Randomized controlled trials
All 7 RCTs used intention-to-treat analyses and had breast cancer mortality as the primary outcome. Two of the trials showed benefit among women aged 40–49; the results of one (the Gothenburg trial5
) were analysed only once, and those of the other (the Malmo trial6,19,20
) were analysed by the study authors twice.
Most of the trials lacked the power to exclude a potentially clinically significant difference, such as a relative risk reduction of 20%. No sample size calculations were published for the Gothenburg Breast Screening Trial,5
the combined Malmo I and II trials,6
the Swedish Two-County Trial21
or the Stockholm Mammographic Screening Trial.40
The sample size for the Malmo I trial was chosen to detect a 25% reduction in mortality (among women aged 45–69) with an α of 0.05 and a β of 0.10. In many cases the calculated power was undermined by poor compliance and contamination (). The Health Insurance Plan (HIP) Trial15
planned for power to detect a 20% reduction in mortality but found a 25% reduction in relative risk that failed to reach statistical significance. The Edinburgh team calculated a sample size of 65 000 women for 80% power to detect a 35% reduction in relative risk at 7 years with a 1-sided p
value of 0.05.27
All of the above trials assessed women aged 40–49 as post hoc subgroups. The NBSS-1, which included only this age group, calculated a sample size to provide 80% power to detect a 40% reduction in 5-year mortality with a 1-sided p
value of 0.05;3
however, the mortality rate in the control group was less than predicted, and contamination was not considered. For the 10.5-year follow-up,38
actual power was estimated to be adequate to detect a mortality reduction of 30% or more, with a 2-sided p
value of 0.05.
The following discussion focuses on recent updates of the individual trials.
HIP Trial (1963–1970):
The most recent update included 18 years of follow-up and considered all deaths from breast cancer diagnosed in the 5 years following the first screening.18
Among women aged 40–49 at the first screen (14 432 invited and 14 701 control subjects), a nonsignificant reduction of 25% in mortality was found, with 50 and 66 deaths observed in the 2 groups respectively. Results of heterogeneity tests by age were negative. Only 25% of the cases were detected by mammography alone.17
Malmo I (1976–1988) and II (1978–1990) Mammographic Screening Trials:
In Malmo I, women born between 1908 and 1932 were randomly assigned to study groups between 1976 and 1978.19
In Malmo II, women born between 1933 and 1945 were randomly assigned between 1978 and 1990.6
A subgroup of 7984 women in the Malmo I cohort were aged 45–49 at entry. In Malmo II, there were 17 786 women aged 45–48 at entry; the women had a mean of 5 rounds of screening, and follow-up was 10 years on average.6
The first combined analysis of Malmo I and II data was limited to women entering the studies at age 44–49. In the screening group (13 528 women, contributing 165 596 woman-years of follow-up) there were 57 deaths, and in the control group (12 242 women, contributing 144 036 woman-years of follow-up) there were 78. The point estimate of relative risk was 0.64 (95% confidence interval [CI] 0.45–0.89, p = 0.0009). From these results, the absolute risk reduction of 0.02% per year translates to a number needed to screen for 1 year to prevent 1 death from breast cancer of 5067.91, which corresponds to the authors' estimate of 500 for screening women aged 45–49 every 18–24 months for 12.5 years.
Swedish Two-County Trial — Ostergotland (1977–1984) and Kopparberg (1978–1985):
The most recent mortality analysis had a follow-up of 13 years on average.25
The subgroup of women aged 40–49 at study entry (19 844 invited and 15 604 control subjects) had a nonsignificant relative risk of 0.87 (95% CI 0.54–1.41). Only 39 and 45 deaths occurred in the 2 groups respectively. Heterogeneity by age at randomization was not found. The point estimates of relative risk differed in the 2 counties, with a subgroup relative risk of 1.02 (95% CI 0.52–1.99) in Ostergotland and 0.73 (95% CI 0.37–1.41) in Kopparberg.
Edinburgh Randomized Trial (1979–1988):
Women aged 45–64 entered the trial between 1979 and 1981,27
with others entering if they turned 45 before 1985.28
No statistically significant differences in mortality were observed between the study and control groups. Subgroup analysis of women aged 45–49 at enrolment (with 14 years of follow-up in the 11 391-person subgroup from the original cohort, and 10–12 years of follow-up in the 10 383-person group entering from 1982 to 1985) showed a relative risk of 0.82 (95% CI 0.51–1.32); no heterogeneity was observed relative to all trial participants. In all, there were 46 deaths from breast cancer among the 11 505 women in the study group and 52 among the 10 269 control subjects.29
Socioeconomic status was a confounding factor in the study. Correction for this factor29
eliminated the difference in mortality from causes other than breast cancer and gave a relative risk of 0.75 (95% CI 0.48–1.18) in the subgroup aged 40–49 at enrolment.
Canadian National Breast Screening Study 1 (1980–1988):
The NBSS-1 is the only study limited to women aged 40–49. Concerns about subversion of randomization32,33,34
have not been supported by internal and external35
reviews or by examination of an alternative data source.36
Participants had a higher socioeconomic status, more risk factors for breast cancer and smoked less heavily than the Canadian population.3
An imbalance in the number of women with more than 3 lymph nodes involved was seen in the study and control groups: 17 versus 5 at randomization (p
= 0.017) and 47 versus 23 at 7 years (p
Fewer lymph node dissections in the control group and more deaths among women with node-negative disease suggest that the difference could have been due to more aggressive surgery among women undergoing screening.38
No difference in mortality was observed between the study and control groups. Follow-up to the end of 1993 (8.75–13 years) revealed 82 deaths from breast cancer among the 25 214 women screened, and 72 among the 25 216 control subjects. The relative risk was 1.14 (95% CI 0.83–1.56).38
Stockholm Breast Cancer Screening Trial (1981–1985):
The subgroup of women aged 40–49 at entry included 14 842 invited to undergo screening and 7108 control subjects.41
In the 173 866 woman-years of follow-up for the screened women, 24 deaths from breast cancer were observed, as compared with 12 deaths in the 87 826 woman-years of follow-up for the control subjects. With a mean follow-up of 11.4 years the relative risk was 1.08 (95% CI 0.54–2.17). No statistically significant differences in mortality were observed between the 2 groups.
Gothenburg Breast Screening Trial (1982–1992):
The long-awaited results of this trial were published in 1997.5
All female residents of Gothenburg, Sweden, born between 1923 and 1944 were randomly assigned to mammography every 18 months for 5 rounds or to control status with a single mammogram at trial completion.5
Cluster randomization was used for the 1923–1936 cohort and individual randomization for the 1936–1944 cohort. Data were analyzed for a subgroup of 25 941 women aged 39–49 at study entry (82% individually randomized) with at least 10 years of follow-up.5
A ratio of 1:1.2 was used to apportion the women, giving 11 724 women in the study group and 14 217 in the control group.
The study group had 18 deaths over 138 402 woman-years of follow-up, and the control group had 40 deaths over 168 025 woman-years of follow-up. The relative risk was 0.55 (95% CI 0.31–0.96, p
The absolute risk reduction of 1.28 per 1000 (mortality rates of 2.8 per 1000 in the control group v. 1.5 per 1000 in the study group) translated into a number needed to screen of 782 for the 5 screens at 18-month intervals.