Effect of mammographic service screening on stage at presentation of breast cancers in Sweden


  • Swedish Organised Service Screening Evaluation Group

    Corresponding author
    • American Cancer Society, Cancer Control Department, 1599 Clifton Rd. NE, Atlanta, GA 30329
    Search for more papers by this author
    • Authorship: Drafting Committee

      Stephen W. Duffy, Cancer Research UK and Queen Mary University of London. László Tabár, Central Hospital, Falun and University of Uppsala, Sweden. Tony H.H. Chen, National Taiwan University. Robert A. Smith, American Cancer Society. Lars Holmberg (Chair of management committee), Regional Oncology Center, Uppsala, Sweden. Håkan Jonsson and Per Lenner, Department of Radiation Sciences, University of Umeå, Sweden. Lennarth Nyström, Department of Public Health and Clinical Medicine, University of Umeå, Sweden. Sven Törnberg, Oncologic Center, Karolinska University Hospital, Stockholm, Sweden. Jan Frisell, Karolinska University Hospital, Stockholm, Sweden.

      Statistical Analysis

      Amy M.F. Yen, National Taiwan University. Li-Sheng Chen, National Taiwan University. Yueh-Hsiah Chiu, National Taiwan University. Chia-Yuan Wu, National Taiwan University. Hui-Min Wu, National Taiwan University. Chih-Chung Huang, National Taiwan University Jane Warwick, Queen Mary University of London. Levent Kemetli, Karolinska University Hospital, Stockholm. Patrick Chou, Queen Mary University of London.

      Project Leaders of the Screening Programs

      Stockholm Region

      Gunilla Svane and Edward Azavedo, Karolinska University Hospital. Helen Grundström and Per Sundén, Danderyd Hospital. Karin Leifland, St. Göran Hospital. Kerstin Moberg, Södersjukhuset. Tor Sahlstedt, Skärholmen.

      Umeå Region

      Pal Bordás, Norrbotten. Leena Starck, Västernorrland. Stina Carlson, Västerbotten. Håkan Laaksonen, Jämtland.

      Uppsala Region

      Shahin Abdsaleh and Erik Thurfjell, Uppsala. Birgitta Epstein and Maria Tholin, Örebro. Ewa Frodis, Västmanland. Ann Sundbom, Värmland. László Tabár, Dalarna. Mika Wiege, Sörmland. Anders Åkerlund and Bengt Lundgren (deceased), Gävleborg.

    • Fax: (404) 325-2548.



Previous results have shown a reduction in mortality with service screening in Sweden on the order of 40%. If the rate of tumors at a later stage were similarly reduced, this would give further support to the mortality findings.


The rates of lymph node-positive cancers, of tumors >2 cm in pathological size, and of tumors of TNM stage II or worse before and after the introduction of screening were compared in 13 areas in Sweden, adjusted for changes in overall incidence during the period of study and stratified by age (40–49 and 50–69 years).


Data were obtained on a total of 23,092 cancers and 10,177,113 person-years of observation. In women exposed to screening in the screening epoch, there was a significant 45% reduction in tumors of size >2 cm compared with the prescreening (relative risk [RR] = 0.55, 95% confidence interval [CI]: 0.46–0.66) in the 40–49 age group, and a 33% reduction in the 50–69 group (RR = 0.67, 95% CI: 0.62–0.72). For lymph node-positive and stage II+ disease, there were smaller but still significant reductions. No reduction in incidence in later-stage disease was observed in the unexposed women in the screening epoch.


Screening has significantly and substantially reduced the rates of larger tumors and lymph node-positive breast cancer in Sweden, and the magnitude of the reduction is consistent with the reduction in breast cancer mortality. Cancer 2007. © 2007 American Cancer Society.

For many years the randomized trials of mammographic screening have shown a substantial and significant reduction in breast cancer mortality in association with invitation to screening. 1, 2 Consequently, mammographic screening is now in phase 4, the era of long-term evaluation of its use in routine healthcare services. The Swedish Organised Service Screening Evaluation Group (SOSSEG) was convened for this purpose and has derived results for 20–40 years of observation on 1.1 million Swedish women, showing around a 30% reduction in breast cancer mortality with invitation to screening and a 40% reduction in those receiving screening.3, 4

Our previous analyses of breast cancer mortality trends took considerable pains to control for other changes over time that might have had confounding effects. Also, our previous results were consistent with other evaluations of service screening. 5–8 It would be of value, however, to investigate the degree to which the observed reductions in breast cancer mortality with screening were accompanied by similar reductions in the incidence of later-stage tumors, because screening works by diagnosing breast tumors earlier in their natural history, before they have grown large or spread to the regional lymph nodes.1, 9 If so, this would indicate that the rate of later-stage tumors could be used as a reliable indicator of the effect of a screening program before mortality results were available, as suggested by the randomized trials.1

In this article we ascertain the effect of offering screening, and of being screened, on the rates of lymph node-positive breast cancers, tumors more than 2 cm, and tumors of TNM-stage 2 or worse in 13 counties in Sweden. We estimate these effects in women aged 40–49, 50–69, and 40–69 combined.


Data on breast cancer incidence by tumor size, node status, and therefore TNM-stage before and after the introduction of mammography screening were obtained from the Regional Oncology Centers, backed up by patient charts held locally. Our 3 primary endpoints were incidence of lymph node-positive breast cancer, incidence of invasive disease with maximum diameter >2 cm, and incidence of cancer with TNM-stage II or worse. In our previous analyses of breast cancer mortality, we had to choose equal duration in the prescreening and screening epochs because of the nonconstant hazard of breast cancer deaths. Equal prescreening and screening time periods were not necessary for this study, which is concerned with the incidence of later-stage disease. Nor was it necessary that the prescreening and screening epochs be contiguous. The prescreening and screening epochs here were determined, first, to reflect actual screening coverage (ie, to have a prescreening epoch with close to zero coverage, and a screening epoch with close to 100% coverage), and second by the availability of reliable pathology data. Thus, for example, the Dalarna analysis omits the period 1978–1987 when around 59% of the eligible population were screened during the course of a randomized trial and instead compares 1988–1997 (88% coverage) with 1968–1977 (0% coverage). The Stockholm analyses omitted the 2 years 1988 and 1989 when the programs were being initiated but when coverage was still less than 10%.

Data on screening exposure in cancer cases and in the population were supplied by the screening services in the study areas. We studied the incidence of later-stage disease by the 3 definitions in women aged 40–49 and 50–69 at diagnosis, because around half of the screening programs in Sweden targeted the age ranges 40–69 and the remainder targeted the age group 50–69. In the current decade, most areas are adopting a lower age limit than 50 years, but this does not apply to the periods studied here. Around half of the areas also include ages 70–74. This age group is not dealt with here.

Table 1 shows the areas and epochs studied, with numbers of cancers and person-years, for ages 40–69 years. Five screening programs within Stockholm (Danderyd, Karolinska, St. Goran, Sodersjukhuset, and Skarholmen) are treated as separate areas. Screening was by 2-view mammography every 2 years in most areas. Further details of the screening programs are given elsewhere. 3, 4 Two areas in our previous analysis, Vastmanland and Gavleborg, were not included as we did not have prescreening pathology data. Two extra areas, Varmland and Vasterbotten, which did have the necessary data, are included.

Table 1. Dates, Total Numbers of Cancers, and Person-Years (Ages 40–69 Combined) for the Prescreening and Screening Epochs
AreaPrescreening epochScreening epoch
DatesCancers (PY)DatesCancers (PY)
Orebro1985–1987245 (142641)1988–20011516 (687023)
Uppsala1979–1989768 (391768)1990–20001254 (530158)
Sodermanland1985–1989380 (221957)1990–20001203 (518587)
Dalarna1968–1977707 (516318)1988–19971114 (513024)
Varmland1989–1993320 (154942)1994–2001869 (264952)
Norrbotten1980–1990679 (492572)1991–20011097 (531255)
Vasternorrland1980–1990818 (523876)1991–20011296 (522918)
Vasterbotten1990–1995341 (161359)1996–2001530 (170763)
Stockholm-Danderyd1977–1987714 (313835)1990–20001230 (361494)
Stockholm-Karolinska1977–1987561 (281072)1990–2000993 (337075)
Stockholm-St. Goran1977–1987858 (376673)1990–20001035 (328260)
Stockholm-Sodersjukhuset1977–19871247 (552993)1990–20001478 (520851)
Stockholm-Skarholmen1977–1987733 (364625)1990–20001096 (396162)

For the 40–49 age group we had data on a total of 926 cancers and 815,341 person-years in the prescreening epochs, and 1980 cancers and 1,279,654 person-years in the screening epoch. The corresponding figures for the 50–69 age group were 7445 cancers and 3,679,290 person-years in the prescreening epoch, and 12,741 cancers and 4,402,828 person-years in the screening epoch.

We calculated relative risks (RR) with 95% confidence intervals (CI) of later-stage disease defined by the 3 primary endpoints (lymph node-positive, >2 cm, and stage II+), for exposed, unexposed, and total populations in the screening epoch compared with the prescreening. The RRs were adjusted for the differing proportions with missing pathology data between epochs and for the increase in incidence over the period of study. The increased incidence was estimated using the unexposed women only in the screening epoch to avoid overestimation of the incidence in the screening epoch due to an excess of early-detected cancers in the exposed group.

We calculated adjusted RRs and 95% CIs as follows: Let u+, u, and um, respectively, denote the number of cases with later-stage disease, without later-stage disease, and with later-stage disease missing in the unexposed women in the screening epoch. Let e+, e, and em denote the corresponding numbers for exposed women and let b+, b, and bm represent the corresponding numbers for the prescreening epoch. Let PYu, PYe, and PYb denote the person-years in the unexposed in the screening epoch, exposed in the screening epoch and the prescreening epoch, respectively. The RR representing the change in underlying incidence between the prescreening epoch and the screening is calculated using the unexposed only in the screening epoch as:

equation image

The RR for the exposed women in the screening epoch relative to the prescreening epoch, assuming missing at random and adjusted for missing data and the change in incidence, is estimated as:

equation image

The variance of the logarithm of the RR is rather complicated due to covariances among the components, but after a little algebra it simplifies to:

equation image

The corresponding RR for the unexposed women is:

equation image

The variance of the log RR is estimated as:

equation image

The RR for exposed and unexposed women combined is:

equation image

The variance of the log RR is estimated as:

equation image

After calculation of the RR and 95% CI for each area, a combined estimate was obtained by inverse variance weighted average of the log RR. 10 The RRs were estimated separately for ages at diagnosis 40–49 and 50–69. We then derived combined estimates for the age group 40–69, first by inverse variance weighted average, and second by weighting the 2 estimates by the Swedish female population sizes in the 2 age groups. For demonstration purposes, we show area-specific results and their combination for the analyses of the incidence of tumors >2 cm in the 40–49 age group comparing exposed women in the screening epoch to the prescreening. Thereafter, we show combined results over all counties by endpoint, exposure group, and age.

We also plotted the cumulative incidence of later-stage tumors as defined by the 3 endpoints for all counties combined, adjusted for proportions with missing data and with the screening epoch data adjusted for the temporal increase in incidence, both similarly combined over all counties.


We first report the details of the analysis of the RR of tumors of size greater than 2 cm in women aged 40–49, in the exposed group in the screening epoch compared with the prescreening epoch. Table 2 shows the calculation of rates of tumors of size >2 cm in the 2 groups adjusted for missing information. Table 3 gives the crude RRs of tumors of size >2 cm and the relative incidence of breast cancer overall for unexposed women in the screening epoch compared with the prescreening epoch, as an estimate of the change in incidence over time. Table 3 also gives the RR of tumors of size >2 cm adjusted for this change in incidence over time. The adjusted RRs are similar for all counties, although for the northern counties (Norrbotten and Vasternorrland) the RR is the result of a substantially increased incidence with time overall, which is not reflected in the incidence of large tumors, whereas in the remaining counties it is a product of a lesser increase in overall incidence accompanied by a noticeable reduction in large tumor incidence. Figure 1 shows the same adjusted RRs, with 95% CIs calculated as described in the Methods section, and the overall estimate for the 6 counties, calculated by the inverse variance weighted average method.

Figure 1.

Meta-analysis of relative risks of tumors of size >2 cm in the age group 40–49, comparing exposed women in the screening epoch vs. prescreening epoch, adjusted for missing data and changes in incidence over time.

Table 2. Calculation of Missing-Data-Adjusted Rates of Cancers >2 cm for the Prescreening Epoch and Screening Epoch, Screening-Exposed Women Aged 40–49
Risk groupCountyNo. of cases >2 cmPerson-yearsProportion of cases with size missingAdjusted* rate/100,000
  • *

    Adjusted rate is the cases divided by the person-years, further divided by (1 − proportion missing).

Prescreening epochOrebro21531690.015440.1
Screening epoch, exposedOrebro541852750.007929.4
Table 3. Calculation of Relative Risk (RR) of Cancers >2 cm for Exposed Women Compared With the Prescreening Epoch, Ages 40–49, Adjusted for the Increase in Incidence
AreaPrescreening epoch − incidence/ 100,000 >2 cm (1)Screening epoch − exposed incidence/100,000 >2 cm (2)Relative overall incidence screening unexposed vs prescreening (3)Adjusted RR* (4)
  • *

    Adjusted RR is calculated as (4) = (2)/[(1)×(3)].


Table 4 shows the corresponding results for the 3 endpoints, the 3 exposure categories, and the 2 age groups. These results are based on all relevant data, including those areas inviting only women aged 50–69. For the 40–49 age group, significant reductions in all 3 endpoints were observed in the exposed women in the screening epoch, ranging from a 45% reduction (RR = 0.55, 95% CI: 0.46–0.66) in tumors of size >2 cm to a 29% reduction (RR = 0.71, 95% CI: 0.59–0.85) in lymph node-positive tumors. Similarly, significant reductions in all 3 endpoints were observed in the exposed women in the 50–69 age group, ranging from a 33% reduction (RR = 0.67, 95% CI: 0.62–0.72) in tumors of size >2 cm to a 16% reduction (RR = 0.84, 95% CI: 0.78–0.90) in lymph node-positive tumors. No significant differences were observed between the unexposed women in the screening epoch and the prescreening epoch. When exposed and unexposed were combined in the screening epoch as a whole, they showed significant reductions in all 3 endpoints, for both age groups. These ranged from a 36% reduction in tumors of size >2 cm in women aged 40–49 (RR = 1/4 0.64, 95% CI: 0.54–0.76) to an 11% reduction in lymph node-positive disease in the 50–69 age group (RR = 1/4 0.89, 95% CI: 0.84–0.95). The results for tumors at stage II or worse were mostly intermediate between those for lymph node-positive disease and those for tumors of size >2 cm.

Table 4. Combined Results for All Areas Studied by Age Group, Endpoint, and Exposure Category (Relative Risks [RRs] Compared With Prescreening Epoch), Adjusted for Missing Data and Changes in Incidence
AgeEndpointExposed RR (95% CI)Unexposed RR (95% CI)All screening epoch RR (95% CI)
  1. 95% CI indicates 95% confidence interval.

  2. Combined estimates over all ages are shown, first weighted by the traditional inverse variance method, and second weighted by the eligible Swedish population size in the two age groups.

40–49Node +0.71 (0.59–0.85)1.12 (0.89–1.42)0.77 (0.65–0.93)
Size >2 cm0.55 (0.46–0.66)1.11 (0.88–1.38)0.64 (0.54–0.76)
stage II +0.68 (0.57–0.81)1.07 (0.88–1.30)0.73 (0.63–0.86)
50–69Node +0.84 (0.78–0.90)1.07 (0.99–1.17)0.89 (0.84–0.95)
Size >2 cm0.67 (0.62–0.72)1.00 (0.91–1.09)0.74 (0.69–0.79)
stage II +0.79 (0.74–0.85)1.00 (0.93–1.07)0.84 (0.79–0.89)
All ages, inverse varianceNode +0.83 (0.77–0.88)1.08 (1.00–1.17)0.88 (0.83–0.94)
Size >2 cm0.65 (0.61–0.70)1.01 (0.93–1.10)0.72 (0.68–0.77)
stage II +0.78 (0.74–0.83)1.01 (0.94–1.08)0.83 (0.78–0.88)
All ages, Swedish populationNode +0.79 (0.73–0.86)1.09 (0.99–1.20)0.85 (0.79–0.92)
Size >2 cm0.63 (0.58–0.69)1.04 (0.94–1.14)0.70 (0.65–0.76)
stage II +0.75 (0.70–0.81)1.02 (0.94–1.11)0.80 (0.75–0.86)

Figure 2A–C shows the cumulative incidence of later-stage tumors by the 3 endpoints over time in the prescreening, exposed, and unexposed groups for ages 40–49. The corresponding results for ages 50–69 are shown in Figure 3A–C.

Figure 2.

(A) Cumulative incidence of lymph node-positive disease, ages 40–49 years. (B) Cumulative incidence of tumors of size >2 cm, ages 40–49 years. (C) Cumulative incidence of tumors stage II or worse, ages 40–49 years.

Figure 3.

(A) Cumulative incidence of lymph node-positive disease, ages 50–69 years. (B) Cumulative incidence of tumors of size >2 cm, ages 50– 69 years. (C) Cumulative incidence of tumors stage II or worse, ages 50–69 years.

Table 4 also shows the combined results for ages 40–69, first calculated by the traditional inverse variance weighted average, and second, weighting by the relative size of the Swedish female population. The methods give very similar results, and show significant 17% to 37% reductions in later-stage tumors in the exposed group, no substantial difference in the unexposed, and significant 12% to 30% reductions in the combined exposed and unexposed groups.


Our results showed significant reductions in the incidence of later-stage tumors in screening-exposed women in the screening epoch compared with the prescreening, adjusting for missing data and changes in overall incidence over time. These reductions varied from around 15% to 45%, depending on the age group and the definition of later-stage tumors. They are consistent with our published mortality reductions. 3, 4 In addition to improving the prognosis of cancers and thus saving lives, the reductions in later-stage disease imply an opportunity for less disfiguring and debilitating treatment. From a methodological point of view, our results also show that relatively simple measures such as the reduction in tumors larger than 2 cm or with positive lymph nodes can be used as early indicators of the effectiveness of screening.

The reductions in later-stage tumors are slightly more conservative than the reductions in breast cancer mortality, as one would expect, because there is considerable scope for variation in prognosis within the later and earlier stage tumors by our definitions. 11 For example, within the lymph node-positive tumors there is a strong relation between survival and number of positive nodes. TNM-stage 2 or worse could mean stage 2, 3, or 4. Also, there will be additional tumors due to lead time in the exposed women in our screening epoch which would have occurred symptomatically after the screening epoch in the absence of screening, and some of these will increase the numbers of later-stage tumors, whereas they would not as yet increase the numbers of breast cancer deaths in the previous mortality analysis. Missing or incomplete axillary node information may introduce a further conservative effect, because better staging practices in more recent years may have led to more tumors being identified as lymph node-positive. Finally, in 1 area, Stockholm Sodersjukhuset, there was a randomized trial of screening during the latter part of the prescreening period, and around 36% of subjects received screening in the course of the trial. This will have diluted the difference between epochs for this area, although the effect on the overall results is likely to be small.

There were greater reduction in rates of tumors larger than 2 cm than in rates of lymph node-positive cancers. This may be partly due to more sensitive lymph node staging in recent years. With increased specialization in breast surgery, improvements in the registration of the number of harvested nodes, and increased demands for more thorough staging for therapeutic decision-making in clinical oncology, it is likely that the lymph node staging is more thorough in the screening than in the prescreening epoch.

There was no reduction in the incidence of later-stage tumors in the unexposed group. Interestingly, there was no significant increase either, suggesting that the self-selection bias which we observed with mortality does not apply to the incidence of later-stage disease. This probably derives from the fact that although women who choose not to be screened have poorer stage at presentation, they also tend to have a lower incidence of breast cancer 12 and these 2 phenomena compensate for each other to some extent. The lack of any significantly increased risk of later-stage tumors in the unexposed group suggested that the correction for self-selection bias, which was necessary in our mortality analyses,3, 4 was not required here. This assumes that the increased incidence over time applied proportionally to attenders and nonattenders for screening. Also, because here we were not dealing with death from the disease but stage at diagnosis, there was no need to consider changes in therapy or other effects on survival from the disease.

Our adjustment for incidence was for incidence of all breast cancers, which implies an assumption that the underlying trend in later-stage breast cancers is the same as that for all breast cancers and that the trend for the unexposed is not affected by self-selection. The fact that the adjusted RRs in the unexposed group (Table 4) are close to unity suggests that this assumption was reasonable. There were, however, small and nonsignificant increases in later-stage tumors in the unexposed group aged 40–49, which were confined to the latter part of the screening epoch (Fig. 2A-C). The timing of this may be pure chance or there may, for example, have been an effect of the inauguration of the screening programs on breast awareness in the entire population, but the effect declined among the nonattender population thereafter.

We assumed that pathology data were missing at random in our adjustment for missing values. We carried out sensitivity analyses, first assuming all missing cases were later-stage and, second, assuming that none of the missing cases were later-stage. The results suggest that the missing at random assumption was on the conservative side. For example, assuming none of the missing cases for node status, age 40–49, was lymph node-positive gives an RR of 0.75, and assuming all were lymph node-positive gives 0.59. The result presented assuming missing at random was 0.71, nearer the conservative end of the range. This reflects the more complete pathology data in more recent years. For example, the proportions of cases with missing tumor size in the 40–49 age group were 13% in the prescreening epoch and 6% in the screening.

We observed greater reductions in later-stage tumor incidence in the 40–49 age group than in the 50–69 group. This is slightly surprising given that the randomized trials observed a greater mortality benefit in women aged 50 or more than in women aged 40–49. It should be noted, however, that in a similar evaluation of service screening in 2 counties in Sweden a slightly more favorable mortality effect was observed in the 40–49 age group. 13 Also, several of the screening centers aimed to screen the under-50 age group more frequently, with a target interval of 18 months, and attendance was better in the younger age group. In the centers studied in this work, screening coverage in the age group 40–49 was 82%. The corresponding figure for the 50–59 age group was 76%. However, the proportions of tumors screen-detected was higher in the 50–69 group, at 57%, compared with 50% in the 40–49 group. Another possible explanation is more careful staging in younger cases, throughout the period of observation, including the prescreening epoch. In terms of missing data, however, this seems to be at most a minor factor. In Dalarna, for example, the percentages with missing node status data in the 50–69 age group were 33% and 9% in the prescreening and screening epochs, respectively. In the 40–49 age group the corresponding figures were 28% and 5%. Work is in hand to analyze age-specific mortality effects in the 40–49 age group in the areas analyzed in this study to see how well they correspond to the later-stage tumor analysis presented here.

In conclusion, in 13 areas in Sweden we observed substantial and significant reductions in later- stage tumor incidence with organized service screening, adjusting for missing data and changes in incidence over time. These reductions were observed for women under age 50 as well as for women aged 50 and above. Reductions in tumors larger than 2 cm or with lymph nodes positive can be used as an early indicator of the effectiveness of screening.


We thank the women who participated in the screening programs and the staff of the screening centers