Background:

Consensus has been reached on the effectiveness of inviting women aged 50 to 69 years to mammography screening, but for older women, the evidence is scarce. The aim of this study was to estimate the marginal effectiveness of inviting women to mammography screening with an upper age limit of 74 years versus stopping at age 69 using data from the Swedish service-screening program.

Methods:

A cohort design was used to compare the breast cancer mortality in the period 1986 to 2012 between geographic areas and periods where women were invited to screening up to the age of 74 years (study group) with those where women were invited up to age 69 (control group). The study group and the control group were compared using the incidence-based breast cancer mortality rate ratio where only breast cancer deaths in cases diagnosed at 70 to 74 years of age were counted.

Results:

After 20 years of follow-up, there were 1,040 and 1,173 breast cancer deaths in the study and the control group, respectively. The breast cancer mortality rate ratio for women invited up to age 74 versus women invited up to age 69 was 0.80 [95% confidence interval (CI): 0.75–0.85] after bias adjustments. The corresponding rate ratio for participating women was 0.73 (95% CI: 0.66–0.81).

Conclusions:

Continuing to screen women up to 74 years of age is effective compared with stopping screening at 69 years.

Impact:

This large long-term study will add to the knowledge of the effect of mammography screening for women 70 to 74 years.

Randomized controlled trials (RCT) have shown that mammography screening of asymptomatic women reduces breast cancer mortality in women aged 50 to 69 years by 20% to 25% (1, 2). Therefore, guidelines (3) on breast cancer screening recommend inviting women aged 50 to 69 years.

Only the Two-County trial (4) invited women aged 70 to 74 years to mammography screening, but the estimates for this age group were inconclusive due to low power. Today, women 70 to 74/75 years and older are invited to mammography screening in France, the Netherlands, Italy, Sweden, United Kingdom, Canada, and United States (5).

The first observational study of the effectiveness of inviting women 70 to 74 years to screening in Sweden showed a nonsignificant 6% reduction in breast cancer mortality and a 24% reduction in excess mortality after 10 years of follow-up (6). A similar study with 11 years of follow-up in northern Sweden found a small nonsignificant effect (7). A study including seven Canadian Breast Cancer Programs between 1990 and 2009 showed that the standardized mortality ratio (SMR) for women aged 70 to 79 years was 0.65 [95% confidence interval (CI): 0.56–0.74; ref. 8].

According to Statistics Sweden, in 2019, women aged 74 years have an expected remaining lifetime of 14.4 years. In general, elderly women with early detected breast cancer have the energy and resilience to tolerate appropriate modalities of treatments (9).

The Swedish guidelines on breast cancer screening recommend screening for women aged 40 to 74 years every 18 to 24 months. Because of local politics and inadequate resources, the guidelines were modified in 1987 and 1988 so that the minimum age range for invitation was 50 to 69 years. This resulted in a variation of age limits, even within counties, such that the upper limit was either 69 or 74 years, and the lower age limit was either 40 or 50 distributed in an almost 1:1 ratio. Since the start of the Swedish service-screening program with mammography in 1986, about half of the screening areas in the country have invited women with an upper age limit 74 years, while the other half have invited women up to 69 years of age. This has created a natural experiment that facilitates an estimation of the effectiveness of inviting women aged 70 to 74 years to screening. A similar natural experiment to evaluate the effectiveness of mammography screening for women aged 40 to 49 was conducted in 2011 (10).

The aim of this study was to investigate the marginal effectiveness of population-based mammography service-screening with an upper age limit of 74 years compared with stopping at age 69 using data from the Swedish service-screening program with mammography.

Service-screening program with mammography in Sweden

In 1986, the Swedish National Board of Health and Welfare issued guidelines on breast cancer screening. They recommended that the county councils invite women aged 40 to 74 years to mammography screening every second year. The program was introduced gradually from 1986 and was implemented nationwide in 1997. Because of local politics and inadequate resources, the guidelines were modified in 1987 and 1988 so that the minimum age range for invitation was 50 to 69 years. This resulted in a variation of age limits, even within counties, such that the upper limit was either 69 or 74 years distributed in an almost 1:1 ratio.

Database

A questionnaire was sent to the screening program in the 21 counties to obtain information on the characteristics of the programs over time such as date of start, geographic coverage, age limits, screening interval, and attendance rate. Individual data on date, age, and place of residence at diagnosis of the breast cancer cases were retrieved from the Cancer Registry, and information on date and cause of death were retrieved from the Cause of Death Registry. Breast cancer was defined according to the International Classification of Diseases (ICD-7 = 170) combined with pathologic codes C24 = 094, 096, 196, or 996 thus including both invasive and noninvasive breast cancers. Breast cancer as the underlying cause of death was defined according to site-codes 174 (ICD-9) and C50 (ICD-10) depending on the year of death. For each breast cancer death, individual data on invitation and attendance to the two latest screening invitations preceding the date of diagnosis were retrieved from the corresponding screening center. Aggregated population data by year, age, and municipality were retrieved from Statistics Sweden and were used to calculate the number of person-years.

Cohort design

The cohorts were defined at population level, and aggregated population data were used for calculation of person-years. Thus, the only individual information used was for the breast cancer cases.

All 21 counties in Sweden were included in the study. Some had to be split due to different upper age limits or time of start of the screening program. Consequently, 15 counties were used at the county level, while the remaining 6 counties were split into 23 areas. For simplicity, the 15 counties and 23 areas are all called areas (Fig. 1).

Figure 1.

Map of Sweden illustrating the counties contributing to the study group, the control group, and both the study group and control group.

Figure 1.

Map of Sweden illustrating the counties contributing to the study group, the control group, and both the study group and control group.

Close modal

To be able to see an effect of screening, we have to wait for cases in the control group to first get the diagnose and then die from breast cancer. It is obvious that choosing a too short follow-up interval would dilute the effect. We chose a minimum of 6 years, which was the same as we used for evaluation of screening in the 40- to 49-year age group (10). For all areas, we defined periods (area-periods, AP) for areas that invited (study group) or did not invite (control group) women in the 70- to 74-year age group to screening. Some areas (Gotland, Kronoberg, and Västerbotten) contributed to two non-overlapping APs, one to the study group and one to the control group. A start point was defined for each AP, which for the study-group APs was the start of screening of the 70- to 74-year age group in that area. In Gävleborg, the service-screening program started in 1974 as a pilot project, but women were not included into the study group until 1997 when the program was adjusted to the Swedish guidelines. For the control group APs, the start point was chosen either as the start of the screening program with an upper age limit of 69 years or at an earlier time chosen with the purpose of achieving similar average follow-up times in the study and control groups. The end points of the APs were defined as either the end of the study period or as the time when screening stopped for the 70- to 74-year age group (study group) or before the start of screening of the 70- to 74-year age group (control group).

For each AP, the cohorts were defined as all women aged 74 years or younger at the start point and at least age 70 at the endpoint. For women aged 70 to 74 years at the start point, this time point was also the cohort entry (start of follow-up), but for women under 70 years the entry was when they reached the age of 70 years (Fig. 2). The study cohorts were based on the APs with screening at ages 70 to 74, and the control cohorts were based on the APs without screening in that age group. The study cohorts were merged into a study group, and the control cohorts were merged into a control group. The numbers of breast cancer deaths and person-years in the cohorts were then aggregated.

Figure 2.

Lexis diagram illustrating AP, cohort definition, and follow-up. S is the start point (start of follow-up), and E is the end point. The dashed area shows the follow-up for breast cancer death. Example of Västernorrland County, where S is 1990 and E is 2012.

Figure 2.

Lexis diagram illustrating AP, cohort definition, and follow-up. S is the start point (start of follow-up), and E is the end point. The dashed area shows the follow-up for breast cancer death. Example of Västernorrland County, where S is 1990 and E is 2012.

Close modal

Follow-up

Women diagnosed with breast cancer in the 70- to 74-year age group were followed through the age of 89 years, death, or end of follow-up of each area (December 31, 2012, at the latest), whichever occurred first (Fig. 2). The weighted average lengths of follow-up for the study and control groups were 20.1 and 19.9 years, respectively, calculated as the total number of person-years divided by the total average population. The weighted average mid-points in the study and control cohorts were December 2000 and April 1998, respectively, calculated using the average population as weights. The length of the study period in the cohorts varied with a maximum of 27 years (1986–2012; Table 1).

Table 1A.

Year of start of screening program of women aged 40/50–69 and 70–74 years; start, end, and length of follow-up; number of person-years; and incidence-based breast cancer deaths in the study period 1986–2012 in the study group by county/subdivision of county.

Year of start of screeningFollow-up periodNo. of
County/subdivision40/50–6970–74StartEndLength (years)Person-yearsBreast cancer deaths
Gävleborg 1974 1977 Jan. 1997 Dec. 2012 16 235,855 32 
Gotland 1997a 2000 Jan. 2000 Dec. 2012 13 33,375 
Halland 1989 1992 May. 1992 Dec. 2012 20.7 313,994 67 
Jönköping (Högland) 1986 1986 Oct. 1986 Dec. 2012 26.3 202,956 56 
Jönköping (Habo, Mullsjö) 1989 1999 Jan. 1999 Dec. 2012 14.0 8,425 
Jönköping (Ryhov) 1987 1987 Apr. 1987 Dec. 2012 25.8 257,806 49 
Jönköping (Värnamo) 1987 1987 Nov. 1987 Dec. 2012 25.2 49,757 10 
Kalmar 1986 1986 Oct. 1986 Dec. 2012 26.3 418,956 69 
Kronoberg 1990 1999 Sep. 1999 Dec. 2012 13.3 113,370 14 
Norrbotten (Luleå) 1989 1989 Mar. 1989 Dec. 2012 23.8 81,350 15 
Norrbotten (Others) 1990 1990 Jan. 1990 Dec. 2012 23.0 253,162 36 
Örebro (Örebro, Kumla, Hallsberg) 1987 1987 Oct. 1987 Apr. 1996 8.6 58,892 
Östergötland 1986 1986 May. 1986 Dec. 2012 26.7 661,630 116 
Skåne (Malmö) 1990 1999 Jan. 1999 Dec. 2012 14 167,106 40 
Skåne (Northeast) 1989 1989 Sep. 1989 Dec. 2012 23.3 243,255 43 
Skåne (Ängelholm) 1989 1990 Apr. 1990 Dec. 2012 22.8 91,504 14 
Skåne (Helsingborg) 1989 1989 Sep. 1989 Dec. 2012 23.3 263,260 78 
Skåne (Southeast) 1987 1987 Apr. 1987 Dec. 2012 25.8 157,357 23 
Skåne (Southwest) 1987 1987 Mar. 1987 Dec. 2012 25.8 112,702 33 
Skåne (Middle) 1989 1989 Mar. 1989 Dec. 2012 23.8 328,094 77 
Södermanland 1989 1989 Nov. 1989 Dec. 2012 23.2 364,977 69 
Uppsala (Enköping, Håbo, Tierp, Älvkarleby) 1988 1988 Oct. 1988 Mar. 2009 20.5 86,834 18 
Uppsala (Östhammar) 1988 1988 Nov. 1988 Mar. 2009 20.4 24,787 
Uppsala (Uppsala, Knivsta) 1988 1988 Feb. 1988 Mar. 2009 21.2 165,454 45 
Västerbotten 1995 2000 Jan. 2001 Dec. 2006 47,124 
Västernorrland 1990 1990 Jan. 1990 Dec. 2012 23 366,026 66 
Västra Götaland (Bohus) 1986 1986 Nov. 1986 Dec. 1998 12.2 159,801 46 
Total      5,267,809 1,040 
Weighted mean     20.1   
Year of start of screeningFollow-up periodNo. of
County/subdivision40/50–6970–74StartEndLength (years)Person-yearsBreast cancer deaths
Gävleborg 1974 1977 Jan. 1997 Dec. 2012 16 235,855 32 
Gotland 1997a 2000 Jan. 2000 Dec. 2012 13 33,375 
Halland 1989 1992 May. 1992 Dec. 2012 20.7 313,994 67 
Jönköping (Högland) 1986 1986 Oct. 1986 Dec. 2012 26.3 202,956 56 
Jönköping (Habo, Mullsjö) 1989 1999 Jan. 1999 Dec. 2012 14.0 8,425 
Jönköping (Ryhov) 1987 1987 Apr. 1987 Dec. 2012 25.8 257,806 49 
Jönköping (Värnamo) 1987 1987 Nov. 1987 Dec. 2012 25.2 49,757 10 
Kalmar 1986 1986 Oct. 1986 Dec. 2012 26.3 418,956 69 
Kronoberg 1990 1999 Sep. 1999 Dec. 2012 13.3 113,370 14 
Norrbotten (Luleå) 1989 1989 Mar. 1989 Dec. 2012 23.8 81,350 15 
Norrbotten (Others) 1990 1990 Jan. 1990 Dec. 2012 23.0 253,162 36 
Örebro (Örebro, Kumla, Hallsberg) 1987 1987 Oct. 1987 Apr. 1996 8.6 58,892 
Östergötland 1986 1986 May. 1986 Dec. 2012 26.7 661,630 116 
Skåne (Malmö) 1990 1999 Jan. 1999 Dec. 2012 14 167,106 40 
Skåne (Northeast) 1989 1989 Sep. 1989 Dec. 2012 23.3 243,255 43 
Skåne (Ängelholm) 1989 1990 Apr. 1990 Dec. 2012 22.8 91,504 14 
Skåne (Helsingborg) 1989 1989 Sep. 1989 Dec. 2012 23.3 263,260 78 
Skåne (Southeast) 1987 1987 Apr. 1987 Dec. 2012 25.8 157,357 23 
Skåne (Southwest) 1987 1987 Mar. 1987 Dec. 2012 25.8 112,702 33 
Skåne (Middle) 1989 1989 Mar. 1989 Dec. 2012 23.8 328,094 77 
Södermanland 1989 1989 Nov. 1989 Dec. 2012 23.2 364,977 69 
Uppsala (Enköping, Håbo, Tierp, Älvkarleby) 1988 1988 Oct. 1988 Mar. 2009 20.5 86,834 18 
Uppsala (Östhammar) 1988 1988 Nov. 1988 Mar. 2009 20.4 24,787 
Uppsala (Uppsala, Knivsta) 1988 1988 Feb. 1988 Mar. 2009 21.2 165,454 45 
Västerbotten 1995 2000 Jan. 2001 Dec. 2006 47,124 
Västernorrland 1990 1990 Jan. 1990 Dec. 2012 23 366,026 66 
Västra Götaland (Bohus) 1986 1986 Nov. 1986 Dec. 1998 12.2 159,801 46 
Total      5,267,809 1,040 
Weighted mean     20.1   
Table 1B.

Year of start of screening program of women aged 40/50–69 and 70–74 years; start, end, and length of follow-up; number of person-years; and incidence-based breast cancer deaths in the study period 1986–2012 in the control group by county/subdivision of county.

Start ofFollow-upNo. of
County/subdivision40/50–6970–74StartEndLength (years)Person-yearsBreast cancer deaths
Blekinge 1988 2009 Oct. 1986 Dec. 2008 22.3 210,642 36 
Dalarna 1986 2011 Oct. 1986 Dec. 2010 24.3 452,851 57 
Gotland 1997 2000 Oct. 1986 Dec. 1998 12.3 33,269 11 
Jämtland 1996 2009 Oct. 1986 Jan. 2009 22.3 189,969 51 
Kronoberg 1990 1999 Oct. 1986 Aug. 1999 12.9 110,473 20 
Örebro 1987 1987/2008 May. 2000 Nov. 2007 7.6 71,627 
Skåne (Malmö) 1990 1999 Oct. 1986 Dec. 1998 12.3 178,036 35 
Stockholm 1989 2012 Oct. 1986 Dec. 2011 25.3 2,302,795 542 
Värmland 1993 2009 Oct. 1986 Dec. 2008 22.3 401,602 91 
Västerbotten 1995 2000 Oct. 1986 Feb. 1999 12.4 147,272 31 
Västmanland 1986 2009 Oct. 1986 Dec. 2008 22.3 337,616 56 
Västra Götaland (Lerum) 1988 1988/2008 Jan. 1995 Dec. 2007 13 13,626 
Västra Götaland (Norra Älvsborg) 1993 2009 Oct. 1986 Dec. 2008 22.3 218,965 54 
Västra Götaland (Södra Älvsborg) 1988 1988/2009 Jan. 1995 Dec. 2008 14 158,851 28 
Västra Götaland (Skaraborgs) 1989 1989/2009 Oct. 1995 Dec. 2008 13.3 170,351 25 
Västra Götaland (Göteborg) 1986 2008 Oct. 1986 Dec. 2007 21.3 566,305 124 
Total      5,564,250 1,173 
Weighted mean     19.9   
Start ofFollow-upNo. of
County/subdivision40/50–6970–74StartEndLength (years)Person-yearsBreast cancer deaths
Blekinge 1988 2009 Oct. 1986 Dec. 2008 22.3 210,642 36 
Dalarna 1986 2011 Oct. 1986 Dec. 2010 24.3 452,851 57 
Gotland 1997 2000 Oct. 1986 Dec. 1998 12.3 33,269 11 
Jämtland 1996 2009 Oct. 1986 Jan. 2009 22.3 189,969 51 
Kronoberg 1990 1999 Oct. 1986 Aug. 1999 12.9 110,473 20 
Örebro 1987 1987/2008 May. 2000 Nov. 2007 7.6 71,627 
Skåne (Malmö) 1990 1999 Oct. 1986 Dec. 1998 12.3 178,036 35 
Stockholm 1989 2012 Oct. 1986 Dec. 2011 25.3 2,302,795 542 
Värmland 1993 2009 Oct. 1986 Dec. 2008 22.3 401,602 91 
Västerbotten 1995 2000 Oct. 1986 Feb. 1999 12.4 147,272 31 
Västmanland 1986 2009 Oct. 1986 Dec. 2008 22.3 337,616 56 
Västra Götaland (Lerum) 1988 1988/2008 Jan. 1995 Dec. 2007 13 13,626 
Västra Götaland (Norra Älvsborg) 1993 2009 Oct. 1986 Dec. 2008 22.3 218,965 54 
Västra Götaland (Södra Älvsborg) 1988 1988/2009 Jan. 1995 Dec. 2008 14 158,851 28 
Västra Götaland (Skaraborgs) 1989 1989/2009 Oct. 1995 Dec. 2008 13.3 170,351 25 
Västra Götaland (Göteborg) 1986 2008 Oct. 1986 Dec. 2007 21.3 566,305 124 
Total      5,564,250 1,173 
Weighted mean     19.9   

Reference period

For a reference period (1971–1985) preceding the study period, cohorts were defined as similar as possible to determine whether there was a baseline difference in breast cancer mortality between the study and the control group. The weighted average follow-up time was 14 years in both the study and control group.

Mortality measures

Incidence-based mortality was used, which means that only breast cancer deaths among women diagnosed at ages 70 to 74 years were included (Fig. 3). The main outcome measure was breast cancer mortality rate ratio (RR) between the study and the control group. We used two different mortality measures–the underlying cause of death and the excess mortality.

Figure 3.

Lexis diagram illustrating incidence-based mortality using an example of 6 breast cancer cases where 2 were included in the mortality and the others were not included. Only cases who died from breast cancer during the follow-up and who were diagnosed at age 70 to 74 years between the start point (S) and the end point (E) were included in the mortality. S is 1986 and E is 2012 in the example.

Figure 3.

Lexis diagram illustrating incidence-based mortality using an example of 6 breast cancer cases where 2 were included in the mortality and the others were not included. Only cases who died from breast cancer during the follow-up and who were diagnosed at age 70 to 74 years between the start point (S) and the end point (E) were included in the mortality. S is 1986 and E is 2012 in the example.

Close modal

A difficulty when evaluating cause-specific mortality is inter-current mortality, and this problem increases with age. Women in this study were followed-up through 89 years of age, which can cause a sticky-diagnosis bias leading to an underestimation of the screening effect (11). We used excess mortality to minimize such bias.

Excess mortality is based on the excess number of all-cause deaths in breast cancer cases calculated as the difference between the observed and the expected number of all-cause deaths among the breast cancer cases. The expected number is calculated as the all-cause mortality rate in the population multiplied by the number of person-years among the breast cancer cases. The calculations were made within cells of birth year and age at death in one-year groups and then accumulated.

The absolute effect of invitation to mammography screening up to age 74 compared with stopping at 69 was calculated as the incidence-based breast cancer mortality for women diagnosed in age 70 to 74 in the control group multiplied by one minus the adjusted RR.

Missing data

Individual screening history data were missing in 320 breast cancer deaths (31%) in the study group, including 102 in Skåne, 66 in Jönköping, 32 in Gävleborg, 46 in Västra Götland Bohus, 69 in Kalmar, 3 in Södermanland, 1 in Östergötland, and 1 in Norrbotten. Furthermore, four cases with information on invitation had no records on participation status at the index screening, and 98 women who participated in the index screen had no information on participation in the penultimate round. The majority of missing data on screening history was not due to individual selection. Instead it was a consequence at clinic level with clinics with no data available or with staff too busy to help us collect their data. Imputation based on the group averages of invitation and participation was used to replace missing data to estimate the total number of nonexposed.

Bias adjustments

A number of women in the study group were diagnosed with breast cancer after the start of follow-up in the area but before they were invited to screening. Including them in the analysis might result in an underestimation of the effectiveness of the program. Cuzick and colleagues' (12) method for adjusting for noncompliance was applied to take this bias into account. The adjusted RR was RR = (O−U)/(E−U), where O is the observed number of breast cancer deaths in the study group and U is the number of nonexposed breast cancer deaths in the study group. E is the observed number of breast cancer deaths in the control group adjusted for the different number of person years in study and control group, that is, the expected number of breast cancer deaths in the study group had there been no screening.

Two types of exposure to screening were studied: invitation to and participation in screening at the age of 70 to 74 years. The nonparticipants chose not to participate and their breast cancer mortality might differ from those who chose to participate, so called, self-selection bias. This was adjusted for using the method of Cuzick and colleagues (12) by regarding women who were invited but did not participate as nonexposed. Participation in the last screening round (the index round) and the last two rounds (the index and penultimate rounds) before diagnosis was considered in the analysis.

Lead time bias leading to an underestimation of the effect of mammography screening might occur if a woman in the control group was diagnosed with breast cancer at age 75 or older and later died from the disease. By definition, her death would not be included in the incidence-based mortality, while a similar woman in the study group might have been diagnosed earlier due to screening detection at age 74 or younger. To correct for this, age at diagnosis was extended to 79 years, and breast cancer deaths in the study group diagnosed at age 75 to 79 were treated as nonexposed to screening and adjustment was made with the method by Cuzick and colleagues (12).

The weighted-average mid–follow-up time was 2.5 years earlier in the control group. From 1990 to 2010, the national average annual decrease in the age-standardized breast cancer mortality among women aged 70 to 84 was 0.92% as estimated by linear regression using data retrieved from the National Board of Health and Welfare. The adjustment for this bias was thus 2.5 × 0.92% = 2.3%.

Screening before the study period can cause a lead time bias, and the RCTs in Östergötland (study group) and in Malmö, and Dalarna (control group) and the pilot service-screening program in Gävleborg (study group) might cause overestimation in the study group and underestimation in the control group. To estimate the magnitude of this possible bias, the number of person-years was adjusted in the study and control group, respectively. The number of person-years was calculated as MST × SD × P, where P is the number of women aged 70–74 years. The mean sojourn time (MST; the average length between screen-detectable time and clinically detectable time) was assumed to be 4 years based on the estimate by Duffy and colleagues (13) and SD the proportion of screen-detected cancers was chosen at 65% based on the proportion of screen-detected cancers out of all diagnosed breast cancers in the 70- to 74-year age group in 2015 (National Breast Cancer Registry).

In 9 of 16 areas in the control group, follow-up started before the screening program (up to age 69 years) was launched. Because the aim of this study was to estimate the marginal effectiveness of screening up to 74 years, this might cause a bias towards overestimation of the effect. An adjustment was made by calculating a new RR based on the mortality measured from the screening start |$( {{s_i}} )$| instead of from the start point |$( {{f_i}} )$| of the follow-up in these 9 areas, where i indicates the area. We calculated the expected number |$( {{E_i}} )$| of breast cancer deaths if the follow-up had started from the start of screening. The adjusted number of breast cancer deaths |$( {{O_{adj}}} )$| in the control group was calculated by |${O_{adj}} = \sum {E_i} = \sum {\frac{{m{s_i}}}{{{\phi _i}}}}{P_i}}$|⁠, where |${P_i}$| is the person-years starting at |${f_i}$| and |$m{s_i}$| is the breast cancer mortality starting at |${s_i}$|⁠. Because of the design of the follow-up of the cohorts and the incidence-based mortality, the observed mortality is dependent on the length of follow-up. To assess this, we simulated the mortality for different lengths of follow-up, which gave us the adjustment factor for different lengths of follow-up |${\phi _i}$| (Supplementary Fig. S1). For the 7 areas where |${s_i} = {f_i}$|⁠, the adjustment factor |${\phi _i} = 1$| and |${E_i} =$| the observed number of breast cancer deaths in the area. The observed number of breast cancer deaths in the control group |${O_c}$| was replaced by |${O_{adj}}$| to adjust RR.

Ethical approval

The study was approved by the Regional Ethics Review Board in Umeå, Sweden.

The reported attendance rate varied from 67% to 87%, with the lowest in the large cities and highest in the more remote areas. In general, the attendance rate increased over the study period.

During the study period, there were 1,040 breast cancer deaths in 5.3 million person-years in the study group and 1,173 breast cancer deaths in 5.6 million person-years in the control group. The crude RR estimate of the incidence-based breast cancer mortality was 0.94 (95% CI: 0.86–1.02) for women in the study group compared with women in the control group (Table 2).

Table 2.

Incidence-based breast cancer mortality and excess mortality for women aged 70–74 years invited to mammography screening in Sweden in the study and control group during the reference and the study period.

Reference periodStudy period
Study groupControl groupStudy groupControl group
No. of person-years 1,965,043 2,385,417 5,267,809 5,564,250 
Breast cancer mortality 
 No. of breast cancer deaths 632 738 1,040 1,173 
 RR (95% CI) 1.04 (0.93–1.16) 0.94 (0.86–1.02) 
Excess mortality 
 No. of breast cancer cases 2,493 3,146 7,680 6,054 
 Total no. of deaths among breast cancer cases 1,102 1,373 3,149 2,753 
 Expected no. of deaths from other causes than breast cancer among breast cancer cases 413.4 514.2 2,119.4 1,533.6 
 Excess no. of deaths (breast cancer deaths) 688.6 858.8 1,029.6 1,219.4 
 RR (95% CI) 0.97 (0.88–1.08) 0.89 (0.82–0.97) 
Reference periodStudy period
Study groupControl groupStudy groupControl group
No. of person-years 1,965,043 2,385,417 5,267,809 5,564,250 
Breast cancer mortality 
 No. of breast cancer deaths 632 738 1,040 1,173 
 RR (95% CI) 1.04 (0.93–1.16) 0.94 (0.86–1.02) 
Excess mortality 
 No. of breast cancer cases 2,493 3,146 7,680 6,054 
 Total no. of deaths among breast cancer cases 1,102 1,373 3,149 2,753 
 Expected no. of deaths from other causes than breast cancer among breast cancer cases 413.4 514.2 2,119.4 1,533.6 
 Excess no. of deaths (breast cancer deaths) 688.6 858.8 1,029.6 1,219.4 
 RR (95% CI) 0.97 (0.88–1.08) 0.89 (0.82–0.97) 

Note: The RR and 95% CI are shown.

The RR adjusted for 61 noninvited women in the study group was 0.93 (95% CI: 0.86–1.02). To adjust for the lead time bias, the upper age for diagnosis was extended up to 79 years. There were 1,808 breast cancer deaths in the study group (768 breast cancer deaths diagnosed in the 75- to 79-year age group were not invited to screening) and 2,197 deaths in the control group. After adjusting for lead time bias, the RR was 0.78 (95% CI: 0.73–0.84). After final adjustment for bias due to different mid-times between the study and control groups, the RR for invitation in the study group versus no invitation in the control group was 0.80 (95% CI: 0.75–0.85; Table 3).

Table 3.

Incidence-based breast cancer mortality for women aged 70–74 years invited to mammography screening in Sweden and in Sweden except for Stockholm County.

EstimatesRR
Crude estimate 0.94 (0.86–1.02) 
Invitation 
 Adjusted for lead time bias 0.78 (0.73–0.84) 
 Final result, adjusted for lead time bias and biasa 0.80 (0.75–0.85) 
Attendance I 
 Adjusted for lead time bias 0.73 (0.66–0.80) 
 Final result, adjusted for lead time bias and biasa 0.74 (0.68–0.82) 
Attendance I + P 
 Adjusted for lead time bias 0.71 (0.64–0.79) 
 Final result, adjusted for lead time bias and biasa 0.73 (0.66–0.81) 
Stockholm county excluded from control group 
Crude estimate 1.02 (0.92–1.13) 
Invitation 
 Adjusted for lead time bias 0.86 (0.79–0.94) 
 Final result, adjusted for lead time bias & biasa & biasb 0.82 (0.75–0.89) 
EstimatesRR
Crude estimate 0.94 (0.86–1.02) 
Invitation 
 Adjusted for lead time bias 0.78 (0.73–0.84) 
 Final result, adjusted for lead time bias and biasa 0.80 (0.75–0.85) 
Attendance I 
 Adjusted for lead time bias 0.73 (0.66–0.80) 
 Final result, adjusted for lead time bias and biasa 0.74 (0.68–0.82) 
Attendance I + P 
 Adjusted for lead time bias 0.71 (0.64–0.79) 
 Final result, adjusted for lead time bias and biasa 0.73 (0.66–0.81) 
Stockholm county excluded from control group 
Crude estimate 1.02 (0.92–1.13) 
Invitation 
 Adjusted for lead time bias 0.86 (0.79–0.94) 
 Final result, adjusted for lead time bias & biasa & biasb 0.82 (0.75–0.89) 

Note: The RR and 95% CIs were adjusted for bias and lead time.

Abbreviations: I, the last screening round before breast cancer diagnosis (index round); P, the penultimate screening round before breast cancer diagnosis.

aImbalance of the population-weighted average mid-time of follow-up between study and control groups.

bImbalance of the population-weighted average length of follow-up between study and control groups.

The absolute incidence-based breast cancer mortality in the control group was 2.11 per 10,000 person-years, while the corresponding mortality in the study group was 1.69 per 10,000 person-years. The difference was thus 0.42 per 10,000 person-years.

The RR for attendance at the index round versus no invitation adjusted for the 253 breast cancer deaths among nonattendees (174 actual nonattendees and 79 imputed nonattendees) and the 61 women diagnosed before their first invitation in the study group was 0.91 (95% CI: 0.82–1.01). After further adjustment for lead time bias, the RR was 0.73 (95% CI: 0.66–0.80), and finally after adjustment for bias due to different mid-times between the study and control groups the RR was 0.74 (95% CI: 0.68–0.82; Table 3).

The RR for attendance in both the index and the penultimate round versus no invitation adjusted for the 311 nonattendees (206 actual nonattendees and 105 imputed nonattendees) at either round or both was 0.91 (95% CI: 0.82–1.02). After further adjustment for lead time bias, the RR was 0.71 (95% CI: 0.64–0.79). After final adjustment for bias due to different mid-times between the study and control group, RR was 0.73 (95% CI: 0.66–0.81; Table 3).

The biases due to screening programs that were ongoing before the start of the follow-up in the study group and the control group almost balanced each other out. The person-years related to the bias were 39,521 and 33,110 in the study and control group, respectively. The adjustment required subtracting 16,669 (0.3%) of the person-years from the study group, and no adjustment was made for this bias.

To assess the possible bias due to follow-up before the start of the screening program in the control group, the estimated adjusted number of breast cancer deaths when follow-up started from screening start, Oadj, was calculated to be 1,229.6, that is, higher than the observed number of breast cancer deaths in the control group (Oc = 1,173). This eventual bias, therefore, did not lead to an overestimation of the effectiveness, and no adjustment was made.

In the study group, there were 3,149 observed and 2,119.4 expected deaths, and in the control group there were 2,753 observed and 1,533.6 expected deaths among the breast cancer cases, thus the excess numbers of deaths were 1,029.6 and 1,219.4 in the study group and the control group, respectively, which resulted in a crude RR of 0.89 (95% CI: 0.82–0.97) for incidence-based excess mortality (Table 2).

During the reference period, there were 632 breast cancer deaths and 1.97 million person-years in the study group compared with 738 breast cancer deaths and 2.39 million person-years in the control group (Table 2). The RR due to breast cancer mortality was 1.04 (95% CI: 0.93–1.16), and for excess mortality the RR was 0.97 (95% CI: 0.88–1.08). Thus, the difference in baseline mortality between the study group and the control group was small and the estimates in the study period were not adjusted for this (Table 2).

Stockholm contributed with 542 cases (44%) to the control group, and estimation of RR was also made without Stockholm. The estimated crude RR after excluding Stockholm was 1.02 (95% CI: 0.92–1.13), and after adjusting for noninvited women RR was 1.02 (95% CI: 0.93–1.13). After adjusting for lead time bias, the estimated RR was 0.86 (95% CI: 0.79–0.94). However, excluding Stockholm changed the balance so that the average length of follow-up became 2.5 years shorter in the control group. Using the adjustment factor|$\ {\rm{\phi \ }}$|= 0.94, the RR became 0.81. Also, the average population-weighted mid-time of follow-up of the control group was changed to December 1998, which was 2 years earlier than in the study group, and the adjustment was 1.8%. After taking these two imbalances into account, the final adjusted RR was 0.82 (95% CI: 0.75–0.89; Table 3).

A summary of the bias adjustments for the estimates of RR for women invited and attending screening is given in Table 4.

Table 4.

Summary of bias adjustments of RR.

Estimate RRBias consideredDirection of bias on the effectMethodAdjusted
Exposure: Invitation I. Selection bias due to not invited BCDs Underestimation Adjusting for the number of BCDs in not invited (known + imputed) using the method by Cuzick et al. Yes 
 II. Lead time bias Underestimation Age at diagnosis extended to 79 years. Adjusting for the number of nonexposed BCDs (75–79 years) in the study group using the method by Cuzick et al. Yes 
 III. Imbalance of mid-time follow-up between study group and control group Overestimation Adjustment based on the change in breast cancer mortality in age 70–74 in Sweden 1990–2010 Yes 
 IV. Screening before the start of follow-up Not obvious Subtracting person-years No, negligible difference 
 V. No screening at start of follow-up in 9 areas in the control group Overestimation Based on mortality measured from screening start combined with a simulated adjustment factor due to imbalance in length of follow-up No 
Exposure: Invitation excluding Stockholm Same as (I–V) Same as (I–V) Same as (I–V) Same as (I–V) 
 VI. Shorter length of follow-up in control group Underestimation Simulated adjustment factor Yes 
Exposure: Attendance in index round VII. Selection bias due to not invited and nonparticipating BCDs Underestimation Adjusting for the number of BCDs not invited or not participating (known + imputed) using the method by Cuzick et al. Yes 
 Same as (II–IV) Same as (II–IV) Same as (II–IV) Same as (II–IV) 
Exposure: Attendance in index and penultimate round Same as VII + (II–IV) Same as VII + (II–IV) Same as VII + (II–IV) Same as VII + (II–IV) 
Estimate RRBias consideredDirection of bias on the effectMethodAdjusted
Exposure: Invitation I. Selection bias due to not invited BCDs Underestimation Adjusting for the number of BCDs in not invited (known + imputed) using the method by Cuzick et al. Yes 
 II. Lead time bias Underestimation Age at diagnosis extended to 79 years. Adjusting for the number of nonexposed BCDs (75–79 years) in the study group using the method by Cuzick et al. Yes 
 III. Imbalance of mid-time follow-up between study group and control group Overestimation Adjustment based on the change in breast cancer mortality in age 70–74 in Sweden 1990–2010 Yes 
 IV. Screening before the start of follow-up Not obvious Subtracting person-years No, negligible difference 
 V. No screening at start of follow-up in 9 areas in the control group Overestimation Based on mortality measured from screening start combined with a simulated adjustment factor due to imbalance in length of follow-up No 
Exposure: Invitation excluding Stockholm Same as (I–V) Same as (I–V) Same as (I–V) Same as (I–V) 
 VI. Shorter length of follow-up in control group Underestimation Simulated adjustment factor Yes 
Exposure: Attendance in index round VII. Selection bias due to not invited and nonparticipating BCDs Underestimation Adjusting for the number of BCDs not invited or not participating (known + imputed) using the method by Cuzick et al. Yes 
 Same as (II–IV) Same as (II–IV) Same as (II–IV) Same as (II–IV) 
Exposure: Attendance in index and penultimate round Same as VII + (II–IV) Same as VII + (II–IV) Same as VII + (II–IV) Same as VII + (II–IV) 

Note: RR is calculated for incidence-based breast cancer mortality during the study period.

Abbreviation: BCD, breast cancer deaths.

The marginal effectiveness of continuing inviting women to mammography screening up to age 74 instead of age 69 was estimated at 20%. Considering actual participation in screening, the effectiveness was 6% to 7% higher. The finding was supported by a 5% higher crude effectiveness using excess mortality as outcome.

The lower crude estimate using excess mortality in comparison with breast cancer mortality (0.89 vs. 0.94; Table 2) indicates a sticky-diagnosis bias towards an underestimation of the effectiveness due to more breast cancer being diagnosed in the study group.

Similar results have been reported from Canada and the Netherlands. Coldman and colleagues (8) included 2.8 million screening participants among women aged 40 to 74 years in seven provinces in Canada with 12 to 20 years follow-up. The study group consisted of women who had at least one screen and the control group consisted of nonparticipants. The SMR for women aged 70 to 79 years was estimated at 0.60 (95% CI: 0.52–0.67; ref. 8). The participation rate varied between 32% and 53%, which was more than 30 percentage points lower than in the current study. De Glas and colleagues (14) studied 25,414 women aged 70 to 75 years who were invited to screening in the Netherlands between 1995 and 2011. Inviting women to mammography screening significantly decreased the incidence of advanced stage breast cancer (RR = 0.88; 95% CI: 0.81–0.97) and increased the incidence of early-stage cancer (RR = 1.46; 95% CI: 1.40–1.52) compared with before the implementation of screening.

The strengths of this study are the length of the study period (20 years) and the size of the study population. Follow-up was based on the nationwide high quality cancer and cause of death register. A reference period was used to assess if there were any baseline differences between the groups before the service screening started. Two outcome measures were used, breast cancer mortality and excess mortality, the latter of which is independent of the individual cause of death determination. We studied both the effect of invitation to screening and the effect of participation, and all known possible biases have been addressed and adjusted for. The national cause of death register has been continuously validated. It has been found to have a high quality and that underlying cause of death could be used as main outcome measure of studies of breast cancer mortality (15).

There were also some limitations. The study and control groups in the reference period were not possible to make as similar as for those in the study period due to RCTs or pilot service-screening in 4 areas. Furthermore, five of the areas in the study period were used in both the study group and the control group in different time periods. They were not included in the reference period. Furthermore, before 1971, the residence code in the cancer register was not complete, thus breast cancer deaths had to be assigned to counties and not to the screening areas. The length of follow-up was therefore 6 years shorter in the reference period. Self-appointment for screening for women over 69 years in the control group might have caused an underestimation of the screening effect. Furthermore, a temporary lack of radiologists may occasionally have caused screening intervals to exceed 24 months in some screening areas, which might lead to an underestimation of the effectiveness. Lead time bias from the RCT in Stockholm before the start of follow-up was not considered because only a very small part of these women could have influenced the result. Stockholm was part of the control group, which means that any bias would be toward underestimation of the effect.

Differences in risk factors (e.g., hormone replacement therapy) and breast cancer treatment may result in differences in breast cancer mortality. Regional care programs were introduced in the 1980s and 1990s to standardize treatment within the six public health regions. However, we believe that the impact of the variation of both risk factors and treatment on breast cancer mortality is more related to calendar time than to geographic areas. Our main concern was Stockholm, which constitute 44% of the control group. However, this was taken into account by a separate analysis where Stockholm was excluded from the control group.

The prevalence of opportunistic screening was in general very low in ages over 70 years and should not have had any impact on the results.

To study the effect of screening on breast cancer mortality, any excess incidence attributable to overdiagnosis is irrelevant because death from breast cancer imply that the cancer is not overdiagnosed. However, to balance benefit and harm overdiagnosis has to be considered. We are not aware of any age-specific estimates of overdiagnosis in women ages 70 to 74 years, which would be relevant for this study. The only randomized trial including women over 70 years is the Two-county trial in Dalarna and Östergötland. Unfortunately, the control group was offered screening in the end of the trial, which makes overdiagnosis estimation difficult.

In conclusion, this nationwide study of the Swedish service screening showed a statistically significant 20% lower incidence-based breast cancer mortality among women diagnosed at age 70 to 74 in areas where women were invited up to the age of 74 years compared with areas where screening stopped at age 69.

No potential conflicts of interest were disclosed.

Z. Mao: Data curation, software, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. L. Nyström: Supervision, validation, investigation, methodology, writing–original draft, writing–review and editing. H. Jonsson: Conceptualization, resources, software, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Barbro Hellqvist, Regional Cancer Center, Umeå, Sweden, for valuable help in interpreting the data from the Screening for Young Women (SCRY) study. They also thank the staff at the mammography clinics who were helpful with data collection. H. Jonsson has received grants from the Swedish Research Council, the Swedish Cancer Society, the Cancer Research Fund in North Sweden, and the County Council of Västerbotten.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Independent UK Panel on Breast Cancer Screening
. 
The benefits and harms of breast cancer screening: an independent review
.
Lancet
2012
;
380
:
1778
86
.
2.
Nyström
L
,
Rutqvist
LE
,
Wall
S
,
Lindgren
A
,
Lindqvist
M
,
Ryden
S
, et al
Breast cancer screening with mammography: overview of Swedish randomised trials
.
Lancet
1993
;
341
:
973
8
.
3.
European Commission Initiative on Breast Cancer; [about 3 screens]
. Available from: https://healthcare-quality.jrc.ec.europa.eu/.
4.
Tabar
L
,
Fagerberg
G
,
Chen
HH
,
Duffy
SW
,
Smart
CR
,
Gad
A
, et al
Efficacy of breast cancer screening by age. New results from the Swedish Two-County Trial
.
Cancer
1995
;
75
:
2507
17
.
5.
Basu
P
,
Ponti
A
,
Anittila
A
,
Ronco
G
,
Senore
C
,
Vale
DB
, et al
Status of implementation and organization of cancer screening in the European Union Member States-summary results from the second European screening report
.
Int J Cancer
2018
;
142
:
44
56
.
6.
Jonsson
H
,
Tornberg
S
,
Nyström
L
,
Lenner
P
. 
Service screening with mammography of women aged 70–74 years in Sweden. Effects on breast cancer mortality
.
Cancer Detect Prev
2003
;
27
:
360
9
.
7.
Jonsson
H
,
Bordas
P
,
Wallin
H
,
Nyström
L
,
Lenner
P
. 
Service screening with mammography in Northern Sweden: effects on breast cancer mortality – an update
.
J Med Screen
2007
;
14
:
87
93
.
8.
Coldman
A
,
Phillips
N
,
Wilson
C
,
Decker
K
,
Chiarelli
AM
,
Brisson
J
, et al
Pan-Canadian study of mammography screening and mortality from breast cancer
.
J Natl Cancer Inst
2014
;
106
:
dju261
.
9.
Van der Walde
A
,
Hurria
A
. 
Early breast cancer in the older woman
.
Clin Geriatr Med
2012
;
28
:
73
91
.
10.
Hellquist
BN
,
Duffy
SW
,
Abdsaleh
S
,
Björneld
L
,
Bordas
P
,
Tabar
L
, et al
Effectiveness of population-based service screening with mammography for women ages 40 to 49 years: evaluation of the Swedish Mammography Screening in Young Women (SCRY) cohort Cancer
2011
;
117
:
714
22
.
11.
Obuchowski
NA
,
Lieber
ML
. 
The effect of misclassification in screening trials: a simulation study
.
Contemp Clin Trials
2008
;
29
:
125
35
.
12.
Cuzick
J
,
Edwards
R
,
Segnan
N
. 
Adjusting for non-compliance and contamination in randomised clinical trials
.
Stat Med
1997
;
16
:
1017
29
.
13.
Duffy
SW
,
Day
NE
,
Tabar
L
,
Chen
HH
,
Smith
TC
. 
Markov models of breast tumor progression: some age-specific results
.
J Natl Cancer Inst Monogr
1997
;
22
:
93
7
.
14.
de Glas
NA
,
de Craen
AJ
,
Bastiaannet
E
,
Op 't Land
EG
,
Kiderlen
M
,
van de Water
W
, et al
Effect of implementation of the mass breast cancer screening programme in older women in the Netherlands: population-based study
.
BMJ
2014
;
349
:
5410
.
15.
Nyström
L
,
Larsson
LG
,
Rutqvist
LE
,
Lindgren
A
,
Lindqvist
M
,
Rydén
S
, et al
Determination of cause of death among breast cancer cases in the Swedish Randomised Mammography Screening Trials: a comparison between official statistics and validation by an endpoint committee
.
Acta Oncol
1995
;
34
:
145
152
.