Int. J. Cancer: 68,693-699 (1996) C 1996 Wiley-Liss, Inc.

Publicatlon of the International U n m Against Cancer !!!@& Publication de I Union Internationale Contre le Cancer

BREAST-CANCER SCREENING WITH MAMMOGRAPHY IN WOMEN AGED 40-49 YEARS Report of the Organizing Committee* and Collaborators, Falun Meeting, Falun, Sweden (21 and 22 March, 1996).

For some years, there has been a perceived need for more information on the effect of screening for breast cancer in women aged 40 to 49. Our approach was to gather the most recent data on screening in this age group, to assess the following quantities: the likely beneffi in mortality terms, measures of screening performance and arrest of tumour progression through screening, costs and public-health implica- tions, and prospects for future screening and research. A collaborative meeting was held in Falun, Sweden, for which data were gathered in advance from all the randomized trials of breast-cancer screening that included women in this age group, and all identifiable substantial databases on service screening of women aged 40 to 49. Updated results from the Swedish overview of mammographic screening trials indicated relative mortality associated with invitation to screening of 0.77 (95% confidence interval 0.59-1.01). Combining all population-based randomized trials gave the relative-mortality figure of 0.76 (0.62-0.93), and combining all trials gave 0.85 (0.7 I- I .O I). Relative-mortality figures in individual trials could be predicted by the rates of advanced cancers in those trials. Detailed analysis suggested faster tumour progression in the age group 40 to 49 compared with groups aged 50 or more. It is likely that mammographic screening of women aged 40 to 49 can reduce subsequent mortality from breast cancer, and studies on tu- mour progression indicate that to obtain substantial benefe it is probably necessary to screen every 12 to 18 months, with 2-view mammography and double reading of films. This repre- sents a greater outlay in resources and in commitment of the population. Further work remains to be done on the time taken for a mortality benefit to emerge, on age at diagnosis in some of the trials, and on costs and public-health aspects. o 1996 Wiley-Liss, Inc.

While the potential benefit of mammographic screen- ing for breast cancer is established for women aged 50 years or more, there remains controversy over the effect of screening women aged under 50. Results for women aged 40 to 49 vary substantially among randomized trials of mammographic screening (Shapiro et al., 1971; Tabar et al., 1985; Anderson et a/., 1988; Frisell et al., 1991 ; Miller et al., 1992; Alexander et al., 1994). In addition, the late appear- ance of a reduction in mortality has led some researchers to question whether the benefit might be due to tumours detected at screening after the individual women have attained age 50 (De Koning et al., 1995). Overview analyses in this age group, whether of individual data or of published results, have not attained the unequivocal statistical signifi- cance achieved in older groups (Wald et al., 1993; Nystrom et al., 1993, 1995; Smart et al., 1995). This had led to considerable debate, some of it quite acrimonious (Mettlin and Smart, 1993; Baines, 1994). There is no prospect of results from new trials aimed specifically at this age group until several years from now. A considerable body of data, however, has been built up over the years, from existing randomized trials and from routine service screening pro- grammes, on all aspects of breast-cancer screening in the age group 40 to 49. It may be that fuller examination of this information than is normally given by a simple meta- analysis of mortality results will settle some of the points of controversy and illuminate previously unremarked areas of common ground.

Accordingly, a collaborative conference on this sub- ject, under the auspices of the Swedish Cancer Society and the Swedish National Board of Health and Welfare was

held in Falun, Sweden, on March 21st and 22nd, 1996. The objectives and preparations were as described below.

1. To gather the most recent data from randomized trials and non-randomized screening programmes world- wide. To this end, data were requested in advance of the meeting from the randomized trials and service screening programmes in women aged 40 to 49. Data solicited included number of women in the screening rogramme

screening regime offered, and numbers and attributes (size, node status and grade where available) of cancers diag- nosed by trial arm or mode of diagnosis (screening or interval). Data were also requested on aspects of screening procedure, including interval between screens, number of views, number of mammogram readers, and use of the grid.

2. To analyze the updated results, in particular those of the Swedish overview. The methodology of the Swedish overview has already been described (Nystrom et al., 1993). At the Falun meeting, results of the overview were u dated to 31 December 1993. In addition, Mantel-Haenszermeth- ods were used for combination of the mortality results of different trials (Mantel, 1963).

3. 'To evaluate parameters of screening practice and outcome and their variability from programme to pro- gramme. The data on the programme populations and the cancers diagnosed were analyzed by Markov chain models (Durn et al., 1995; Chen et al., 1996) where sufficient information was available, and by generalized linear mod- els otherwise (Paci and D u e , 1991), to produce estimates of the following screening efficacy and natural-history parameters: mean sojourn time (duration of the period in which a tumour is asymptomatic but screen-dctectable), average lead time attained, incidence of advanced cancers (e.g:, node-positive, diameter 2 cm or more), sensitivity, positive predictive value, predicted mortality based on the size node status and (where available) malignancy grade of tumours diagnosed, and rates of tumour progression in terms of node status and malignancy grade. Predicted mortality is based on the prognostic characteristics of the tumours observed within each study applied to the survival rates observed for those tumours in the 2467 cancers in the Two-County Study. All other parameters are estimated from data internal to each study. Further details of the statistical analysis are available from HHC or SWD.

In relation to these parameters it should be noted that, in this context, positive predictive value means the propor- tion of tumours diagnosed at the first screen which would subsequently have arisen clinically had screening not taken

lace. It does not refer to the proportion of positive giopsies. Thus it has the meaning usually assigned in the context of diagnostic test assessment. The estimates of tumour progression rates, in conjunction with the survival

'Names and affiliations of the committee members and the collabo-

(and control group in the case of trials), atten B ance rates,

rators are listed in the appendix.

*Correspondence and reprint requests should be addressed to Dr L. Tabar, Director, Mammography Department, Central Hospital, 79 182 Falun, Sweden.

Received: July 16,1996.

694 FALUN MEETING

data from the Swedish Two-County Study, enable us to predict the effect on mortality of different screening re- gimes (Tabar ef al., 19956; Day and Durn, 1996). The potential for de-differentiation or phenotypic drift was also addressed. According to this hypothesis, a tumour with a heterogeneous malignancy-grade pattern has more rapid progression of the more poorly differentiated com onent, which in turn becomes the dominant part, ancfhence “de-differentiates” (Tabar ef al., 19966).

4. To discuss public-health aspects and cost-benefit considerations. This included financial and human costs of screening and of not screening, and the differential in the resource and human costs and benefits between screening women under age 50 and screening women aged 50 years or more.

5. To summarize the current state of knowledge and discuss implications and guidelines for the future. This would include establishing issues on which there can be agreement at the moment, and enumerating which issues remain unresolved. The aim was to contribute practical information to aid decisions concerning age groups to be screened and which (if any) screening regimen is indicated for the 40 to 49 age group. Other questions addressed included the likely mortality benefit in this age group and when it was likely to be achieved, also what indications existed of succcss or failure in advance of the mortality results?

PROCEEDINGS OF THE MEETING

1 . Assimilating the most recent data. Table I shows all the screening studies, trials and service programmes, which

formed the database for the meeting, with the screening regimes. These are from data sent to the committee in advance of the meeting. This constitutes a formidable body of information on 244,000 women aged 40 to 49 invited to screening, although the items on which data are available vary from project to project. Table I1 shows breast-cancer deaths and person-years, and relative risks (RR) of breast- cancer mortality for the trials only.

2. Updated results of trials, in particular those of the Swedish overview. Figure 1 shows the updated results of the randomized trials of screening invitation against no invitation, with overall results, first for the Swedish over- view, then for all population-based trials, then for all trials. The Swedish overview yields an estimated R R of breast- cancer mortality of 0.77 (95% CI 0.59-1.01), in association with invitation to screening. The recent reductions in mortality in Gothenburg and reported in the Two-County Study are mainly responsible for this result. Note that the R R differ slightly from those in Table I1 because of the recoding of causes of death in the Swedish overview and the fact that the Swedish overview data is updated uniformly to the end of 1993. Table I1 data are from the original trials, and follow-up varies, for some studies being as recent as mid-1995. The result for all population-based randomized trials is an RR of 0.76 (95% C.I. 0.62-0.93). The result for all randomized trials (note that the data from the Canadian NBSS are interim and subject to confirmation) is an RR of

The results overall are strongly suggestive of a benefit in the region of a 15 to 25% reduction in mortality, although the confidence intervals are wide.

0.85 (95% C.I. 0.71-1.01).

TABLE 1 - DETAILS OF SCREESING PROGRAMMES PARTICIPAIING

Use of grid Views’ Readers’ Age range Number invitedlcontrol (trials only) Programme

.-

(a) Trials3 2-County 4 0 4 9 19844/15604 No 24 1 1 Malmo 4 5 4 9 3795/3769 No 21 2/1 2 Stockholm 4 0 4 9 14842/7 103 No 28 1 1 Gothenburg 4 0 4 9 10821/13101 Yes 18 2/1 1 Edinburgh 45-49 11370/10269 Yes 24 2/1 1 HIP 4 0 4 9 14432/14701 No 12 2 2 NBSS 4 0 4 9 25214/25216 Partly4 12 2 112

Turku 4 0 4 9 10568/8791 Yes 12/36 2 2

Uppsala 4 0 4 9 30657 Yes 20 2/1 1 Vastmanland 4 0 4 9 18462 No 24 2/1 2 Florence 75-85 4 0 4 9 12604 No 24 2 1 British Columbia 4 0 4 9 63196 Yes 15 1 1

(b) Trials3

(c) Service

‘2/ 1,2-view on first screen, 1-view subsequently, unless first screen indicates 0therwise.-~1/2, one reader at screening, 2 readers for suspicious rnammogram~.-~(a) Trials are invited vs. not invited, with the exception of the NBSS, in which a control group with a single baseline physical examination is compared with a study group with annual mammography and physical examination. (b) Trials are of annual vs. 3-yearly ~creening.-~In the latter part of the trial and not in all centres.

TABLE I1 - BREAST-CANCER MORTALITY REDUCTIOSS IK TRIALS IN AGE GROUP 4049

Years of Study-group Study-group Conlrol-group Control-group Relative Trial follow-up deaths person-years deaths person-years mortality

2-County 15 45 264,059 39 207,725 0.91 Malmo 15 15 61,000 23 62,000 0.66 Stockholm 12 25 173,866 12 87,826 1.05 Gothenburg 10 19 106.000 37 129.000 0.62 Edinburgh 10 25 971206 31 881766 0.73 HIP 18 49 248,454 65 253,085 0.77 NBSS’ 10 73 252.060 66 251.814 1.10 Turku’ 7 3 48;068 9 411532 0.29

’Interim data.-2Annual vs. 3-yearly screening.

9596 C1

0.59-1.39

0.53-2.09 0.36-1.08

0.34-1.27

0.43-1.25 0.53-1. I 1 0.78-1.54 0.07-1.08

BREAST-CANCER SCREENING, AGES 4049 695

3. Parameters of screening practice and outcome and their variability. The parameters of screening practice in Table I do not correlate particularly well with the mortality reductions observed in the trials, although there is a suggestion that 2-view mammography and double reading are associated with slightly better results on average; since only 2 of the trials comparing no-invitation with invitation used the grid, it is difficult to assess its effect on outcome. A shorter interval between screens may be associated with a better result, although the NBSS is an exception to this.

Malmo

Koppakrg (W)

Ostergotland (E)

Siockholm

Gorhenburg

AU Swedish trials

Edinburgh

HIP

AU population-based trials

Canadian NBSS

All triah

FIGURE 1 -Relative mortality in the age group 40 to 49 from breast cancer in randomized trials of breast-cancer screening (invitation vs. no invitation), with overall results from the Swedish trials, all population-based trials, and all trials.

Estimated measures of efficacy and tumour progres- sion are shown in Tables 111 and IV. Note that these, while varying substantially among programmes, differ markedly from those typically observed when screening women aged 50 or more (Paci and D u e , 1991; Tabar et al., 19956, 19966; Chen, 1995; Brekelmans et al., 1996). Sensitivity estimates tended to be poorer and tumour progression in the pre-clinical phase to be faster in women aged 40 to 49. The latter suggests that these measures are informative in relation to the tumours and their biological background, rather than to the screening. Progression by size or node status is also more rapid in this age group. Markov modelling of the Two-County Study data indicated that a

re-clinical node-negative tumour has a 16% probability of gecoming a clinical node-positive tumour within one year in the age group 40 to 49. For the age groups 50 to 59 and 60 to 69, the corresponding probabilities are 7% and 5%.

The measures of efficacy, which correlated strongly with the effect on mortality in the randomized trials, were the relative incidence rate of node-positive tumours, rela- tive incidence of tumours 2 cm or larger in diameter, and the predicted effect on mortali from the size, node status

trials. The agreement between observed and predicted relative mortality is shown in Figure 2. The results show good agreement, whether or not the Two-County Study, from which the survival data were derived, is included. More complete modelling of tumour progression in the Two-County Study by size, node status and malignancy grade, and subsequent suwival (Tabar et al., 19956; Chen, 1995) yielded the predicted effects on mortality by age group and interscreening interval shown in Table. Clearly, for any substantial benefit to be achieved in women aged 40 to 49, screening should be more frequent than every 2 years.

Results also indicate that there is greater potential for turnour de-differentiation in women a ed 40 to 49 than in

service screening programme and the British Columbia Service screening programme, the estimated proportions of tumours in women aged 40 to 49 with the potential to de- differentiate were 81%, 100% and 88% respectively. The corres ondin proportion in women aged 50 to 59 in the

of the interval between screens in women aged 40 to 49. 4. Public health aspects and costs. Although screening

with a frequency of 18 to 24 months has a potential to

and grade (where available) o ? tumours diagnosed in the

older women. In the Two-County 8 tudy, the Uppsala

Two- 8 8 ounty tudy was 48%. This suggests the importance

TABLE 111 - SELECTED MEASURES OF EFFICACY IN THE PARTICIPATIKG PROGRAMMES'

Programme

(a) Trials 2-County Malmo Stockholm Gothenburg Turku Edinburgh HIP NBSS

(b) Service Uppsala Vastmanland Florence British Columbia

Range trials) Range [service) Tvoical5O+ values

Sensifivity

83% 81% 65 %

100% 100% 76% 75 % 77%

58% 75 % 69% 61%

65-100% 58-75% 90-100%

ppv2 Lead time RII Node- (months) positive

. -

85 % 23 0.84 82% 21 0.56 59% 7 0.98

100% 17 0.61 100% 20 - 61% 18 0.82 85 % 14 77% 18 1.40

68% 11 0.754 81% 19 0.754

61% 9 0.804 59-100% 7-23 0.56-1.55 61-100%) 7-19 0.75-0.80

-

100% 7

90-100% 24-36 0.65-0.80

RI 2cm+

0.90 0.56 0.87 0.59

0.76

1 .oo 0.774 0.764

1.174 0.56-1.38 0.76-1.17 0.65-0.80

- -

Predicted relafive mortality -

0.97 0.61 1.19 0.76

0.80

1.23

O X i 4

- -

0.904

0.85-0.90 0.65-0.80

0.61-1.22

'Left blank if unknown or not applicable.-2PPV, positive redictive value.-'RI, relative incidence rate study/~ontrol.-~No control group: estimated from MarKov chain model.

696 I;AI.UN M

TABLE N - MEASURES OF TUMOUR PROGYESSION I V 1 THE SCKEENING PROGRAMMES

IEETING

Mean sojourn Progression Progression Programme time to node-posit~ve~ 10 grade 3’ __ -~

(a) Trials 2-County Malmo

30 0.10 0.29 28 0.18

Stockholm 15 0.43

Turku (1-vr) 24 0.06 Gothenburg 22 0.02

Edinburgh ‘ HIP

26 0.14 22

NRSS 22 (b) Service

Uoosala 18 0.06 0.31 V%manland 27 0.12 Florence 17 British Columbia 20 0.12 0.88

Range (trials) 15-30 0.02-0.43

TvDical50+ values 36-48 0.05-0.10 0.05-0.10 Range (service) 17-27 0.06-0.12

‘Left blank if unknown.-2Probability of a node-negative tumour becoming node-positive within one year.-3Probability of a grade-1 or -2 tumour becoming grade-3 within one year.

1.2 I /--- I

0.4 1 0.2 ~

0 - - ._~_._________. 1 0.5 0.7 0.9 1.1 1.3

Predicted RM

FIGURE 2 - Relative mortality observed in the MalmO, Gothen- burg, Edinburgh, Two-County, Stockholm and Canada trials, plotted against the mortality predicted from the size, node status and malignancy grade (where available) of the cancers diagnosed in the trials. X, Two-County Study.

reduce mortality from breast cancer in the age group 40 to 49 by 15 to 25% (see Fig. l ) , results in Tables I1 and V suggest that the mortality reduction in the age group 40 to 49 is more dependent on the screening interval than is the case in older women. A screening interval of 12 to 18 months appears to be necessary to achieve a substantial effect. In addition, 2-view mammography and double read- ing are indicated. This requires a substantial outlay of resources and considerable commitment from the invited population, to undergo intensive screening and selective diagnostic activity in return for the mortality benefit. It has been argued that the side effects in terms of false positives, the detection and treatment of biologically insignificant cancer, mainly ductal carcinoma in situ, and radiation- induced breast cancer (although this last is controversial) may outweigh the benefit. In the Swedish trials, the recall rate at first screen ranged from 4.2 to 6.7% and at second screen from 4.1 to 4.9%. The proportion of benign biopsies ranged from 46 to 74% at first screening and from 36 to 50% at the second. These results are not particularly alarming, but there is clearly room for improvement.

The mortality benefit in the age group 40 to 49 is smaller in absolute terms of number of deaths avoided than that derived from less intensive screening in older women.

TABLE V - EXPECTED RELATIVE MORTALITY FROM 3-YEARLY. 2-YEAR1.Y AND ANNUAL SCREENING BY AGE GROYP

(ASSUMING 10% KEFUSAL RATES TIIROUGHOUT)

Interval W Y 50-59 60-69 between screens (83% sensitivitv) ~100% scnsitiviiv) (100% sensitivitvl

1 year 0.64 0.54 0.56 2 years 0.82 (0.87) 0.61 0.61 3 years 0.96 0.66 (0.66) 0.66 (0.60)

‘Figures in parentheses are relative risks actually observed in the Two-County Study.-*Sensitivity estimates are those observed in the Two-County Study.

There are, however, other considerations, such as the number of life-years gained and the quality of life achieved through early detection in younger women. The decision whether or not to offer screenin in this age group will be based on availability

benefit in relation to the human and financial costs. 5. Current state of knowledge and implications for the

future. The results described above have 2 major implica- tions. Firstly, they strongly suggest a mortality benefit in the age group 40 to 49, albeit a smaller and more variable one than that observed in older women. Secondly, the mortality benefit takes longer to appear in younger women. This is consistent with results at different times of follow-up in the HIP study (Shapiro et al., 1988). In the Swedish overview, after around 12 years of follow-up, a 13% reduction in mortality was observed (Nystrom el al., 1993). In the present results, after an additional 3 to 4 years of follow-up, the reduction in women aged 40 to 49 was 23%. This is due mainly to changes in the results from Gothenburg and the Two-County Study. The major contribution of Gothenburg to these results may be due to the relatively short screening interval of 18 months.

An ex lanation for the delayed benefit in this age

cancer, the more rapid progression of tumours in the age group 40 to 49, and in the failure of a relatively long interscreening interval to cope with these, as can be seen from the histology-specific mortality. Figure 3 shows cumu- lative mortality from (a) ductal grade-3 breast cancer, (b ductal grade-2, lobular and medullary carcinoma, and (c all breast cancer by study group in Kopparberg County of the Swedish Two-County Trial, in the age group 40 to 49. The corresponding results for the age group 50 to 74 are given in Fi ure 4. The mortalily benefit for grade-3 ductal

age roup, and does not appear convincingly in the 40 to 49. Foraoth age groups, the mortality benefit in terms of ductal grade-2, lobular and medullary carcinomas a ears at 7 to 8

mortality benefit starts to appear at about 8 years, when the benefit in terms of the ductal grade-2, lobular and medul- lary carcinomas emerges. In women aged 50 and over, the benefit in terms of all breast-cancer mortality appears at 5 years, when the reduction in mortality from ductal grade-3 tumours appears in this age group. Thus, if mortality from grade-3 carcinoma is reduced by early detection, the deaths prevented include some that would have occurred relatively soon after diagnosis. Reducing mortality from the less aggressive ductal grade-2, lobular and medullary carcino- mas prevents deaths that would have occurred in the longer term. The implication of Figures 3 and 4 is that screening with the average 2-year interval used in the Two-County trial for the 40 to 49 age group is not, generally, preventing early deaths from grade-3 carcinoma, but is preventing later deaths from less aggressive turnours.

Further examination of age-specific results from the Swedish overview suggests that the problem is in the early

of resources an d subjective judgement of the value of the

group can g e found in the heterogeneous nature of breast

1

carcinoma % egins to appear after only 5 years in the 50 to 74

years. In women aged 40 to 49, the overal PE reast-cancer

697 BREAST-CANCER SCREENING, AGES 40-49

6- 4 8 140 L 120

0

7

a, -G 100 ._ - +- 80

60 ' a, 40 + 20 2 0

m

> m ._ -

1 2 3 4 5 6 7 8 9 1011 1 2 1 3 1 4

- . /---- f I 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 0

(4 350 I I

300 250 200 150 100 50 0

I . - ]

1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4

Years since randomization

FIGURE 3 - Cumulative mortality from (a) ductal grade-3 breast cancer, (b) ductal grade-2, lobular and medullary breast cancers and (c) all breast cancers in women aged 40 to 49, Kopparberg County of the Two-County trial, b trial arm (ASP active study population, invited to screening; PS6 passive study population, not invited).

fifties rather than in the age group 40 to 49, and will be the subject of a scparate publication. This is consistent with the findings of the Two-County Study, in which substantial reductions in mortality were observed in women aged 40 to 44 and 55 to 74, but not 45 to 54 (Tabar et al., 199%).

One unresolved issue identified as a target for future investigation within the current data was the quantification of that part of the mortality benefit observed in women aged 40 to 49 at randomization due to screening tests after age 50 years (De Koning et al., 1995). This was not the case, however, in the empirical results of the Two-County Study (Tabar et al., 1996~). Also, in the HIP Study, most of the women in the invited group aged 40 to 49 at entry with tumours diagnosed in the trial period were still aged 40 to 49 at diagnosis (Shapiro et al., 1988). It remains to be seen whether the results of other trials with respect to age at

/ I

loo 0 I /--/ 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4

?5 9 0" 300

& 250

0

7

a v

> 200 + ._ -

150

E 100

..

I 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 t 3 0

(4 800 I I

700 600 500 400 300 200 100

0 0 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 i 4

Years since randomization

FIGURE 4 - Cumulative mortality from (a) ductal grade-3 breast cancer, (b) ductal grade-2, medullary and lobular breast cancers and (c) all breast cancers in women agcd 50 to 74 in Kopparberg County of the Two-County trial, by trial arm (ASP, active study population, invited to screening; PSP, passive study population, not invited).

diagnosis agree with the Two-County Study and HIP Study results. Circumstantial evidence on the subject of age at diagnosis can be derived by considering age groups 40 to 44 and 45 to 49 separately. This has been done for the Two-County Study (Tabar el al., 1995a), in which there was a benefit in the age group 40 to 44 but not in the 45 to 59 group; a similar analysis is planned for the Swedish over- view as a whole.

Another issue o f major importance in the monitoring and evaluation of breast-cancer screening is the assessment of the effectiveness of screening in advance of the results on mortality. This has implications both for ongoing and for planned trials, and for the evaluation of service screening programmes. The results with respect to predicted mortal- ity from size, node status and malignancy grade show that the effect of a screening programme can be accurately

698 FALUN MEETING

predicted before mortality results are available (Fig. 2), although prediction can be conservative if there is an excess of tumours in the study group, as in the NBSS and the Two-County Study.

A similar level of prediction can be achieved using a single measure, such as the incidence of node-positive tumours or of large tumours, but the inclusion of the histological type and malignancy grade gives information on when the mortality benefit is likely to occur.

It was commented that these results at least increas- ingly suggest that a benefit in terms of mortality can accrue from screening in this age group. There is evidence that, to be successful, the screening technique must be of high quality, should use 2-view mammography and double read- ing of films, and should have an interscreening interval of no more than 18 months, preferably one year. The decision whether or not to screen this age group will depend on

subjective views of the importance of the benefit and the ability to provide the resources involved. It seems clear, however, that while the size and timing of the mortality reduction require further research, the existence of such a reduction is no longer in question.

ACKNOWLEDGEMENTS We thank the Swedish Cancer Society and the Swedish

National Board of Health and Welfare for their support. We also thank all those working in the participating screening programmes. Thanks are also due to the staff of the Mammography Department, Falun Central Hospital, and to Mr. T. Tabar for their help in organizing the meeting. Every effort has been made to reflect the proceed- ings of the meeting accurately in the conclusions above, but these do not necessarily accord with the opinions of all the committee members or collaborators.

REFERENCES

ALEXANDER, F.E., ANDERSON, T.J., BROWN, H.K., FORREST, A.P.M., HEPBURN, W., KIRKPATRICK. A.E., MCDONALD, C., MUIR, B.B., PREscorr, R.J., SHEPHERD, S.M., SMI-~H. A. and WARNER, J., The Edinburgh randomized trial of breast-cancer screening: results after 10 years of follow-up. Brit. J. Cancer, 70,542-548 (1994). ANDERSON, I., ASPEGRLN, K., JANZON, L., LANDBERG, T., LINDHOLM, K., Liwi-i-, F., LJUNGBERG, 0.. RANSTAM, J. and SIGFUSSON, B., Mammographic screening and mortality from breast cancer: the Malmo mammographic screening trial. Brit. rned. J., 297, 943-948 (1988). BAISES. C.J., The Canadian National Breast Screening, Study: a perspective on criticisms. Ann. intern. Med., 120, 326334 (1994). BREKELMANS. C.T.M., WESTERS, P., FABER, J.A.J., PEETERS, P.H.M. and COLLETTE, Il.J.A., Age-specific sensitivity and sojourn time in a breast-cancer screening programme (DOM) in the Netherlands: a comparison of different methods. J. Epidemiol. Comm. lllth., 50,68-71 (1996). CHEN, H.H., Mathematical models for progression of breast cancer and evaluation of breast-cancer screening. PhD Thesis, University of Cam- bridge (1995). CHEN, H.H., DUFI:Y, S.W. and TABAK, L., A Markov chain method to estimate the tumour-progresson rate from pre-clinical to clinical

hase, sensitivity and positive predictive value for mammography in E reast-cancer screening. Statistician 45,307-3 17, (1996). DAY, N.E. and DUFFY, S.W., Trial design based on surrogate end points - ap lication to comparison of different breast-screening frequen- cies. J. R. Jatist. SOC. A , , 159,4960 (1996). DEKONING, H.J., BOER, R., WARMERDAM, P.G., BEEMSTERBOER, P.M. and VAN DER MAAS, P.J.. Quantitative interpretation of age-specific mortality reductions from the Swedish breast-cancer screening tria1s.J. nut. Cancerlnst., 87,1217-1223 (1995). DUFFY, S.W., C I I ~ N , H.H., TABAR, L. and DAY, N.E., Estimation of mean so'ourn time in breast-cancer screenin using a Markov chain model 0) both entry to and exit from the pre-cfinical detectable phase. Statist. Med., 14,1531-1543 (1995).

L.E. and SOMELL, A., Randomisation study of mammogra hic screen- ing - preliminary report on mortality in the Stockholm frial. Breast CancerRes. Treat., 18,49-56 (1991). MANTEL, N., Chi-square tests with one degree of freedom: extensions of the Mantel-Haenszel procedure. J. Amer. stat. Ass., 58, 690-700 (1963). MEITLIN, C.J. and SMART, C.R., The Canadian National Breast Screening Study: an appraisal and implications for early detection policy. Cancer, 72,1461-1465 (1993). MILLER, A.B., BAINES, C.J., TO. T. and WAIL, C. , Canadian National Breast Screening Study: 1. Breast-cancer detection and death rates amon women aged 40 to 49 years. Can med. Assoc. J., 147,1459-1476 (1992f Erratum Published in Can. med. Ass. J., 148,718 (1993).

FRISELI., J., EKLUND, G. , HELLSTROM, L., LIDBRISK, E., RUTOVIST,

NYSTROM, L., LARSSON, L.G., RWOVIST, L.E., LINDGREN, A., LISDOVIST, M., RYDEN, S., ANDERSON, I., BJURSTAM, N., FAGERRERG,

G., FRISELL, J. and TABAR, L., Determination of cause of death among breast-cancer cases in the Swedish randomized mammography screen- ing trials. Acta oncologica, 34, 145-152 (1995).

M., RYDEN, S., ANDERSSON, I . , BJURSTAM, N., FAGERBERG, G., FKISEIL, J., TABAR, L. and LARSSON, L.G., Breast-cancer screening with mammography: an overview of the Swedish randomised trials. Lancet, 341,973-978 (1993). PACI, E. and DUFFY, S.W., Modelling the analysis of breast-cancer screenine oroerammes: sensitivitv. lead time and oredictive value in

NYSTROM, L., RCIOVIST, L.E., WALL, S., LINDGREN, A., LINDOVIST,

the Flo&ke "District Programme (1975-86). Int.'J. Epidemiol., 20, 852-858 (1991). SHAPIRO, S., STRAX, P. and VENET, L., Periodic breast-cancer screen- ing in reducin mortality from breast cancer. J. Amer. rned. Ass., 215, 1777-1785 (19%). SHAPIRO, S., VENET, W., STRAX, P. and VENET, L., Periodic screening for breast cancer: the health insurance plan project and irs sequelae 1963-86. Johns Hopkins University Press, Baltimore (1988). SMART, C.R., HESDRICK, R.E., RUTLEDGE, J.H. and SMITH, R.A., Benefit of mammo raphy screening in women ages 40 to 49 years - current evidence from randomized controlled trials. Cancer, 75,

TABAR, L., DUFFY, S.W. and CHEN, H.H., Re: quantitative interpreta- tion of age-specific mortality reductions from the Swedish breast- cancer screening trials. J. nat. Cancer Inst., 88,52-53 (1996~). TABAR, L., FAGERBERG, G.. CHEN, H.H. and DCFPY, S.W., Screening for breast cancer in women aged under 50: mode of detection, incidence, fatality and histo1ogy.J. rned. Screening, 2,94-98 (1995a). TABAR, L., FAGERHERG, G., CHEN, H.H., DCFFY, S.W. and GAD, A., Tumour development, histology and grade of breast cancers: prognosis and progression. Int. J. Cancer, 66,413419 (19966). TABAR, L., FACERBERG, G., CHES, H I I . . DUFFY, S.W., SMART, C.R., GAD, A. and SMITH, R.A., Efficacy of breast-cancer screening by a e new results from the Swedish two-county trial. Cancer, 75, 2507-2f17 ( 19956). TABAR, L., FAGERBERG, C.J.G., GAD, A., BALDETORP, L., HOLMBERG,

EKLUSD, G., DAY, N.E. and PETTERSON, F., Reduction in mortality from breast cancer after mass screening with mammography. Lancer, I,

WALD, N., CHAMBERLAIS, J. and HACKSHAW, A., Report of the European Society for Mastology: Breast-Cancer Screening Evaluation Committee. Breast, 2,209-216 (1993).

1619-1626 (1995).

L.H.,GRONTOFT, 0.. LJCSGOUIST, u., LL;NI)STROM, B., MKNsoN,J.C.,

829-832 (1985).

APPENDIX

Organizing committee

Sweden. Laszlo Tabar (convenor), Central Hospital, Falun,

Lars-Gunnar Larsson, University of Umei, Sweden.

BREASI-CANCER SCREENING, AGES 40-49 699

Ingvar Andersson, Malmo General Hospital, Malmo,

Stephen W. Duffy, MRC Biostatistics Unit,

Lennarth Nystrom, University of Umed, Sweden. Lars Erik Rutqvist, Karolinska Hospital, Stockholm,

Sweden.

Cambridge, UK.

Sweden.

Collaborators Freda Alexander, University of Edinburgh, UK. Cornelia J. Baines and Anthony B. Miller, University

Nils Bjurstam, Sahlgrenska Hospital, Gothenburg, of Toronto, Canada.

Sweden.

Hsiu-Hsi Chen and Nicholas E. Day, MRC Biostatis-

Gunnar Fagerberg, University of Linkoping, Sweden. Jan Frisell, South Hospital, Stockholm, Sweden. Pirjo Immonen-Raiha, City of Turku, Finland. Eva Frodis, Vastmanland Breast Screening

Eugenio Paci, Centro per lo Studio e la Prevenzione

Sam Shapiro, HIP trial, New York, USA. Erik Thurfjell, University of Uppsala, Sweden. Stig Wall, University of UmeA, Sweden. Linda Warren Burhenne, University of British

tics Unit, Cambridge, UK.

Programme, Sweden.

Ongologica, Florencc, Italy.

Columbia, Canada.