Re: Twenty five year follow-up for breast cancer incidence and mortality of the Canadian National Breast Screening Study: randomised screening trial
Miller’s recent article on the results of the Canadian National Breast Screening Study [1] has been heavily criticized for errors in the randomisation employed, leading to more challenging patients being disproportionately assigned to the mammography arm of the trial [2, 3]. Miller has made the claim that randomisation was done appropriately [4]. However, it should be noted that Miller’s own mortality data [1] is supportive of the criticisms raised against his study [2, 3]. The hazard ratio comparing breast cancer mortality from women in the mammography arm to women in the control arm during the first round of screening was reported as 1.47 [1], implying that being assigned to the mammography arm was associated with substantially increased risk of dying of breast cancer (based on the results of the first round of screening only). This is a remarkable imbalance regarding the hazard associated with death from breast cancer between the mammography arm and the control arm and the most likely explanation is that the criticisms regarding the randomisation employed in this trial are valid [2, 3]. Women presenting to the trial with pre-existing concerns regarding their breast health were most likely regularly assigned to the mammography arm of the trial for this study’s results to demonstrate such a marked increase in the hazard of breast cancer mortality in the mammographic screening arm based exclusively on the first year of the trial.
It is not ideal to assess a disease screening technology based on the first round of screening (even in trials that have randomized their subjects appropriately). When we consider a patient who died of cancer detected in the first round of a screening trial, we simply won’t know whether the technology would have contributed to saving that patient’s life had the patient been enrolled in screening earlier on in the progression of their disease. As such, it is more reliable to form conclusions regarding the efficacy of a screening technology based on its results from the second round of screening onwards. Incidentally, the authors provide this data which indicates a mortality benefit (hazard ratio 0.9) from the use of mammographic screening after the first year of the trial. This indicates that mammography has contributing to lowering breast cancer mortality after the initial overly challenging cases were attended to in the first year of screening. This hazard ratio (0.9) reporting a mortality benefit from mammography after the first year of screening did not achieve a p-value below 0.05 and as such was not deemed statistically significant and was apparently ignored when establishing the study’s conclusions. However, it is doubtful that this calculation included enough breast cancer mortality samples to achieve sufficient statistical power for the statistical significance test to be reliably relied upon. Had the authors collected many more samples that continued to produce an identical hazard ratio of 0.9, the finding would have been statistically significant as it is known that the amount of separation required between two groups in order to achieve statistical significance diminishes as the number of samples increases. It is unfortunate that the total number of samples contributing to this study’s mortality calculations from cancer detected in screening rounds 2 through 5 is so limited, as this study’s mortality results (much like this study’s survival results) appear positive for mammography.
Jacob Levman, PhD
Institute of Biomedical Engineering
University of Oxford
[1] A B 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.
[2] D B 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, Feb. 12th 2014.
[3] L K Tabar, 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, Feb. 18th 2014.
[4] A B Miller, 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, Feb. 17th 2014.
Competing interests:
No competing interests
10 March 2014
Jacob Levman
Researcher
Insitute 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
Miller’s recent article on the results of the Canadian National Breast Screening Study [1] has been heavily criticized for errors in the randomisation employed, leading to more challenging patients being disproportionately assigned to the mammography arm of the trial [2, 3]. Miller has made the claim that randomisation was done appropriately [4]. However, it should be noted that Miller’s own mortality data [1] is supportive of the criticisms raised against his study [2, 3]. The hazard ratio comparing breast cancer mortality from women in the mammography arm to women in the control arm during the first round of screening was reported as 1.47 [1], implying that being assigned to the mammography arm was associated with substantially increased risk of dying of breast cancer (based on the results of the first round of screening only). This is a remarkable imbalance regarding the hazard associated with death from breast cancer between the mammography arm and the control arm and the most likely explanation is that the criticisms regarding the randomisation employed in this trial are valid [2, 3]. Women presenting to the trial with pre-existing concerns regarding their breast health were most likely regularly assigned to the mammography arm of the trial for this study’s results to demonstrate such a marked increase in the hazard of breast cancer mortality in the mammographic screening arm based exclusively on the first year of the trial.
It is not ideal to assess a disease screening technology based on the first round of screening (even in trials that have randomized their subjects appropriately). When we consider a patient who died of cancer detected in the first round of a screening trial, we simply won’t know whether the technology would have contributed to saving that patient’s life had the patient been enrolled in screening earlier on in the progression of their disease. As such, it is more reliable to form conclusions regarding the efficacy of a screening technology based on its results from the second round of screening onwards. Incidentally, the authors provide this data which indicates a mortality benefit (hazard ratio 0.9) from the use of mammographic screening after the first year of the trial. This indicates that mammography has contributing to lowering breast cancer mortality after the initial overly challenging cases were attended to in the first year of screening. This hazard ratio (0.9) reporting a mortality benefit from mammography after the first year of screening did not achieve a p-value below 0.05 and as such was not deemed statistically significant and was apparently ignored when establishing the study’s conclusions. However, it is doubtful that this calculation included enough breast cancer mortality samples to achieve sufficient statistical power for the statistical significance test to be reliably relied upon. Had the authors collected many more samples that continued to produce an identical hazard ratio of 0.9, the finding would have been statistically significant as it is known that the amount of separation required between two groups in order to achieve statistical significance diminishes as the number of samples increases. It is unfortunate that the total number of samples contributing to this study’s mortality calculations from cancer detected in screening rounds 2 through 5 is so limited, as this study’s mortality results (much like this study’s survival results) appear positive for mammography.
Jacob Levman, PhD
Institute of Biomedical Engineering
University of Oxford
[1] A B 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.
[2] D B 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, Feb. 12th 2014.
[3] L K Tabar, 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, Feb. 18th 2014.
[4] A B Miller, 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, Feb. 17th 2014.
Competing interests: No competing interests