Confounders versus Mediators: An Important Distinction

We thank Kashy et al.¹  for their interesting analysis (“Effect of Hydroxyethyl Starch on Postoperative Kidney Function in Patients Having Noncardiac Surgery).” However, we disagree with some of their methods/assumptions and, in fact, reached different conclusions with the same data. As shown in table 1,¹  the authors did not match patients on intraoperative characteristics. Accordingly, the two groups being compared (intraoperative hydroxyethyl starch [HES] recipients vs. noncolloid recipients) were significantly different. After propensity matching, nearly twice as many HES recipients were hypotensive (37 vs. 20%), nearly three times as many received blood transfusions intraoperatively (14 vs. 5%), and nearly one-and-a-half times as many were likely to have received vasopressors (70 vs. 45%). In addition, blood loss was twice as much as among noncolloid recipients than among HES recipients (on average 200 vs. 100 ml). Hence, these groups are not comparable “at baseline.”

As shown in figure 2,¹  blood loss and hypotension are correctly considered confounders (i.e., may be associated with both predictor and outcome) and are controlled for in analysis. In contrast, the authors state that intraoperative vasopressor use and intraoperative blood product transfusion might be mediators (i.e., “mechanisms by which HES administration might cause increased risk of AKI [acute kidney injury]”), implying a position in the causal pathway. Are the authors claiming that AKI (occurring as a result of intraoperative exposure to HES) might occur via HES leading to intraoperative vasopressor use and/or blood product transfusion? HES has been shown to influence hemostasis adversely.²  Are the authors saying that HES-associated AKI may be a result of HES-induced coagulopathy (leading to increased blood product transfusion)?

We believe that intraoperative HES therapy (among patients undergoing major noncardiac surgery) is probably related to clinician-perceived hypovolemia (absolute or relative). Such hypovolemia (rather than receipt of HES per se) may lead to both vasopressor use (secondary to hypotension), transfusion, and to AKI, that is, residual confounding is very likely. This is apparent in the sensitivity analysis (table 3),¹  as the model with transfusion and vasopressor use as potential confounders showed no effect of HES on AKI (odds ratio, 1.10; 95% CI, 0.96 to 1.25; P = 0.12).

The authors might also consider an instrumental variable approach (with calendar time as the instrument). The discontinuation of intraoperative HES use is essentially a “pseudorandom event” such that patients presenting for noncardiac surgery before the HES withdrawal date are probably very similar to patients presenting after this date (of course, all relevant baseline characteristics need to be tabulated to ensure that comparability exists, and where it does not, the parameter that is dissimilar between groups needs to be controlled for if it is a confounder). Such an observational study would emulate the “ideal” randomized controlled trial where essentially similar patients receive different interventions.

The authors declare no competing interests.