Authors' response

David A Freedman1, Diana B Petitti2 and James M Robins3

1 University of California, Berkeley, CA 94720, USA
2 Kaiser Permanente Southern California, Pasadena, CA 91188, USA
3 Harvard School of Public Health, Boston, MA 02115, USA

We reply to a few of the points raised in the letters, starting with Baines. Kappa is a widely-used measure of inter-observer agreement, which corrects raw agreement rates for coincidental matches.1 For example, if a coin lands heads with probability 50%, two flips of the coin will match 50% of the time, just by chance. The raw agreement rate is 50%. Kappa is 0, because there is no agreement above the chance level.

Baines et al.2 used Kappa to measure agreement rates between the ‘reference radiologist’ and radiologists at the centres where screening was actually done. We followed suit.3 Baines4 now seems to deny that raw agreement rates are inflated, or that Kappa is a measure of agreement.

The radiology in the Canadian National Breast Screening Study (CNBSS) has been widely criticized. Poor radiology would explain the lack of effect in CNBSS. Extensive disagreements between the reference radiologist and centre radiologists support that explanation. The critics have additional data, and the CNBSS investigators defend the study.3 On balance, we think CNBSS had problems.3

Baines4 reiterates a point made by Miller,5 that we ignored later CNBSS publications. This is wrong (ref. 3, p. 51; ref. 6, p. 70). We ask Baines to consider our discussion of such publications. There is an excess of advanced cancer detected by physical examination at baseline, among women age 40–49 assigned to screening in CNBSS. Miller5 and Gøtzsche7 explain the excess on the theory that these were small cancers, which were hard to find when women were referred onwards for diagnostic work-up; and diagnostics were more thorough in the screening arm. Given recently published CNBSS data on tumour size, this explanation is not tenable.6 The tumours were not small.

Gøtzsche8 says ‘434 more women with breast cancer prior to randomization were excluded from the study group than from the control group in the Health Insurance Plan (HIP) trial...’ He says this is our discovery. He forgets that four years ago, he wrote (ref. 9, pp. 129–30):

Women were excluded if breast cancer had been diagnosed before entry to the trial, and this status was more completely ascertained for the screened women; thus, the final study cohort was smaller than the control cohort (30 131 vs 30 565).

The 434 is the difference between 30 565 and 30 131.

Gøtzsche's interpretation of the 434 does not come to terms with the HIP design. Women with a diagnosis of breast cancer before entry were excluded when they were screened, or when the disease recurred, or at death.3,5,6,10 Bias in this design is (1) small and (2) against mammography.3,6,10 If anything, the 434 shows that HIP adhered to its design, because the difference in cohort size is close to actuarial estimates based on the design. (ref. 3, p. 45)

Gøtzsche8 points to ‘the large discrepancy in the Two-County study involving a benefit of 24% reported by the trialists versus only 10% reported in the Swedish overview.’ The ‘discrepancy’ comes from having two different groups independently determining cause of death (ref. 6, p. 70). It is not apparent that the Swedish overview determinations are more accurate, or that the differences approach statistical significance (ref. 6, p. 70). The overview finds ‘a significant 21% reduction in breast cancer mortality (RR = 0.79, 95% CI: 0.70–0.89)’ when data from the Swedish trials are combined (ref. 11, p. 909). The overview concludes (ref. 11, p. 909) that

The advantageous effect of breast screening on breast cancer mortality persists after long-term follow-up. The recent criticism against the Swedish randomized controlled trials is misleading and scientifically unfounded.

Gøtzsche8 writes that we ignore length bias, and
do not acknowledge... error in comparing total mortality among breast cancer cases in the study group versus the control group.... It is well-known that case-survival is a highly misleading outcome...

Apparently, he has not considered the technical appendix to our rejoinder (ref. 6, pp. 72–73). The appendix discusses length bias, pinpoints the statistical errors made by Gøtzsche in his commentary7—some of which are repeated in his letter8—and demonstrates the validity of statistical procedures used in our article.3 The data show that screening saves lives.3,6,11,12


    References
 Top
 References
 
1 Gordis L. Epidemiology. Philadelphia: WB Saunders, 1996.

2 Baines CJ, McFarlane DV, Miller AB. The role of the reference radiologist. Estimates of inter-observer agreement and potential delay in cancer detection in the national breast screening study. Invest Radiol 1990;25:971–76.[ISI][Medline]

3 Freedman DA, Petitti DM, Robins JM. On the efficacy of screening for breast cancer. Int J Epidemiol 2004;33:43–55.[Abstract/Free Full Text]

4 Baines CJ. In search of the best available version of the truth. Int J Epidemiol 2004; 33:1404–05.

5 Miller AB. Commentary: A defence of the Health Insurance Plan (HIP) study and the Canadian National Breast Screening Study (CNBSS). Int J Epidemiol 2004;33:64–65.[Free Full Text]

6 Freedman DA, Petitti DM, Robins JM. Rejoinder. Int J Epidemiol 2004;33:69–73.[Free Full Text]

7 Gøtzsche PC. On the benefits and harms of screening for breast cancer. Int J Epidemiol 2004;33:56–64.[Free Full Text]

8 Gøtzsche PC. Misleading quotations and other errors persist in rejoinder on breast cancer screening. Int J Epidemiol 2004; 33:1404.

9 Gøtzsche PC, Olsen O. Is screening for breast cancer with mammography justifiable? Lancet 2000;355:129–34. Discussion, Lancet 2000;355:747–52.[CrossRef][ISI][Medline]

10 Miller AB. Screening for breast cancer with mammography. Letter. Lancet 2001;358:2164.

11 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 Lancet 2002;359:909–19. Discussion, Lancet 2002;360:337–40.[CrossRef][ISI][Medline]

12 Smith RA. Ideology masquerading as evidence-based medicine: The Cochrane review on screening for breast cancer with mammography. Breast Diseases Quart 2003;13:298–307.





This Article
Extract
Full Text (PDF)
All Versions of this Article:
33/6/1405    most recent
dyh242v1
Alert me when this article is cited
Alert me if a correction is posted
Services
Email this article to a friend
Similar articles in this journal
Similar articles in PubMed
Alert me to new issues of the journal
Add to My Personal Archive
Download to citation manager
Request Permissions
Google Scholar
Articles by Freedman, D. A
Articles by Robins, J. M
PubMed
PubMed Citation
Articles by Freedman, D. A
Articles by Robins, J. M