Effect of service screening mammography on population mortality from breast carcinoma

Authors


  • See referenced original article on pages 458–69, this issue.

Abstract

Breast cancer deaths have been reduced as much as 44% in Swedish counties that offer service screening mammography. In the United States, death rates from breast cancer are now approximately 39 percent lower than expected given the increased incidence of invasive carcinoma. This estimated reduction may be attributed to heightened awareness of breast cancer, increased utilization of screening mammography, and improved treatment. See also pages 458–69.

Randomized clinical trials conducted over the past 40 years have produced more than ample proof of substantially reduced mortality rates from breast carcinoma among women offered screening mammography. For example, the Swedish Two-County Trial (1978–1985) found a 32% reduction in breast carcinoma mortality on 20-year follow-up of 77,080 women ages 40–74 years at the time of their initial invitation to undergo screening.1

However, randomized clinical trials (RCTs) cannot directly answer several important questions. Consider the potential differences in interpretive expertise and quality control between the research setting of the RCT and the environment of everyday clinical practice. Can service screening performed in a clinical practice setting duplicate results obtained during the RCT's? Realize the limitations in the design and analytic methods of the RCTs. Could these have led to underestimations of reductions in breast carcinoma mortality? Appreciate the improvements in mammographic equipment and technique since the era when RCTs were conducted. Will these allow for even greater benefits from screening performed today?

Aside from addressing these questions, service screening studies can effectively refute the controversial claim by Gøtzsche and Olsen2, 3 that RCT results were not valid because of supposed “flaws” in randomization and ascertainment of cause of death. By reducing breast carcinoma mortality rates in the Swedish population, service screening also can negate the much criticized study by Sjonell and Stahle4 that found no such outcome.

Results from RCTs

In addition to the Swedish Two-County Trial, two other RCTs have individually reported statistically significant reductions in breast carcinoma mortality among all women offered screening. The Health Insurance Plan (HIP) of Greater New York Trial (1963–1969) found a 23% reduction among women ages 40–64 years who were screened with a combination of annual mammography and clinical examination.5 The Edinburgh Scotland Trial (1979–1988) found a 29% decrease among women ages 45–64 at the time of entry into screening.6 Three other Swedish trials found reductions in breast carcinoma mortality that individually were not statistically significant. These were the Malmo Trial (MMST 1) (1976–1986), which screened women ages 45–70 years; the Stockholm trial (1981–1985), which screened women ages 45–65 years; and the Gothenburg trial (1982–1988), which screened women ages 40–59 years at entry. However, when their individual results were combined with those from the Two-County Trial, a statistically significant 24% reduction in breast carcinoma mortality was found.7 Only one trial, the National Breast Screening Study (NBSS) of Canada (1980–1985), failed to demonstrate any benefit from screening mammography.8, 9 Their results have been attributed to poor mammography technique and faulty trial design, which may have led to randomization errors.10

Significance of Improvements in Mammographic Technique since the Randomized Trial Era

Many substantial improvements in mammographic X-ray equipment, technique, screen-film processing systems, viewboxes, and viewing conditions have been made over the past 25 years.11 Because the mammographic findings associated with early breast carcinoma may be extremely subtle, the technical quality of mammographic images has a major effect on detection rates and the stage of disease detected.12 The use of a moving grid to improve image contrast came into widespread use in the mid 1980s and was associated with a fourfold increase in the detection of ductal carcinoma in situ at the time of screening in Kopparberg, Sweden.13 Adequate exposure is equally important. In the United Kingdom National Health Service Screening Program, the detection rate for small invasive tumors (greatest dimension ≤ 10 mm) was 50% higher at centers in which film densities were ≥ 1.2 density units.14 As a result of this study, published in 1994, the U.K. National Health Service and the American College of Radiology subsequently recommended a film density of ≥ 1.4 for mammography. Although in many RCTs screening was performed with only a single mediolateral oblique view, later studies showed that tumor detection rates increased by a mean of 7% with the use of an additional craniocaudal view, a practice that now is routine.15 One study found that women whose mammograms were performed with suboptimal breast positioning technique had a higher likelihood of interval tumors (i.e., tumors that are missed at the time of screening and surface clinically before the next screen).16 Thus, improvements in the practice of breast positioning in recent years may result in increased tumor detection rates. In summary, advances in mammographic equipment and technique should allow for greater mortality reductions than demonstrated in RCTs that were conducted in past years.

Limitations of RCTs in Quantifying Mortality Reduction

Compliance and contamination rates

An RCT is a prospective comparison of breast carcinoma mortality rates among a study group of women offered screening and an otherwise comparable control group of women not offered screening. There are several reasons why RCTs underestimate the true benefit from screening. First, to avoid selection bias, breast carcinoma mortality rates are tabulated for the entire study group offered screening. However, not all study group women accept the invitation to be screened. The mean attendance rate for all screening rounds at each trial was 66% in the HIP study, 85% in the Swedish Two-County trial, 78% in the Gothenburg trial, 81% in the Stockholm trial, 71% in the Malmo trial, and 55% in the Edinburgh trial.5–7 Second, some control group women obtain screening on their own. The reported percentage of control group women receiving mammograms outside each trial was 13% in the Swedish Two-County trial, 20%in the Gothenburg trial, and 25% in the Malmo trial.7 Both the “noncompliance” of some study group women and “contamination” of the control group reduce the calculated benefit from a trial. Thus, RCTs may underestimate the average benefit for an individual woman who accepts the invitation to screening.

Prevalence screen and number of screening rounds

A third reason why RCTs may underestimate the benefit from screening relates to the prevalence screen. Because the threshold for screening is below the usual clinical threshold in the absence of screening, the detection rate for an initial (prevalence) screen will be higher than the incidence of breast carcinoma in the control group. A prevalence screen will not only detect more tumors but also more advanced tumors than a subsequent incidence screen. Detection rates for subsequent (incidence) screens will be lower than those for the prevalence screen, unless the screening interval is longer than the earliest detection lead time.

When the last screening round of an RCT has been completed, the breast carcinoma mortality rates of the study group and control group are compared. Deaths from breast tumors that were diagnosed in both study and control groups before the screening period began are excluded from this tabulation so as not to dilute the calculated benefit. More cases of breast carcinoma are diagnosed in the study group than in the control group during the screening years, mainly because of the prevalence screen. Prevalence screens also may detect a larger number of poor-prognosis tumors than occur in normal incidence. Comparable cases may not all emerge in the control group until after the screening period. For these reasons, the use of the mortality rate from breast carcinoma cases diagnosed only during the screening period will underestimate the benefit from screening. This effect will be more substantial in trials having a shorter duration and fewer screening rounds.

One method with which to reduce this effect was suggested by Chu et al.17 in their reanalysis of data from the HIP trial, which had four annual screening rounds. Shapiro et al., the original HIP trial investigators, compared deaths from study and control group tumors diagnosed within 5 years of entry.5 Instead, Chu et al. compared death rates within 6 years of entry, the “catch-up point” at which the number of cases diagnosed in the control group equaled those in the study group. The limitation of the method of Chu et al. is that deaths from 2 years' worth of nonscreen-detected tumors were added to both the study and control group death rates. This action dampened the relative mortality reduction calculated from screening.

The five Swedish trials performed a single screen on the control group around the time of completion of screening of the study group. This initiative provided some screening benefit to the control group. It also brought the number of cancers in the control group in line with those in the study group. An appropriate number of deaths from breast cancers detected in this screen could then be included in the control group calculations. Despite this adjustment, the benefit from screening was still underestimated by these trials.

There still is another reason why an RCT, especially one with a limited number of screening rounds, will underestimate the benefit from longer-term periodic screening. Early tumors are more likely to be missed on an initial (prevalence) screen than on a subsequent (incidence) screen. Subtle mammographic findings are appreciated more easily on a subsequent screen because there are prior mammograms for comparison. Thus, trials with fewer screening rounds are more likely to underestimate the benefit from screening.

Among the randomized trials, the HIP trial screened annually for 4 rounds, the Gothenburg trial screened every 18 months for 4 rounds, the Two-County trial screened every 23–33 months for 4 rounds, the Malmo trial screened every 18–24 months for 5 rounds, and the Edinburgh trial screened every 24 months for 4 rounds. None of these trials had enough rounds to demonstrate the full benefit from screening.

Miettinen et al. used data from the Malmo study, which continued screening through Years 10–11 of follow-up, to estimate mortality reduction from continuous longer-duration screening.18 The authors hypothesized that when screening continues for a sufficiently long duration, benefits reach a “steady state.” Considering the limitations of conventional methods for the estimation of mortality reduction from screening, they instead tracked 3-year moving averages of mortality rate ratios (mortality rates for study group women/control group women) during 11 years of follow-up. The authors found that these ratios changed over time and were lowest during Years 8–11 of follow-up for women ages 55–69 years at the time of study entry. This amounted to a 55% “steady state” reduction in breast carcinoma mortality. This benefit may be compared with the lower 26% mortality reduction that Andersson and Nystrom found for 11 years of follow-up in women in the Malmo trial using conventional methods of calculation.19 Miettinen et al. also found a 60% mortality reduction for women ages 45–57 years at the time of study entry that began to appear at 10 years of follow-up and barely missed reaching statistical significance.18

Length of follow-up

Long-term follow-up of screening trials also is necessary because breast carcinoma is a chronic disease. Death from invasive disease may occur as long as 20 years after initial treatment.20 Benefits from the detection of an in situ tumor may not become apparent until long after those from detection of invasive disease.21 The latest lengths of follow-up reported from RCTs were as follows: 20 years in the Two-County trial, 18 years in the HIP trial, 14 years in the Edinburgh trial, 12 years in the Malmo trial, 8 years in the Stockholm trial, and 7 years in the Gothenburg trial. Even at 20 years of follow-up, the mortality difference between the study and control groups in the Two-County trial had widened from the previous 15-year follow-up.1

Length of screening intervals

Excessively long screening intervals may also prevent RCTs from demonstrating the full potential benefit from screening. The Two-County trial produced a 34% mortality reduction in women ages 50–69 years at entry who were screened every 33 months. The investigators estimated that if screening intervals had been every 2 years or every 1 year, mortality reductions of 39% and 45%, respectively, would have been achieved.22 For women ages 40–49 years at entry, a 13% mortality reduction was observed from screening every 2 years. The authors estimated a 36% mortality reduction if these younger women had been screened annually.22 Several other modeling studies also indicated substantially greater benefits from shorter screening intervals of 1 year.15, 22–24

Breast Carcinoma Mortality Reduction from Service Screening

After the success of the Swedish randomized trials, organized service screening mammography became routine in nearly all 24 Swedish counties by 1990. Based on very early results, Tornberg et al. projected in 1994 that within 10 years these programs would lead to a 19% decrease in breast carcinoma mortality among Swedish women ages 50–74 years.25

Between 1977–1992, the period of the Malmo randomized trial and subsequent service screening program, breast carcinoma mortality among women ages 45–69 years in Malmo decreased by 43% compared with 12% in the rest of Sweden, according to a study by Garne et al.26 The observation that the mortality reduction for the women in Malmo County was greater than that for study group women of the same age range in the Malmo trial was attributed by Moss to the 24% contamination rate for control group women in the trial.27

Beginning in 1990, screening mammography at intervals of 20 months was offered to women ages 40–74 years in Norrbotten and Västernorrland, 2 counties in northern Sweden. Two other adjacent counties that until that time had an identical breast carcinoma mortality rate served as controls. The mean attendance rate at the 3 screening rounds conducted between 1990–1995 was 86%. Lenner and Jonsson28 found a 28% decrease in breast carcinoma mortality among women ages 40–74 years and a 47% decrease among those ages 50–69 years. In 1995, breast carcinoma mortality among women ages 50–69 years was reported to have decreased by 67%, suggesting that a benefit of this size or even greater would have been sustained if periodic screening continued. In this study, the benefit was measured as a reduction in excess mortality (i.e., death from any cause in a woman diagnosed with breast carcinoma rather than death with breast carcinoma as the immediate cause).29

In Finland, nationwide population-based breast carcinoma screening every 2 years for women ages 50–59 years was introduced gradually between 1987–1991. Women born in even years began screening in 1987 or 1988. Women born in odd years began screening between 1989–1991 and served as controls. Attendance at screening was 85%. An effect from screening emerged after 3–4 years of follow-up and rapidly diluted as controls were screened. For this narrow window of time, Hakama et al. found that mortality from breast carcinoma was 24% lower among those women offered screening and 33% lower among those who actually were screened.30 Because control group women began screening only 2.5 years later on average than study group women, these results underestimate the effect of screening.

In Sweden, population-based service screening started in 1986 and by 1997 had been introduced in all counties. Jonsson et al. compared excess breast carcinoma mortality rates for study and control group counties.31 Study group counties were those that in which screening was initiated between 1986–1987. Control group counties were those in which screening was initiated in 1993 or later.

Only women ages 50–69 years were evaluated because only half the counties began screening at age 40 years and some counties did not invite women to undergo screening after age 69 years; women ages 50–69 years thus were invited to undergo screening in all counties. A 20% reduction in excess mortality from breast carcinoma was evident in the study group after a mean individual follow-up time of 8.4 years.31

In an earlier study involving Östergötland County and Kopparberg (Dalarna) County, (1 of the 7 Swedish Counties reported in the current issue of Cancer), Tabar et al.32 compared breast carcinoma mortality rates over 3 time periods: 1) 1968–1977, when virtually no women were screened; 2) 1978–1987, when half of the population was offered screening in the Two-County Trial; and 3) 1988–1996, after the completion of the trial when service screening was offered to all women ages 40–69 years and 85% of these women were screened.

When compared with breast carcinoma mortality rates among women ages 40–69 years in the prescreening era, breast carcinoma mortality rates for 1988–1996 were reduced by 50% for the entire population, which was comprised of 85% screened women and 15% nonscreened women. A 63% reduction in breast carcinoma mortality was found for screened women, similar to the 57% reduction in mortality for women screened during the Two-County Trial. During that era, when only approximately 50% of all women were offered screening, the reduction in the breast carcinoma mortality rate in the entire population was only 21%.

It is likely that screening, rather than advances in treatment, was responsible for nearly all the benefit. The relative risk of breast carcinoma death among nonscreened women ages 40–69 years was similar (1.0, 1.17, and 1.19, respectively) during the 3 consecutive time periods.

The study by Duffy et al. that appears in this issue of Cancer33 assesses the effect of service screening mammography on breast cancer mortality in seven Swedish counties where the radiologists had been trained by Tabar. Breast cancer mortality was reduced 44% from the prescreening era among women ages 40–69 years in 6 counties and those ages 50–69 years in 1 county. Rates for noncompliance with screening recommendations ranged from 8–36%. For compliers and noncompliers combined, mortality rates from breast carcinoma were 30% lower than those from the prescreening era. Counties with > 10 years of screening demonstrated a significant 32% reduction in mortality while counties with ≤ 10 years of screening demonstrated a significant 18% reduction in mortality from breast carcinoma.

The results from the study by Duffy et al.,33 as well as those from seven other reports regarding mammography service screening in Sweden and Finland,25, 26, 28, 32 confirm beyond any doubt that the substantial reductions in breast carcinoma mortality found in the RCTs also can be obtained in nonresearch, organized screening settings. These eight studies all contradict the conclusion of Sjonell and Stahle4 that screening had no apparent effect on breast carcinoma mortality rates in Sweden. Sjonell and Stahle concluded that the 1985 decision of the Swedish National Board of Health to invite women ages 40–70 years to undergo screening mammography every other year was not followed by any reduction in breast carcinoma mortality between 1987–1996.4 The conclusion of Sjonell and Stahle was incorrect because their analysis contained at least 4 fatal errors: 1) only 4 of the 17 counties in their study had initiated screening by the starting point of January 1987; 2) at least half of the breast carcinoma deaths that they included were from tumors diagnosed before the screening programs were initiated; 3) their study had insufficient follow-up time to demonstrate the effect of screening (at least 5–8 years are needed to observe even the beginning of any effect); and 4) the authors ignored any effect from the increased incidence of breast cancer during the study period.34, 35

Reduction in Breast Carcinoma Mortality in the U.S.

In addition to the Scandinavian counties, several other national healthcare services now perform screening mammography as a preventive healthcare measure. In the United Kingdom, women ages 50–69 years are offered screening every 3 years. In the Netherlands, women ages 50–74 years may obtain screening every 2 years. Based on initial reductions in breast carcinoma mortality, van den Akker-van Marle et al. projected long-term mortality reductions of 24% in the U.K. and 29% in the Netherlands.36

In the U.S., the use of screening mammography has increased continuously since 1975. Surveys performed by the National Center for Health Statistics found that the percentage of women age ≥ 40 years who reported having undergone a mammogram within the past 2 years was as follows: 28.8% in 1987, 55.8% in 1992, and 66.9% in 1998.37

Data from the Surveillance, Epidemiology, and End Results (SEER) 2002 Cancer Statistics Review, which encompasses the period 1973–1999, can be used to analyze changes in breast carcinoma incidence and mortality in the U.S.20 Cancer rates in this report are adjusted to a single population “standard” that is based on the Year 2000 U.S. census.38 Although the incidence of invasive breast carcinoma was reported to have increased markedly by 30.9% between 1980–1990, breast carcinoma mortality rates increased only 4.4% over that period. Breast carcinoma incidence rose slightly by 4.5% between 1990–1999, whereas the mortality rate from breast carcinoma was reported to have decreased moderately by 17.0% during that decade (Table 1).

Table 1. U.S. Age-Adjusted Breast Carcinoma Incidence and Mortality Ratesa
  • a

    The rate is per 100,000. The incidence is for invasive breast carcinoma only. Adjusted for 2000 U.S. population.

  • Data taken from Ries LAG, Eisner MP, Kosary CL, et al., editors. SEER cancer statistics review, 1973–1999. Bethesda, MD: National Cancer Institute, 2002.

Year19801985199019951999
Incidence101.6121.6133.0133.3139.0
Mortality31.732.933.130.827.5

Relative survival rate curves for invasive breast carcinoma cases, as well as cumulative mortality curves from screening mammography trials, indicate that the majority of deaths from breast carcinoma occur within the first 10–15 years after detection and treatment.1, 20 Changes in breast carcinoma mortality rates reflect the effects of detection and treatment on tumors that comprised the invasive breast carcinoma incidence over the previous 15 years.

The 17.0% decrease in breast carcinoma mortality noted between 1990–1999 may be compared with the 36.8% increase in the incidence of invasive breast carcinoma reported between 1980–1999. If there had been no improvements in detection, treatment, and breast cancer awareness between 1980–1999, mortality rates for the period 1990–1999 would be expected to increase by 36.8% rather than decrease by 17.0%. This would suggest that the actual reduction in breast carcinoma mortality between 1990–1999 was approximately 39% for the average patient with invasive breast carcinoma.

The following is an explanation of calculations of estimated 1990–1999 mortality reduction, based on Table 1. The increased incidence = 139.0/101.6 = 1.37; the expected mortality rate = 1.37 × 33.1 = 45.3; and the absolute difference between the expected and observed (1999) mortality rates = 45.3 – 27.5 = 17.8. The relative difference = 17.8/45.3 = 39%.

Stated otherwise, I have estimated that death rates from breast cancer in the United States may now be approximately 39 percent lower than expected on the basis of increased incidence of invasive carcinoma. Reasons for this estimated decline may include increased breast cancer awareness, increased use of screening mammography, and improved treatment. It may be that early detection may be responsible for a substantial portion of the decline in breast cancer mortality. Studies on the treatment of invasive breast cancer have found that at most, Tamoxifen reduces death rates by 22%.39 Furthermore, if my calculations had assumed that some cases of invasive breast cancer were prevented through mammographic detection of ductal carcinoma in situ, the estimated benefit attributed to mammography would be even greater.

CONCLUSIONS

Already proven by RCTs, breast cancer mortality reduction through screening mammography has now been confirmed with organized service screening programs as well. Both types of studies have limitations that result in underestimation of benefit for the individual woman who is screened. Nevertheless, the benefits calculated from service screening studies are often greater than those from RCTs. Swedish counties with 10 or more years of services screening have shown a significant 44 percent decrease in breast cancer deaths among women who were screened. In the U.S., analysis of breast cancer incidence and mortality rates suggest that since 1980, there has been an approximate 39 percent reduction in breast cancer mortality for the average patient with invasive breast cancer. The proportions of this estimated gain attributable to greater breast health awareness, increased utilization of screening mammography, and better treatment cannot be precisely determined. However, further gains may be anticipated as more women comply with American Cancer Society screening guidelines.

Ancillary