We thank Dr. Maxwell et al. for their interest and comments on our study,1 and we will try to address some of their concerns regarding the methodology that was used in the study and our interpretation of the findings.
Dr. Maxwell et al. express concern about the propensity score–based matching process. We were able to match 56% of the patients treated with inotropic therapy with a nontreated patient. The fact that it was not possible to match 100% of the patients indicates that the distribution of the propensity score among the treated and nontreated patients did not fully overlap. However, we do not agree that discarding patients in the matching process from the original cohort per se implies that the internal validity of the study is affected. It may, however, imply that concern should be taken before extrapolating the findings to patients that differ from the characteristics of our matched population. When matching our patients, we assessed the balance using both absolute standardized differences and variance ratios.
The variance ratios of the individual covariates in the matched population ranged from a low of 0.86 (critical preoperative state) to high of 1.18 (off pump surgery; table 1). Along with the standardized differences of less than 10%, we find strong indications of a well-balanced matching. Notably, the covariate postinfarct septal rupture had a very low variance ration of 0.6, which is probably due to the very few patients characterized by this covariate (five patient in treated group and three with no inotropic therapy).
Furthermore, the order of the patients was randomized before matching although this was not specified in the article.
As we already discussed in our article, we were not able to control for residual confounding relating to intraoperative events not accounted for by procedure scoring.
We agree that extracorporeal circulation (ECC) time and perhaps especially cross-clamp (CC) time have some impact on whether patients are treated with inotropes and with which dose. We also agree that there might be a correlation between ECC/CC times and mortality. Thus, it is possible that the dichotomization with on/off bypass and then ECC time greater than/less than 120 min was too “tight.” However, we could not demonstrate any statistically significant difference in the mortality when the treated and nontreated was handled individually according to ECC time in either 30- or 60-min intervals. Including CC time in a conditional linear regression analysis on the matched cohort reduced the risk estimate of 30-day mortality from adjusted hazard ratio 3.71 (2.11–6.53, 95% CI) to a hazard ratio 2.39 (1.57–3.63, 95% CI).
The authors likewise express concern about our dichotomization of ventricular function into less than or equal to 30% and greater than 30%. It could be argued that a more detailed categorization of patients with normal left ventricular function would be relevant.
However as table 2 indicates, there was no significant difference in mortality between different states of left ventricular function but a constant difference between treated and nontreated patients.
As we wanted to primarily investigate the effect of inotropic therapy, we excluded the patients who had exclusively received vasopressors and we think it is well established that norepinephrine has only minimal inotropic effect, as it is primarily an α-receptor agonist and only have very little, if any, direct β-effect.
Dr. Maxwell and colleagues suggest studying the dose-response of inotropic therapy by a vasoactive-active scoring system. To our knowledge the scoring system in concern has been validated only on a neonatal and infant population, and we do not find it appropriate to apply on an adult population. Our data were not granular enough to distinguish whether inotropes were given in parallel or sequential therapy, but we have future studies on a larger population size that will focus on the association between specific inotropes, dosages, and outcome.
The problem of capturing the anesthesiologists “gut feeling” will exist in both propensity score matching studies and in a randomized clinical trial, unless a very fixed protocol. We know that initiation of inotropes is highly dependent on the name of the anesthetist,2 thus we tried to adjust for the provider effect in the conditional regression analysis.
Although an observational study as the present always will be challenged with the risk of confounding by indication, we believe that the design and statistical analyses of our study are enough robust to interpret data as to raise concern about a possible harmful effect of inotropic therapy in cardiac surgery. The concern of a possible risk that could exceed the beneficial effect of inotropic therapy to some patients is not novel. There is a growing body of literature, especially from large-scale randomized trials of nonsurgical heart failure patients that indicate that inotropic therapy may cause more harm than good. In the light of the lack of randomized clinical trials demonstrating any improvement of clinical outcomes including mortality from perioperative inotropic therapy,3–5 we are restraint to ignore or explain away the underlying signal our data raise.
The authors declare no competing interests.