Study population and data sources
The study included all women aged 49–84 residing in Norway during the period 1987–2019, and the associated first cases of invasive breast cancer or ductal carcinoma in situ (DCIS). The Cancer Registry of Norway provided the data aggregated by the women’s birth cohort, calendar year and county of residence.
Reporting to the Cancer Registry of Norway is mandatory, and diagnostic information is obtained separately from clinicians, pathologists, and death certificates, with only 0.2% of all cancers ascertained from death certificates alone [
8]. Reporting of DCIS became mandatory from 1993, but the stable DCIS incidence rates around 1993 suggest that the registration was nearly complete also before mandatory registration (Additional file
1: Fig. S1).
BreastScreen Norway, administered by the Cancer Registry of Norway, targets women 50–69 years of age in biennial mammography screening. The program was initiated in four large counties (approximately 40% of eligible women in Norway) in late 1995 and early 1996, and was rolled out in the remaining 14 Norwegian counties during the period 1999–2005 (Table
1). Women are invited according to birth cohort to county-wise screening rounds, and we used the exact start and end dates of screening invitations for the respective birth cohorts in each county. The overall attendance to the program has been relatively stable at around 76% [
9]. Full-field digital mammograms gradually replaced screen film mammograms from the year 2000 [
10].
Table 1
The BreastScreen Norway mammography screening program and women included in the study
Study data | |
Area | Norway |
Time period | 1987–2019 |
Age rangea | 49–84 |
Design | Dynamic cohort study |
Person-years under study | 23,709,444 |
DCIS or invasive breast cancer cases | 63,378 |
BreastScreen Norway | |
Implementation period | 1995–2005 |
Target age range and frequency | 50–69 every second year |
Attendance rate | 76% |
Screening test | Two-view mammography |
Evaluation | Two independent readers |
Women in invited birth cohorts | 1,063,409 (approximated) |
Proportion of cases screening-detected aged 50–69 years | 68% (year 2016–2019) |
Menopausal hormone therapy is associated with increased risk of breast cancer [
11], and was extensively used in Norway during the period coinciding with the first introductions of screening [
12]. However, its use declined sharply from around 2002 after hormone therapy was linked to an increased risk of cardiovascular disease [
13]. To avoid potential biases by variation in the use of hormone therapy, we applied county-specific sales figures from the Norwegian Drug Wholesales Statistics of preparations containing both estrogen and estrogen-progestogen combinations, scaled according to the age distribution among women with a prescription for hormone therapy in the Prescription Database of Norway. Both sources are available at the Norwegian Institute of Public Health (See Additional file
1).
Statistical analysis
To estimate the frequency of non-progressive breast carcinomas, we followed women up to 85 years of age, and compared the cumulative rate of breast carcinoma between screened women and women without screening. In the lack of a randomized non-invited comparison group, and since no birth cohorts invited to screening from age 50 have yet reached the age of 85 years, we took advantage of the gradual county-wise introduction of BreastScreen Norway to distinguish the effects of screening by age from temporary incidence changes (Additional file
1: Fig. S2). Using the Norwegian data, we estimated likely age-specific incidence rates in the presence and absence of screening, where the cumulative rate is the sum of the age-specific incidence rates.
We modeled the incidence of first cases of breast carcinoma among Norwegian women using an extended Age-Period-Cohort (APC) Poisson regression model [
14,
15], similar to our previous modeling work [
12,
16]. The gradual county-wise introduction of BreastScreen Norway provided data with different screening program status at the same calendar times, which could be exploited in the APC Poisson regression model. We included variables for different effects of the screening program. For the screening period, we added separate variables for the initial screening round at 50–51 years of age, for the second screening round at 52–53 years of age, and for the subsequent eight screening rounds, modeled as the proportion of the calendar year women in a birth cohort were covered by the specific parts of the program. The separate variable for the second screening round was added to account for cases that might have been overlooked at the initial screening. We applied natural cubic splines to allow for nonlinear effects by age for the screening period (inner knot at age: 60). After the screening period, we allowed for a gradual declining effect of previous screening by applying natural cubic splines for the time since screening cessation (inner knots at year: 1, 2, 5, 10, and 13) in addition to a variable for no longer being covered by the screening program.
Since changes in incidence typically appear gradually [
17], we also used natural cubic splines to smooth the effects of age (inner knots at age: 50, 52, 54, 56, 60, 70, and 80), period (inner knots at year: 1997 and 2009) and birth cohort (inner knots at birth cohort: 1929 and 1943), in order to limit the number of variables in the model. For the age component, we added more knots around the age of menopause to increase modeling flexibility for ages at which the effects of certain breast cancer risk factors tend to change [
18]. Each county was assigned its own breast carcinoma incidence level. As in our previous modeling of breast cancer incidence, we applied a one-year time lag for the hormone therapy variable [
12] to reflect the likely time lag between use of hormone therapy and increased risk of breast carcinoma [
11]. Other changes in risk factors were taken into account by the age, period and cohort components.
The women contributed person-years until they were censored due to death or end of follow-up. At screening implementation, women in the entire target age range of 50–69 years were invited to their first screening. To limit model complexity, data regarding initial or second screening at higher ages, above the age of 53 or 55 years, respectively, were not used in the modeling. It was thus assumed that the relative risk of breast carcinoma from the third screening round onwards, and for post-screening women, was similar to the incidence for women screened since the age of 50. The full specification of the incidence model is given in the Additional file
1.
We calculated the average estimated age-specific incidence rates across all Norwegian counties, both in the presence and absence of invitations to the full screening program, by applying the incidence model on calendar year 2019 for the 1969 birth cohort, with hormone therapy set at the national 2019 level.
We estimated the frequency of non-progressive screening-detected breast carcinomas as the difference in the cumulative rates at 85 years of age between screening and non-screening scenarios. The age of 85 years was chosen to allow for disease progression within a reasonable time (15 years), while still having a substantial number of person years under study.
To estimate the proportion of non-progressive cases related to screening, we calculated the frequency of screening-detected breast carcinomas from the APC model. We multiplied the cumulative rate for ages 50–69 years in the presence of screening with the observed proportion of screening-detected carcinomas among women 50–69 years of age in the years 2016–2019 (Table
1, Additional file
1: Fig. S4).
To facilitate comparison with other studies, we performed analyses without inclusion of DCIS. Active treatment of screening-detected DCIS might have prevented the transition of DCIS to invasive cancer [
19]. Thus, to estimate invasive non-progressive cases only, using population data, will likely lead to a result biased toward the null.
To assess statistical uncertainty, we calculated 95% bootstrap percentile confidence intervals (CI) based on 10,000 repetitions. For sensitivity analyses we either removed the hormone variable, the period variable or the interaction between age and subsequent screening. As a sensitivity analysis, we further extended follow-up to 90 and 95 years of age, under the conservative assumption that the difference in age-specific incidence at 85 years of age between the screening and non-screening scenario remained constant at higher ages.
All statistical analyses were conducted using the R statistical package (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria) [
20].