Meta-analyses of clinical trials have shown reductions in breast cancer and overall mortality in women randomized to receive screening mammography.1 Consequently, screening mammography has become a recommended method for the early detection of breast cancer in many countries.2 However, uncertainties still exist regarding the value of screening at different ages3 and at different frequencies.1, 4 Trials conducted in Canada failed to find an effect of screening including mammography in women aged 40–495 or comparing mammography with physical examination in women 50–59.6 Further trials are currently underway7, 8 to try to provide further evidence. The existence of service screening in different countries over the last 2 decades has provided added information on the impact of screening mammography.9, 10, 11, 12 The overall contribution of mammographic screening to declines in population breast cancer mortality rates is an area of some uncertainty, with some authors estimating effects as large as 65%13 and others arguing for little or no effect.14
The Screening Mammography Program of British Columbia (SMPBC) was established in 1988 to provide breast screening to women in British Columbia.15 Women aged 40–79 were eligible to self-refer and received screening mammograms free of charge through community clinics with interpretation by radiologists affiliated to the program.16 Currently, women under the age of 50 are rescreened annually and those over 50 every 2 years. As the program grew, participation rates have increased so that by 2003 over 50% of British Columbia women aged over 40 had received at least one screening mammogram through the program.
To provide additional information on the potential of mammography screening to reduce deaths from breast cancer, we undertook an analysis of mortality from breast cancer in women who participated in the SMPBC and compared their observed outcomes with what would be expected based upon incidence and survival rates for other British Columbia women who have not participated.
Material and methods
Data were obtained from the SMPBC database of women receiving screening mammography through the program, the British Columbia Cancer Registry (BCCR) and the Vital Statistics Agency death file (VSA) comprising death notifications in British Columbia residents.
The study cohort consisted of women aged 40 or more first screened by SMPBC between January 1988 and December 2003. Women entered the cohort at the date of their first SMPBC mammogram. Breast cancer cases and deaths in the cohort were ascertained through linkage with the VSA and BCCR. Linkage was performed in 2 steps. First the VSA and BCCR were linked to identify the cause and dates of death in all breast cancer cases diagnosed in the study period. A second linkage was performed of the study cohort to the VSA and BCCA to identify dates of death and cases of breast cancer in the study cohort. Combination of the resulting files permitted the identification of breast cancer cases and deaths for British Columbia women, in the cohort and not in the cohort, throughout the study period. Linkage was performed using Automatch,17 with personal health number (a unique identifier of the publicly funded medical plan), address, name (first, middle, last and birth) and dates of birth and diagnosis as linking variables as appropriate. This linkage also identified deaths from all causes in the cohort.
The resulting data for the cohort consisted of dates of birth, death, entry to the cohort, and first breast cancer, postal code of residence at last screen, details of breast cancer and cause of death. The data on breast cancer cases not in the cohort consisted of dates of birth, death and first breast cancer, details of breast cancer and cause of death. The number of expected deaths from breast cancer in the cohort was calculated using the method described by Sasieni,18 which utilizes the age-specific time-at-risk of the cohort and requires the specification of age-specific incidence and survival rates expected without screening.
Age-specific incidence rates of breast cancer were calculated using data from the SMPBC, BCCR and published population counts for British Columbia. Rates were calculated in the usual way, with those in nonparticipants of SMPBC being calculated by subtracting SMPBC cancers and populations counts from the respective age-specific totals for the whole population by study year (1988–2003). In situ cancers were not included in the calculation of incidence rates. Statistical comparison of cancer incidence rates was made using a general linear model19 with a Poisson error distribution.
Breast cancer specific survival rates were estimated and analyzed using Cox regression20 on cancers diagnosed in the population between 1985 and 2003 with data taken from the BCCR and VSA. Both age, in decade categories, period of diagnosis and average personal income of area of residence in intervals defined by quartiles were included in the Cox analysis. Median family income was obtained from tax return information for 1996 and is aggregated for a unit based on the postal code which contains a maximum of 440 households per unit.21
Cases of in situ cancer and those registered by death certificate alone were excluded from the calculation of incidence and survival in nonparticipants. All deaths attributed to breast cancer occurring within the study cohort were included in the calculation of observed deaths if they occurred within the study period.
The SMPBC has a high retention rate, with 80% (90%) of women returning within 30 (60) months for another mammography screen.22 However, subjects leaving the province may not have their residence status updated. To estimate the effect of undocumented emigration on the time-at-risk of the cohort, a random sample of 200 women was drawn from the SMPBC database among those not having returned in 4 years after their last screen. Their last known address and their general practitioners office were contacted by telephone to ascertain their current residency. The result was then used to adjust the years-at-risk.
Observed and expected breast cancer mortality rates were calculated with cohort members considered to be at risk from entry until the first of death or December 2003. The calculation of expected used incidence and survival rates derived from nonparticipants. Calculation was also performed excluding deaths, both observed and expected, associated with nonscreen detected cancers diagnosed within 6 months of date of first screening. The purpose of this exclusion was to remove any effect of preexisting symptomatic cases, which are present in population rates but unlikely to be present in screening populations at entry. For women first screened before age 50, calculations were repeated excluding deaths, both observed and expected, associated with cancers diagnosed after age 50.
Observed and expected breast cancer mortality was compared assuming observed deaths to be Poisson distributed with mean value given by the expected mortality. The mortality ratio was calculated as the ratio of observed to expected mortality and confidence intervals were based upon Poisson statistics with allowance for sampling variation of the expected.
There were 598,690 women in the cohort who underwent a total of 2,196,441 mammograms in the period 1988–2003 for an average of 3.7 screens per woman. Record linkage identified 14,247 breast cancers in the cohort in the study period (Table I). There were a further 19,913 invasive breast cancers reported to the BCCR in British Columbia women not in the cohort during the study period.
Table I. Number of Women by Age and Year, and Number of Cancer Cases by Age, Year and Mode of Detection in the Cohort of Women Screened at Least Once through the Screening Mammography Program of British Columbia between 1988 and 2003
Number of women entering cohort
Breast cancer diagnoses
Mode of detection of cancers
Post screen ≤12 months
Post screen 13–24 months
Post screen 25–36 months
Post screen 37+ months
In the random sample of 200 women who had not returned for screening in 4 or more years, current residence could not be ascertained in 17 and among the known 15% (27/183) were no longer BC residents. Assuming the probability of migration to be uniform in time, the above information along with the observed pattern of return for screening indicated that the average annual out-migration rate for the cohort yielded 8.1 per 1,000 women per year. This rate was used to correct the person-years of follow-up in the cohort.
The age-specific incidence rates of breast cancer for women not in the study cohort in 1988–2003 Non-SMP and in the British Columbia population in 1985–1987 Pre-SMP are given in Figure 1. The age-specific population rates for 1985–1987 and for the nonparticipants were similar and the latter was used in the calculation of expected mortality of the cohort.
The survival rates of breast cancer in women aged 40–79 at diagnosis not in the study cohort in 1988–2003 Non-SMP and in the British Columbia population in 1985–1987 Pre-SMP are given in Figure 2. Survival rates were higher in those women not in the cohort 1988–2003 than in all British Columbia breast cancer cases diagnosed in the period 1985–1987 and thus the 1988–2003 cases were used for the calculation of expected mortality. To control for potential survival confounders, a Cox model was fit to the 1988–2003 nonmembers of the cohort with factors for age at diagnosis (40–49, 50–59, 60–69 and 70+), calendar period of diagnosis (pre- versus post-1996) and median income level of area of residence (<40, 40–50, 50–60, 60–70, 70+ thousand dollars). Age and area significantly effected survival and the categories were collapsed into homogeneous categories (Table II). The results from the resulting survival model were then applied, with the incidence rate, to calculate expected mortality for cohort members based on their age and region of residence while at risk in the cohort using the method described by Sasieni.18
Table II. Survival Estimates, Hazard Ratios and Confidence Intervals Obtained from Proportional Hazards Analysis of Invasive Cancers Diagnosed between 1988 and 2003 in Women not Participating in the Screening Mammography Program of British Columbia
Number of cases
95% confidence interval for hazard ratio
Median family income
Baseline disease specific survival estimates
The observed and expected number of deaths and mortality ratio by age-decade of starting screening are given in Table III. The mortality ratio was similar for each of the age ranges (p = 0.90). Observed and expected mortality rates were low in the initial years after entry into the cohort and the ratio of observed to expected breast cancer deaths stabilized beyond 5 years. Exclusion of deaths associated with cancers occurring within 6 months of first screening increased the mortality ratio for all age groups by 0.03–0.05 for all age groups (Table III). The observed and expected cumulative mortality rates by time since entry into the cohort (first screen) for women aged 40–49 at first screen is shown in Figure 3. Figure 3 illustrates that observed and expected breast cancer mortality increases slowly in the first 4 years after cohort entry reflecting the absence of symptomatic disease at first screening, but subsequently increases more rapidly. Expected mortality rates exceed observed from 2 years after first screening.
Table III. Observed and Expected Deaths from Breast Cancer, Mortality Ratio with Confidence Intervals (CI) and p-Value for Differences in Ratios by Age at First Screening and Assumed Time at Risk for Cancer Diagnosis
The observed number of cancers occurring in the cohort prior to age 50, 2,469, exceeded the expected, 1,492, reflecting the effect of earlier detection. In contrast, observed numbers (Table III), and rates (Fig. 3), of breast cancer deaths for cases diagnosed prior to age 50 were less than expected. Restriction to cases diagnosed before age 50 caused the mortality ratio to increase from 0.61 to 0.65 but it remained significant (p < 0.0001), indicating that the observed mortality reduction was not due to screening after the age of 50 in this group. Figure 3 also shows that observed mortality rates are less than the expected rates after limiting deaths to women in whom diagnosis occurred prior to age 50.
It is generally accepted that mammographic screening reduces breast cancer mortality in women over the age of 50 and it is recommended in most western countries. Mammography screening in women under the age of 50 has been more controversial regarding its efficacy and effectiveness. Recent meta-analyses of randomized screening trials of women under the age of 50 provide evidence of a significant 18% mortality reduction23 a little smaller than that found after age 50.1, 24 A population study in Sweden found a greater magnitude of reductions in both age groups.10 Analysis using a simulation model, based on data from screening trials conducted in Sweden, suggested that as much as 70% of the observed reduction in women aged 40–49 at start of screening may be due to screening performed after the age of 50.25 A subsequent analysis based on the same trials came to a somewhat different conclusion, finding almost all of the observed mortality reduction resulted from screening before the age of 50.26 The results of the study presented are supportive of the conclusion that mammographic screening does reduce mortality from breast cancer and that the proportionate mortality reduction is similar at ages above 40, although the magnitude of effect may be dependent upon the screening frequency. In current British Columbia practice, the shorter sojourn time (the time spent in an asymptomatic mammographically detectable state) in younger women is compensated for by screening more frequently,4, 27 with women being recommended to return annually before age 50 and biennially thereafter.
The analysis of recent service data from population-based screening programs can provide information upon the impacts of screening in settings where adjuvant systemic therapy is commonplace. By restricting attention to women entering screening it is easier to identify the effects of screening when population uptake has gradually increased over a long period, such as in British Columbia. As such, this study measures efficacy and not effectiveness. The cohort nature of this study design also permits the effects of screening to be evident earlier than using population mortality statistics since these include deaths from nonparticipants and cases diagnosed prior to the introduction of screening. Analyses have demonstrated temporal changes in breast cancer mortality after the introduction of mammographic screening and have inferred that much of this change is due to the impact of screening.9, 11, 28 Previous publications using SMPBC data have shown that women participating in screening are diagnosed with smaller tumors and survive longer.15 However, findings from observational studies are subject to potential bias associated with self-selection of screening by the study subjects. A review comparing the results from randomized trials and observational studies in 5 areas of medicine, including screening mammography, concluded that well-designed observational studies did not systematically overestimate the magnitude of treatment effects.29 Also, observational studies based on screening use estimate efficacy while randomized population based trials estimate effectiveness.30 Adjusting for compliance in randomized trials generally increases the effect size of screening.31 The results presented in this paper are consistent with those reported from case-control studies.30
The magnitude of the benefit estimated in this study is dependent upon the appropriateness of the incidence and survival rates used in the calculation of expected mortality. It was assumed that women who chose to participate had the same risk of developing breast cancer as those who did not. This was supported by the observation that nonparticipants in SMPBC had very similar breast cancer incidence rates to that of the total population prior to the creation of SMPBC. However, incidence rates in British Columbia women under age 40, where screening is uncommon, have declined by approximately one half of 1% per year over the study period. If a similar decline in breast cancer risk had occurred in screened women over the age of 40, then the calculation of expected mortality presented in Table III and Figure 3 would be approximately 5% too high. An inaccuracy of this magnitude does not substantially effect the conclusions of the analysis. Incidence rates would have had to decline by an average of 40% in the study period to reduce the measured effect to zero.
Assuming that survival rates of nonparticipants are appropriate for participants, in they had not been screened, is more problematic. This analysis found that income level of region of residence influenced survival and it was adjusted for in the analysis of mortality. However, participation in screening is influenced by personal health perception and other factors that may have effect independent of socioeconomic status. Previous research in the UK has found that breast cancer mortality was 17% higher among women offered but nonparticipating in screening than it was in a similar base population not offered screening.31 From the participation rate in the UK study (71%) and the reasoning contained in the paper by Duffy et al.,32 the expected breast cancer mortality ratio of nonparticipants to participants can be calculated, which yields a ratio of 1.26. If this mortality ratio applies to the data in this study, it would imply that the expected mortality calculated is 26% too high so that the breast mortality ratio associated with screening would be 0.76 (0.60 × 1.26).
Clinical trials of screening have employed standardized methods to assess cause of death to accurately identify women who die from breast cancer. This study relied on the death certificate cause of death as the outcome. While this is likely to contain some inaccuracy, it seems unlikely that this would result in any biased assessment in favor of screened women since both screened and unscreened women receive care through a common medical system. Outcomes among SMPBC participants could also be improved if they received superior treatment to nonparticipants. Previous publications have found breast cancer treatment patterns in British Columbia women to show a high degree of consistency during the study period33, 34 so that a large effect is unlikely. Although it can be argued that modern systemic therapy may have eroded the value of breast screening,14 large differences still persist in prognosis by extent of disease at diagnosis.35 In fact, during the study period, metastatic breast cancer remained essentially incurable while survival from early stage cancer improved.35
It appears that the majority of the mortality reduction seen in women participating in the SMPBC is due to the effect of mammography screening and that it is shared by all women over age 40. The results for women aged 40–49 contrasts with findings of a Canadian trial conducted in the same age group that found no mortality reduction.5 However, the results presented here for age group 40–49 are compatible with reviews of trial results2, 23, 24 for this age group after allowance of a 26% mortality advantage due to self-selection. The absolute benefit of screening for women aged 40–49 is less because the risk of breast cancer, and subsequent death, is lower than for older women. However, their life expectancy is higher, so that more years of life are potentially saved for every death prevented.
The authors acknowledge the helpful comments from the reviewers and associate editor, which substantially improved the quality of this manuscript.