Re: Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial
Mr. Hefti has challenged my previous letters [1], my rebuttal follows. He begins with the issue I have raised that lethal tumours caught in the first round of screening degrade mortality results unfairly between the two arms of the trial: “If … this prevalence bias were significantly relevant … and if mammography were so much more effective in detecting the lethal advanced breast cancers … you'd expect to find a large highly significant reduction in breast cancer mortality after excluding the data of the first round of screening”. Mr. Hefti is mistaken, if a given screening technology is very good at catching lethal stage cancers it will not necessarily be good at reducing the mortality of the disease it is detecting. Mortality will not improve from the detection of disease that has already reached a lethal stage of development. Mr. Hefti continues: “the Miller group study reports no statistically significant decrease in the mortality rate after omitting the data of the prevalent screening”. Miller’s reported hazard ratio for death from breast cancer in subsequent screening rounds is 0.9 which indicates a mortality benefit from mammographic screening. That their results are not statistically significant is not surprising given the limited number of samples, as previously discussed [2]. A lack of statistical significance is not the same as an insignificant effect.
Mr. Hefti writes: “other studies (including trials without the confounding factor of a mixed experimental group) found no significant reduction in lethal late-stage breast-cancer incidence” citing [3-5]. Citation [3] refers to a textbook on radiation oncology, focused on the therapeutic applications of radiation and is not an appropriate reference towards discrediting mammography. Mr. Hefti’s following citations [4, 5] are population based studies that do not control the mammographic status of the subjects. It is misleading to cite these three references [3-5] at the end of a statement that implies they avoid the confounding factor of having mixed experimental groups, they most certainly do not.
Mr. Hefti writes in reference to the Marmot review [6]: “the panel combined the more reliable and the less reliable mammogram studies to estimate the effect on breast cancer mortality” and continues with “Isn't this the same type of confounding factor, "mixing" diverging entities together and deriving conclusions from it, Levman had dismissed my Lousdal et al. citation? Wouldn't it then also call for Levman's dismissal (instead of his advocation) of the review panel's investigation based on the same methodological argumentation?” The answer is no, this is not the same type of confounding factor that contributed to dismissing the Lousdal study [7]. The Lousdal study’s experimental group was not purely patients who received mammography. The randomized controlled trials from Marmot’s review [6] all had experimental groups controlled to only include subjects receiving mammography. The Marmot review’s combination of multiple randomized controlled trials is not needed to assess the effect of breast cancer screening as the review included the results of each trial individually, indicating cancer screening mortality benefits from almost every trial, with this Canadian trial being the exception.
Mr. Hefti continues: “Many of the large prominent pro-mammography trials have severe methodological limitations and flaws” citing [8, 9]. He continues: “other independent mammogram scientists, evaluating several randomized studies that compared screened versus non-screened groups, reported that the most adequate, well-designed of those studies denoted no "statistically significant reduction in breast cancer mortality"” citing [9-12]. These main criticisms have been raised by Gøtzsche whose own competing reviews published with the Nordic Cochrane Database [9, 12] combine three randomized controlled trials he assessed as having adequate randomization. Two of Gøtzsche’s three adequate trials reported a mortality benefit from mammographic breast cancer screening (the UK Age trial and Malmo I), the remaining trial was the Canadian study we are presently discussing. Gøtzsche’s combination of these three trials is misleading given that the Canadian trial compares mammography to breast examination while the other two trials compare mammography to nothing. Thus their combination is inappropriate towards the independent assessment of the mortality benefits of mammography. Citation [10] indicates that “the trials suggest only a 12% reduction in breast cancer mortality” in contradiction to Mr. Hefti’s statement of no mortality effect. Hefti’s citation [11] is another population based analysis that does not control the subject’s mammography status and was not a randomized trial contrary to Mr. Hefti’s statement.
Mr. Hefti writes: “Aggressive lethal breast cancers tend to grow fast and frequently at a size too small for mammography to timely detect in the preclinical stages” and cites two academic papers in support [4, 11]. These papers are discussion pieces that are far less reliable than a randomized controlled trial. Mr. Hefti continues: “Levman mentions an independent review … as proof that mammographic screening evokes a significant mortality benefit”. There appears to be a misunderstanding here as to what constitutes scientific proof. Randomized controlled trials are considered the gold standard evaluation method. While I certainly believe they are the strongest source available on which to base our conclusions, by no means do I consider their results proof. However, the alternative population based studies that do not control the mammographic screening status of each subject are a much less reliable source on which to base conclusions.
Mr. Hefti writes: “another investigator … remarked that the mammographic benefit is outstripped by harm if the data on "death caused by treatment is included" ” citing [13]. This article [13] is based on an overestimate of the rate of mammographic overdiagnosis provided by Bleyer [14], which is based on the absurd assumption that every in situ cancer detected by mammography is a case of overdiagnosis. While it is inevitable that some in situ cancers need not be treated, assuming that they are all indolent ignores the standard model of malignant tumour progression whereby invasive life-threatening tumours are expected to develop through an in situ stage. Baum’s work [13] is self-described as a crude estimate and its criticism of mammographic screening is based on the shortcomings of subsequent radiation treatments which could be replaced with alternative therapies, thus the author is tarnishing mammography with the deficiencies of different technology.
Mr. Hefti writes: “Levman discredits one of my citations (while overlooking the others)”. I will endeavour to be more thorough given the journal’s space allotment rules. Autier’s study [15] suffers from the same type of problem as the Lousdal study [7] that I have previously discussed [16] in that the mammographic status of the subjects was not controlled, which is not a reliable type of experiment to rely on. The remaining citation [10] was addressed above. Mr. Hefti writes: “Furthermore, an analysis of the population-based mortality data pertaining to national breast screening programs found no effect from mammographic screening”. The study cited [17] suffers from the same class of shortcoming as the Lousdal study [7], whereby the mammographic status of the patients wasn’t controlled.
Hefti has written that I am “silent about the fact that the Miller study included and calibrated for a vast multitude of confounding factors or variables”. Miller at al. [18] and many studies attempt to control for confounding factors without providing a complete set of details regarding the exact adjustments that have been employed in their calculations. On this point I am sympathetic to the authors [18] as journal publication rules regularly exclude the possibility of providing every relevant detail due to limited publication space. Given the complexity of modern research, it is a challenge to present one’s work thoroughly while simultaneously balancing clarity and abiding by the journal’s publication space requirements. A large amount of relevant calculations are typically not provided in studies such as this, preventing informed commentary on much of the methods employed. Thus, this task is typically left to the publishing journal’s peer reviewers.
Hefti has written: “As Miller had pointed out, Levman, oddly, had also ignored crucial confounding factors, such as lead and length biases”. I merely don’t think it is safe to ignore the study’s survival data which indicates benefits from mammographic screening, especially in favour of mortality data that penalizes mammography for having been presented with so many detectable lethal stage tumours in the first round of screening.
Mr. Hefti writes: “a review of several randomized mammogram trials also found no screening effect from high sensitivity of tumor detection and breast cancer mortality” citing [8]. The following passage [8] conflicts with Mr. Hefti’s assessment of the article: “Comprehensive systematic reviews have suggested that mammography screening reduces breast cancer mortality by 15-16%” [8] citing [9].
In summary, Mr. Hefti’s arguments rely heavily on population based research that doesn’t control the mammographic screening status of each patient. Mr. Hefti’s arguments also rely heavily on the work of Gøtzsche. However, Gøtzsche concedes that mortality benefits are obtainable from mammographic screening [8, 10] and of the three randomized controlled trials he has deemed acceptable in his reviews on the subject [9, 12], two reported mortality benefits from mammography. Combining this Canadian trial’s negative results with the remaining two that he deemed acceptable was misleading due to differences between the trials. Pooling these three randomized controlled trials is not appropriate towards the independent assessment of the mortality benefits of mammographic screening. Furthermore, whether randomization was implemented appropriately in this Canadian trial remains an open question with poignant criticisms raised by Dr. Kopans at Harvard University who was involved early on in the trial’s review process [19]. It is not safe to conclude no mortality benefits are obtainable from mammographic breast cancer screening based on the results of this study [18] or any review that includes it in a pooled analysis [9, 12].
Jacob Levman, PhD
Institute of Biomedical Engineering
University of Oxford
References
[1] Hefti, Accusations of Bias Versus Actual Bias Regarding Mammography, British Medical Journal, 348:g366, March 31, 2014.
[2] Levman, Subject randomisation and mortality in mammographic breast cancer screening, British Medical Journal, 348:g366, March 10, 2014.
[3] Perez et al., Principles and Practice of Radiation Oncology, 4th Edition, 2003.
[4] Esserman et al., Rethinking screening for breast cancer and prostate cancer, Journal of the American Medical Association, 2009;302(15):1685-92.
[5] Kalager et al., Overdiagnosis of invasive breast cancer due to mammography screening: results from the Norwegian screening program, Annals of Internal Medicine, 2012;156(7):491-9.
[6] Marmot, et al, The benefits and harms of breast cancer screening: an independent review, British Journal of Cancer, 2013;108:2205-2240.
[7] Lousdal et al., Trends in breast cancer stage distribution before, during and after introduction of a screening programme in Norway, European Journal of Public Health, Published online ahead of print edition, March 4th, 2014, doi:10.1093/eurpub/cku015.
[8] Gøtzsche, Relation between breast cancer mortality and screening effectiveness: systematic review of the mammography trials, Dan Med Bul, 2011;58(3):A4246.
[9] Gøtzsche, Nielsen, Screening for breast cancer with mammography, Cochrane Database Syst Rev. 2009, Issue 4.
[10] Gøtzsche et al., Why mammography screening has not lived up to expectations from the randomised trials, Cancer Causes Control, 2012;23(1):15-21.
[11] Jørgensen et al., Is mammographic screening justifiable considering its substantial overdiagnosis rate and minor effect on mortality, Radiology, 2011;260(3):621-7.
[12] Gøtzsche, Jørgensen, Screening for breast cancer with mammography, Cochrane Database Syst Rev. 2013, Issue 6.
[13] Baum, Harms from breast cancer screening outweigh benefits if death caused by treatment is included, British Medical Journal, 2013;346:f385.
[14] Bleyer, Welch, Effect of three decades of mammography on breast-cancer incidence, New England Journal of Medicine, 2012;131:949-55.
[15] Autier et al, Advanced breast cancer incidence following population-based mammographic screening, Annals of Oncology, 2011;22(8):1726-35.
[16] Levman, Hazard ratios and bias against mammography, British Medical Journal, 2014;348:g366, March 25, 2014.
[17] Mukhtar et al., Breast cancer mortality trends in England and the assessment of the effectiveness of mammography screening: population-based study, J R Soc Med. 2013 Jun;106(6):234-42.
[18] Miller et al., Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial, British Medical Journal, 2014;348:g366.
[19] Kopans, Re: Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial, British Medical Journal, 2014;348:g366, February 12, 2014.
Competing interests:
No competing interests
07 April 2014
Jacob Levman
Researcher
Institute of Biomedical Engineering, University of Oxford
Rapid Response:
Re: Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial
Mr. Hefti has challenged my previous letters [1], my rebuttal follows. He begins with the issue I have raised that lethal tumours caught in the first round of screening degrade mortality results unfairly between the two arms of the trial: “If … this prevalence bias were significantly relevant … and if mammography were so much more effective in detecting the lethal advanced breast cancers … you'd expect to find a large highly significant reduction in breast cancer mortality after excluding the data of the first round of screening”. Mr. Hefti is mistaken, if a given screening technology is very good at catching lethal stage cancers it will not necessarily be good at reducing the mortality of the disease it is detecting. Mortality will not improve from the detection of disease that has already reached a lethal stage of development. Mr. Hefti continues: “the Miller group study reports no statistically significant decrease in the mortality rate after omitting the data of the prevalent screening”. Miller’s reported hazard ratio for death from breast cancer in subsequent screening rounds is 0.9 which indicates a mortality benefit from mammographic screening. That their results are not statistically significant is not surprising given the limited number of samples, as previously discussed [2]. A lack of statistical significance is not the same as an insignificant effect.
Mr. Hefti writes: “other studies (including trials without the confounding factor of a mixed experimental group) found no significant reduction in lethal late-stage breast-cancer incidence” citing [3-5]. Citation [3] refers to a textbook on radiation oncology, focused on the therapeutic applications of radiation and is not an appropriate reference towards discrediting mammography. Mr. Hefti’s following citations [4, 5] are population based studies that do not control the mammographic status of the subjects. It is misleading to cite these three references [3-5] at the end of a statement that implies they avoid the confounding factor of having mixed experimental groups, they most certainly do not.
Mr. Hefti writes in reference to the Marmot review [6]: “the panel combined the more reliable and the less reliable mammogram studies to estimate the effect on breast cancer mortality” and continues with “Isn't this the same type of confounding factor, "mixing" diverging entities together and deriving conclusions from it, Levman had dismissed my Lousdal et al. citation? Wouldn't it then also call for Levman's dismissal (instead of his advocation) of the review panel's investigation based on the same methodological argumentation?” The answer is no, this is not the same type of confounding factor that contributed to dismissing the Lousdal study [7]. The Lousdal study’s experimental group was not purely patients who received mammography. The randomized controlled trials from Marmot’s review [6] all had experimental groups controlled to only include subjects receiving mammography. The Marmot review’s combination of multiple randomized controlled trials is not needed to assess the effect of breast cancer screening as the review included the results of each trial individually, indicating cancer screening mortality benefits from almost every trial, with this Canadian trial being the exception.
Mr. Hefti continues: “Many of the large prominent pro-mammography trials have severe methodological limitations and flaws” citing [8, 9]. He continues: “other independent mammogram scientists, evaluating several randomized studies that compared screened versus non-screened groups, reported that the most adequate, well-designed of those studies denoted no "statistically significant reduction in breast cancer mortality"” citing [9-12]. These main criticisms have been raised by Gøtzsche whose own competing reviews published with the Nordic Cochrane Database [9, 12] combine three randomized controlled trials he assessed as having adequate randomization. Two of Gøtzsche’s three adequate trials reported a mortality benefit from mammographic breast cancer screening (the UK Age trial and Malmo I), the remaining trial was the Canadian study we are presently discussing. Gøtzsche’s combination of these three trials is misleading given that the Canadian trial compares mammography to breast examination while the other two trials compare mammography to nothing. Thus their combination is inappropriate towards the independent assessment of the mortality benefits of mammography. Citation [10] indicates that “the trials suggest only a 12% reduction in breast cancer mortality” in contradiction to Mr. Hefti’s statement of no mortality effect. Hefti’s citation [11] is another population based analysis that does not control the subject’s mammography status and was not a randomized trial contrary to Mr. Hefti’s statement.
Mr. Hefti writes: “Aggressive lethal breast cancers tend to grow fast and frequently at a size too small for mammography to timely detect in the preclinical stages” and cites two academic papers in support [4, 11]. These papers are discussion pieces that are far less reliable than a randomized controlled trial. Mr. Hefti continues: “Levman mentions an independent review … as proof that mammographic screening evokes a significant mortality benefit”. There appears to be a misunderstanding here as to what constitutes scientific proof. Randomized controlled trials are considered the gold standard evaluation method. While I certainly believe they are the strongest source available on which to base our conclusions, by no means do I consider their results proof. However, the alternative population based studies that do not control the mammographic screening status of each subject are a much less reliable source on which to base conclusions.
Mr. Hefti writes: “another investigator … remarked that the mammographic benefit is outstripped by harm if the data on "death caused by treatment is included" ” citing [13]. This article [13] is based on an overestimate of the rate of mammographic overdiagnosis provided by Bleyer [14], which is based on the absurd assumption that every in situ cancer detected by mammography is a case of overdiagnosis. While it is inevitable that some in situ cancers need not be treated, assuming that they are all indolent ignores the standard model of malignant tumour progression whereby invasive life-threatening tumours are expected to develop through an in situ stage. Baum’s work [13] is self-described as a crude estimate and its criticism of mammographic screening is based on the shortcomings of subsequent radiation treatments which could be replaced with alternative therapies, thus the author is tarnishing mammography with the deficiencies of different technology.
Mr. Hefti writes: “Levman discredits one of my citations (while overlooking the others)”. I will endeavour to be more thorough given the journal’s space allotment rules. Autier’s study [15] suffers from the same type of problem as the Lousdal study [7] that I have previously discussed [16] in that the mammographic status of the subjects was not controlled, which is not a reliable type of experiment to rely on. The remaining citation [10] was addressed above. Mr. Hefti writes: “Furthermore, an analysis of the population-based mortality data pertaining to national breast screening programs found no effect from mammographic screening”. The study cited [17] suffers from the same class of shortcoming as the Lousdal study [7], whereby the mammographic status of the patients wasn’t controlled.
Hefti has written that I am “silent about the fact that the Miller study included and calibrated for a vast multitude of confounding factors or variables”. Miller at al. [18] and many studies attempt to control for confounding factors without providing a complete set of details regarding the exact adjustments that have been employed in their calculations. On this point I am sympathetic to the authors [18] as journal publication rules regularly exclude the possibility of providing every relevant detail due to limited publication space. Given the complexity of modern research, it is a challenge to present one’s work thoroughly while simultaneously balancing clarity and abiding by the journal’s publication space requirements. A large amount of relevant calculations are typically not provided in studies such as this, preventing informed commentary on much of the methods employed. Thus, this task is typically left to the publishing journal’s peer reviewers.
Hefti has written: “As Miller had pointed out, Levman, oddly, had also ignored crucial confounding factors, such as lead and length biases”. I merely don’t think it is safe to ignore the study’s survival data which indicates benefits from mammographic screening, especially in favour of mortality data that penalizes mammography for having been presented with so many detectable lethal stage tumours in the first round of screening.
Mr. Hefti writes: “a review of several randomized mammogram trials also found no screening effect from high sensitivity of tumor detection and breast cancer mortality” citing [8]. The following passage [8] conflicts with Mr. Hefti’s assessment of the article: “Comprehensive systematic reviews have suggested that mammography screening reduces breast cancer mortality by 15-16%” [8] citing [9].
In summary, Mr. Hefti’s arguments rely heavily on population based research that doesn’t control the mammographic screening status of each patient. Mr. Hefti’s arguments also rely heavily on the work of Gøtzsche. However, Gøtzsche concedes that mortality benefits are obtainable from mammographic screening [8, 10] and of the three randomized controlled trials he has deemed acceptable in his reviews on the subject [9, 12], two reported mortality benefits from mammography. Combining this Canadian trial’s negative results with the remaining two that he deemed acceptable was misleading due to differences between the trials. Pooling these three randomized controlled trials is not appropriate towards the independent assessment of the mortality benefits of mammographic screening. Furthermore, whether randomization was implemented appropriately in this Canadian trial remains an open question with poignant criticisms raised by Dr. Kopans at Harvard University who was involved early on in the trial’s review process [19]. It is not safe to conclude no mortality benefits are obtainable from mammographic breast cancer screening based on the results of this study [18] or any review that includes it in a pooled analysis [9, 12].
Jacob Levman, PhD
Institute of Biomedical Engineering
University of Oxford
References
[1] Hefti, Accusations of Bias Versus Actual Bias Regarding Mammography, British Medical Journal, 348:g366, March 31, 2014.
[2] Levman, Subject randomisation and mortality in mammographic breast cancer screening, British Medical Journal, 348:g366, March 10, 2014.
[3] Perez et al., Principles and Practice of Radiation Oncology, 4th Edition, 2003.
[4] Esserman et al., Rethinking screening for breast cancer and prostate cancer, Journal of the American Medical Association, 2009;302(15):1685-92.
[5] Kalager et al., Overdiagnosis of invasive breast cancer due to mammography screening: results from the Norwegian screening program, Annals of Internal Medicine, 2012;156(7):491-9.
[6] Marmot, et al, The benefits and harms of breast cancer screening: an independent review, British Journal of Cancer, 2013;108:2205-2240.
[7] Lousdal et al., Trends in breast cancer stage distribution before, during and after introduction of a screening programme in Norway, European Journal of Public Health, Published online ahead of print edition, March 4th, 2014, doi:10.1093/eurpub/cku015.
[8] Gøtzsche, Relation between breast cancer mortality and screening effectiveness: systematic review of the mammography trials, Dan Med Bul, 2011;58(3):A4246.
[9] Gøtzsche, Nielsen, Screening for breast cancer with mammography, Cochrane Database Syst Rev. 2009, Issue 4.
[10] Gøtzsche et al., Why mammography screening has not lived up to expectations from the randomised trials, Cancer Causes Control, 2012;23(1):15-21.
[11] Jørgensen et al., Is mammographic screening justifiable considering its substantial overdiagnosis rate and minor effect on mortality, Radiology, 2011;260(3):621-7.
[12] Gøtzsche, Jørgensen, Screening for breast cancer with mammography, Cochrane Database Syst Rev. 2013, Issue 6.
[13] Baum, Harms from breast cancer screening outweigh benefits if death caused by treatment is included, British Medical Journal, 2013;346:f385.
[14] Bleyer, Welch, Effect of three decades of mammography on breast-cancer incidence, New England Journal of Medicine, 2012;131:949-55.
[15] Autier et al, Advanced breast cancer incidence following population-based mammographic screening, Annals of Oncology, 2011;22(8):1726-35.
[16] Levman, Hazard ratios and bias against mammography, British Medical Journal, 2014;348:g366, March 25, 2014.
[17] Mukhtar et al., Breast cancer mortality trends in England and the assessment of the effectiveness of mammography screening: population-based study, J R Soc Med. 2013 Jun;106(6):234-42.
[18] Miller et al., Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial, British Medical Journal, 2014;348:g366.
[19] Kopans, Re: Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial, British Medical Journal, 2014;348:g366, February 12, 2014.
Competing interests: No competing interests