in 89 of these the disagreement was whether breast cancer was underlying cause or
had
Reviewing The EPC consisted of a pathologist (A. L.), a diagnostic radiologist (M. L.), a surgeon (S. R.), and an oncologist (L. E. R.), none of whom had been involved in the screening trials. From the 27 582 deaths during follow-up to Dec 31, 1989, the review was restricted to the 1301 women who were reported to the Swedish Cancer Register with breast cancer after the date for randomisation and who died before Jan 1,1990 according to the Swedish Cause of Death Register. Women whose breast cancer was not reported to the Swedish Cancer Register, but who had breast cancer (ICD 174) as underlying or contributory cause of death were also reviewed if the death happened after randomisation but before Jan 1, 1990.
Retrieval of information A list of cases was sent to each screening centre asking for copies of medical records including necropsy reports, death certificates, and histopathology reports. Medical records and/or necropsy protocols with clinical summary were available for 99% of the breast cancer deaths, medical records for 86%, necropsy protocols for 39%, death certificates for 99%, and histopathology reports for 90%. Information relevant for the assessment of cause of death was blinded in respect of identity and mammography screening and the material on every case was checked by an oncologist (L.-G. L.) before submission to the EPC. All four members of the EPC then individually and independently of each other reviewed every case to determine the underlying cause of death and whether breast cancer was present at death. The rationale for having these two definitions of "breast cancer death" was that the second one had been used in previous analyses of the WE and Stockholm trials, whereas the former definition was used in the HIP and Mahnö trials. Histopathological slides or X-ray images were re-examined if requested by any of the committee members. Further clinical or other information could also be requested. If there was disagreement between the committee members the case was discussed at closed meetings until a consensus was reached. For the 1301 cases agreement was immediate for 1170 (89-9%). 131 cases
Fig
. Breast
cancer
0 Breast
cancer as
not.
Models In four of the five screening centres the controls were later also invited to screening. It can be argued that deaths in women whose cancer was diagnosed after completion of the screening of the control group cannot contribute useful information about the effects of screening. For this reason two models were used-a "follow-up" and an "evaluation" model (fig 1). The follow-up model included all breast cancer deaths among women with a primary diagnosis after the date of randomisation, and before a common fixed study end-point at Dec 31, 1989. The evaluation model ignored breast cancer deaths among women whose primary tumour, according to the cancer register, was diagnosed after completion of the screening of the controls. Any dilution of screening effects originating from cancers detected after the completion of screening of the control group was thus minimised. This design implies that both analyses will have the same study base
(person-year distribution). Statistical methods We used the QUEST software program." RRs were calculated by the density method, with total person-time experience of the cohort by time interval of follow-up as a basis for estimating the mortality rates. Weighted RRs and CIs were calculated by Mantel-Haenszel procedures.
Results
Study end-points On the follow-up model the results for the two study end-points are shown in table II. In the study group (1429 880 person-years) and control group (1138 850
accordmg to the Cancer Register underlying cause of death accordmg
to EPC
1-lnclusion and exclusion criteria in follow-up and evaluation models.
976
TABLE III-PERSON-YEARS OF FOLLOW-UP AND END-POINTS WITH RELATIVE RISKS, ON BOTH MODELS I
I
I
person-years) there were 418 and 425 deaths, respectively, with "breast cancer as the underlying cause of death", according to the EPC. The corresponding figures for
age groups.
"breast cancer present at death" were 440 and 442. The RR estimates were, however, similar (0-77 and 0-78) and the large numbers yielded narrow 95% CIs of 0 67-0-88 and 0’68-0’89, respectively. RRs for the two study end-points were similar in individual studies too. In what follows the only outcome measure will be "breast cancer as an underlying cause of death".
all centres and for all age groups. The largest reduction of breast cancer mortality (29%) was among women aged 50-69 at randomisation. Reductions among women aged 40-49 and 70-74 were lower, although a statistical test of heterogeneity between age groups was not significant. Among women aged 40-49 when invited to screening there was a non-significant 13% reduction in breast cancer mortality. Point estimates and 95% CI of RRs by screening centre and by age group are summarised in fig 2.
"Follow-up" and "evaluation"models Table III presents the results for screening centres separately, on both models and the age-specific results based on all centres. The weighted RR (and 95% CI) for all centres combined differed only marginally between the two models and the pattern was similar for specific centres and
Fig 2-Point estimates and 95% CI of RR, by screening
centre
and by age.
Although the weighted result was highly significant, individually only the Kopparberg data were significant (p < 0-05). However, RRs were below unity for
Cumulative mortality Cumulative mortality per 1000 person-years by time since randomisation was estimated on the follow-up model (fig 3). About 13 breast cancer deaths per 1000 occurred within 6 years in the invited group compared with 1-6 in the controls. The corresponding figures after 9 years were 26 and 3-3 and after 12 years they were 3-9 and 5-1. Fig 4 shows cumulative mortality by time since randomisation for each centre. Below the figure for each centre the cumulative proportion of women invited to screening of the control group is shown to illustrate its successive "contamination". Screening outside the protocols is not included in these figures. In the Maimo study the control group was not invited to screening during the study period. The cumulative mortality curves for the invited and control groups showed a similar pattern. However, the curves for the Gothenburg trial diverged earlier than in other trials. Cumulative mortality according to age is shown in fig 5. Curves for age groups 50-69 and 60-69 follow a similar pattern; the curves for the age group 70-74 years intersect several times; and those for the age group 40-49 seem to diverge after about 8 years.
Discussion Overview design
Fig 3-Cumulative breast cancer mortality per 1000 in invited group (-)and control group (- - - -) for breast cancer as underlying cause of death according to EPC follow-up model.
To avoid the potential risk of differential misclassification of breast cancer death we set up an extramural review by an independent committee whose members were unaware of whether a woman had been randomised to screening or not. Record linkage with the Swedish cancer and cause-of-death registries will have ensured almost complete follow-up of all women randomised and a near-complete identification of incident breast cancers. The potential risk of bias resulting from any differences in follow-up routines between the
977
Fig 4-Cumulative breast
proportions screened
cancer mortality per 1000 in invited groups (-) and control groups in controls by year since randomisation for individual screening centres.
(- - - -)
and cumulative
invited and control groups should thus have been minimised. The mortality reduction was similar irrespective of whether the end-point used was "breast cancer" as the underlying cause of death or "present at death". The estimated RR for total mortality associated with screening was 1-00. This result is as expected from the relative importance of breast cancer as a cause of death in the female population and the observed effect of screening.
"Vo//ow-1//?" a/?
"evaluation"models
In four centres control women were invited to screening 3-6 years after the trials began (fig 4) and with time this would tend to have obscured any difference in breast cancer mortality between the groups. This is why we used two
different models-namely, "follow-up", which included all breast cancer deaths before the common end-date for follow-up, and "evaluation", which excluded breast cancer deaths in women whose breast cancer was detected after the completion of the screening of the controls. However, the RR associated with screening was much the same with either model, probably because benefit did not emerge until about 4 years after screening was initiated. Since only 2-7 years (table I) had elapsed between the invitation to control groups and the common end-date for follow-up most control women will not yet have gained from screening. On the evaluation model the reduction of breast cancer mortality was estimated at 24%. This is a conservative estimate of the effect of mammography because the calculation includes those who chose not to attend. Moreover, some controls had mammography outside the trials. However, 24% represents an unbiased estimate of the effect of invitation of screening in comparison with no invitation. Years after randomisation
Fig 5-Cumulative breast cancer mortality per 1000 in invited group (-) and control group (- - - -) by year since randomisation and by age at randomisation.
Benefit by centre and by age There was a risk reduction associated with invitation to screening in all five studies though the RR for all ages varied,
978
the evaluation model, between 0-68 and 084. The individual studies differed in setting, attendance, penetrance of the intervention into the control group, and length of follow-up and we cannot tell whether there was any true variation in the efficacy between centres. A statistical test of heterogeneity of RRs for the five centres was not significant, implying that the variation observed may have been due to chance. The effect of screening by age has received considerable attention. For instance, it has been suggested that the more dense breast parenchyma in young women may contribute to a lower screening sensitivity and hence a smaller benefit than in older women. In a previous analysis of the Maimo trial there was even a slight numerical increase of breast cancer mortality among young women invited to screening. Our overview points to considerable age-related variation in estimated screening benefit even though a statistical test of heterogeneity between the age groups (40-49, 50-59, 60-69, 70-74) was not significant. The largest benefit in terms of the proportionate breast cancer mortality reduction was observed among the women aged 50-69 years at randomisation. Among women aged 40-49 the benefit was small and delayed (cumulative mortality curves did not diverge until after about 8 years). In the pooled material there was no evidence of an early increase of breast cancer mortality in this age group among those invited to screening. Among women aged 70-74 the cumulative breast cancer mortality was similar among the invited and control group during the entire period of observation. Only the WE trial included women aged 70 or more but the study was closed after only two screening rounds. This closure and the limited number of women (less than 18 000) make it difficult to draw any conclusions about the benefit of screening in women aged 70 years and above. on
Screening interval A previous analysis of the cost of breast cancer screening in Sweden suggested that the screening interval was a more important determinant of programme costs than the age groups selected." The screening interval in these studies varied from 18 to 33 months and no study included a randomised comparison for different intervals. However, the distribution of interval cancers in the WE study indicates that the incidence of such cancers approaches that expected without screening during the third year after the initial screening round-leading to the suggestion that the screening interval should not exceed two years. However, this conclusion should be considered tentative because interval cancers are only a proxy measure of the efficacy of screening in terms of reduced breast cancer mortality.
Conclusion
Many factors other than breast cancer mortality should be considered in a comprehensive evaluation of screening-for instance, the benefit of being able to use breast-conserving surgery in patients with small tumours, a decreased need of systemic treatment, a reduced fear of breast cancer among women with a negative screening result, the risk of anxiety caused by screening, the risk of diagnosing biologically benign lesions as cancer, the cost, inconvenience, and harm of negative biopsies, and the cost of the screening programme. This overview has been focused on breast cancer mortality. The greatest benefit was among women aged 50-69, who showed a proportionate breast cancer mortality reduction of 29%. However, longer follow-up
than the 7-12 years so far is important because the analysis suggests that a putative benefit for younger women may not emerge until after 8-10 years of observation. We thank Prof Jan Ponten, one of the initiators of this project and chairman of the Research Board of the Swedish Cancer Society which generously supported this project; Anna-Lena Johansson (department of epidemiology and public health, Umea University) who handled the medical records, necropsy reports and other documents for blind reviewing by the End Point Committee; and Vivi-Maj Miren and Gunilla Andersson in Malino, Barbro Ahlqvist in Kopparberg, Gunnel Nygren in Ostergotland, Birgit Angelholm in Stockholm, and Agneta Eliasson in Gothenburg who collected the information required.
REFERENCES
Shapiro S, Strax P, Venet L. Periodic breast screening in reducing mortality from breast cancer. JAMA 1971; 215: 1777-85. 2. Rutqvist LE, Miller AB, Andersson I, Hakama M, Hakulinen T, Sigfrisson BF, Tabar L. Reduced breast cancer mortality with mammography screening: an assessment of currently available data. Int J Cancer 1990; 5 (suppl): 76-84. 3. Bailar JC III. Mammography: a contrary view. Ann Intern Med 1976; 84: 1.
77-84,2783. 4.
Jacobsson S, Lundgren B, Melander O, Noren T. Mass-screening of a female population for detection of early carcinoma of the breast. Acta
5.
Lundgren B, Jacobsson S. Single view mammography. Cancer 1976; 38:
Radiol 1975; 14: 424-32. 1124-29. 6. Tabár L, Fagerberg CJG, Gad A, Baldetorp L, Holmberg LH, Gröntoft O, et al. Reduction in mortality from breast cancer after mass screening with mammography. Lancet 1985; i: 829-32. 7. Andersson I, Aspegren K, Janzon L, Landberg T, Lindholm K, Linell F, et al. Mammographic screening and mortality from breast cancer: the Malmö mammographic screening trial. BMJ 1988; 297: 943-48. 8. Frisell J, Eklund G, Hellström L, Lidbrink E, Rutqvist LE, Somell A. Randomized study of mammography screening: preliminary report on mortality in the Stockholm trial. Breat Cancer Res Treat 1991; 18: 49-56. 9. Duffy S, Tabár L, Krusemo UB. Randomization by cluster in the Swedish two-county trial: results from Kopparberg and implications for interpretation. Presentation at Nordic Cancer Union Symposium
(Stockholm, Aug 17-19, 1989). L. QUEST: a program system for statistical and epidemiological data analysis. Umeå: Umeå University, 1990. 11. Anon. Mammographic screening for early detection of breast cancer. Stockholm: National Board of Health and Welfare, 1990. 10. Gustafsson
From The Lancet A
screw too
few
The mishap which overtook the Umbria in mid-Atlantic and occasioned widespread alarm both here and in America, though now happily passed, has emphasised the serious peril which is involved in trusting to the capabilities of a single propeller however excellently constructed and skilfully worked. A break-down must occasionally happen whatever measures may have been taken to prevent it and then if no provision has been made for efficiently navigating the disabled ship an emergency arises which is most painful to contemplate. In some of the best ships afloat the propeller consists of twin screws, and if one should become incapacitated the other can still be relied upon with no worse result than some small delay. We believe that we are correct in stating that upon one of the voyages which the City of New York made last year one of her screws became disabled, but the ship, instead of becoming unmanageable, was propelled by the uninjured screw alone at a rate of seventeen knots an hour. If this be correct, surely no sea-going steamer should be built without this most necessary equipment of
duplicated propelling machinery. (Jan 7, 1893)