Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Observed and Predicted Risk of Breast Cancer Death in Randomized Trials on Breast Cancer Screening

  • Philippe Autier ,

    philippe.autier@i-pri.org

    Affiliations Strathclyde Institute of Global Public Health at iPRI, Lyon, France, International Prevention Research Institute (iPRI), Lyon, France

  • Mathieu Boniol,

    Affiliations Strathclyde Institute of Global Public Health at iPRI, Lyon, France, International Prevention Research Institute (iPRI), Lyon, France

  • Michel Smans,

    Affiliation International Prevention Research Institute (iPRI), Lyon, France

  • Richard Sullivan,

    Affiliation Institute of Cancer Policy, Kings Health Partners Cancer Centre, Bermondsey Wing, Guy’s Campus, London, United Kingdom

  • Peter Boyle

    Affiliations Strathclyde Institute of Global Public Health at iPRI, Lyon, France, International Prevention Research Institute (iPRI), Lyon, France

Observed and Predicted Risk of Breast Cancer Death in Randomized Trials on Breast Cancer Screening

  • Philippe Autier, 
  • Mathieu Boniol, 
  • Michel Smans, 
  • Richard Sullivan, 
  • Peter Boyle
PLOS
x

Abstract

Background

The role of breast screening in breast cancer mortality declines is debated. Screening impacts cancer mortality through decreasing the number of advanced cancers with poor diagnosis, while cancer treatment works through decreasing the case-fatality rate. Hence, reductions in cancer death rates thanks to screening should directly reflect reductions in advanced cancer rates. We verified whether in breast screening trials, the observed reductions in the risk of breast cancer death could be predicted from reductions of advanced breast cancer rates.

Patients and Methods

The Greater New York Health Insurance Plan trial (HIP) is the only breast screening trial that reported stage-specific cancer fatality for the screening and for the control group separately. The Swedish Two-County trial (TCT)) reported size-specific fatalities for cancer patients in both screening and control groups. We computed predicted numbers of breast cancer deaths, from which we calculated predicted relative risks (RR) and (95% confidence intervals). The Age trial in England performed its own calculations of predicted relative risk.

Results

The observed and predicted RR of breast cancer death were 0.72 (0.56–0.94) and 0.98 (0.77–1.24) in the HIP trial, and 0.79 (0.78–1.01) and 0.90 (0.80–1.01) in the Age trial. In the TCT, the observed RR was 0.73 (0.62–0.87), while the predicted RR was 0.89 (0.75–1.05) if overdiagnosis was assumed to be negligible and 0.83 (0.70–0.97) if extra cancers were excluded.

Conclusions

In breast screening trials, factors other than screening have contributed to reductions in the risk of breast cancer death most probably by reducing the fatality of advanced cancers in screening groups. These factors were the better management of breast cancer patients and the underreporting of breast cancer as the underlying cause of death. Breast screening trials should publish stage-specific fatalities observed in each group.

Introduction

In 1985, publications of the Greater New-York Health Insurance Plan and of the Two-County Trial suggested that 2 to 4 rounds of breast screening could reduce the risk of breast cancer death by 30% [1, 2]. Following these publications, mammography screening expanded rapidly in the USA, Sweden, Finland, the United Kingdom, and in the Netherlands. Three other randomized trials conducted in the 1980-90s in Sweden also obtained risk reductions ranging from 10 to 22% [3]. Two randomized trials conducted in Canada obtained no reduction in the risk of breast cancer death with mammography screening [4, 5].

Meta-analyses of these trials have suggested that a breast cancer mortality reduction of about 20% could be expected if at least 70% of women aged 40 to 74 years attended two to five rounds of mammography screening [6]. However, studies in Europe that compared changes in breast cancer mortality between areas with similar economic level and access to treatment, but with early or with late (i.e., 10 to 15 years later) implementation of mammography screening found no difference in changes of breast cancer mortality rates over time [79]. The progressive introduction of mammography screening in Swedish counties had no impact on breast cancer mortality trends at county level [10]. In the USA, participation to screening ranging from 40 to 90% did not affect rates of breast death after 30 years of screening [11]. No association was found between the timing and magnitude of declining trends in breast cancer mortality and the timing of mammography screening implementation in the various States of the USA [12]. This apparent absence of impact of mammography screening on breast cancer mortality was in sharp contrast with studies that clearly showed quicker and steeper reductions in the risk of cervical and colorectal cancers in areas where screening is widespread, compared to areas where screening is not common [1315].

How can such contrast between results of meta-analyses and of population studies on breast screening be most logically explained? We hypothesize that randomized trials overestimated reductions in risks of breast cancer death associated with screening. The hypothesis can be tested in a simple way. Advanced cancers are highly fatal and screening reduces cancer mortality through decreasing numbers of patients diagnosed with an advanced cancer. Hence, the reduction in the risk of cancer death observed in a trial should be equivalent to the reduction of the risk of cancer deaths associated with the lower rate of advanced cancer in the screening group as compared to the control group (i.e., the predicted reduction). If the observed risk reduction is greater than the predicted risk reduction, then it is likely that factors other than screening have diminished the fatality of advanced cancers in the screening group.

The objective of this study is to test this hypothesis. For didactic purpose, we first tested the hypothesis for the Prostate Lung Colorectal and Ovarian (PLCO) trial on colorectal cancer (CRC) screening with flexible sigmoidoscopy that reported the relevant data [16].

Methods

Cancer fatality

In a randomised trial, cancer fatality is the proportion of cancer patients who, from randomisation until the end of follow-up, die because of their cancer. Cancer fatality can be readily deduced from trial data (Table 1). The early stage cancers are usually small, local cancers, and the advanced cancers are cancers with regional extension (e.g., in regional lymph nodes) or with metastases in distant organs. At end of follow-up, for each category of cancer stage, the numbers of cancer patients and of cancer deaths are totalised. Stage specific fatality can be deduce by dividing numbers of cancer deaths by numbers of cancer patients (i.e., f = d/n). Conversely, the number of cancer deaths in each stage category can be derived from the number and the cancer-specific fatality of cancer patients (d = n*f).

thumbnail
Table 1. Typical results of randomised trials testing the efficacy of interventions aiming at reducing the risk of cancer death.

https://doi.org/10.1371/journal.pone.0154113.t001

Assuming that numbers of subjects randomised in the intervention and control group are identical, an efficient screening method would reduce the number of advanced cancers in the intervention group (nai<nac), but their fatality would remain the same (fai = fac). As a consequence there would be fewer cancer deaths due to late diagnosis in the intervention than in the control group (dai<dac), and the relative risk computed as Di/Dc would be smaller than 1.0. In contrast, an efficient treatment would reduce the fatality of advanced cancer in the intervention group (fai<fac), but their numbers would remain the same (nai = nac). As a consequence among subjects with advanced cancer, there would be fewer cancer deaths in the intervention than in the control group (dai<dac), and the relative risk computed as Di/Dc would be smaller than 1.0. It follows that numbers of cancers deaths and relative risks observed in a trial can be predicted from knowledge of the number of cancer patients in each stage category and the fatality specific to each stage category.

Literature search

We performed a systematic search on PubMed for articles on breast screening trials. We also searched reference lists in major reviews [1719] and in publications on breast screening trials we already had. We retrieved all articles reporting original data of these trials. Three of us (PA, MB, MS) read the articles with looking for cancer-specific fatality data reported by categories of size or lymph-node status or stage. For the sake of finding comparable data on screening for other cancers, we performed a similar literature search for colorectal cancer. Fatality could be reported as a proportion of cancer patients who died because of the cancer or as a survival statistics, which is the reverse of fatality. It could be reported as a measure of the risk (e.g., the hazard rate) to die from a large (more advanced) cancer compared to the risk to die from a small (less advanced) cancer. We also looked for breast screening trials that performed their own assessment of predicted numbers of breast cancer deaths.

Statistical analysis

The statistical methods we used were the same as those used by Morrison [20] and by Tabar et al, 1995 [21]. When fatality rates were reported for the screening and control group separately, we computed the predicted numbers of cancer deaths in the screening group using the size-, or stage-, or lymph-node specific fatality rates of the control group. We then computed the predicted relative risk and 95% confidence interval using the predicted number of cancer deaths in the screening group and the number of observed cancer deaths in the control group. When fatality rates were calculated after pooling breast cancer patients found in the screening and the control group, we computed the predicted numbers of cancer deaths in the screening and in the control group using the size-, or stage-, or lymph-node specific fatality rates reported for both groups. We then computed the predicted relative risk and 95% confidence interval using the predicted numbers of cancer deaths in the screening and in the control group.

For the Two-County trial, two articles reported numbers of breast cancers by size categories diagnosed in women 40 to 74 years of age during the intervention period [22, 23]. Another article reported the 11-year size-specific crude hazard rates of breast cancer death in women 40 to 69 years of both screening and control groups [24]. Hazard rates represented the risk of breast death of women diagnosed with a cancer 10 mm size or more relative to the risk of breast cancer death of women diagnosed with a small (1–9 mm) breast cancer. The transformation of hazard rates into fatality (in percent) can be done using the cumulative mortality formula displayed in the Appendix of Tabar et al, 1995 [21], i.e.,

The value 0.052 is the 11-year fatality rate for the reference category of cancers 1–9 mm size, that was derived from Figure 7 of Tabar et al, 1992 [22].

As an example, we first applied the statistical method on data reported by the PLCO trial on flexible sigmoidoscopy [16].

Results

The PLCO colorectal cancer screening trial

The PLCO trial reported an observed relative risk of colorectal cancer death of 0.74 (95% CI: 0.63 to 0.87) associated with screening with flexible sigmoidoscopy [16](Table 2). There were 225 fewer stage II-IV CRCs in the screening than in the control group. There were also fewer CRCs in the screening than in the control group because sigmoidoscopy allows the removal of adenomas. Stage-specific fatality rates in the screening and in the control group were comparable, but for the carcinoid tumours, probably because screening could detect slow-growing carcinoid tumours. We computed the predicted number of colorectal cancer death in the screening group by multiplying cancer numbers in each stage category of the screening group by the fatality of corresponding categories of the control group. We obtained a predicted relative risk of 0.76 that was close to the observed relative risk of 0.74.

thumbnail
Table 2. Reported and predicted risk of colorectal cancer (CRC) death in the PLCO trial*.

https://doi.org/10.1371/journal.pone.0154113.t002

Breast cancer screening trials

The Greater New-York Health Insurance Plan (HIP) trial and the Two-County Trial (TCT) reported relevant data. Other studies also computed the predicted relative risks of breast cancer death for these trials [20, 21]. The Age trial performed a comparison of observed and of predicted breast cancer mortality but did not report fatality rates observed during the trial [25]. We therefore just summarized the key findings of this trial. For the Malmö, Stockholm and Goteborg trials, we found no article that reported on size, lymph-node or stage-specific fatality.

The Greater New-York Health Insurance Plan (HIP) trial

After 7 year-follow-up, the HIP trial reported 97 breast cancer deaths in the screening group for 131 in the control group, resulting in an observed relative risk of 0.72 (95% CI: 0.56–0.94) (Table 3)[26].

thumbnail
Table 3. Reported and predicted breast cancer deaths in the HIP trial*.

https://doi.org/10.1371/journal.pone.0154113.t003

At first sight the trial fulfilled the objective to reduce breast cancer deaths associated with advanced cancer because there were 31 (i.e., 79–48) fewer breast cancer deaths in women with positive lymph-node in the screening as compared to the control group. However, there were only 19 (i.e., 121–102) fewer women with positive lymph-node in the screening as compared to the control group. So, the number of breast cancer deaths among lymph-node positive women had reduced more than the number of women with positive lymph-node. This result was possible because the 7-year fatality in the screening group was 27% lower (i.e., 0.475/0.650) than in the control group. Similarly, the 7-year fatality of lymph-node negative women was 28% lower (i.e., 0.185/0.256) in the screening than in the control group. As a consequence, when 7-year fatality rates observed in the control group were used for predicting breast cancer deaths in the screening group, there were 66 predicted instead of 48 observed breast cancer deaths among lymph-node positive women, and 44 predicted instead of 33 observed breast cancer deaths in lymph-node negative women.

Overall, using fatality rates reported for the control group, we predicted 5 fewer breast cancers associated with screening, equating to a predicted relative risk of 0.98 (95% CI: 0.77–1.24). There were 14 (5%) more invasive breast cancers in the screening than in the control group, and the predicted relative risk computed after exclusion of extra cancers was similar. Hence, reductions in breast cancer deaths in the HIP trial were essentially the consequence of reductions in fatality rates that were due to factors other than screening.

Morrison who had direct access to HIP data [27] used 10-year fatality rates of breast cancers in the control group for predicting reduction in breast cancer deaths in the screening group. The prediction was 8 fewer breast cancer deaths, equating to a predicted relative risk of 0.94 (95% CI: 0.75–1.19).

The Two-County trial

Size-specific numbers of breast cancer reported by the Two-county trial after 11 years of follow-up are displayed in columns (1) and (2) of Table 4 [22]. In column (3), we corrected the numbers of breast cancer in the control group for the imbalance in group size. There were 85 fewer breast cancer deaths in the screening than in the control group and the relative risk of breast cancer death was 0.73 (95% CI: 0.62–0.87).

thumbnail
Table 4. Computations of predicted numbers of breast cancer deaths in the Two-County trial.

https://doi.org/10.1371/journal.pone.0154113.t004

A comparison of columns (1) and (3) shows that there were 135 less invasive cancers having a size of 20 mm or more in the screening than in the control group, and 385 more invasive cancers 1 to 19 mm size in the screening than in the control group(column (8) of Table 4). A part of extra cancers 1 to 19 mm size proceeds from the advance in diagnosis of cancer that would have been larger in the absence of screening (i.e., lead time effect) and another part are screen-detected cancers that would probably have never become clinical (i.e., length time effect), thus representing overdiagnosis.

According to the Two-county trial investigators, only 1% of in situ and invasive breast cancer found in the screening group of this trial would represent overdiagnosis [28]. So, considering that practically no extra invasive breast cancer would be overdiagnosed, and using the hazard ratios reported in Duffy et al (1991),[24] we first computed the predicted 11-year fatalities (Column (5)), and then the numbers of predicted cancer deaths in the screening and in the control group (columns (6) and (7)). The predicted relative risk was 0.89 (95% CI: 0.75–1.05). This predicted relative risk suggests that the reduction in the risk of breast of cancer death due to down-staging was 11%, and that the remaining reduction of 16% (i.e., (1–0.73)-0.11) would be due to factors other than screening.

As a sensitivity analysis, we performed computations considering that all extra cancers represented overdiagnosis. To this end, we redistributed the 135 cancers 20 mm or more deemed to have been detected earlier thanks to screening (column (9)) in size-specific categories of cancers less than 20 mm of the control group. We performed these transfers of cancer cases with keeping the ranking of cancers across size categories. For instance, the 108 cancers 1 to 9 mm in column (10) were the sum of the 69 cancers 1 to 9 mm of the control group (column (3)) and of the 39 fewer cancers 20 to 29 mm in the screening group (column (8)). The predicted relative risk was 0.83 (95% CI: 0.70–0.97).

Using the same fatality data and statistical formula we used, the Two-county trial investigators computed predicted relative risks of breast cancer death for women 40 to 49 and 50 to 74 years, equating to a predicted relative risks of 0.76 (95% CI: 0.64–0.89) for women 40 to 74 years (S1 Table)[21]. This predicted relative risk is smaller than the ones we obtained but still at a distance from the relative risk of 0.69 reported by the Two-county trial investigators [21]. The dissimilarity between our predicted relative risk and that of Two-county trial investigators proceeds from the total number of breast cancer in the control group (not adjusted for the difference in group size): 1041 in the S1 Table for 729 in Table 3. The difference of 312 cancers in the control group corresponds to the addition by the Two-county trial investigators of 266 invasive and 46 in situ breast cancers found at first screening of the control group that took place in years following the end of the intervention period [22, 23]. The 266 invasive cancers included 73 cancers of 20 mm or more, which boosted the numbers of predicted breast cancer deaths in the control group, and led to a predicted relative risk of 0.76 much smaller than the one of 0.89 we obtained. The article of Tabar et al (1995)[21] does not allude to the addition of breast cancers found at first screening of control women to breast cancers found in this group during the screening period.

The Age trial

After a mean follow-up of 10.7 years, the observed relative risk of breast cancer death in the Age trial was 0.83 (95% CI: 0.66–1.04)[29]. The cumulative incidence of breast cancers 20 mm or more was 3.2 per 1,000 women in the screening group for 3.6 per 1,000 women in the control group [30]. For women who were part of the Age trial for at least 10 years, the observed relative risk of breast cancer death was 0.79 (95% CI: 0.60–1.06)[29].

The Age trial investigators computed predicted relative risks using tumour size, grade, and lymph-node status for a 10-year period. Predicted relative risks were 0.90 (95% CI: 0.80–1.01) when the Nottingham Profile Index was used and 0.89 (95% CI: 0.78–1.01) when the Two-County trial procedure was used [21, 25]. There were 8% more invasive breast cancers in the screening than in the control group, and a predicted relative risk computed after exclusion of extra cancers was likely to be similar.

Discussion

Our analysis found that in the randomized trial on colorectal cancer screening with flexible sigmoidoscopy, the observed and predicted relative risks were equivalent, indicating that the reduction in colorectal cancer mortality obtained by the trial was entirely due to stage-shift associated with screening. In contrast, in breast screening trials, the relative risks reported in publications were substantially smaller than the predicted relative risks, which suggests that a substantial part of reductions in breast cancer mortality obtained by these trials were due to factors other than reductions in the incidence rate of advanced breast cancer associated with screening. No other breast screening trial reported data allowing the comparison of observed and of predicted risk reductions. However, a recent study [31] computed an overall predicted relative risk of breast cancer death in women 50–74 years of age using the evidence that the mean cumulative rates of advanced cancer in the screening groups of all breast screening trials was on average 15% lower than in control groups [32]. The Birnbaum et al. study estimated that reductions in rates of advanced cancer would translate into a mean 10% reduction in the risk of breast cancer death. This 10% reduction is in sharp contrast with meta-analyses of breast screening trials have reported relative reductions of 23% in the risk of breast cancer death in women aged 50 years or more [19].

Published data of the HIP trial provide clues about the non-screening factors implicated in this trial. The fatality of breast cancer patients with none or with one or more positive lymph nodes was 27% lower in the screening than in the control group (Table 2) [33]. In the 1960s, no efficient chemotherapeutic adjuvant or therapeutic regimen existed, and the HIP cause of death committee was totally blinded as to the randomization status of women [34]. The more plausible explanation for the lower fatality proceeds from data showing first that although women who did not participate to screening represented 35% of women in the screening group, only 25% (i.e., 74/299) of breast cases and 29% (i.e., 28/96) of breast cancer deaths occurred in these non-participants (S2 Table). Second that the breast cancer fatality in women invited to screening but who did not attend screening was lower than the fatality of interval cancers and of cancers in the control group (S2 Table). These data in non-participants are clearly anomalous. In the Two-County trial, 27% of breast cancer deaths occurred in the 15% of women who did not participate to screening [22] and an abundant literature documents that breast cancers diagnosed in women not attending screening are more advanced and have a fatality rate 1.5 to 2 times greater than that of interval cancers or of cancers in control women [24, 3539]. Investigations of characteristics of women included in the HIP trial revealed that the tracing of women who refused screening had been arduous, which may have led to information gaps on cancer occurrence and causes of death for many of these women [40]. So, the breast cancer mortality reduction in the HIP trial was most probably due to the failure to register a substantial number of breast cancer deaths in non-participants.

In the Two-county trial, the observed relative risk was 0.73 (0.62–0.87), while the predicted relative risk was 0.89 (0.75–1.05) if, as suggested by investigators, [28] overdiagnosis was assumed to represent 1% of all cancers. The predicted relative risk was 0.83 (0.70–0.97) if extra cancers were excluded. Similarly to the HIP trial, could the size-specific fatalities in the Two-county trial have been lower in the screening than in the control group? The Two-county trial never reported size-specific fatalities for the screening and for the control group separately. It is therefore impossible to formally establish that in this trial, the size-specific fatalities of the large breast cancers were smaller in the screening than in the control group. But an indirect proof exists. If fatalities of large cancers were actually smaller in the screening than in the control group, then the reported mean fatalities (noted “fam”) should be greater than the actual (unknown) fatalities in the screening group, and smaller than the actual (unknown) fatalities in the control group (i.e., fai <fam<fac in Table 1). As a consequence, computations of predicted breast cancer deaths in the screening group using the mean fatalities (i.e., nai*fam) will end up in a predicted number of breast cancer deaths greater than the observed number. Conversely, computations of predicted breast cancer deaths in the control group using the mean fatalities (i.e., nac*fam) will end up in a predicted number of breast cancer deaths smaller than the observed number. As a matter of fact, in Table 4, the reported and predicted numbers of breast cancer deaths in the screening group were 234 and 267 respectively, whilst in the control group these numbers were 319 and 299, respectively. These numerical relationships accredit the hypothesis that in the Two-county trial, size-specific fatalities in the screening group were lower than in the control group.

The other Swedish trials and the UK Age trial did not report breast-cancer specific fatality according to size, stage, or lymph node status for the screening and the control group separately. Moss et al did not provide a firm explanation for discrepancies they found in the UK Age trial [29]. A legitimate concern is thus that in all trials that found a reduced risk of breast cancer death associated with screening, the stage (or size or lymph-node) specific fatality of breast cancer women was lower in the screening than in the control group.

The quest for factors that may have contributed to a lower fatality in screening groups needs to consider the “left-to-nature” design adopted by all trials that found reduced risk of breast cancer death associated with screening. Typically, parallel group randomized trials first recruit a group of eligible subjects that are informed on trial objectives, on potential health benefits and probable side effects. Subjects agreeing to participate must first sign an informed consent form after which they are randomized in a screening or in a control group. This typical procedure was used in the Canadian trials [4, 5]. In left-to-nature trials, only women invited to participate in breast screening knew they were part of a clinical trial. Women allocated to control groups were never contacted, did not sign an informed consent and were completely ignorant they were part of a trial. All the follow-up of control women was purely administrative, with linkage between the trial and the cancer registries, and also linkage with the cause of death registry if causes of death were those reported on death certificates. Health professionals could thus know or detect which women were invited to screening. In contrast, health professionals could not identify which women in the population were allocated to control groups.

Size-specific fatality is derived from numbers of breast cancer patients in each size category found during the intervention period and from numbers of breast cancer deaths observed in each size category (Table 1). A first question is whether the assessment of cancer size (or stage, or lymph-node status) could have differed between the screening and the control group, with the consequence that compared to the control group, breast cancer patients of the screening group would have been systematically classified in larger size categories. This hypothesis assumes that compared to the control group, reductions in large size cancers in the screening group were underestimated. This hypothesis is unlikely because cancer size is measured during the histological examination of the surgically removed breast tissues. The same laboratories analysed surgical materials of women invited and not invited to screening, and there is no reason to believe that the histological analyses could have been different for women invited and not invited to screening.

In the Two-county trial, causes of death were established by local endpoint committees, and in the UK Age trial, death certificates were used. The absence of blinding of health professionals may have caused underreporting of breast cancer as the underlying cause of death of women invited to screening, which could have biased results in favour of screening. This hypothesis is supported by knowledge that relative risks of breast cancer death in the Two-county trial were always lower when causes of death were derived from death certificates than from the local endpoint committee [41, 42]. But death certificates are also prone to reporting bias, as shown by an intervention study on skin screening in Germany where the absence of blinding of doctors ended in the underreporting of melanoma as the underlying cause of death on death certificates [43, 44].

The lower fatality in the screening group may also have been due to the better medical management of women invited to screening (i.e., a performance bias). Better management means quicker referral to work-up procedures of women found with a breast lump, better access to specialized care, and more intense therapy.

In conclusion, our study indicates that in three randomized trials for which relevant data were available, the actual ability of mammography screening to decrease the risk of breast cancer death had been overestimated. Breast screening trials should make available the cancer-specific survival or fatality rates by stage or by size for the screening and control groups, separately.

Supporting Information

S1 STROBE Checklist. Checklist of items that should be included in reports of observational studies.

https://doi.org/10.1371/journal.pone.0154113.s001

(DOC)

S1 Table. Predicted numbers of breast cancer deaths in the two-County trial reported in Tabar et al, Cancer, 1995, table 6.

* Data in italicized letters were not reported table 6 of Tabar et al, 1995.[1]. † Includes invasive and in situ cancers.

https://doi.org/10.1371/journal.pone.0154113.s002

(DOCX)

S2 Table. Reported invasive breast cancer patients and deaths in the HIP trial.

BC: breast cancer. *Data from Table 9 of Shapiro et al, 1977 [1].

https://doi.org/10.1371/journal.pone.0154113.s003

(DOCX)

Author Contributions

Conceived and designed the experiments: PA. Performed the experiments: PA MB. Analyzed the data: PA MS MB. Contributed reagents/materials/analysis tools: PA. Wrote the paper: PA MB RS MS PB.

References

  1. 1. Tabar L, Fagerberg CJ, Gad A, Baldetorp L, Holmberg LH, Grontoft O, et al. Reduction in mortality from breast cancer after mass screening with mammography. Randomised trial from the Breast Cancer Screening Working Group of the Swedish National Board of Health and Welfare. Lancet. 1985;1(8433):829–32. pmid:2858707
  2. 2. Shapiro S, Venet W, Strax P, Venet L, Roeser R. Selection, follow-up, and analysis in the Health Insurance Plan Study: a randomized trial with breast cancer screening. Natl Cancer Inst Monogr. 1985;67:65–74. pmid:4047153
  3. 3. Nystrom L, Andersson I, Bjurstam N, Frisell J, Nordenskjold B, Rutqvist LE. Long-term effects of mammography screening: updated overview of the Swedish randomised trials. Lancet. 2002;359(9310):909–19. pmid:11918907
  4. 4. Miller AB, To T, Baines CJ, Wall C. The Canadian National Breast Screening Study-1: breast cancer mortality after 11 to 16 years of follow-up. A randomized screening trial of mammography in women age 40 to 49 years. Ann Intern Med. 2002;137(5 Part 1):305–12. pmid:12204013
  5. 5. Miller AB, To T, Baines CJ, Wall C. Canadian National Breast Screening Study-2: 13-year results of a randomized trial in women aged 50–59 years. J Natl Cancer Inst. 2000;92(18):1490–9. pmid:10995804
  6. 6. Myers ER, Moorman P, Gierisch JM, Havrilesky LJ, Grimm LJ, Ghate S, et al. Benefits and Harms of Breast Cancer Screening: A Systematic Review. Jama. 2015;314(15):1615–34. pmid:26501537
  7. 7. Autier P, Boniol M, Gavin A, Vatten LJ. Breast cancer mortality in neighbouring European countries with different levels of screening but similar access to treatment: trend analysis of WHO mortality database. BMJ. 2011;343:d4411. pmid:21798968
  8. 8. Jorgensen KJ, Zahl PH, Gotzsche PC. Breast cancer mortality in organised mammography screening in Denmark: comparative study. Bmj. 2010;340:c1241. pmid:20332505
  9. 9. Broeders MJ, Peer PG, Straatman H, Beex LV, Hendriks JH, Holland R, et al. Diverging breast cancer mortality rates in relation to screening? A comparison of Nijmegen to Arnhem and the Netherlands, 1969–1997. Int J Cancer. 2001;92(2):303–8. pmid:11291061
  10. 10. Autier P, Koechlin A, Smans M, Vatten L, Boniol M. Mammography screening and breast cancer mortality in Sweden. J Natl Cancer Inst. 2012;104(14):1080–93. pmid:22811439
  11. 11. Harding C, Pompei F, Burmistrov D, Welch HG, Abebe R, Wilson R. Breast Cancer Screening, Incidence, and Mortality Across US Counties. JAMA Intern Med. 2015.
  12. 12. Bleyer A, Baines C, Miller AB. Impact of Screening Mammography on Breast Cancer Mortality. Int J Cancer. 2015.
  13. 13. Ait Ouakrim D, Pizot C, Boniol M, Malvezzi M, Boniol M, Negri E, et al. Trends in colorectal cancer mortality in Europe: retrospective analysis of the WHO mortality database. BMJ. 2015;351:h4970. pmid:26442928
  14. 14. Laara E, Day NE, Hakama M. Trends in mortality from cervical cancer in the Nordic countries: association with organised screening programmes. Lancet. 1987;1(8544):1247–9. pmid:2884378
  15. 15. Miller AB, Lindsay J, Hill GB. Mortality from cancer of the uterus in Canada and its relationship to screening for cancer of the cervix. Int J Cancer. 1976;17(5):602–12. pmid:1270176
  16. 16. Schoen RE, Pinsky PF, Weissfeld JL, Yokochi LA, Church T, Laiyemo AO, et al. Colorectal-cancer incidence and mortality with screening flexible sigmoidoscopy. N Engl J Med. 2012;366(25):2345–57. pmid:22612596
  17. 17. International Agency for Research on Cancer. Breast Cancer Screening: IARCPress; 2002.
  18. 18. Humphrey LL, Helfand M, Chan BK, Woolf SH. Breast cancer screening: a summary of the evidence for the U.S. Preventive Services Task Force. Ann Intern Med. 2002;137(5 Part 1):347–60. pmid:12204020
  19. 19. Gotzsche PC, Jorgensen KJ. Screening for breast cancer with mammography. Cochrane Database Syst Rev. 2013;6:Cd001877. pmid:23737396
  20. 20. Morrison AS, Brisson J, Khalid N. Breast cancer incidence and mortality in the breast cancer detection demonstration project [published errtum appears in J Natl Cancer Inst 1989 Oct 4;81(19):1513]. J Natl Cancer Inst. 1988;80(19):1540–7. pmid:3193469
  21. 21. 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(10):2507–17. pmid:7736395
  22. 22. Tabar L, Fagerberg G, Duffy SW, Day NE, Gad A, Grontoft O. Update of the Swedish two-county program of mammographic screening for breast cancer. Radiol Clin North Am. 1992;30(1):187–210. pmid:1732926
  23. 23. Tabar L, Fagerberg G, Day NE, Duffy SW, Kitchin RM. Breast cancer treatment and natural history: new insights from results of screening. Lancet. 1992;339(8790):412–4. pmid:1346670
  24. 24. Duffy SW, Tabar L, Fagerberg G, Gad A, Grontoft O, South MC, et al. Breast screening, prognostic factors and survival—results from the Swedish two county study. Br J Cancer. 1991;64(6):1133–8. pmid:1764377
  25. 25. Moss S, Waller M, Anderson TJ, Cuckle H, Trial Management G. Randomised controlled trial of mammographic screening in women from age 40: predicted mortality based on surrogate outcome measures. Br J Cancer. 2005;92(5):955–60. pmid:15726103
  26. 26. Shapiro S. Evidence on screening for breast cancer from a randomized trial. Cancer. 1977;39(6 Suppl):2772–82. pmid:326378
  27. 27. Morrison AS. Intermediate determinants of mortality in the evaluation of screening. Int J Epidemiol. 1991;20(3):642–50. pmid:1955248
  28. 28. Duffy SW, Agbaje O, Tabar L, Vitak B, Bjurstam N, Bjorneld L, et al. Overdiagnosis and overtreatment of breast cancer: estimates of overdiagnosis from two trials of mammographic screening for breast cancer. Breast Cancer Res. 2005;7(6):258–65. pmid:16457701
  29. 29. Moss SM, Cuckle H, Evans A, Johns L, Waller M, Bobrow L. Effect of mammographic screening from age 40 years on breast cancer mortality at 10 years' follow-up: a randomised controlled trial. Lancet. 2006;368(9552):2053–60. pmid:17161727
  30. 30. Anderson TJ, Waller M, Ellis IO, Bobrow L, Moss S. Influence of annual mammography from age 40 on breast cancer pathology. Human Pathology. 2004;35(10):1252–9. pmid:15492993
  31. 31. Birnbaum J, Gadi V.K., Markowitz E., Etzioni R. The impact of treatment advances on the mortality results of breast cancer screening trials: A microsimulation model. Annals of Internal Medicine. 2015.
  32. 32. Autier P, Hery C, Haukka J, Boniol M, Byrnes G. Advanced breast cancer and breast cancer mortality in randomized controlled trials on mammography screening. J Clin Oncol. 2009;27(35):5919–23. pmid:19884547
  33. 33. Shapiro S, Venet W, Strax P, Venet L. Current results of the breast cancer screening randomized trial: the health insurance plan (HIP) of greater New York study. Screening for breast cancer. Toronto: Huber Toronto; 1987. p. 3–15.
  34. 34. Miller AB. Screening for breast cancer with mammography. Lancet. 2001;358(9299):2164; author reply 7–8.
  35. 35. Andersson I, Aspegren K, Janzon L, Landberg T, Lindholm K, Linell F, et al. Mammographic screening and mortality from breast cancer: the Malmo mammographic screening trial. BMJ. 1988;297(6654):943–8. pmid:3142562
  36. 36. Bjurstam N, Bjorneld L, Warwick J, Sala E, Duffy SW, Nystrom L, et al. The Gothenburg Breast Screening Trial. Cancer. 2003;97(10):2387–96. pmid:12733136
  37. 37. Biesheuvel C, Czene K, Orgeas CC, Hall P. The role of mammography screening attendance and detection mode in predicting breast cancer survival-is there added prognostic value? Cancer Epidemiol. 2011;35(6):545–50. pmid:21470933
  38. 38. Mook S, Van 't Veer LJ, Rutgers EJ, Ravdin PM, van de Velde AO, van Leeuwen FE, et al. Independent prognostic value of screen detection in invasive breast cancer. J Natl Cancer Inst. 2011;103(7):585–97. pmid:21350218
  39. 39. Peer PG, Holland R, Hendriks JH, Mravunac M, Verbeek AL. Age-specific effectiveness of the Nijmegen population-based breast cancer-screening program: assessment of early indicators of screening effectiveness. J Natl Cancer Inst. 1994;86(6):436–41. pmid:8120918
  40. 40. Fink R, Shapiro S, Lewison J. The Reluctant Participant in a Breast Cancer Screening Program. Rep Pub Health. 1968;83(6):479–90.
  41. 41. Nystrom L. Breast cancer screening with mammography: overview of Swedish randomised trials. Lancet. 1993;341.
  42. 42. Nyström L, Andersson I, Bjurstam N, Frisell J, Nordenskjöld B, Rutqvist LE. Long-term effects of mammography screening: updated overview of the Swedish randomised trials. The Lancet. 2002;359(9310):909–19.
  43. 43. Boniol M, Autier P, Gandini S. Melanoma mortality following skin cancer screening in Germany. BMJ Open. 2015;5(9):e008158. pmid:26373399
  44. 44. Stang A, Jockel KH. Does skin cancer screening save lives? A detailed analysis of mortality time trends in Schleswig-Holstein and Germany. Cancer. 2015.