In Reply:—
We thank Drs. Haller and Myles for their comments regarding our article.1They suggest that a reanalysis comparing our paravertebral and general anesthesia groups treating histologic grade as a categorical factor shows the groups to differ on histologic grade, thus implying an overlooked problem with confounding.
This is simply not the case: A chi-square analysis comparing the groups on histologic grade III versus the combined I/II, as suggested, leads to a P value of 0.19 (not the < 0.001 stated by Haller and Myles). This is less significant than the P = 0.16 that we reported when considering the variable to be ordinal. The primary reason is that there is a loss of information on histologic grade by collapsing the first and second levels, which is why we analyzed all three categories. Furthermore, our analysis using the Mann–Whitney test is a more powerful way to detect group differences on severity because it uses the natural ordering, as opposed to simply considering the grades as nominal categories such as red, white, and blue, as the chi-square test does.
Most importantly, adjustment for histologic grade in our multivariable analyses of cancer recurrence obviates concern for the potential confounding due to this factor. Our results are thus interpreted as the hazard ratio of recurrence for paravertebral versus general anesthesia for patients at the same histologic grade, and similarly for other factors in the model. This sort of multivariable analysis compensates for small, or even moderate, imbalances at baseline. We adjusted for this factor because of the retrospective nature of the study, even though we did not have evidence of it being a true confounder because it was not associated with the treatment groups (P = 0.16) or the outcome (P = 0.25), both of which are required by the classic definition of confounding.
As specified in the article, a single surgeon performed all cases in both groups. And again as specified, all paravertebral anesthesia was performed by a single anesthesiologist (D.J.B.), who also performed some of general anesthesia alone cases. The remainder were performed by three other attending anesthesiologists. The cases were similar, and the primary determinant of anesthetic type was assignment to D.J.B., who was the only anesthesiologist in the group familiar with the paravertebral technique.
The substantial limitations of observational studies are well known and were discussed in our article. For example, we specified: “Patients were not randomized and clinical care was not standardized, so that selection bias and the effects of unmeasured confounding variables cannot be excluded. For example, patients in the general anesthesia group had slightly larger tumors, smaller margins, and higher chemotherapy rates than patients in the paravertebral group, factors that could affect mortality, although these differences did not reach statistical significance. Relevant information such as the amount of morphine given and the type of chemotherapy used in each group was not available in the records.”
Under no circumstances should a small retrospective study be the basis for practice, and we suggested no such thing in our report. In contrast, the conclusion of our article was that “this study should be viewed as generating a hypothesis and an estimated effect size for future large randomized controlled trials, which are being planned and which will require several years for execution and analysis.” A prospective trial is now in progress (ClinicalTrials.gov No. NCT00418457).
*Mater Misericordiae University Hospital and National Breast Screening Programme, Dublin, Ireland. donal.buggy@nbsp.ie
