Correspondence
Drug-Eluting versus Bare-Metal Stents in Acute Myocardial Infarction
N Engl J Med 2009; 360:300-302January 15, 2009
- Article
To the Editor:
Mauri et al. (Sept. 25 issue)1 report lower 2-year mortality rates among patients presenting in 2003–2004 with acute myocardial infarction who were selected for treatment with drug-eluting stents than among those selected for treatment with bare-metal stents. However, the fundamental question of causality between use of drug-eluting stents and mortality cannot be answered in this observational study. The rate of use of drug-eluting stents in the study was 55.6%, which is much lower than the national average of 90% in the same period.2 Although the recognition and avoidance of off-label use by academic cardiologists in Massachusetts may be part of the reason for the lower rate in the study by Mauri et al., it is also possible that drug-eluting stents were systematically avoided in patients who were perceived to have a poor prognosis and therefore less potential gain, such as those who were terminally ill, had a poor general health status, or had advanced coexisting conditions. Thus, before one speculates about the putative mechanisms underlying the observed survival advantage of drug-eluting stents over bare-metal stents, perhaps one should first investigate the reason behind the below-average use of drug-eluting stents in this cohort.
2 ReferencesBrian Y.L. Wong, M.D.
Sudbury Regional Hospital, Sudbury, ON P3E 3B6, Canada1
Mauri L ,Silbaugh TS ,Garg P , et al. Drug-eluting or bare-metal stents for acute myocardial infarction. N Engl J Med 2008;359:1330-1342
Full Text | Web of Science | Medline2
Shuchman M . Trading restenosis for thrombosis? New questions about drug-eluting stents. N Engl J Med 2006;355:1949-1952
Full Text | Web of Science | Medline
To the Editor:
Mauri et al. assessed the long-term outcomes with drug-eluting stents as compared with bare-metal stents after acute myocardial infarction, using propensity-score matching to minimize bias resulting from nonrandom treatment assignment. A propensity model should include the variables that affect both the treatment assignment and outcomes in order to avoid imbalances in prognostically important variables between the treatment groups.1 The authors included a list of patient-level and lesion-specific characteristics in their propensity models. However, there were 21 hospitals in the Massachusetts Data Analysis Center database, and the practice pattern (e.g., selection between drug-eluting stents and bare-metal stents) and outcomes may well vary among different hospitals, a factor that should also be taken into consideration in constructing the propensity models. Otherwise, this could be a source of residual bias.
1 ReferencesDae Hyun Kim, M.D., M.P.H.
Beth Israel Deaconess Medical Center, Boston, MA 02215
dkim2@bidmc.harvard. edu 1
Austin PC ,Grootendorst P ,Anderson GM . A comparison of the ability of different propensity score models to balance measured variables between treated and untreated subjects: a Monte Carlo study. Stat Med 2007;26:734-753
CrossRef | Web of Science | Medline
To the Editor:
Mauri et al. report that the 2-year risk of death among patients receiving drug-eluting stents was 16% lower than the risk among patients receiving bare-metal stents. To control for confounding, the authors used propensity-score matching, but because they were uncertain whether their data accounted for all confounding, they conducted a sensitivity analysis, looking at 2-day mortality after stent insertion, a period during which baseline risk factors would affect the risk of death but the drug elution from a stent would not. During the 2 days after the stent procedure, patients receiving drug-eluting stents had a risk of death that was only 58% of the risk among those receiving bare-metal stents, after adjustment for measured confounders. This finding confirms that confounding was not fully controlled. The authors should have then adjusted the observed findings through 2 years by the factor of 58% from their sensitivity analysis, a method known as bias analysis.1 Had they done so, the observed risk ratio over the 2-year period would have been a 44% increase in the risk of death among patients receiving drug-eluting stents, rather than the reported 16% lower risk.
1 ReferencesKenneth J. Rothman, Dr.P.H.
RTI Health Solutions, Research Triangle Park, NC 27709Author/Editor Response
Wong suggests that the relationship between stent choice and outcome cannot be evaluated in our study because physicians selected the type of stent according to perceptions of relative risk and benefit in individual patients. In fact, the purpose of propensity-score matching, which we performed, is to adjust for exactly the sort of selection bias that Wong describes. Patients with similar observed characteristics are matched, and patients with a composite of characteristics not represented in both treatment groups are excluded. An inference regarding the treatment effect is therefore made on the basis of patient groups with similar observed characteristics. Like other groups reporting on the use of drug-eluting stents in the United States, we found that the rate of use of drug-eluting stents during the latest period in our study was above 90% overall.1 We purposely selected a treatment period that did not involve the use of one type of stent in the majority of patients but that instead spanned a period when stent use changed from predominantly bare-metal stents to predominantly drug-eluting stents; we did this to reduce selection bias, which would not be possible in a population in which 10% or less of the patients received one type of stent. Finally, our study population represented a range of practices, including community hospitals performing primary angioplasty, not only academic hospitals.
Kim suggests that one must consider how treatment selection and outcomes may vary across different hospital sites, and we agree. Over our study period, we detected no measurable differences in in-hospital mortality across hospitals.2,3 We confirmed that every hospital used each type of stent. We reported propensity models that adjusted for hospital type, time, and the interaction between hospital type and time with regard to treatment selection, with results that were consistent with the primary observations in our study.
In reply to Rothman: our causal estimate was the absolute difference in risk, not the relative risk. Two days after stent placement, the absolute difference was –0.5 percentage point (95% confidence interval [CI], –1.0 to 0.0), favoring drug-eluting stents. If, as Rothman suggests, this absolute difference of 0.5 percentage point were applied to the 2-year mortality difference (–2.1 percentage points; 95% CI, –3.8 to –0.4), the point estimate we observed would continue to favor drug-eluting stents rather than demonstrate an increased risk.
Our study is observational, and we acknowledge that one cannot rule out residual confounding. To understand the effect of stent choice in myocardial infarction, we have made our best attempts with this rich data set to eliminate confounding, and we welcome replication in other studies.
3 ReferencesLaura Mauri, M.D., M.Sc.
Brigham and Women's Hospital, Boston, MA 02115
lmauri1@partners.org Sharon-Lise T. Normand, Ph.D.
Harvard Medical School, Boston, MA 021151
Mauri L ,Silbaugh TS ,Wolf RE , et al. Long-term clinical outcomes after drug-eluting and bare-metal stenting in Massachusetts. Circulation 2008;118:1817-1827
CrossRef | Web of Science | Medline2
Massachusetts Data Analysis Center. Percutaneous coronary intervention in the commonwealth of Massachusetts: April 1, 2003–December 31, 2003. Boston: Department of Health Care Policy, 2005. (Accessed December 22, 2008, at http://www.massdac.org/reports/PCI%20Public%202003.pdf.)
3
Idem. Percutaneous coronary intervention in the commonwealth of Massachusetts: January 1, 2004–December 31, 2004. Boston: Department of Health Care Policy, 2006. (Accessed December 22, 2008, at http://www.massdac.org/reports/PCI%20Public%202004.pdf.)
