Many thanks for your comments on our recent retrospective study.1
With regard to the first point posed by Drs. Ali and Ghori, concerning the use of the term “long-term cancer survival,” we agree that together with 1-yr survival, 5- and 10-yr survival rates are commonly used when reporting cancer survival. However, contextually, mortality rates for perioperative interventions are commonly reported as either 30 days or length of stay, and as such the reported follow-up of between 18 months and 4.5 yr would qualify as long term.
With regard to the use of the propensity model and all-cause mortality data, we agree that a better approach would have been to consider cancer-attributable mortality. However, these data are not reliably available in the United Kingdom. National cancer registries do not cover the broad span of cancers we considered, and in addition often have incomplete data for the early years covered by the study.
We agree that Kaplan–Meier curves for the propensity-matched groups should have been included in the study. These are now included in figure 1, A–C, and as you can see are very similar to those for the nonmatched groups.
Drs. Doleman, Lun, and Williams raise interesting queries regarding our analysis, much of which we are in agreement with.
There is little doubt that cancer type and stage have a profound effect on outcome, and the lack of accurate data within our study for the latter in particular is a potential major confounder. However, as we state in the discussion, the very lack of availability of staging data to the practitioners administering the anesthesia lessens this fact since it could not have been a deciding factor in the choice of anesthetic.
As far as including cancer types in either the propensity model or the multivariate analysis (beyond the broad groups that have already been included in the analysis, see tables 1 and 2), the major issue is the numbers of different types and subtypes, with consequent substantial implications for outcome. A look through the data reveals more than 20 broad cancer types. Within those types are further subdivisions, for example, triple-negative breast cancer has a very different outcome from estrogen receptor–positive breast cancer, as do the different types of thyroid cancer. The numbers of individual cancers are often small, and as a result, statistical power would be lost if we were to subdivide beyond the point that has already been undertaken.
Dr. Dawson makes specific reference to the effects of transfusion, and we agree that there are a significantly greater number of patients who underwent transfusion in the group receiving volatile anesthesia. The fundamental reason for this relates to the allowance margin in the propensity scoring and to the numbers of factors included in the propensity model. While statistically significant, the actual difference in numbers of transfused patients is low at only 40. Given that the unadjusted morality rate for transfused patients is 51%, this would have little impact on the overall mortality difference between the 2 groups (190 patients), but we would agree that it is a confounding factor. Eliminating the difference between groups completely would have resulted in reducing the number of patients in the propensity-matched model further.
The strength of our analysis lies in the fact that it considered a large number of unselected cancer patients admitted for elective surgery. The weakness is that it is a retrospective study with all the inherent problems and unaccounted for confounders that come with that. The lack of data around staging and the small numbers of individual cancer subtypes prohibit further analysis without loss of meaning. As stated in our conclusion, the only assertion we make is that our study found an association between mode of anesthesia administration and mortality. Further adequately powered prospective studies of specific cancer types with comprehensive staging data now need to be undertaken to confirm or refute our findings.
The authors declare no competing interests.