To the Editor:
Wallace et al. reported the effects of the implementation of a hospital protocol for the addition or continuation of perioperative β-blockade in almost 40,000 patients at risk for myocardial ischemia and operated on between 1996 and 2008.1The addition of perioperative β-blockade in eligible patients was associated with a significant reduction in 30-day and 1-yr mortality. Continuation of existing β-blockade in these patients also was beneficial, whereas withdrawal was reported to be associated with increased mortality (almost 400% increased 30-day mortality and almost 200% increased 1-yr mortality). Wallace and coworkers should be commended for this important and large study, which seems to confirm existing evidence about the continuation and withdrawal of perioperative β-blockade using “real world” data.
However, as also acknowledged by the authors, confounding by indication and selection bias are likely to have influenced the results of this retrospective analysis considerably. Therefore, we have some important questions regarding this study.
First, the authors tried to adjust for these potential sources of bias by collecting confounders and performing a propensity analysis. However, the logistic regression model described in table 5 of the article seems to include only previous coronary artery disease and peripheral vascular disease and not age, sex, and other potential confounders mentioned in table 4 of the article. We would like to see a table with the β-blockade effect measures adjusted for all potential confounders, because it is unclear whether these have been taken into account in table 5. Moreover, the methods of the propensity analysis are poorly described, which makes it difficult to interpret the results of these analyses.
Second, figure 3 of the article shows a markedly decreased mortality rate over time, which hardly can be attributed solely to the β-blockade protocol. It may also reflect a change in other practice patterns over time. This problem with retrospective studies with long duration is also recognized in the accompanying editorial.2In this case, clonidine was added to the protocol in 2004.1Likely, however, other drugs, such as statins or aspirin, were continued or prescribed more often as well in more recent years. Furthermore, there may have been improvements in surgical care, such as an increase in minimally invasive surgery. Apparently, these variables were not available to adjust for as confounders. However, adding “time” to the multivariable analysis as a proxy for a change in these variables may partially adjust the β-blockade effect measures for this potential confounding and could at least have been conducted as a sensitivity analysis.
Finally, previous comparable studies showed that including nadir and postoperative hemoglobin both in regression analysis and propensity analysis significantly influenced the β-blockade effect measures.3,4If available, including these hemoglobin values in the analyses may therefore reduce the remarkably strong reported association between β-blockade withdrawal and outcome (odds ratio, 3.9; 95% CI, 2.6–6.0).
In conclusion, we would like to see the results of a regression model that includes both the variable of interest (pattern of β-blockade use) as well as all potential confounders, including time and, if available, both nadir and postoperative hemoglobin values, in a proper and crystal-clear analysis.
*University Medical Center Utrecht, Utrecht, The Netherlands. w.a.vanklei@umcutrecht.nl
