Volume 193 - Issue 10
The ABC breast cancer cluster: the bad news about a good outcome
Author: Michael D Coory
Med J Aust 2010; 193 (10): 620-621. || doi: 10.5694/j.1326-5377.2010.tb04081.x
Published online: 15 November 2010
Published online: 15 November 2010
To the Editor: An editorial by Stewart alludes to the problem of silent multiple comparisons when interpreting P values from cancer cluster investigations.1 Visible multiplicities such as occur with pre-specified subgroup analyses or sequential monitoring of trials are difficult enough, but at least in these circumstances we know how many multiple comparisons are under consideration. More difficult are silent multiplicities such as occur with cluster investigations (and also with publication bias2 or reporting bias3) where we do not know how many multiple comparisons should be considered.
The P value is intended to be an objective measure of the play of chance, and this is (arguably) the case when applied to a pre-specified primary hypothesis in a randomised trial. But this is not the case for cluster investigations, in which the number of multiple comparisons can never be known with any certainty. Statisticians analysing data from a cluster could obtain any P value they wanted by calibrating it against an arbitrary number of multiple comparisons.
Where does this leave scientific reasoning in cluster investigations? All cases of cancer have causes; the key question in a cluster investigation is whether the cases have a common cause related to the neighbourhood or workplace from which the cluster was reported. Only rarely is an obvious common cause identified, and a decision to take some action (eg, evacuate the workplace) needs to be based on expert opinion.
For the ABC cluster, no obvious common cause was identified. However, the expert panel was concerned that the women with breast cancer were relatively young and were long-term employees at the site, suggesting that there might be an unidentified common cause related to the site.4 This concern, based on expert opinion, is (arguably) enough evidence to evacuate the site.
Investigation of cancer clusters is a difficult task. If a common cause cannot be identified, then there is no objective evidence on which to obtain agreement among experts about the importance of the cluster. Specifically, we need to be very clear that, for cluster investigations, a P value (even when adjusted for multiple comparisons) does not provide an objective measure of whether the cluster is due to chance. In the end, an expert group has to make a decision in the presence of uncertainty. When communicating the results to the public, the uncertainty should be acknowledged — as should the fact that experts sometimes disagree.
References
- Stewart B. The ABC breast cancer cluster: the bad news about a good outcome [editorial]. Med J Aust 2010; 192: 629-631. 0_CBBJHBFG
- Easterbrook P, Berlin J, Gopalan R, Matthews D. Publication bias in clinical research. Lancet 1991; 337: 867-872. 0_CBBJJAAG
- Chan A, Hróbjartsson A, Haahr M, et al. Empirical evidence for selective reporting of outcomes in randomized trials: comparison of protocols to published articles. JAMA 2004; 291: 2457-2465. 0_i1095404
- Armstrong B, Aitken J, Sim M, et al. Breast cancer at the ABC Toowong Queensland: final report of the Independent Review and Scientific Investigation Panel. 2007. http://abc.net.au/corp/pubs/documents/Breast_Cancer_Toowong_Final_Report.pdf (accessed Jun 2010) .