Current issues of ACP Journal Club are published in Annals of Internal Medicine


Rosiglitazone increased heart failure but did not differ from metformin plus sulfonylurea for other CV outcomes at interim analysis


ACP J Club. 2007 Nov-Dec;147:67. doi:10.7326/ACPJC-2007-147-3-067

Related Content in this Issue
• Companion Abstract and Commentary: Review: Rosiglitazone increases risk for MI but does not differ from other drugs for CV death in type 2 diabetes

Clinical Impact Ratings

GIM/FP/GP: 6 stars

Cardiology: 6 stars

Endocrinology: 6 stars

Source Citation

Home PD, Pocock SJ, Beck-Nielsen H, et al. Rosiglitazone evaluated for cardiovascular outcomes—an interim analysis. N Engl J Med. 2007;357:28-38. [PubMed ID: 17551159]



In patients with type 2 diabetes, is rosiglitazone (RGZ) as add-on therapy noninferior to metformin (MFN) plus sulfonylurea (SFU) for cardiovascular (CV) outcomes?


Design: Randomized controlled trial (RCT) (Rosiglitazone Evaluated for Cardiac Outcomes and Regulation of glycemia in Diabetes [RECORD] trial).

Allocation: Unclear allocation concealment.*

Blinding: Blinded (outcome assessors).*

Follow-up period: Mean 3.75 years.

Setting: 338 centers in 23 countries in Europe and Australasia.

Patients: 4458 patients 40 to 75 years of age (mean age 58 y, 52% men, 99% white, based on 4447 patients) who had type 2 diabetes, body mass index > 25 kg/m2, and hemoglobin (Hb) A1c level > 7% to 9% while taking maximum doses of MFN or SFU. Exclusion criteria were use of other glucose-lowering drugs, hospitalization for major CV events in the past 3 months, planned CV intervention, heart failure (HF), hepatic disease, renal impairment, and uncontrolled hypertension.

Intervention: RGZ, 4 mg/d, plus MFN or SFU, with starting doses determined by local practices (n = 2220); or MFN plus SFU (control) (n = 2227). After 8 weeks, if HbA1c levels were > 7%, patients were given maximum daily drug doses (RGZ, 8 mg/d; MFN, 2550 mg/d; glyburide, 15 mg/d; gliclazide, 240 mg/d; and glimepiride, 4 mg/d). If HbA1c levels were > 8.5% with maximum doses, the RGZ group received a third agent and the control group started insulin therapy (IT). In the RGZ group, if HbA1c levels were > 8.5% with triple therapy, RGZ therapy was replaced with IT.

Outcomes: Composite endpoint of hospitalization or death from CV causes. Secondary outcomes were a composite endpoint of CV death, myocardial infarction (MI), or stroke, and individual outcomes of CV death, all-cause death, MI, and congestive HF. The study needed 4000 patients followed for a median 6 years to have 99% power to detect noninferiority (upper limit of the 2-sided 95% CI for the hazard ratio of the primary endpoint < 1.2 at study completion) when the control group had an annual event rate of 11% (3% CV death and 8% hospitalizations).

Patient follow-up: 90% (intention-to-treat analysis).

Main results

The interim analysis at a mean 3.75 years showed that the RGZ and control groups did not differ for the composite endpoint of hospitalization or death from CV causes (Table). The RGZ group had a higher incidence of congestive HF than did the control group, but groups did not differ for other secondary outcomes (Table).


At interim analysis (mean 3.75 y), rosiglitazone as add-on therapy increased risk for heart failure but did not differ from metformin plus sulfonylurea for other cardiovascular outcomes in patients with type 2 diabetes .

*See Glossary.

Source of funding: GlaxoSmithKline.

For correspondence: Dr. P.D. Home, SCMS-Diabetes Medical School, Newcastle upon Tyne, England, United Kingdom. E-Mail

Table. Rosiglitazone (RGZ) as add-on therapy to metformin or sulfonylurea vs metformin plus sulfonylurea (control) in type 2 diabetes at interim analysis (mean 3.75 y)†

Outcomes RGZ Control RRI (95% CI) NNH (CI)
Composite endpoint of death or hospitalization from CV causes 9.8% 9.1% 7.6% (−10 to 29) Not significant
Acute MI‡ 1.9% 1.7% 16% (−25 to 80) Not significant
Congestive heart failure‡ 1.7% 0.8% 123% (27 to 293) 107 (45 to 488)
Composite endpoint of CV death, MI, or stroke 4.2% 4.3% 2.9% (−28 to 27) Not significant
CV death 1.3% 1.6% 17% (−36 to 49) Not significant
All-cause death 3.3% 3.6% 6.9% (−26 to 33) Not significant

†CV = cardiovascular; MI = myocardial infarction; other abbreviations defined in Glossary. RRI, RRR, NNH, NNT, and CI calculated from control event rates and hazard ratios in article based on adjudicated events.
‡Included hospitalizations and deaths.


Recent controversy over the CV safety of RGZ was sparked by the online publication of the meta-analysis by Nissen and Wolski, which received considerable attention in the lay press and led to a congressional hearing and calls to reform drug regulation. It also spurred Home and colleagues to publish the interim analysis of the RECORD trial. Both studies have substantial limitations, so no simple answers are forthcoming.

The meta-analysis by Nissen and Wolski was based on a large number of low-quality studies and used flawed methods. Almost all studies analyzed were small, short-duration RCTs that were not designed to assess CV outcomes and therefore did not prospectively ascertain or blindly adjudicate cardiac events. Most of these studies are unpublished, so the data (which were not peer-reviewed) were abstracted from documents on the Internet.

The analysis of these low-quality data had numerous methodological flaws. First, 6 of 48 eligible studies were excluded because they reported no cardiac deaths or MIs, which biases estimated CV risk upward. Second, because cardiac events were rare, most RCTs reported zero outcomes in 1 or both groups, which may lead to inaccurate risk estimates with the meta-analysis method used. Third, a fixed-effects model was used despite obvious differences between RCTs, which exaggerate statistical significance. Finally, a separate U.S. Food and Drug Administration meta-analysis found only a nonsignificant trend toward more CV deaths, MIs, and strokes (odds ratio 1.2, 95% CI 0.7 to 1.8) (1). The low-quality data and technical flaws make the meta-analysis by Nissen and Wolski far from conclusive—at best, it is a hypothesis-generating study.

Even if the review by Nissen and Wolski had stronger methods, it would need to be interpreted cautiously because meta-analyses based on many small studies may not predict the results of subsequent definitive trials (2). In general, replicated results from well-designed, adequately powered RCTs provide the most reliable guide to clinical practice. Unfortunately, there are few such results available to assess the effect of RGZ on cardiac death and MI.

The DREAM trial (3), 1 of the studies included in the meta-analysis, randomized 5269 patients without CV disease to RGZ or placebo and found a nonsignificant trend toward more CV events with RGZ (hazard ratio 1.37, CI 0.97 to 1.94). The RECORD trial by Home and colleagues used CV events as the primary outcome and found the risk for MI to be slightly higher in the RGZ group (hazard ratio 1.23, CI 0.81 to 1.86), whereas risk for CV death was slightly lower (hazard ratio 0.80, CI 0.52 to 1.24); neither result was statistically significant, but the interim analysis had low statistical power.

The results on the cardiac risk associated with RGZ are suggestive but not definitive—more data from other large, well-conducted RCTs are needed. The ACCORD trial of > 10 000 patients with diabetes and high CV risk and the BARI-2D trial of > 2300 patients with diabetes and coronary disease are ongoing, but results will not be available for several years. Importantly, the data and safety monitoring boards of ACCORD and BARI-2D examined their interim outcomes in light of these 2 publications about RGZ and voted to continue their trials without modification; because patient safety is their paramount responsibility, presumably these data safety and monitoring boards did not detect significant excess CV risk.

The possible adverse effects of RGZ on cardiac death and MI should be distinguished from its established effect on exacerbation of HF, which is because of fluid retention and is reversible on discontinuation of the drug.

So, what can we conclude? Clearly, RGZ can worsen HF, but available evidence is not yet definitive about risk for cardiac death and MI. In light of safety concerns, it is reasonable to consider alternative agents until this question can be better resolved by ongoing RCTs.

The RGZ controversy underscores the limitations of using surrogate outcomes. Studies of diabetes therapy use glucose control as the endpoint, but better glucose control does not translate into reduced macrovascular events. Because CV disease is the major cause of death in diabetes, the endpoint of RCTs for diabetes drugs should include CV outcomes in addition to glucose control. Similarly, developers of clinical guidelines and performance measures should weigh the risk for increasing CV events when setting benchmarks for tighter glucose control.

Mark A. Hlatky, MD
Dena M. Bravata, MD, MS
Stanford University School of Medicine
Stanford, California, USA


1. Rosen CJ. The rosiglitazone story—lessons from an FDA Advisory Committee Meeting. N Engl J Med. 2007;357:844-6. [PubMed ID: 17687124]

2. LeLorier J, Grégoire G, Benhaddad A, Lapierre J, Derderian F. Discrepancies between meta-analyses and subsequent large randomized, controlled trials. N Engl J Med. 1997;337:536-42. [PubMed ID: 9262498]

3. Gerstein HC, Yusuf S, Bosch J, et al for the DREAM (Diabetes REduction Assessment with rampiril and rosiglitazone Medication) Trial Investigators. Effect of rosiglitazone on the frequency of diabetes in patients with impaired glucose tolerance or impaired fasting glucose: a randomised controlled trial. Lancet. 2006;368:1096-105. [PubMed ID: 16997664]