Molecular profiling of breast cancer: clinical implications

Abstract

Breast cancers are routinely subcategorised on the basis of clinical stage, cellular morphology and immunohistochemical analysis of a small number of markers. The recent development of gene expression microarray and related technologies provides an opportunity to perform more detailed profiling of the disease. It is anticipated that the molecular classification arising from such studies will be more powerful than its pathological predecessor at confining treatment to those patients who are most likely to benefit. It is likely that this will result in a much less frequent use of adjuvant chemotherapy. Furthermore, of those who do receive it, a higher proportion will benefit. If adopted, this will offer considerable patient benefits in terms of reducing unnecessary toxicity and have major health economic implications.

Figure 1

Model for the effect of molecular profiling on numbers of premenopausal women with node negative breast cancer receiving chemotherapy, and associated benefit at 5 years. (A) The Oxford overview of polychemotherapy (Early Breast Cancer Trialists' Collaborative group, 1998b) indicated that for 100 node-negative, premenopausal women receiving chemotherapy according to standard criteria, at 5 years three are cured by chemotherapy, 83.5 would have been alive without chemotherapy and 13.5 die despite chemotherapy. With application of molecular profiling to predict the outcome, these figures (for the same 100 people) would become 3, 22.5 (false-positive rate of 27% from van ‘t Veer et al, 2002) and 13.5, respectively. If response could be predicted, the third column would be reduced further to the false-positive rate for the predictive profile used (here we have assumed that this is approximately as for outcome profiling, that is 27%). The second column may be reduced but this is difficult to estimate. This illustrates that much less chemotherapy would be given. (B) This shows the proportional benefit of women receiving chemotherapy according to the same criteria. Considering the same 100 women, if outcome prediction with molecular profiling of breast cancer was employed, the number treated would be reduced to 39, resulting in an increase in the proportion cured (from three out of 100 to three out of 39 or 8%). If it were possible to predict chemoresponsiveness, it is possible that the number receiving chemotherapy would reduce further from 39 to 29.5 (allowing for a false-positive rate equivalent to that seen in the van ‘t Veer study). A reduction in the second column would probably occur but is not shown. In this scenario, the proportion cured by chemotherapy would be three out of 29.5, approximately 10% (a three-fold increase), and the number of women treated has been reduced by 70%. Note that in neither figure has consideration been given to the false-negative rate inherent in molecular profiling. It has been assumed that all deaths occurring were breast cancer related.