We appreciate Dr. Raghunathan’s and Dr. Van Aken’s interest in our recent article.1 Our analysis was based on approximately 29,000 propensity-matched patients who were or were not given intraoperative 6% hydroxyethyl starch (HES) 670/0.75. As Raghunathan et al. note, we did not match on intraoperative factors such as hypotension, vasopressor use, and transfusions, and these factors were thus unbalanced. However, we adjusted for hypotension—an obvious confounder—in our subsequent multivariable analysis. A second propensity match that includes all intraoperative factors except vasopressor use and transfusions produced nearly identical results, a roughly 22% increase in the odds of acute kidney injury.
We chose to consider vasopressor use and transfusions to be mediators in our primary analysis “based on our assumption that the administration of hetastarch is mainly triggered by blood loss and that the administration of transfusions and vasopressors happens thereafter and thus might not influence the decision to administer colloids. Thus, vasopressor use and blood transfusions might be mediators.”
However, we recognize that vasopressor use and transfusions could also be confounders or (perhaps most likely) both confounders and mediators. We therefore conducted and presented sensitivity analyses in which various factors were considered to be either confounders or mediators. Whereas the conclusion of our primary analysis was that hetastarch administration increases the odds of acute kidney injury approximately 21%, the increases were no longer statistically significant when transfusions or the combination of transfusions and vasopressor use were included as confounding factors. We note, though, that among the eight sensitivity analyses presented in our article, all others remained statistically significant and had roughly comparable treatment effects. Difficulty distinguishing confounders from mediators, and the fact that some factors are surly both, is one of the many reasons why the results of large trials are more reliable than the retrospective analyses.
Raghunathan et al. suggest using calendar time as an instrumental variable for clinician decision making on HES usage in an environment of changing Food and Drug Administration regulations. Study patients’ had noncardiac surgeries between January 2005 and September 2012; during this time period, regulatory changes might have affected overall HES usage. Therefore, we intentionally propensity-matched patients on the year of surgery to make sure most of the times we compared surgeries close in time to each other. Although, among nonmatched patients, the proportion of those receiving HES decreased in years 2011 and 2012 compared to previous years.
Ertmer and Van Aken are also concerned that unadjusted confounding may have contributed to our conclusion that 6% HES 670/0.75 promotes acute renal injury, which would suggest that the product is actually safe. Curiously, they then express surprise that high-molecular-weight starches, which they claim to be “unsuitable for modern perioperative care,” are still used at the Cleveland Clinic. It is not just at the Clinic. The high-molecular-weight starch we used remains by far the most commonly used plasma expander in the United States, even after Food and Drug Administration approval of low-molecular-weight starches in December 2007.
Continued use of 6% HES 670/0.75 is hardly unreasonable. There is little previous evidence that the intraoperative use is harmful and there has never been a large trial comparing high- and low-molecular-weight starches. Ertmer and Van Aken cite a meta-analysis to support their assertion that low-molecular-weight starches are safer than higher-molecular preparations.2 However, that study did not compare low-molecular-weight starches to 6% HES 670/0.75, the preparation we used. In fact, a more recent meta-analysis in cardiac surgical patients who presumably are at high risk for acute kidney injury concludes that “no reliable analysis for separate hetastarch generations compared to albumin, gelatin, or crystalloids was possible.”3
In summary, retrospective analyses are complicated by factors that are not clearly confounders or mediators. As illustrated by the comparison between our primary and sensitivity analyses, the distinction matters and can profoundly influence conclusions—and is one reason why randomized trials are so important. No large study has compared high- and low-molecular-weight starches; whether one is safer than the other for perioperative use thus remains unclear. In the mean time, the conclusion that 6% HES 670/0.75 is “unsuitable for modern perioperative care” seems premature.
The authors declare no competing interests.