- Email: [email protected]
Mammographic screening: no reliable supporting evidence?
CORRESPONDENCE
6
Potential viral pathogenic mechanism for new variant inflammatory bowel disease. http://mp.bmjjournals.com/cgi/content/full/ 54/6/DC1 (accessed on Feb 19, 2002). Morris A, Aldulaimi D. New evidence for a viral pathogenic mechanism for new variant inflammatory bowel disease and developmental disorder? http://mp.bmjjournals.com/cgi/content/full/ 54/6/DC1 (accessed on Feb 19, 2002).
Mammographic screening: no reliable supporting evidence? Sir—Olli Miettinen and colleagues (Feb 2, p 404),1 in their re-analysis of the Malmö trial,2 use the asymptotes of the breast-cancer mortality curves in the study and control groups instead of total numbers of deaths. This method seems appealing, since a possible effect of screening would take some time to emerge. However, their analysis is flawed. The researchers accept our results that there are two good screening trials, but they looked only at one of them. They argue that the screening period and the follow-up period in the Canadian trial were too short. However, four to five cycles of screening were completed in this trial, with 1-year intervals; in Malmö, it was five to six cycles, and the interval was 18–24 months.3 These differences do not justify exclusion of the Canadian trial, which, furthermore, has now been followed up for 13 years. Taken together, these trials provide no evidence for an effect on breast-cancer mortality (relative risk after 13 years 0·97, 95% CI 0·82–1·14).3 Miettinen and colleagues not only looked at a subset of the trials, but at a subset of the selected trial, namely individuals aged at least 55 years. This cut-off point is arbitrary and probably data driven, since it was not prespecified by the Malmö investigators,2 which raises the risk of bias. It is generally judged poor science to base the conclusion on a subgroup when the overall result is negative, as in this instance (63 vs 66 breast-cancer deaths, or 84 vs 69 deaths if preliminary data are included, as done by Miettinen and colleagues). Finally, the researchers overlook entirely that breast-cancer mortality is a flawed outcome that is biased in favour of screening.3–5 This misclassification bias would be expected to increase when the necropsy rate goes down. Since this effect was seen in Malmö,3 it provides an alternative explanation for the small difference in breast-cancer mortality after long follow-up that Miettinen and
706
colleagues used in their statistical model. Peter C Gøtzsche Nordic Cochrane Centre, Rigshospitalet Dept 7112, DK-2100 Copenhagen, Denmark (e-mail: [email protected]) 1
2
3
4
5
Miettinen OS, Henschke CI, Pasmantier MW, Smith JP, Libby DM, Yankelvitz DF. Mammographic screening: no reliable supporting evidence? Lancet 2002; 359: 404–05. Andersson I, Aspegren K, Janzon L, et al. Mammographic screening and mortality from breast cancer: the Malmö mammographic screening trial. BMJ 1988; 297: 943–48. Olsen O, Gøtzsche PC. Systematic review of screening for breast cancer with mammography. http://image.thelancet.com/lancet/extra/fullr eport.pdf (accessed Feb 14, 2002). Gøtzsche PC. Screening for breast cancer with mammography. Lancet 2001; 358: 2167–68. Black WC, Haggstrom DA, Welch HG. Allcause mortality in randomized trials of cancer screening. J Natl Cancer Inst 2002; 94: 167–73.
the [cause-specific] mortality curves in the study and control groups instead of total numbers of deaths”; we do not even believe that those curves (however defined) have asymptotes. We illustrated the application of these principles, and Gøtzsche’s criticisms of this are unjustifiable, since his portrayal of our method is distorting. The addressing of these principles one by one here would detract from the main points presented—re-presented—here. Suffice to note that we did not conclude that there are two good trials on breast cancer screening; we merely accepted the conclusion of Olsen and Gøtzsche that there are two valid studies, for the sake of argument, which clause was, alas, edited out of the published version of the report. We added our criterion of sufficient-duration of screening (and follow-up), and this led, as we said, to the exclusion of the Canadian trial. Olli S Miettinen, *Claudia I Henschke Weill Medical College of Cornell University, 1300 York Avenue, New York, NY 10021, USA (e-mail: [email protected])
Authors’ reply Sir—Peter Gøtzsche criticises us on two fronts. First, he says that what we present on the level of principles seems appealing, thereby (and otherwise) implying that in substance it is not. Second, he categorically asserts that our application of the approach we advocate is flawed on several grounds. So grossly does Gøtzsche misrepresent the principles we adduced that they need to be reasserted. Those principles have to do with trials contrasting a particular regimen of screening with no screening for a cancer. The most fundamental principle is not a matter of method but concerns the design of the object of study: the object must be the screening regimen’s usefulness in terms of the degree of proportional reduction in the cancer’s case-fatality rate, resulting from the earlier interventions after the earlier diagnoses afforded by the screening. This fundamental principle was supplemented by two subordinate principles of methods design, concerning provisions for valid estimation of that parameter of principal interest. For one, the screening must continue long enough (we specified how long, in principle); and for another, attention to causespecific deaths must be focused on a suitably select segment of follow-up during and after the screening (also specified, in principle). These principles are not fairly expressed by saying that we “use the asymptotes of
Bovine tuberculosis: milk and meat safety Sir—Progress in politics and science frequently happens by serendipity, and, therefore, it is a pleasant irony that the foot and mouth disaster might bring about a rethink over a warning from 1847, when it was difficult to find milk in London that was not contaminated with pus or blood and that these substances may be linked to scrofula infection. After discovering the tubercle bacillus in 1882, Koch suggested bovine tuberculosis was not a risk to human beings—an error that took 15 years to rectify. Two world wars delayed eradication of bovine tuberculosis, although milk pasteurisation had been advocated before 1900. Although pasteurisation of milk became mandatory in many countries, it is still not so in England and Wales, where there are 400 producers of non-pasteurised milk. Lobbying has thwarted a legal ban, although annual tests and labelling changes have been tightened. The last cases involved school children in Yorkshire in 1959,1 but with the rise in bovine tuberculosis, further cases seem inevitable, despite closer liaison between health and veterinary authorities. The first two farmer cases happened in 1999, and the long
THE LANCET • Vol 359 • February 23, 2002 • www.thelancet.com
For personal use. Only reproduce with permission from The Lancet Publishing Group.
6
Potential viral pathogenic mechanism for new variant inflammatory bowel disease. http://mp.bmjjournals.com/cgi/content/full/ 54/6/DC1 (accessed on Feb 19, 2002). Morris A, Aldulaimi D. New evidence for a viral pathogenic mechanism for new variant inflammatory bowel disease and developmental disorder? http://mp.bmjjournals.com/cgi/content/full/ 54/6/DC1 (accessed on Feb 19, 2002).
Mammographic screening: no reliable supporting evidence? Sir—Olli Miettinen and colleagues (Feb 2, p 404),1 in their re-analysis of the Malmö trial,2 use the asymptotes of the breast-cancer mortality curves in the study and control groups instead of total numbers of deaths. This method seems appealing, since a possible effect of screening would take some time to emerge. However, their analysis is flawed. The researchers accept our results that there are two good screening trials, but they looked only at one of them. They argue that the screening period and the follow-up period in the Canadian trial were too short. However, four to five cycles of screening were completed in this trial, with 1-year intervals; in Malmö, it was five to six cycles, and the interval was 18–24 months.3 These differences do not justify exclusion of the Canadian trial, which, furthermore, has now been followed up for 13 years. Taken together, these trials provide no evidence for an effect on breast-cancer mortality (relative risk after 13 years 0·97, 95% CI 0·82–1·14).3 Miettinen and colleagues not only looked at a subset of the trials, but at a subset of the selected trial, namely individuals aged at least 55 years. This cut-off point is arbitrary and probably data driven, since it was not prespecified by the Malmö investigators,2 which raises the risk of bias. It is generally judged poor science to base the conclusion on a subgroup when the overall result is negative, as in this instance (63 vs 66 breast-cancer deaths, or 84 vs 69 deaths if preliminary data are included, as done by Miettinen and colleagues). Finally, the researchers overlook entirely that breast-cancer mortality is a flawed outcome that is biased in favour of screening.3–5 This misclassification bias would be expected to increase when the necropsy rate goes down. Since this effect was seen in Malmö,3 it provides an alternative explanation for the small difference in breast-cancer mortality after long follow-up that Miettinen and
706
colleagues used in their statistical model. Peter C Gøtzsche Nordic Cochrane Centre, Rigshospitalet Dept 7112, DK-2100 Copenhagen, Denmark (e-mail: [email protected]) 1
2
3
4
5
Miettinen OS, Henschke CI, Pasmantier MW, Smith JP, Libby DM, Yankelvitz DF. Mammographic screening: no reliable supporting evidence? Lancet 2002; 359: 404–05. Andersson I, Aspegren K, Janzon L, et al. Mammographic screening and mortality from breast cancer: the Malmö mammographic screening trial. BMJ 1988; 297: 943–48. Olsen O, Gøtzsche PC. Systematic review of screening for breast cancer with mammography. http://image.thelancet.com/lancet/extra/fullr eport.pdf (accessed Feb 14, 2002). Gøtzsche PC. Screening for breast cancer with mammography. Lancet 2001; 358: 2167–68. Black WC, Haggstrom DA, Welch HG. Allcause mortality in randomized trials of cancer screening. J Natl Cancer Inst 2002; 94: 167–73.
the [cause-specific] mortality curves in the study and control groups instead of total numbers of deaths”; we do not even believe that those curves (however defined) have asymptotes. We illustrated the application of these principles, and Gøtzsche’s criticisms of this are unjustifiable, since his portrayal of our method is distorting. The addressing of these principles one by one here would detract from the main points presented—re-presented—here. Suffice to note that we did not conclude that there are two good trials on breast cancer screening; we merely accepted the conclusion of Olsen and Gøtzsche that there are two valid studies, for the sake of argument, which clause was, alas, edited out of the published version of the report. We added our criterion of sufficient-duration of screening (and follow-up), and this led, as we said, to the exclusion of the Canadian trial. Olli S Miettinen, *Claudia I Henschke Weill Medical College of Cornell University, 1300 York Avenue, New York, NY 10021, USA (e-mail: [email protected])
Authors’ reply Sir—Peter Gøtzsche criticises us on two fronts. First, he says that what we present on the level of principles seems appealing, thereby (and otherwise) implying that in substance it is not. Second, he categorically asserts that our application of the approach we advocate is flawed on several grounds. So grossly does Gøtzsche misrepresent the principles we adduced that they need to be reasserted. Those principles have to do with trials contrasting a particular regimen of screening with no screening for a cancer. The most fundamental principle is not a matter of method but concerns the design of the object of study: the object must be the screening regimen’s usefulness in terms of the degree of proportional reduction in the cancer’s case-fatality rate, resulting from the earlier interventions after the earlier diagnoses afforded by the screening. This fundamental principle was supplemented by two subordinate principles of methods design, concerning provisions for valid estimation of that parameter of principal interest. For one, the screening must continue long enough (we specified how long, in principle); and for another, attention to causespecific deaths must be focused on a suitably select segment of follow-up during and after the screening (also specified, in principle). These principles are not fairly expressed by saying that we “use the asymptotes of
Bovine tuberculosis: milk and meat safety Sir—Progress in politics and science frequently happens by serendipity, and, therefore, it is a pleasant irony that the foot and mouth disaster might bring about a rethink over a warning from 1847, when it was difficult to find milk in London that was not contaminated with pus or blood and that these substances may be linked to scrofula infection. After discovering the tubercle bacillus in 1882, Koch suggested bovine tuberculosis was not a risk to human beings—an error that took 15 years to rectify. Two world wars delayed eradication of bovine tuberculosis, although milk pasteurisation had been advocated before 1900. Although pasteurisation of milk became mandatory in many countries, it is still not so in England and Wales, where there are 400 producers of non-pasteurised milk. Lobbying has thwarted a legal ban, although annual tests and labelling changes have been tightened. The last cases involved school children in Yorkshire in 1959,1 but with the rise in bovine tuberculosis, further cases seem inevitable, despite closer liaison between health and veterinary authorities. The first two farmer cases happened in 1999, and the long
THE LANCET • Vol 359 • February 23, 2002 • www.thelancet.com
For personal use. Only reproduce with permission from The Lancet Publishing Group.