Bias in revascularization study

[Two of the authors respond:]

Carl van Walraven and Alan Forster are correct. In our study,1 the medical treatment group included patients who were truly selected for medical management and those for whom revascularization was planned initially but not carried out owing to early death or patient or provider preference. Thus, some of the patients in the medical treatment group would have been in the revascularization group if information on their initial therapy plan had been available. The bias, then, is perhaps more correctly labelled misclassification bias rather than time-dependent bias.

Thompson and colleagues have elegantly demonstrated the potential effect of such a misclassification in observational studies.2 In this work, 4 groups were analyzed: patients who received coronary artery bypass grafting as recommended, patients who received percutaneous coronary intervention as recommended, patients who received medical management as recommended, and patients who received medical management although percutaneous coronary intervention or coronary artery bypass grafting had been recommended (this group is comparable to the group misclassified in our work). Indeed, this final group had the poorest survival rate.

Unfortunately, in our study we were unable to differentiate between patients who received medical therapy as a chosen therapy and patients who were treated medically, although the initial plan was for revascularization. Thus, early deaths in the medical management group may have been events that occurred while patients were waiting for a planned revascularization procedure that did not occur. In this case, the issue is not one of time-dependent covariates but rather one of knowing the true intention at t = 0, an issue not easily addressed using observational data. In our case, the separation between the survival curves does occur early on, when this bias would be at play. However, our curves continue to separate over time, indicating a longer term survival advantage that is possibly attributable to revascularization. We thank Walraven and Forster for shedding light on this important issue.