We thank Dr. Sanders et al. for their interest in our work.1 We address their comments with additional summaries and responses to specific comments.
Sanders et al. requested that drug doses be used in our analyses. We recognize the lack of drug dose as a potential weakness to our analysis; however, all induction drugs, including etomidate (median dose of 0.15 mg/kg), were administered to achieve hypnosis. We did not model the induction dose of etomidate because etomidate suppresses adrenal function at concentrations less than 10 ng/ml, which are one-twentieth of the concentration associated with hypnosis (200 ng/ml; 1 µM).2–4 We therefore believe it is reasonable to assume that the overwhelming majority of patients who received a hypnotic dose of etomidate achieved concentrations well above the adrenal suppression threshold.
Sanders et al. requested that we compare among all induction drugs and that we exclude those patients who did not receive an induction drug due to already having been intubated. Table 1 shows that it does not necessarily appear that attendings so much as choose among single induction agents (etomidate, midazolam, or propofol); approximately 20% (n = 619) of all patients received only one agent. In fact, 93% received midazolam (n = 2,906). Among those receiving midazolam, 648 also received propofol only, 1,572 received etomidate only, and 220 received both propofol and etomidate. We therefore did not necessarily see the decision to use etomidate as a choice between it and another agent because it was most often given with another agent (midazolam with or without propofol). Our analysis sought to examine whether adding etomidate to an induction regimen was associated with harm. In fact if one sought to compare the different agents, one could have used the results provided in the figures in the article. For example, from figure 1 in our original article, if one sought to contrast the addition of etomidate to that of propofol, the odds ratio for vasoplegia is estimated to be 0.80/0.73 = 1.09 (standard errors are more difficult to calculate).
The number of patients who did not receive an induction agent was 1.2% (38 of 3,127). As an additional analysis, we removed those patients and reran our regression model as suggested by Sanders et al. Our results shown in table 2 suggest that etomidate may be associated with longer lengths of hospital stay when compared with midazolam and propofol at the unadjusted significance level of 0.05. Statistical significance is not preserved with Bonferroni adjustments to control family-wise error rates. That said, we again question whether interdrug comparisons are most appropriate. As can also be seen from table 2, adding etomidate to an existing regimen is associated with nonsignificant decreases in length of stay (hazard ratio, 1.05; 95% CI, 0.96 to 1.16).
Sanders et al. suggest conducting subgroup analyses “for elective and emergency surgery as these factors may affect treatment propensity.” We dealt with differential treatment propensity by controlling for “emergent surgery” in our regression analyses. Whether there exists an interaction between treatment and many other possible subgroups, as implied by the suggestion of subgroup analyses, was not a goal of our study.
Sanders et al. suggest a propensity score-matching analysis approach for “etomidate only” versus “propofol only” or “midazolam only”: As stated above, only 20% of all surgeries used only one induction agent. The analysis does not represent how attendings at our institution conduct their practices. For the vast majority of surgeries, attendings coinduce patients.
Sanders et al. suggest a propensity score-matching analysis for midazolam plus etomidate versus midazolam plus propofol. We have subset the original dataset to the 2,210 patients who received one of those regimens and report a regression-based analysis as well as a propensity-matching analysis. Table 3 shows the comparison between regimens midazolam plus etomidate and midazolam plus propofol among the subset of patients who received one of the regimens using the suggested propensity-matching approach and a regression modeling approach. In neither case would we conclude any difference between the treatments.
Finally, Sanders et al. suggested that we adjust for hypertension as a potential confounder for the observed angiotensin-converting enzyme inhibitor effect seen in figure 3 of the original article. We agree that if interest was in angiotensin-converting enzyme inhibitor effects, controlling for hypertension would certainly be warranted. That said, even though we were explicit about showing all modeling results, our interest in including covariates was to control for confounding of etomidate effects. We are aware that one could always improve modeling approaches; however, ours was a prespecified model that we thought would be adequate (not perfect) in its capacity to control for confounding of etomidate associations with outcomes.
The authors declare no competing interests.