Use of Thiazolidinediones and Risk of Heart Failure in People With Type 2 Diabetes: A Retrospective Cohort Study : Response to Karter et al.

We thank Karter et al. (1) for their comments regarding our study (2). We agree that its principal weakness is the possibility of residual confounding, and we acknowledge this limitation in our article. We also agree that comparison of patients initiating thiazolidinediones (TZDs) with all those receiving other oral antidiabetic agents (new starters plus those on maintenance therapy) may have biased our findings against TZDs. On the other hand, the alternative of comparing patients initiating therapy with TZDs with those initiating other oral antidiabetic agents could have imparted a potentially more serious bias to the study because (as Karter et al. note) the former may be more likely to be initiated later in the course of the disease than the latter. Lacking data on the duration of diabetes, we chose an approach that we hoped would minimize bias and we attempted to control for confounding using multivariate analysis and propensity matching.

Karter et al. (1) present a number of intriguing preliminary results from the Kaiser Permanente Northern California Diabetes Registry. Although the Registry provides a richer clinical picture than the claims dataset we employed in terms of patient characteristics such as glycemic control, it too has its limitations. The Registry represents a group of patients receiving treatment in a single, group practice, integrated delivery system in which processes and outcomes of care for diabetes patients may differ substantially from those among the more diverse group of patients represented in our study. In Kaiser Permanente Northern California, TZDs may be more likely to be reserved for the “sickest” patients. Also, the effects of TZDs on risk of heart failure may be less pronounced in settings where patients are monitored more frequently and edema is therefore less likely to progress to overt heart failure before arousing clinical suspicion and action.

Thus, although we eagerly await the results of the analysis by Karter et al., we suspect that it may not provide a definitive answer with respect to the effects of TZDs on risk of heart failure because their analysis will suffer from the same fundamental weakness (i.e., possibility of residual confounding) that they correctly identified in ours, although perhaps to a lesser degree. Moreover, their study may suffer from the additional potential limitation of lack of generalizability to the overall population of patients receiving TZDs. As we note in our study, definitive conclusions must await the results of long-term, randomized, controlled trials (although these too may suffer from problems of generalizability).

So where does this leave us? While awaiting the results of ongoing studies, clinicians must make use of the best available data to guide their decisions. We agree with Karter et al. that the results of our study do not warrant changes in clinical practice guidelines. Our recommendation that physicians use these drugs with caution in patients with heart failure is entirely consistent with warnings set forth in U.S. Food and Drug Administration-approved labeling for rosiglitazone and pioglitazone (3,4). Our recommendation that physicians seek alternatives for patients with shortness of breath is only common sense in light of the strength and consistency of the association that we observed, the known physiologic effects of these agents, and published reports of TZD-induced heart failure and pulmonary edema resolving after discontinuation of such therapy (5).

Unfortunately, even well-established treatment guidelines are not consistently followed in typical clinical practice, as demonstrated by a recent study (6) that found that patients hospitalized for heart failure frequently receive TZDs despite explicit warnings against this practice. We hope that our study will increase physician awareness of the potential risk of heart failure associated with the use TZDs so that it may be weighed against the potential benefits of these agents in improving clinical outcomes in patients with diabetes (7–8).