The Swedish Two-County Trial of mammographic screening: cluster randomisation and end point evaluation

S. W. Duffy1,+, L. Tabar2, B. Vitak3, M. F. Yen1, J. Warwick1, R. A. Smith4 and H. H. Chen5

1 Queen Mary University of London, London, UK; 2 Falun Central Hospital, Falun, Sweden; 3 University of Linköping, Linköping, Sweden; 4 American Cancer Society, Atlanta, GA, USA; 5 Institute of Preventive Medicine, National Taiwan University, Taipei, Taiwan

Received 19 July 2002; accepted 3 April 2003


    Abstract
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
Background:

The Swedish Two-County Trial has been criticised on the grounds of the cluster randomisation and alleged bias in classification of cause of death.

Patients and methods:

In the Two-County Trial, 77 080 women were randomised to regular invitation to screening (active study population, ASP) and 55 985 to no invitation (passive study population, PSP), in 45 geographical clusters. After ~7 years, the PSP was invited to screening and the trial closed. We analysed data using hierarchical statistical models to take account of cluster randomisation, and performed a conservative analysis assuming a systematic difference between ASP and PSP in baseline breast cancer mortality in one of the counties. We also analysed deaths from causes other than breast cancer and from all causes among breast cancer cases diagnosed in the ASP and PSP.

Results:

Taking account of the cluster randomisation there was a significant 30% reduction in breast cancer mortality in the ASP. Conservatively, assuming a systematic difference between ASP and PSP clusters in baseline breast cancer mortality, there was a significant 27% reduction in mortality in the ASP. Ignoring classification of cause of death, there was a significant 13% reduction in all-cause mortality in breast cancer cases in the ASP.

Conclusions:

Breast cancer mortality is a valid end point and mammographic screening does indeed reduce mortality from breast cancer. The criticisms of the Swedish Two-County Trial are unfounded.

Key words: breast screening, mammography, randomised trial


    Introduction
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
The randomised trials of breast screening have demonstrated a substantial and significant reduction in breast cancer mortality in association with invitation to mammographic screening [1, 2]. Recently, the evidence has been called into question in a meta-analysis by Olsen and Gøtzsche [3, 4], who maintained that poor quality of the trials and lack of validity of breast cancer mortality as an end point cast doubt on the evidence for a benefit. In terms of the Swedish Two-County Trial, Olsen and Gøtzsche asserted, on the basis of a slight age imbalance between study and control groups, that the cluster randomisation was inadequate and that the cause of death classification was invalid, on the basis of an alleged excess of deaths from other causes among the breast cancer cases in the study group.

Although these and other alleged shortcomings in the Two-County Trial have been shown to be ill-founded [5, 6], it is worthwhile to revisit the issues of randomisation and end point classification, as these are most crucial to interpretation and because the issue of classification of cause of death has been taken up by others [7]. We therefore address the issues of potential biases resulting from the cluster randomisation and the effect of invitation to screening on other causes of death among the breast cancer cases. In approaching the latter problem, we take account of the additional time for study group cases to die of other causes due to lead time. Previous analyses have omitted to do so and have consequently come to erroneous conclusions [4, 6].


    Patients and methods
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
The Swedish Two-County Trial is a cluster-randomised trial of mammographic screening for breast cancer, conducted in Östergötland (E) and Kopparberg (W) counties in Sweden. Excluding those with breast cancer before randomisation, 77 080 women aged 40–74 years were randomised in geographical clusters to invitation to screening (active study population, ASP), and 55 985 to no invitation (passive study population, PSP). Clusters were randomised within strata, comprising 12 pairs of geographical clusters in E county and seven triples in W county. The strata were designed so that clusters within a stratum were similar in socioeconomic terms. In E county, one cluster from each pair was randomised to ASP and one to PSP. In W county, two clusters from each triple were randomised to ASP and one to PSP. Randomisation was by traditional mechanical methods and took place under the supervision of the trial statistician but in the absence of the principal investigators from each county.

The ASP were invited to single-view mammography every 33 months for women aged 50–74 years at randomisation, and every 24 months for women aged 40–49 years. At the end of 1984, a significant 31% mortality reduction was observed in the ASP [8]. The PSP was invited to screening and the trial closed after the first screen of the PSP. All cancers diagnosed in both groups up to and including the end of the first screen of the PSP were then followed-up for mortality from breast cancer. The most recent follow-up is to the end of 1998 [1].

To take account of variation between strata or clusters, we quote results from our analysis of the 1996 follow-up data by a series of hierarchical models [9]. Model 1: Poisson regression, ignoring cluster randomisation; model 2: allowing baseline breast cancer mortality to vary from stratum to stratum; model 3: allowing variation in the effect of invitation to screening from stratum to stratum; model 4: allowing variation between clusters within strata in baseline breast cancer mortality. We also estimated the effect of invitation to screening adjusting for observed differences between clusters in breast cancer mortality in the 10 years before the trial took place.

To investigate possible biases in cause of death, we estimated the relative risk of death from other causes and from all causes within the breast cancer cases, taking into account the additional time for ASP cases to die from causes other than breast cancer (due to early detection, the ASP cases were diagnosed almost 2 years earlier than the PSP cases). We only considered deaths within the breast cancer cases, since only those with breast cancer can die of breast cancer, and because mammographic screening cannot affect risk of death in women who do not have breast cancer. For further details, see Tabar et al. [10].


    Results
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
Up to the end of the first screen of the PSP there were 1426 cancers diagnosed in the ASP and 1042 in the PSP. Table 1 shows the age distribution of women in the ASP and PSP, and the numbers of breast cancer deaths to 1998 by age. Although the age distributions are very similar, and the difference would tend to bias the result against screening, the small difference is significant, due to the large numbers. This may result from the cluster randomisation, so analysis that takes account of the cluster randomisation is advisable. Adjusting for age in the analysis is also necessary. The age-adjusted reduction in breast cancer mortality associated with screening was 31% [relative risk (RR) = 0.69, 95% CI 0.58–0.80, P <0.001].


View this table:
[in this window]
[in a new window]
 
Table 1. Numbers of women randomised and of breast cancer deaths by age group and study population, Swedish Two-County Trial
 
Table 2 shows the results from the hierarchical models 1–4, age-adjusted. The results of all four models are very similar, all significant and showing an ~30% mortality reduction, indicating that the cluster randomisation was efficient. To take account of possible systematic differences in baseline mortality between clusters, we considered breast cancer deaths within clusters in the 10 years before the trial. There was no difference in the rates between ASP and PSP clusters in E county [2]. Due to changes in the tax districts, we could not obtain pre-trial mortality by cluster within three of the strata representing large municipalities in W county, but we did have the data for the 12 small municipality clusters within the remaining four strata. These showed a non-significant 15% lower rate of breast cancer mortality in the ASP clusters in the 10 years before the trial. Although this is compatible with pure chance, we assumed conservatively that it was real and that it also applied to the three remaining strata. Adjusting for the pre-trial differences in mortality, there remained a significant reduction in breast cancer mortality in the ASP, albeit slightly attenuated at 27% (RR = 0.73, 95% CI 0.63–0.85, P <0.001).


View this table:
[in this window]
[in a new window]
 
Table 2. Results of hierarchical models in terms of relative risk (RR) of breast cancer death
 
Adjusting for age and taking account of the longer follow-up time in the ASP, there was no significant difference between deaths from other causes among the breast cancer cases in the ASP and PSP (RR = 1.12, 95% CI 0.96–1.31, P = 0.14). There was a significant 13% reduction in deaths from all causes in breast cancer cases in the ASP (RR = 0.87, 95% CI 0.78–0.99, P = 0.02).


    Discussion
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
The above results show no evidence of bias, either from the cluster randomisation or from misclassification of cause of death. The most conservative analysis based on an assumed systematic difference in baseline mortality in W county still gave a significant 27% reduction in breast cancer mortality in the ASP. Ignoring the cause of death classification, there was a significant 13% reduction in all-cause mortality among breast cancer cases in the ASP. This suggests that the classification of cause of death is reliable, and that death from breast cancer is the appropriate end point.

These results are not surprising. We have dealt with the issues of cluster randomisation and cause of death in our previous publications [8, 9, 1113]. In addition, the Swedish overview has published an excess mortality analysis that did not rely on classification of cause of death and which came to the same conclusions as the breast cancer mortality analysis [14].

Concerns about the screening of the PSP or its timing should be allayed by two factors. First, the published result in 1985 (of breast cancer mortality to the end of 1984) is essentially the pre-PSP screening result. Although the planned first screen of the PSP had started by the end of 1984, it was only 5% complete and had no bearing on mortality at the time. Also, while 13% of the control group had experienced mammography by this time, <2% had received screening mammography (i.e. true contamination). Secondly, the Swedish overview [15] analysed the results by the follow-up model including all breast cancers diagnosed up to the end of 1989, and obtained a similar result to ours for the Two-County Trial.

We therefore conclude that the major criticisms of the Two-county Trial are unfounded. The trial is not perfect, nor is any other study. Breast cancer mortality is a valid end point and mammography screening does reduce deaths from breast cancer.


    Acknowledgements
 
We thank the women who took part in the Two-County Trial and the large number of clinical and support staff who contributed their time and talents to it.


    Footnotes
 
+ Correspondence to: Professor S. W. Duffy, Department of Epidemiology, Mathematics and Statistics, Wolfson Institute of Preventative Medicine, Charterhouse Square, London EC1M 6BQ, UK. Tel: +44-020-7882-6202; Fax: +44-020-7882-6096; E-mail: stephen.duffy{at}cancer.org.uk Back


    References
 Top
 Abstract
 Introduction
 Patients and methods
 Results
 Discussion
 References
 
1. Tabar L, Vitak B, Chen HH et al. The Swedish Two-County Trial twenty years later: updated mortality results and new insights from long-term follow-up. Radiol Clin North Am 2000; 38: 625–651.[ISI][Medline]

2. Nyström L, Andersson I, Bjurstam N et al. Long-term effects of mammography screening: updated overview of the Swedish randomised trials. Lancet 2002; 359: 909–919.[CrossRef][ISI][Medline]

3. Gøtzsche PC, Olsen O. Is screening for breast cancer with mammography justifiable? Lancet 2000; 355: 129–134.[CrossRef][ISI][Medline]

4. Olsen O, Gøtzsche PC. Cochrane review on screening for breast cancer with mammography. Lancet 2001; 358: 1340–1342.[CrossRef][ISI][Medline]

5. Duffy S. Interpretation of the breast screening trials: a commentary on the recent paper by Gøtzsche and Olsen. Breast 2001; 10: 209–212.[CrossRef][ISI]

6. Duffy SW, Tabar L, Smith RA. The mammographic screening trials: commentary on the recent work by Olsen and Gøtzsche. CA Cancer J Clin 2002; 52: 68–71.[Free Full Text]

7. Black WC, Haggstrom DA, Gilbert Welch H. All-cause mortality in randomized trials of cancer screening. J Natl Cancer Inst 2002; 94: 167–173.[Abstract/Free Full Text]

8. Tabar L, Fagerberg CJG, Gad A et al. Reduction in mortality from breast cancer after mass screening with mammography. Lancet 1985; 1: 829–832.[CrossRef][Medline]

9. Nixon RM, Prevost TC, Duffy SW et al. Some random-effects models for the analysis of matched-cluster randomised trials: application to the Swedish Two-County Trial of breast cancer screening. J Epidemiol Biostat 2000; 5: 349–358.[Medline]

10. Tabar L, Duffy SW, Yen MF et al. All-cause mortality among breast cancer patients in a screening trial: support for breast cancer mortality as an endpoint. J Med Screen 2002; 9: 159–162.[CrossRef][ISI][Medline]

11. Tabar L, Fagerberg G, Duffy SW, Day NE. The Swedish two county trial of mammographic screening for breast cancer: recent results and calculation of benefit. J Epidemiol Community Health 1989; 43: 107–114.[Abstract]

12. Tabar L, Fagerberg G, Duffy SW et al. Update of the Swedish Two-County program of mammographic screening for breast cancer. Radiol Clin North Am 1992; 30: 187–210.[ISI][Medline]

13. Tabar L, Fagerberg G, Chen HH et al. Efficacy of breast cancer screening by age: new results from the Swedish Two-County Trial. Cancer 1995; 75: 2507–2517.[ISI][Medline]

14. Larsson LG, Nyström L, Wall S et al. The Swedish randomised mammography screening trials: analysis of their effect on the breast cancer related excess mortality. J Med Screening 1996; 3: 129–132.[Medline]

15. Nyström L, Rutqvist LE, Wall S et al. Breast cancer screening with mammography—overview of the Swedish randomized trials. Lancet 1993; 341: 973–978.[ISI][Medline]