To the Editor: The study by Nissen and Wolski (June 14 issue)1shows increased rates of myocardial infarction and death fromcardiovascular causes associated with rosiglitazone treatmentfor type 2 diabetes. The authors' results are intriguing becausea simple calculation of the pooled data shows a higher rateof myocardial infarction in the control group than in the rosiglitazonegroup (0.59% vs. 0.55%). In contrast, the Peto odds ratio formyocardial infarction of 1.43 (95% confidence interval [CI],1.03 to 1.98) that is reported in the article shows a significantlyincreased risk in the rosiglitazone group.
Since it is unusual for a weighted analysis to be so differentfrom that of simple pooled data, particularly when events areuncommon, I repeated the analyses with the use of more commonstatistical methods. The findings reported for the Peto oddsratio were replicated, but this statistic was the only one thatwas significant. Neither the relative risk nor the common oddsratio showed a significant increase in either myocardial infarctionor death from cardiovascular causes. The addition of findingsfrom the recently reported Rosiglitazone Evaluated for CardiacOutcomes and Regulation of Glycaemia in Diabetes (RECORD) trial2did not materially influence these results (Table 1).
Table 1. Comparison of Statistical Analyses of 42 Trials Involving Rosiglitazone.
The use of the Peto odds ratio is not recommended when thereis substantial imbalance in the numbers of patients in manytrials,4 as was the case in the study by Nissen and Wolski.When allocation ratios between the rosiglitazone group and thecontrol group were 2:1 or higher, the Peto odds ratio was substantiallyhigher than either the risk ratio or the common odds ratio.Every trial with an allocation ratio of 3:1 or 4:1 and a relativerisk of more than 0.7 showed similarly extreme results for thePeto odds ratio. The risk difference showed little evidenceof large effects, but this procedure is also not recommendedfor meta-analyses of rare events.4
Unfortunately, the Peto odds ratio was the only statistic reportedby Nissen and Wolski. My additional analyses show that the estimatedeffect of rosiglitazone on myocardial infarction is sensitiveto the choice of the treatment-effect estimator, especiallyat high allocation ratios and risk ratios above 1. These findingsweaken the inference that rosiglitazone causes a significantincrease in the risk of myocardial infarction and an increasein the risk of death from cardiovascular causes that has borderlinesignificance.
Michael B. Bracken, Ph.D., M.P.H. Yale University Schools of Public Health and Medicine New Haven, CT 06510 michael.bracken{at}yale.edu
Dr. Bracken reports receiving fees from GlaxoSmithKline forconsulting on an unrelated matter. No other potential conflictof interest relevant to this letter was reported.
References
Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial infarction and death from cardiovascular disease. N Engl J Med 2007;356:2457-2471. [Erratum, N Engl J Med 2007;357:100.] [Free Full Text]
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. [Free Full Text]
Bradburn MJ, Deeks JJ, Berlin JA, Russell Localio A. Much ado about nothing: a comparison of the performance of meta-analytical methods with rare events. Stat Med 2007;26:53-77. [CrossRef][ISI][Medline]
To the Editor: The meta-analysis by Nissen and Wolski includestwo large trials with primary end points that did not includecardiovascular events. In the Diabetes Reduction Assessmentwith Ramipiril and Rosiglitazone Medication (DREAM) trial,1the observation was stopped when diabetes developed in studypatients. In the A Diabetes Outcome Prevention Trial (ADOPT),2observation was stopped when unsatisfactory glycemic levelsoccurred.
Since rosiglitazone provided better results than comparatorsin both trials, the mean follow-up in the rosiglitazone groupwas longer than that in the control group. These differenceswere not accounted for in the meta-analysis. For example, inthe ADOPT trial, the yearly incidences of myocardial infarctionin the rosiglitazone group and the control group were 0.46%and 0.39%, respectively, with an estimated increase in riskin the rosiglitazone group of 18%, rather than 33%, as reportedin the meta-analysis. Data on mean follow-up in the DREAM trialwere not reported. However, the reported number of patientsat all points of follow-up was higher in the rosiglitazone groupthan in the placebo group. Therefore, the effect of rosiglitazoneon cardiovascular end points in longer-term trials could havebeen overestimated.
Edoardo Mannucci, M.D. Matteo Monami, M.D., Ph.D. NiccolòMarchionni, M.D. University of Florence 50141 Florence, Italy edoardo.mannucci{at}fastwebnet.it
Dr. Mannucci reports receiving lecture fees from Abiogen Pharma,GlaxoSmithKline, Guidotti, Eli Lilly, Menarini, Merck, NovoNordisk, Sanofi-Aventis, and Takeda, consulting fees from Sanofi-Aventis,and research grants from Novartis, Novo Nordisk, Sanofi-Aventis,and Takeda; Dr. Monami, lecture fees from Guidotti, Eli Lilly,Merck, Menarini, and Takeda and consulting fees from Sanofi-Aventisand Menarini; and Dr. Marchionni, lecture fees from GlaxoSmithKline,Guidotti, and Menarini and research grants from Novartis, NovoNordisk, Sanofi-Aventis, and Takeda. No other potential conflictof interest relevant to this letter was reported.
References
The DREAM (Diabetes REduction Assessment with ramipril 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;368:1096-1105. [Erratum, Lancet 2006;368:1770.] [CrossRef][Medline]
Kahn SE, Haffner SM, Heise MA, et al. Glycemic durability of rosiglitazone, metformin, or glyburide monotherapy. N Engl J Med 2006;355:2427-2443. [Erratum, N Engl J Med 2007;356:1387-8.] [Free Full Text]
To the Editor: Nissen and Wolski state that patients with diabeteswho are treated with rosiglitazone face a "worrisome" risk ofmyocardial infarction (P=0.03) and death from cardiovascularcauses (P=0.06). They base their conclusion on a meta-analysisof 42 trials showing an absolute difference in risk of only0.2%, but they excluded 4 trials from the infarction analysisand 19 trials from the mortality analysis in which the rateswere zero. Did these exclusions influence the results?
To find out, we performed a Bayesian meta-analysis1,2 of theentire data set with the use of a standard continuity correctionthat adjusts for values of zero.3 The resultant odds ratio forinfarction dropped from 1.43 to 1.26 (95% CI, 0.93 to 1.72),and the odds ratio for death from cardiovascular causes droppedfrom 1.64 to 1.14 (95% CI, 0.74 to 1.74). Similarly correctedfixed-effects and random-effects analyses had similar results,and none of the resultant odds ratios remained significant.Although additional studies might be warranted, there is littlehere to justify what the authors (not to mention the media,the U.S. Congress, and worried patient groups) decry as an "urgentneed for comprehensive evaluations."
George A. Diamond, M.D. Sanjay Kaul, M.D. Cedars–Sinai Medical Center Los Angeles, CA 90048 gadiamond{at}pol.net
References
Jaynes ET. Probability theory: the logic of science. Cambridge, England: Cambridge University Press, 2003:257-60.
Brophy JM, Joseph L, Rouleau JL. Beta-blockers in congestive heart failure: a Bayesian meta-analysis. Ann Intern Med 2001;134:550-560. [Free Full Text]
Friedrich JO, Adhikari NKJ, Beyene J. Inclusion of zero total event trials in meta-analyses maintains analytic consistency and incorporates all available data. BMC Med Res Methodol 2007;7:5-11. [CrossRef][Medline]
To the Editor: In the editorial accompanying the article byNissen and Wolski, Psaty and Furberg1 disparage physicians whohave prescribed rosiglitazone, saying that "physicians who choseto prescribe rosiglitazone perhaps focused on the single dimensionof glycemic control," and that many physicians "did not requireproof of health benefits as a criterion for selecting rosiglitazone."If physicians had required this higher standard, the authorswrite, "they would have been at a loss for evidence from large,long-term trials."
Practicing physicians should not be faulted for prescribingrosiglitazone. Experts and clinical guidelines repeatedly emphasizethat glycemic control is the important thing, no matter howyou accomplish it. In a recent guideline, the American DiabetesAssociation states that "the goal of therapy is to achieve anA1c [glycated hemoglobin level] as close to normal as possible,"without acknowledging that clinical outcomes may depend on bothglycemic control and the therapies used to achieve it.2 In addition,targets for glycated hemoglobin are increasingly used to judgethe quality of care, without regard to specific therapies.3It is unreasonable to expect that practicing physicians wouldbe more knowledgeable about fine distinctions between outcomesand surrogate end points than are the experts who guide clinicalpractice.
Allan S. Brett, M.D. University of South Carolina School of Medicine Columbia, SC 29203
References
Psaty BM, Furberg CD. Rosiglitazone and cardiovascular risk. N Engl J Med 2007;356:2522-2524. [Free Full Text]
American Diabetes Association. Standards of medical care in diabetes -- 2007. Diabetes Care 2007;30:Suppl 1:S4-S41. [Free Full Text]
Pogach L, Engelgau M, Aron D. Measuring progress toward achieving hemoglobin A1c goals in diabetes care: pass/fail or partial credit. JAMA 2007;297:520-523. [Free Full Text]
The authors reply: Dr. Bracken expresses concern that the analysisof relative risk with the simple pooled trial data does notshow an increased risk for rosiglitazone. Because the 48 analyzedstudies included trials with sizes and randomization ratiosthat varied widely, performing a simple pooled analysis is nota statistically meaningful approach to integrating the trialresults. We used the Peto odds ratio because this method iswidely viewed as the optimal approach when there are relativelyfew events in individual trials.1 In two of the trials in ourstudy, 49653/234 and SB-712753/009, there were no myocardialinfarctions or deaths from cardiovascular causes, but thesetrials were included in Table 3 of our article because theycontributed to the analysis of death from any cause.
Dr. Bracken also expresses concern regarding potential amplificationof odds ratios by the inclusion of trials with unbalanced allocationratios. Combining all trials with the same allocation ratios(including those with zero events in both groups), we find noevidence of such an effect (Table 1). In fact, the exclusionof trials with a 3:1 or 4:1 allocation ratio gives a resultnearly identical to that we reported, regardless of the methodused (Peto odds ratio, 1.43 [95% CI, 1.02 to 1.98]; Mantel–Haenszelodds ratio, 1.42 [95% CI, 1.02 to 1.98]).
Table 1. Results from Trials Grouped According to Allocation Ratio.
Mannucci et al. express concern about the potential for differentialexposure for comparators in the DREAM and ADOPT trials. We findno evidence of such an effect. The "number at risk" values inthe Kaplan–Meier plots for these trials refer to patientswho did not reach a primary glycemic end point. Safety dataare reported for the entire population with the use of an intention-to-treatapproach. For the safety analysis, randomization ensures thatexposure is balanced.
Drs. Diamond and Kaul suggest the use of a Bayesian approachto analyze the data from our study. However, this method remainscontroversial and is sensitive to the previous probability distributionselected by the investigators.2 For Bayesian meta-analyses,confidence intervals are almost always wider than those calculatedby other methods.
In selecting statistical methods for a meta-analysis, we believeit is important to approach the problem with the same rigoras that used in a randomized trial.2 Accordingly, investigatorsshould prespecify the approach and not apply a series of alternativemethods in a post hoc manner.
In considering all of these alternative methods, it should benoted that the maker of rosiglitazone, GlaxoSmithKline, conductedits own analysis of "14,237 subjects in 42 controlled double-blindstudies," using logistic regression with adjustment for covariatesand exposure.3 This analysis showed a significant increase of31% in the risk of "myocardial ischemic events." Presumably,the company did not deliberately select statistical methodsthat were likely to maximize the estimated hazard.
Steven E. Nissen, M.D. Kathy Wolski, M.P.H. Cleveland Clinic Cleveland, OH 44195 nissens{at}ccf.org
References
Bradburn MJ, Deeks JJ, Berlin JA, Russell Localio A. Much ado about nothing: a comparison of the performance of meta-analytical methods with rare events. Stat Med 2007;26:53-77. [CrossRef][ISI][Medline]
Egger M, Smith GD, Phillips AN. Meta-analysis: principles and procedures. BMJ 1997;315:1533-1537. [Free Full Text]
The editorialists reply: Before the Food and Drug Administrationapproved rosiglitazone in 1999, the agency, which was concernedabout the drug's effects on body weight and lipids, recommendedthat a post-marketing study of cardiovascular events be conductedto document the presumed health benefits of rosiglitazone-associatedglucose lowering.1 Despite the absence of such an events trial,the rapid expansion of the use of rosiglitazone depended onprescriptions written by physicians, who had a choice amongavailable agents. Although the American Diabetes Associationrecommends the lowering of glycated hemoglobin levels, the recommendationsidentify the possibility of the prevention of macrovasculardisease as grade B, or supported by epidemiologic studies ratherthan by randomized clinical trials.2 The guidelines also featuremetformin as the preferred first-line agent in the treatmentof type 2 diabetes.
Treatment decisions based on surrogate end points should bemade with caution. Insofar as the prescribing patterns of physicianscoincide with high-quality evidence provided by large, long-termtrials, they may also serve as an incentive for industry toconduct trials of sufficient range and scope to provide high-qualitycare to patients.
Bruce M. Psaty, M.D., Ph.D. University of Washington Seattle, WA 98101
Curt D. Furberg, M.D., Ph.D. Wake Forest University Winston-Salem,NC 27106
References
Misbin RI. Medical office review of Avandia: application number: 021071. Center for Drug Evaluation and Research. (Accessed August 9, 2007, at http://oversight.house.gov/story.asp?ID=1325.)
American Diabetes Association. Standards of medical care in diabetes -- 2007. Diabetes Care 2007;30:Suppl 1:S4-S41. [Free Full Text]
WALKER, E., HERNANDEZ, A. V., KATTAN, M. W.
(2008). Meta-analysis: Its strengths and limitations. Cleveland Clinic Journal of Medicine
75: 431-439
[Abstract][Full Text]
Dahabreh, I. J
(2008). Meta-analysis of rare events: an update and sensitivity analysis of cardiovascular events in randomized trials of rosiglitazone. Clin Trials
5: 116-120
[Abstract]
Previous
