OUP user menu

Trends in breast cancer stage distribution before, during and after introduction of a screening programme in Norway

Mette L. Lousdal, Ivar S. Kristiansen, Bjørn Møller, Henrik Støvring
DOI: http://dx.doi.org/10.1093/eurpub/cku015 1017-1022 First published online: 4 March 2014

Abstract

Background: Screening is intended to advance diagnosis thereby shifting the stage distribution towards more locally confined stages. Consequently we aimed to estimate trends in stage-specific breast cancer in relation to the introduction of population-based screening. Methods: From the Cancer Registry of Norway we retrieved cancer stage, age and year of diagnosis on all women aged 20 or older diagnosed with breast cancer during the period 1987–2010 in Norway (approximate source population: 1.8 million). Three calendar-time periods were defined: before (1987–95), during (1996–2004), and after (2005–10) screening was introduced; and two age groups: women eligible for screening (50–69 years) or younger (20–49 years). Poisson regression was used to estimate the incidence of localized (stage I) and more advanced cancer (stages II+), respectively, and logistic regression to estimate the proportion of localized cancer. Results: The annual incidence of localized breast cancer among women aged 50–69 years rose from 63.9 per 100 000 before the introduction of screening to 141.2 afterwards, corresponding to a ratio of 2.21 (95% confidence interval: 2.10; 2.32). The incidence of more advanced cancers increased from 86.9 to 117.3 per 100 000 afterwards, corresponding to a 1.35 (1.29; 1.42)-fold increase. Advanced cancers also increased among younger women not eligible for screening, whereas their incidence of localized cancers remained nearly constant. Conclusion: Incidence of localized breast cancer increased significantly among women aged 50–69 years old after introduction of screening, while the incidence of more advanced cancers was not reduced in the same period when compared to the younger unscreened age group.

Introduction

Breast cancer constitutes a major health problem with currently more than 18 000 women diagnosed per year in the Nordic Countries leading to 15% of all female cancer deaths.1 From a public health perspective primary prevention is desirable, but in practice not feasible due to the wide range of risk factors of which many are intertwined with Western lifestyle such as reproductive factors.2 As a key component of secondary prevention, breast cancer screening is intended to advance the time of diagnosis and thereby improve prognosis.3 Treating the cancer at a more curable, locally confined stage should ultimately lead to a reduction in cause-specific mortality.

Indeed, randomized controlled trials conducted in Western countries during the 1970s and 1980s indicate that screening reduces cause-specific mortality by 25–30% among women aged 50–69 years.3,4 Although the randomization process has been questioned in some of the studies,5,6 a recent review concluded that there is general agreement that invitation to screening reduces breast cancer mortality by ∼20%.7 After the introduction of national population-based screening programmes in European countries, only observational studies can be used to estimate the reduction in mortality. A recent review considering all published European studies concludes that the reduction in breast cancer mortality associated with population-based screening programmes is in the range of 25–31% for women invited to screening.8 Whether this screening benefit of reduced mortality outweighs the potential harm in terms of overdiagnosis is however still debated.7,9 Overdiagnosis of breast cancer is defined as a cancer that would not have been diagnosed in a woman’s lifetime, had she not been screened.3 Several studies have investigated the mortality reduction, but to our knowledge only a few studies10,11 have considered stage distribution as an outcome of interest, and then only in subpopulations, although this is crucial for understanding whether screening shifts stage distribution downwards as intended. A decrease in advanced stage cancers is the best early indicator of the effect of screening.12 As stage distribution is an intermediary outcome for improving survival, trends in the stage distribution might help reassure screening as an effective prevention, if found to increase the incidence of localized cancers by decreasing the incidence of more advanced stages.

We therefore estimated the absolute and relative trends in stage-specific breast cancer incidence before, during, and after the gradual introduction of a national screening programme in Norway. We compared trends among women aged 50–69 years eligible for screening to the corresponding trends among younger women ineligible for screening.

Methods

Population

The study was designed as a population-based, open cohort study including all women, aged 20 years or older, diagnosed with a first-time, invasive breast cancer (ICD-10 code: C50) during the period 1987–2010 in Norway. Precursor lesions such as ductal carcinoma in situ are not systematically and consistently recorded by cancer registries and therefore not included.13 The Norwegian population was fairly stable over this period of time, and the size of the source population in 2010 was 1.8 million. The criterion for first-time incidence was that the woman did not occur with an earlier cancer in the Norwegian Cancer Registry, which holds information on all incident cancers back to 1953. Women who emigrated prior to diagnosis or had a missing end of follow-up date were excluded (0.2%). Data including stage at diagnosis, age at diagnosis and year of diagnosis were obtained from the Cancer Registry of Norway. Data on the size of the source population were drawn from Statistics Norway. All data were de-identified to avoid ethical issues.

Study variables

The Norwegian Breast Cancer Screening Programme started as a pilot project in one of 19 counties in November 1995 and was gradually expanded during the following 9 years to a nationwide programme, wherein all women aged 50–69 years are invited to biennial mammography screening.14 Three calendar-time periods were defined: before (1987–95), during (1996–2004) and after (2005–10) the complete introduction of screening. As the screening programme targeted the specific age group of women aged 50–69 years, they were chosen as the group of exposed, while women aged 20–49 years served as a control group for estimating temporal trends in background incidence and stage distribution. Since information on individual exposure status (being invited to the screening programme) was not available, we used age group and time period as proxies to exposure status (figure 1). Outcome was stage at diagnosis classified into localized (stage I) or more advanced: regional lymph node involvement (II), local infiltration into skin or chest wall (III), or distant metastatic spread (IV).15

Figure 1

Time axis showing the experimental nature of the study design in Norway, 1987–2010

Statistical analysis

The stage distribution in the study population was estimated for each age group and time period. Absolute annual stage-specific incidences per 100 000 women were calculated by age-stratified Poisson regression with time period as covariate and the average annual population size as time at risk. Corresponding incidence rates ratios were estimated with time period, age period and their potential interaction as covariates. Odds ratios for localized stage vs. more advanced stages were calculated using logistic regression with age group, time period, and potential interaction as covariates and converted to proportions to facilitate interpretation. In a supplementary analysis we expanded the group of exposed to all women aged 50–79 years i.e. it included the entire age group that over the study period could potentially have been invited to screening and benefitted from it. This gives a more fair comparison of the development over time between the two age groups, since adding the initial increase in incidence due to lead-time and the subsequent decrease among older women no longer offered screening, the so-called compensatory drop, reduces lead-time bias.16 In the absence of overdiagnosis the initial increase would be fully compensated for by a later decrease.16 We further performed a sub-analysis to address the potential impact of missing information on stage at diagnosis for 14% of the women. The missing stage information was imputed in a multiple logistic model with survival time, age at diagnosis, and year of diagnosis as predictors. To allow use of survival time as predictor we imputed it when right censored using stage, age and year of diagnosis as predictors. Additionally, we conducted two extreme sensitivity analyses replacing missing information as either all localized or all more advanced stage cancer, respectively. Estimates are reported with 95% confidence intervals and are deemed significant at a 5% level. All statistical analyses were conducted in STATA® version 12.

Results

A total of 56 277 first-time, invasive breast cancers were identified among women aged 20 years or older in Norway during the period 1987–2010. Among them 105 women who emigrated prior to diagnosis and one with missing end of follow-up date were excluded, as were, for the main analysis, 7905 women for whom recorded stage at diagnosis was unknown. Having missing information on cancer stage was more common among elderly and in the time period 1996–2004 (table 1). The study population for the main analysis thus consisted of 48 266 cases.

View this table:
Table 1

Distribution of stage at diagnosis in each age group and time period of 56 171a first-time, invasive breast cancers in Norway, 1987–2010

Age1987–951996–20042005–10
(Years)n(%)n(%)n(%)
20–49I1368(36.0)1383(31.6)1073(33.2)
II1731(45.6)2198(50.3)1678(51.9)
III152(4.0)164(3.8)160(5.0)
IV206(5.4)184(4.2)90(2.8)
Unknown341(9.0)441(10.1)230(7.1)
Total3798(100.0)4370(100.0)3231(100.0)
Source population (Person-years)8 182 5768 446 3915 731 493
50–69I2338(38.4)4622(41.9)4499(52.5)
II2461(40.4)4135(37.5)3161(36.9)
III279(4.6)257(2.3)321(3.7)
IV438(7.2)421(3.8)255(3.0)
Unknown570(9.4)1594(14.5)340(4.0)
Total6086(100.0)11,029(100.0)8576(100.0)
Source population (Person-years)3 657 8654 065 1723 186 331
70+I2543(35.7)1871(25.5)1301(28.1)
II2289(32.2)2440(33.2)1772(38.4)
III464(6.5)381(5.2)287(6.2)
IV508(7.1)543(7.4)293(6.3)
Unknown1315(18.5)2107(28.7)967(20.9)
Total7119(100.0)7342(100.0)4621(100.0)
Source population (Person-years)2 657 0402 784 1161 804 736
  • a48 075 cases classified as stage I–IV, 191 cases classified due to a more general stage classification as either I or IV, and 7905 missing.

The annual incidence of localized breast cancer among women aged 50–69 years increased from 63.9 per 100 000 before the introduction of screening to 141.2 per 100 000 afterwards, corresponding to an incidence rate ratio of 2.21 (95% confidence interval: 2.10; 2.32) (table 2). The incidence of more advanced stages increased from 86.9 before to 117.3 per 100 000 afterwards, corresponding to an increase by a factor 1.35 (1.29; 1.42). This overall increase reflects that the incidence of both stage II and III increased, whereas the incidence of stage IV decreased. There was no indication of an increase in incidence resulting from a prevalence peak due to initial screening rounds in women aged 50–54 years, as their incidence increase was similar to the one among 55–69 years. In the younger age group the annual incidence of localized cancer increased from 16.7 per 100 000 before screening to 18.7 afterwards with an incidence rate ratio of 1.12 (1.03; 1.21), and for more advanced cancers the incidence increased from 25.5 before to 33.6 afterwards with a ratio of 1.32 (1.24; 1.40). The increases in incidence were similar among 20–39 vs. 40–49, which we therefore combined into one control group. The change in incidence of localized stage was significantly higher in the age group eligible for screening than in the younger age group ineligible for screening with a relative ratio of 1.97 (1.80; 2.17), or P < 0.001, comparing 50–69 vs. 20–49, whereas the changes for more advanced stages were similar for the two with a ratio of 1.02 (0.95; 1.11), or P = 0.542. During the study period the proportion of localized stage cancers among all incident breast cancers rose in the age group invited to screening from 42% to 55%, while the proportion in the younger age group decreased from 39% to 35% (figure 2).

Figure 2

Trends in stage-specific absolute incidence in Norway, 1987–2010, with stages indicated by I, II, III and IV

View this table:
Table 2

Absolute annual incidence per 100 000 women, incidence rate ratios and the relative ratio in Norway, 1987–2010

50–69 years20–49 yearsRelative ratio
1987–951996–20042005–101987–95 vs. 2005–10a1987–951996–20042005–101987–95 vs. 2005–10a50–69 vs. 20–49b
IR before (95% CI)IR during (95% CI)IR after (95% CI)IRR 50–69 (95% CI)IR before (95% CI)IR during (95% CI)IR after (95% CI)IRR 20–49 (95% CI)IRR50–69 / IRR20–49 (95% CI)
Localized63.9 (61.4; 66.6)113.7 (110.5; 117.0)141.2 (137.1; 145.4)2.21 (2.10; 2.32)16.7 (15.9; 17.6)16.4 (15.5; 17.3)18.7 (17.6; 19.9)1.12 (1.03; 1.21)1.97 (1.80; 2.17)
More advanced86.9 (83.9; 90.0)118.4 (115.1; 121.8)117.3 (113.6; 121.1)1.35 (1.29; 1.42)25.5 (24.5; 26.6)30.1 (29.0; 31.3)33.6 (32.2; 35.2)1.32 (1.24; 1.40)1.02 (0.95; 1.11)
– II67.3 (64.7; 70.0)101.7 (98.7; 104.9)99.2 (95.8; 102.7)1.47 (1.40; 1.55)21.2 (20.2; 22.2)26.0 (25.0; 27.1)29.3 (27.9; 30.7)1.38 (1.29; 1.48)1.07 (0.98; 1.16)
– III7.6 (6.8; 8.6)6.3 (5.6; 7.1)10.1 (9.0; 11.2)1.32 (1.13; 1.55)1.9 (1.6; 2.2)1.9 (1.7; 2.3)2.8 (2.4; 3.3)1.50 (1.20; 1.88)0.88 (0.67; 1.16)
– IV12.0 (10.9; 13.1)10.4 (9.4; 11.4)8.0 (7.1; 9.0)0.67 (0.57; 0.78)2.5 (2.2; 2.9)2.2 (1.9; 2.5)1.6 (1.3; 1.9)0.62 (0.49; 0.80)1.07 (0.80; 1.43)
Total150.8 (146.9; 154.8)232.1 (227.5; 236.8)258.5 (253.0; 264.1)1.71 (1.66; 1.77)42.2 (40.9; 43.7)46.5 (45.1; 48.0)52.4 (50.5; 54.3)1.24 (1.18; 1.30)1.38 (1.30; 1.47)
  • aRatio describing the development, when comparing the time period before with the period after screening has been implemented.

  • bRelative ratio comparing the development among 50–69 vs. the development among 20–49.

When the group of exposed was expanded to include women aged 50–79 years, the increase in incidence of localized stage changed to an incidence rate ratio of 1.72 (1.65; 1.79) and remained significantly higher than the corresponding change in the younger age group with a relative ratio of 1.53 (1.40; 1.68), or P < 0.001. The change in more advanced stages was reduced to an increase of 1.22 (1.17; 1.27) indicating a reduction in incidence when compared to the younger age group—the relative ratio comparing changes over time for 50–79 years vs. 20–49 years was 0.93 (0.86; 1.00), or P = 0.040.

The multiple imputation of missing data led to virtually identical results as the complete case analysis. Even in the two extreme cases the increase in localized cancers among women aged 50–69 years remained significantly higher, whereas the increase in more advanced cancers was similar to the increase in the younger age group. Results are presented in the Supplementary Data.

Discussion

Findings

The annual incidence of localized breast cancer among women invited to screening more than doubled from 63.9 per 100 000 before the introduction of a national population-based screening programme to 141.2 per 100 000 afterwards in Norway. This increase in early stage cancers among women aged 50–69 years was twice as high as the corresponding development in the younger age group. By contrast the changes in incidence of more advanced stage cancers were nearly identical for the two age groups with a ratio of 1.02 (0.95; 1.11). When corrected for lead-time bias, trends persisted although the increases in both localized and more advanced stages were lower among women aged 50–79 years. Before screening was introduced 42% of all incident breast cancers were of localized stage rising to 55% afterwards in the age group eligible for screening. This increase was observed despite the concurrent increase in the number of more advanced stages.

Strength and limitations

A total of 14% of the study population had no recorded stage at diagnosis, which potentially could affect validity, because we found it to be related to increasing age as have others.13 When however, we allowed for this in a multiple imputation analysis, results were virtually identical to the results of the analysis based only on cases with recorded cancer stage, and even in the extreme sensitivity analyses the trends persisted. As exposure we considered invitation to the screening programme, and not attendance, since this yields estimates of effectiveness in a real world setting with non-compliance. This analytic approach is preferable from a population perspective. An evaluation of the Norwegian screening programme estimated that 76.2% of invited women attended screening in the period 1996–2005.14 This implies that the intended effect of shifting the stage distribution is likely diluted in our study because the group of exposed comprises women not participating in the screening programme. Likewise, any overdiagnosis will tend to be underestimated in our study in comparison to the hypothetical situation with full participation by all invited.

Due to the high quality of the Cancer Registry of Norway information on breast cancer incidence is 99.95% complete.17 The stage classification system has been changed over time in order to agree with international standards when converting TNM-classification into stage I–IV,18 which could lead to time dependent misclassification of the outcome, but would however be expected to apply equally across age groups. Improved diagnostics have likewise lead to changes over time in the way tumours are classified, known as stage migration,19 but this misclassification is also likely to apply independently of age.

Data did not allow us to correct for the use of hormone replacement therapy (HRT), although it is known to be associated with a higher risk of breast cancer.20 Women aged 55–59 years have the highest consumption of HRT preparations in Norway, and the sale reached a peak plateau in 1997–2002.21 Therefore the increased breast cancer incidence observed in women aged 50–69 years might in part be explained by HRT.21 Another factor that affects the stage-specific incidence is the prevalence of opportunistic screening before and concurrently with the introduction of the national routine screening programme. At least 40% of Norwegian women regularly underwent mammography prior to their first invitation to the screening programme,22 which will diminish the intended effect of shifting the stage distribution as well as any overdiagnosis, when comparing the periods afterwards and before the introduction of the national programme. Additionally the stage-specific incidence may be influenced by other temporal changes like public awareness and enhancements in mammographic imaging. None of these changes would however be expected to apply in isolation for only one of the age groups.

Comparison with other studies

In 2006, Hofvind et al.23 investigated how stage-specific incidence changed in relation to the introduction of the screening programme in two different areas in Norway. The incidence of localized cancer among women aged 50–59 years increased markedly at the time when screening was introduced. Although decreasing in the following years the incidence remained at a higher level. The incidence of more advanced cancers was constant in one area, while in the other area it first increased at the introduction of screening and then later decreased. These results are consistent with the findings of our study, but in addition our study analysed the stage distribution in the time period after the nationwide introduction of screening.

Esserman et al.10 estimated stage distribution in the USA, where screening was introduced in 1983 and more widely used from 1986 and onwards. When comparing the distributions in 1982 and 1998 the proportion of localized stage increased over time, while the incidence of more advanced cancer dropped as a fraction of all cancers detected, but the absolute numbers did not decrease as much as hoped. The explanation offered is that screening increases the detection of indolent cancers, but might miss the most aggressive ones. Bleyer et al.11 investigated trends in stage distribution from 1976 through 2008 in the USA. The incidence of early-stage breast cancer including ductal carcinoma in situ doubled, while the rate of late-stage cancers only decreased by 8%. This imbalance suggests substantial overdiagnosis.11

A recent review16 included 13 primary studies that estimated the magnitude of overdiagnosis in population-based screening in seven Western European countries. Studies, with plausible correction for underlying breast cancer risk and lead-time bias, estimated the magnitude of overdiagnosis to range from 1% to 10% of the expected incidence in the absence of screening. Another recent study,24 not included in the review, estimated the magnitude of overdiagnosis in exactly the Norwegian setting. The staggered implementation was used to correct for contemporaneous trends, and lead-time bias was corrected through two different approaches. The magnitude of overdiagnosis was estimated at 15–25% of the actual observed cancer incidence. A further finding was that the incidence of stage III and IV decreased among women aged 50–69 years after the introduction of screening, but this change was similar to the development in the non-screened group of equal age. They did not consider the stage distribution in the younger age group. Their findings correspond with our findings of a similar development for more advanced stages in women aged 50–69 years and the younger group of women not eligible for screening. Falk et al.25 investigated overdiagnosis in the same Norwegian setting, but used a different approach based on individual data with either actual attendance or invitation to the screening programme as exposure. They estimated overdiagnosis at 10–20% of the expected incidence of breast cancer in the absence of screening. Autier et al.12 investigated in a review the incidence of advanced breast cancer in areas where mammography screening has been practised for at least 7 years with a high degree of participation. However they found no noticeable change in incidence and consequently concluded that screening did not play a substantial role in decreasing the mortality.

Interpretation

This study suggests that the introduction of a national screening programme leads to more localized stage cancers being diagnosed, but this is not mirrored by a corresponding decline in more advanced cancers among women aged 50–69 years, even when compared to corresponding trends among women aged 20–49 years ineligible for screening. It seems unlikely that this increase in localized cancers can be explained by changes in lifestyle such as nutrition and physical activity or environmental carcinogens, since no such increase is seen in the younger age group influenced by the same temporal trends. While the group eligible for screening and the younger control group differ in terms of age and reproductive factors, it should be noted that we do not compare the absolute incidences as such, but rather the development over time in each group. Other explanations for the increase in localized stage among women aged 50–69 years such as opportunistic screening, higher public awareness, enhancements in mammographic imaging, or changes in classification do not seem plausible, as they would apply equally across age groups. The remaining main differences between groups are thus the use of HRT and enrolment in the screening programme. This corresponds with a recent study modelling trends in invasive breast cancer between 1987 and 2008, which claims that a combination of HRT and mammography screening can explain recent changes in incidence.26 The incidence of breast cancer is expected to decrease in response to the declining use of HRT from 2003 and onwards,21 but we did not find such a reduction, possibly because the effect of declining HRT use was counterbalanced by other factors. The introduction of the national screening programme remains as an explanation for the continued increase in localized cancers among women aged 50–69 years. Still uncertainty remains as to what extent the increase in localized cancers is due to overdiagnosis. Although the expected decrease in more advanced cancers has not yet become manifest, one cannot rule out that this change might appear in years to come if one assumes that the tumours are very slow-growing.

Conclusion

These results suggest that the implementation of a national screening programme in Norway led to a significant increase in incidence of localized stage cancers, while the incidence of more advanced stages did not change significantly among women aged 50–69 years, when compared to a younger control group ineligible for screening. Whether this trend in stage distribution is a common phenomenon needs to be studied in other similar settings. While we have estimated the population impact of introducing a breast cancer screening programme, a comparison of trends between participants and non-participants in the age group eligible for screening warrants further investigation. Also the causal link between stage distribution and mortality needs to be investigated in the context of screening.

Supplementary data

Supplementary data available at EURPUB online.

Conflicts of Interest

None declared

Key points

  • Implementation of a national screening programme for breast cancer in Norway led to a significant increase in localized stage cancers.

  • The incidence of breast cancer in more advanced stages was not reduced, also when compared to the younger age group.

  • Data are compatible with overdiagnosis.

  • Women invited to screening should be informed about potential benefits and harms to be able to make an informed choice about participation.

Acknowledgements

Previously orally presented at the European Congress of Epidemiology in Aarhus, Denmark, 12/8-2013.

References

View Abstract