We appreciate the opportunity to reply. In his letter, Dr. Glaser correctly notes that, in our study cohort, women undergoing mastectomy had a greater proportion of patients in the highest socioeconomic quintile compared with women undergoing lumpectomy. Furthermore, he suggests that the higher socioeconomic status (SES) among women undergoing mastectomy may in part explain a worse outcome in this group due to the association between increased use of alcohol and/or hormone-replacement therapy and higher income. Although these factors have been shown to contribute to increased risk of breast cancer incidence, there are no data suggesting that they lead to worse survival. Moreover, there are substantial data to support worse cancer outcomes among patients of lower SES, and this, in fact, would be expected to bias the results to produce an effect opposite to what we observed.
In a separate editorial, de Glas and colleagues raise an important potential problem of selection bias, or “confounding by indication,” which precludes causal inferences from observational data. Indeed, we have cautioned against making such conclusions in our discussion, although we analyzed and interpreted the data to ensure careful adjustment for the key factors known to most strongly impact cancer outcome. However, we note in the article that we could not exclude the possibility that unmeasured (and indeed unknown) residual confounding could exist based on factors that could impact both breast cancer-specific survival as well as surgery type. Such factors could include tumor multifocality, body mass index, and distance between residence and radiation facility, the latter which may serve as a proxy for poor access to health care overall. However, we suggest that the aggregate impact of these unmeasured factors is unlikely to negate the findings of this analysis and others such that mastectomy would be a superior treatment in any subgroup.
We acknowledge that presenting this observation is only one, albeit important, step to better understand the relation between locoregional treatment and disease outcome in early stage breast cancer. Further work evaluating the potential impact of confounding by indication will require alternate approaches using different statistical tools, each of which carries its own assumptions and limitations. The expected impact of such additional analyses on the reported point estimates is unclear, and the magnitude of this correction may in fact be small. A recent study suggests that such a correction for selection bias using a propensity score adjustment did not significantly change the findings of the unadjusted survival outcomes. Although the propensity score method cannot account for unmeasured confounders, this type of analysis will help establish whether the imbalance of measured factors between the 2 groups contributed to differences in outcome.
In summary, we entirely concur with de Glas et al that women “with early stage breast cancer should be given the opportunity to make a well balanced choice for either 1 of these treatment modalities,” but we emphasize that this decision should not be made based on the incorrect perception that mastectomy is a more effective therapy than BCT for early stage disease. Women of all age groups and estrogen receptor status choosing BCT should be confident that they are not opting for an inferior treatment.