IMPROVING PATIENT CARE
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure Among Patients Having Percutaneous Coronary Intervention Hitinder S. Gurm, MD; Carrie Hosman, PhD; David Share, MD; Mauro Moscucci, MD; and Ben B. Hansen, PhD, for the Blue Cross Blue Shield of Michigan Cardiovascular Consortium
Background: The role of vascular closure devices (VCDs) in patients having percutaneous coronary intervention (PCI) is controversial, and recommendations for use vary. Objective: To examine the use of and outcomes associated with VCDs in real-world practice. Design: Observational cohort study. Setting: 32 hospitals in Michigan that participate in a large multicenter quality improvement collaborative. Patients: Consecutive patients having emergent and nonemergent PCI from 2007 to 2009. Measurements: transfusion.
Vascular
complications
and
the
need
for
Results: Of the 85 048 PCIs performed during the study that met the inclusion criteria, 28 528 (37%) procedures used VCDs. In propensity score–matched analysis, VCDs were associated with reductions in vascular complications (odds ratio [OR], 0.78 [95% CI, 0.67 to 0.90]; P ⫽ 0.001) and postprocedure transfusions (OR, 0.85 [CI, 0.74 to 0.96]; P ⫽ 0.011). These findings were consistent across many prespecified subgroups except for patients with a body
V
ascular complications related to arterial access are a common cause of morbidity, mortality, and health care cost in patients having percutaneous coronary intervention (PCI) (1, 2). Radial artery access is the most effective strategy to prevent vascular complications (3); however, most patients in the United States have transfemoral PCI and require other strategies to reduce vascular complications. Interest has been growing in access-site management as a factor in reducing vascular complications. Vascular closure devices (VCDs) permit closure of the arteriotomy site using sutures, plugs, or metallic clips. The role of these devices in preventing vascular complications remains controversial. Most of the randomized trials evaluating VCDs have been small and underpowered, and the largest metaanalysis on the subject raised concerns that these devices may be associated with an increase in vascular complications (4). They are commonly used in clinical practice, and a recent large observational study suggested that they may be associated with a reduction in bleeding complications (5).
See also: Web-Only Supplements
mass index (BMI) less than 25 kg/m2 and those treated with platelet glycoprotein (GP) IIb/IIIa inhibitors, in whom the benefit of VCDs over manual closure was attenuated. When the specific subtypes of vascular complications were evaluated, VCDs were associated with fewer hematomas (OR, 0.69 [CI, 0.58 to 0.83]; P ⬍ 0.001) or pseudoaneurysms (OR, 0.54 [CI, 0.38 to 0.76]; P ⬍ 0.001) but an increase in the odds of retroperitoneal bleeding (OR, 1.57 [CI, 1.12 to 2.20]; P ⫽ 0.009). Limitation: Unmeasured confounding cannot be excluded despite the study having measured and balanced many confounders. Conclusion: Vascular closure devices were associated with a significant reduction in vascular complications and need for transfusion in this large cohort of patients having transfemoral PCI. This benefit was lost in patients receiving GP IIb/IIIa inhibitors and those with normal or lean BMI and was counterbalanced by a small increase in the more serious risk for retroperitoneal bleeding. Primary Funding Source: Blue Cross Blue Shield of Michigan and the National Science Foundation. Ann Intern Med. 2013;159:660-666. For author affiliations, see end of text.
www.annals.org
Data are scarce on the comparative efficacy of VCDs in real-world practice. The aim of our study was to use advanced matching techniques to assess the association of VCDs with vascular complications. Furthermore, we specifically focused on patient subgroups that may be at particularly high risk for these complications but in whom data have been lacking.
METHODS Data Sources and Study Population
We included data from patients having PCI at 32 hospitals participating in the Blue Cross Blue Shield of Michigan Cardiovascular Consortium (BMC2). The details of the BMC2 registry and its data collection and auditing processes have been described previously (6, 7). In brief, procedural data on all patients having PCI at participating hospitals are collected using standardized data collection forms. Baseline data include clinical, demographic, procedural, and angiographic characteristics; medications used before, during, and after the procedure; and in-hospital outcomes. All data elements have been prospectively defined, and the protocol is approved by local institutional review boards at each hospital. In addition to a random audit of 2% of all cases, medical records of all patients having multiple procedures or coronary artery bypass graft-
660 © 2013 American College of Physicians
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Comparative Safety of Vascular Closure Devices and Manual Closure
ing and of those who died in the hospital are reviewed routinely to ensure data accuracy. The population for this analysis included all consecutive patients (n ⫽ 92 673) who had PCI between 1 January 2007 and 31 December 2009. We excluded patients having PCI via a radial or brachial route (n ⫽ 1643); patients who presented with cardiac arrest (n ⫽ 1404) or cardiogenic shock (n ⫽ 1635); patients who required cardiopulmonary resuscitation in the catheterization laboratory (n ⫽ 2019); patients in whom a support device, such an intraaortic balloon pump or left ventricular assist device, was used (because the complication from the support device could not be differentiated from that associated with the access site) (n ⫽ 2967); patients who died in the catheterization laboratory (n ⫽104); and patients with missing data on VCD use (n ⫽ 1275 [1.3%]). Study End Points
Our study had 3 end points: vascular complication, transfusion, and in-hospital death. Vascular complication was defined as the occurrence of one of the following in relation to the access site: acute thrombosis, loss of limb, retroperitoneal bleeding, need for surgical repair, pseudoaneurysm, or hematoma requiring transfusion or arteriovenous fistula. In-hospital death was defined as death from any cause before discharge from the hospital after PCI. Statistical Analysis
We separated patients into 2 categories according to whether or not they were treated with VCDs after PCI. The 2 sets of patients were then compared on baseline variables, summarizing continuous variables in terms of their means and discrete variables using percentages. To account for the nonrandom assignment of a closure device to patients having PCI, we used a combination of exact matching, caliper matching, and optimal full matching on propensity scores (8). The propensity scores (conditional probabilities of receiving a VCD) were estimated using nonparsimonious logistic regression models incorporating various patient characteristics and therapies before PCI, including weight, height, race, cardiac status, presenting symptoms or diagnosis, and comorbid conditions and therapies received before and during PCI. Appendix Table 1 (available at www.annals.org) presents the distribution of a selected subset of 45 of these variables among patients with and without VCDs. Supplements 1 to 4 (available at www.annals.org) provide the details of the statistical procedure. In brief, to account for the many clinical variables and respect the grouping of patients by hospital, we matched simultaneously on 2 distinct propensity scores: an “inclusive” propensity score comprising all 206 clinical variables and a “multilevel” score incorporating fewer variables (the 45 appearing in Appendix Table 1 plus 3 constructed variables described in Supplement 1) but addressing the multilevel structure by permitting the variables’ coefficients to vary by hospital. To facilitate subsequent subgroup analyses, this www.annals.org
Original Research
Context Vascular closure devices (VCDs) are designed to prevent arterial bleeding, especially after percutaneous coronary intervention (PCI) by the transfemoral route. Their use is controversial.
Contribution The investigators studied a large number of patients in Michigan hospitals who had PCI. Patients with VCDs had fewer hematomas and pseudoaneurysms but more retroperitoneal bleeding. The benefits were confined to patients with higher body mass indexes and those who were not receiving glycoprotein IIb/IIIa inhibitors.
Caution Such observational studies can establish associations but not cause and effect.
Implication Vascular closure devices should be considered only in patients who are most likely to benefit. —The Editors
matching was performed within exact matched subsets defined by age, sex, body mass index (BMI) (lean or normal vs. overweight or obese), myocardial infarction within 7 days of presentation, use of a platelet glycoprotein (GP) IIb/IIIa inhibitor, use of bivalirudin, and presence of peripheral vascular disease. To further address the possibility of confounding by cluster, we exact-matched within groups of hospitals experiencing lower, moderate, and higher volumes of PCI during the study, having categorized hospitals in such a way that the 3 groups would contain roughly equal numbers of patients (9). Outcome models that we fit subsequently not only accounted for the matches but also, following a common method to address confounding by cluster in unmatched analysis, included fixed effects for hospitals (10). To assess the stability of our results to the choice of match, we adapted identical outcome models to an alternative match produced by a simplified version of our procedure that gave less attention to hospital differences. Supplement 1 describes the details of this match. Adequacy of covariate balance after matching was assessed using 2 strategies (8, 11). First, we assessed paired differences on the variables contributing to either of the 2 propensity scores to ensure that few large imbalances, or trends favoring the VCD or the non-VCD halves of matched pairs, exist after matching and that remaining imbalances are not statistically significant (that is, larger than would be expected if VCD use were distributed at random in each matched set). The same calculations can be viewed as using the varyingly sized matched sets that full matching had generated to specify a standard population, then comparing VCD and non-VCD segments of the matched sam19 November 2013 Annals of Internal Medicine Volume 159 • Number 10 661
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure
ple in terms of weighted means and percentages, weighting both subsamples to reflect the same standard population (12). The Appendix Figure and Appendix Table 2 (both available at www.annals.org) report descriptive statistics for the subsamples that have been standardized in this way. A similar pattern of close agreement at baseline between matched VCD and non-VCD samples was observed for the alternate, simplified match. However, for brevity, we omitted those comparisons. Second, we assessed differences in an unrelated outcome (contrast-induced nephropathy [CIN]) that would not be expected to be influenced by VCD use (13). Presence of a differing outcome between the 2 groups in the absence of biological plausibility would suggest residual confounding. Outcome analyses used conditional logistic regression to model vascular complications, the need for subsequent transfusion, or in-hospital death as a parsimonious function of VCD use and selected variables preceding the VCD intervention; the 3 analyses each incorporated matching results using conditioning for matched sets (14). The additional independent variables (besides VCD use and the propensity score matches) included in these outcome regressions were glomerular filtration rate, ST-segment elevation myocardial infarction, prior congestive heart failure, congestive heart failure on admission, prior PCI, and the hospital in which the procedure was performed. Accordingly, VCD coefficients estimated with these models are interpretable as increments to the log odds of vascular complications, transfusion need, or death, respectively, associated with VCD use, after taking into account these clinical characteristics and the hospital in which the procedure was performed. Subgroup analyses considered elaborations of these models to include interactions of the VCD variables with exact matching variables; evidence of differences across a particular categorization is reported only if its interacting with VCD usage significantly improved the model’s overall fit. The association of VCDs with these complications was thus assessed in the entire population and within the prespecified subgroups. Subgroup analyses are reported only for outcomes with a sufficient number of events that permitted fitting of the models, including interactions. A sensitivity analysis was performed to assess the effect of an unmeasured confounder as previously described (15). Analysis was performed using R, version 2.15.1, and matching was performed using R package optmatch, version 7.3 (R Foundation for Statistical Computing, Vienna, Austria). Role of the Funding Source
The BMC2 registry is funded by Blue Cross Blue Shield of Michigan. Funding was also provided by the National Science Foundation. The funding source had no role in the analysis, study design, or decision to publish these results.
RESULTS Our study examined 85 048 PCI procedures, of which 28 528 (37%) used a VCD. The use of VCDs ranged from 0% to 83% across the participating institutions. The most commonly used device was the Angio-Seal (St. Jude Medical, Minneapolis, Minnesota), which was used in 56.5% of the VCD users, followed by Perclose (Abbott, Santa Clara, California), which was used in 9.7% of VCD users, and Starclose (Abbott, Santa Clara, California), which was used in 9.5% of VCD users. Supplements 5 and 6 (available at www.annals.org) depict an artist’s rendition of the 2 most commonly used devices. In unadjusted analysis, several significant differences between patients who were treated with manual closure and those treated with VCDs were found (Table 1 and Appendix Table 1). Patients treated with VCDs were generally younger and less likely to have comorbid conditions or primary PCI, but the matched population was wellbalanced (Table 1, Appendix Table 1, and Appendix Figure). Of the 85 048 patients, 68 874 were included in the matched analysis; of these, 1314 (1.9%) experienced vascular complications and 1893 (2.7%) required a postprocedure transfusion. Vascular closure devices were associated with a significant reduction in the odds of vascular complications (odds ratio [OR], 0.78 [95% CI, 0.67 to 0.90]; P ⫽ 0.001) and need for postprocedure transfusion (OR, 0.85 [CI, 0.74 to 0.96]; P ⫽ 0.011) (Table 2). The effect of VCDs on reduction in vascular complications was evident in all subgroups except those with a BMI less than 25 kg/m2 and those who received a platelet GP IIb/IIIa inhibitor (Figure). The reduction in vascular complications was especially marked in patients who received bivalirudin. When the specific subtypes of vascular complications were evaluated, the most common complication was hematoma, which occurred in 924 patients (1.3%), whereas pseudoaneurysms occurred in 325 (0.5%) and retroperitoneal bleeding developed in 239 (0.3%). Vascular closure devices were associated with a reduction in hematomas (OR, 0.69 [CI, 0.58 to 0.82]; P ⬍ 0.001) or pseudoaneurysms (OR, 0.57 [CI, 0.41 to 0.78]; P ⫽ 0.007), whereas the odds of retroperitoneal bleeding (OR, 1.57 [CI, 1.13 to 2.16]; P ⫽ 0.001) were higher. The other outcomes were too infrequent to provide reliable estimates of the comparative efficacy of VCDs and manual compression. These trends were preserved in all subgroups except patients with normal or lean BMI and those who received platelet GP IIb/IIIa inhibitors in whom no benefit of VCDs was noted. Among patients who received GP IIb/ IIIa inhibitors, despite an overall lack of benefit of VCDs with respect to vascular complications, pseudoaneurysms were significantly reduced (n ⫽ 143; OR, 0.53 [CI, 0.31 to 0.90]; P ⫽ 0.019), which was counterbalanced by significantly increased odds of retroperitoneal bleeding (n ⫽ 171; OR, 1.88 [CI, 1.25 to 2.83]; P ⫽ 0.003). There was
662 19 November 2013 Annals of Internal Medicine Volume 159 • Number 10
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
www.annals.org
Comparative Safety of Vascular Closure Devices and Manual Closure
Original Research
Table 1. Covariate Balance Before and After Matching for Select Variables* Variable
Age, y Female Weight, kg White Hypertension Type 1 diabetes Type 2 diabetes Congestive heart failure Peripheral vascular disease Preprocedure hemoglobin level, g/L GFR, mL/min/1.73 m2 Primary PCI for STEMI Bivalirudin Platelet GP IIb/IIIa inhibitor Ejection fraction, %
Before Matching
After Matching
No VCD Use
VCD Use
P Value
No VCD Use
VCD Use
P Value
64.27 0.33 89.43 0.87 0.85 0.13 0.23 0.15 0.14 136 78.89 0.12 0.35 0.42 52.18
63.87 0.33 89.86 0.84 0.85 0.13 0.24 0.16 0.15 135 78.97 0.10 0.30 0.44 51.99
0.00 0.88 0.00 0.00 0.01 0.06 0.14 0.00 0.00 0.00 0.75 0.00 0.00 0.00 0.03
64.18 0.35 89.00 0.86 0.85 0.13 0.23 0.16 0.17 135 78.41 0.10 0.32 0.42 52.16
64.34 0.35 88.81 0.86 0.85 0.13 0.23 0.16 0.17 135 78.14 0.10 0.32 0.42 52.02
0.15 1.00 0.19 0.95 0.71 0.83 0.95 0.67 1.00 0.28 0.36 0.46 1.00 0.40 0.18
GFR ⫽ glomerular filtration rate; GP ⫽ glycoprotein; PCI ⫽ percutaneous coronary intervention; STEMI ⫽ ST-segment elevation myocardial infarction; VCD ⫽ vascular closure device. * Distribution of selected baseline demographic variables, comorbid conditions, and preprocedural and procedural therapy in patients who were treated with VCDs compared with those treated with manual compression in unmatched and matched data.
no significant reduction in general hematomas (n ⫽ 572; OR, 0.82 [CI, 0.65 to 1.04]; P ⫽ 0.102). No difference in overall mortality was observed with VCDs (n ⫽ 112; OR, 0.82 [CI, 0.45 to 1.48]; P ⫽ 0.51), but the small number of events made this comparison imprecise. The findings were consistent across institutions for vascular complications and transfusion (data not shown). There was no association between VCD use and the risk for CIN (OR, 0.94 [CI, 0.86 to 1.02]; P ⫽ 0.14). Findings about vascular complications and transfusion were moderately robust to the presence of unmeasured confounding. When a hypothetical unmeasured confounder was posited to have a moderate effect size, the significance of the effect of VCD on vascular complications and transfusion remained unchanged in each scenario considered (the scenarios varied according to prevalence of the unmeasured confounder in patients with a VCD as opposed to patients without a VCD). Allowing a hypothetical omitted confounder to have a large effect size and to differ in its prevalence in the 2 comparison groups (VCD and no VCD) in a manner unlike most of the measured variables in the data led to a loss of significance of the effect of VCD on transfusion and, for a specific set of prevalence rates, vascular complications (Supplement 1). Further, the stability check of substituting the simplified, alternate match into the analysis left the estimation results largely unchanged with respect to directionality of treatment effect estimates, although differences in significance were noted (Appendix Table 3, available at www .annals.org). First, although no significant association for in-hospital death was observed in our results, with use of the alternate simplified match, this finding would nearly have reversed for in-hospital mortality (P ⫽ 0.056 based on the analysis using the alternate match). Second, CIN is www.annals.org
associated with nonuse of VCDs after either match but significantly so only under the simplified match (P ⫽ 0.14 vs. 0.005).
DISCUSSION The key finding of our study is that VCD use is associated with a significant reduction in vascular complications and need for transfusion in patients having transfemoral PCI. The benefits of VCDs were consistently observed in multiple subgroups except for patients with a lean or normal BMI or those who received platelet GP IIb/IIIa inhibitors, in whom the clinical efficacy of these devices was attenuated. The results of our study are important for many reasons. Vascular closure devices are better-tolerated and permit more rapid ambulation than manual closure. Although VCD use is not reimbursed by most third-party payers, their relatively low cost (approximate list price of $275 to $350 per device) combined with these factors probably account for the widely prevalent use of VCDs despite a relative paucity of high-quality data supporting their efficacy.
Table 2. Association of Vascular Closure Device Use With Key Clinical Outcomes Variable
Total Patients, n
Events, n
Odds Ratio (95% CI)
P Value
Vascular complications Transfusion needed In-hospital death
68 874 68 874 68 874
1314 1893 112
0.78 (0.67–0.90) 0.85 (0.74–0.96) 0.82 (0.45–1.48)
0.001 0.011 0.51
19 November 2013 Annals of Internal Medicine Volume 159 • Number 10 663
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure
Figure. Propensity score–matched odds of vascular complications and transfusion in the overall cohort and subgroups of patients at varying risk for access site complications. Odds Ratio (95% CI) for Vascular Complications
P Value Events, n (for Interaction)
Women (n = 22 827)
775
Men (n = 46 047)
539
BMI
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure Among Patients Having Percutaneous Coronary Intervention Hitinder S. Gurm, MD; Carrie Hosman, PhD; David Share, MD; Mauro Moscucci, MD; and Ben B. Hansen, PhD, for the Blue Cross Blue Shield of Michigan Cardiovascular Consortium
Background: The role of vascular closure devices (VCDs) in patients having percutaneous coronary intervention (PCI) is controversial, and recommendations for use vary. Objective: To examine the use of and outcomes associated with VCDs in real-world practice. Design: Observational cohort study. Setting: 32 hospitals in Michigan that participate in a large multicenter quality improvement collaborative. Patients: Consecutive patients having emergent and nonemergent PCI from 2007 to 2009. Measurements: transfusion.
Vascular
complications
and
the
need
for
Results: Of the 85 048 PCIs performed during the study that met the inclusion criteria, 28 528 (37%) procedures used VCDs. In propensity score–matched analysis, VCDs were associated with reductions in vascular complications (odds ratio [OR], 0.78 [95% CI, 0.67 to 0.90]; P ⫽ 0.001) and postprocedure transfusions (OR, 0.85 [CI, 0.74 to 0.96]; P ⫽ 0.011). These findings were consistent across many prespecified subgroups except for patients with a body
V
ascular complications related to arterial access are a common cause of morbidity, mortality, and health care cost in patients having percutaneous coronary intervention (PCI) (1, 2). Radial artery access is the most effective strategy to prevent vascular complications (3); however, most patients in the United States have transfemoral PCI and require other strategies to reduce vascular complications. Interest has been growing in access-site management as a factor in reducing vascular complications. Vascular closure devices (VCDs) permit closure of the arteriotomy site using sutures, plugs, or metallic clips. The role of these devices in preventing vascular complications remains controversial. Most of the randomized trials evaluating VCDs have been small and underpowered, and the largest metaanalysis on the subject raised concerns that these devices may be associated with an increase in vascular complications (4). They are commonly used in clinical practice, and a recent large observational study suggested that they may be associated with a reduction in bleeding complications (5).
See also: Web-Only Supplements
mass index (BMI) less than 25 kg/m2 and those treated with platelet glycoprotein (GP) IIb/IIIa inhibitors, in whom the benefit of VCDs over manual closure was attenuated. When the specific subtypes of vascular complications were evaluated, VCDs were associated with fewer hematomas (OR, 0.69 [CI, 0.58 to 0.83]; P ⬍ 0.001) or pseudoaneurysms (OR, 0.54 [CI, 0.38 to 0.76]; P ⬍ 0.001) but an increase in the odds of retroperitoneal bleeding (OR, 1.57 [CI, 1.12 to 2.20]; P ⫽ 0.009). Limitation: Unmeasured confounding cannot be excluded despite the study having measured and balanced many confounders. Conclusion: Vascular closure devices were associated with a significant reduction in vascular complications and need for transfusion in this large cohort of patients having transfemoral PCI. This benefit was lost in patients receiving GP IIb/IIIa inhibitors and those with normal or lean BMI and was counterbalanced by a small increase in the more serious risk for retroperitoneal bleeding. Primary Funding Source: Blue Cross Blue Shield of Michigan and the National Science Foundation. Ann Intern Med. 2013;159:660-666. For author affiliations, see end of text.
www.annals.org
Data are scarce on the comparative efficacy of VCDs in real-world practice. The aim of our study was to use advanced matching techniques to assess the association of VCDs with vascular complications. Furthermore, we specifically focused on patient subgroups that may be at particularly high risk for these complications but in whom data have been lacking.
METHODS Data Sources and Study Population
We included data from patients having PCI at 32 hospitals participating in the Blue Cross Blue Shield of Michigan Cardiovascular Consortium (BMC2). The details of the BMC2 registry and its data collection and auditing processes have been described previously (6, 7). In brief, procedural data on all patients having PCI at participating hospitals are collected using standardized data collection forms. Baseline data include clinical, demographic, procedural, and angiographic characteristics; medications used before, during, and after the procedure; and in-hospital outcomes. All data elements have been prospectively defined, and the protocol is approved by local institutional review boards at each hospital. In addition to a random audit of 2% of all cases, medical records of all patients having multiple procedures or coronary artery bypass graft-
660 © 2013 American College of Physicians
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Comparative Safety of Vascular Closure Devices and Manual Closure
ing and of those who died in the hospital are reviewed routinely to ensure data accuracy. The population for this analysis included all consecutive patients (n ⫽ 92 673) who had PCI between 1 January 2007 and 31 December 2009. We excluded patients having PCI via a radial or brachial route (n ⫽ 1643); patients who presented with cardiac arrest (n ⫽ 1404) or cardiogenic shock (n ⫽ 1635); patients who required cardiopulmonary resuscitation in the catheterization laboratory (n ⫽ 2019); patients in whom a support device, such an intraaortic balloon pump or left ventricular assist device, was used (because the complication from the support device could not be differentiated from that associated with the access site) (n ⫽ 2967); patients who died in the catheterization laboratory (n ⫽104); and patients with missing data on VCD use (n ⫽ 1275 [1.3%]). Study End Points
Our study had 3 end points: vascular complication, transfusion, and in-hospital death. Vascular complication was defined as the occurrence of one of the following in relation to the access site: acute thrombosis, loss of limb, retroperitoneal bleeding, need for surgical repair, pseudoaneurysm, or hematoma requiring transfusion or arteriovenous fistula. In-hospital death was defined as death from any cause before discharge from the hospital after PCI. Statistical Analysis
We separated patients into 2 categories according to whether or not they were treated with VCDs after PCI. The 2 sets of patients were then compared on baseline variables, summarizing continuous variables in terms of their means and discrete variables using percentages. To account for the nonrandom assignment of a closure device to patients having PCI, we used a combination of exact matching, caliper matching, and optimal full matching on propensity scores (8). The propensity scores (conditional probabilities of receiving a VCD) were estimated using nonparsimonious logistic regression models incorporating various patient characteristics and therapies before PCI, including weight, height, race, cardiac status, presenting symptoms or diagnosis, and comorbid conditions and therapies received before and during PCI. Appendix Table 1 (available at www.annals.org) presents the distribution of a selected subset of 45 of these variables among patients with and without VCDs. Supplements 1 to 4 (available at www.annals.org) provide the details of the statistical procedure. In brief, to account for the many clinical variables and respect the grouping of patients by hospital, we matched simultaneously on 2 distinct propensity scores: an “inclusive” propensity score comprising all 206 clinical variables and a “multilevel” score incorporating fewer variables (the 45 appearing in Appendix Table 1 plus 3 constructed variables described in Supplement 1) but addressing the multilevel structure by permitting the variables’ coefficients to vary by hospital. To facilitate subsequent subgroup analyses, this www.annals.org
Original Research
Context Vascular closure devices (VCDs) are designed to prevent arterial bleeding, especially after percutaneous coronary intervention (PCI) by the transfemoral route. Their use is controversial.
Contribution The investigators studied a large number of patients in Michigan hospitals who had PCI. Patients with VCDs had fewer hematomas and pseudoaneurysms but more retroperitoneal bleeding. The benefits were confined to patients with higher body mass indexes and those who were not receiving glycoprotein IIb/IIIa inhibitors.
Caution Such observational studies can establish associations but not cause and effect.
Implication Vascular closure devices should be considered only in patients who are most likely to benefit. —The Editors
matching was performed within exact matched subsets defined by age, sex, body mass index (BMI) (lean or normal vs. overweight or obese), myocardial infarction within 7 days of presentation, use of a platelet glycoprotein (GP) IIb/IIIa inhibitor, use of bivalirudin, and presence of peripheral vascular disease. To further address the possibility of confounding by cluster, we exact-matched within groups of hospitals experiencing lower, moderate, and higher volumes of PCI during the study, having categorized hospitals in such a way that the 3 groups would contain roughly equal numbers of patients (9). Outcome models that we fit subsequently not only accounted for the matches but also, following a common method to address confounding by cluster in unmatched analysis, included fixed effects for hospitals (10). To assess the stability of our results to the choice of match, we adapted identical outcome models to an alternative match produced by a simplified version of our procedure that gave less attention to hospital differences. Supplement 1 describes the details of this match. Adequacy of covariate balance after matching was assessed using 2 strategies (8, 11). First, we assessed paired differences on the variables contributing to either of the 2 propensity scores to ensure that few large imbalances, or trends favoring the VCD or the non-VCD halves of matched pairs, exist after matching and that remaining imbalances are not statistically significant (that is, larger than would be expected if VCD use were distributed at random in each matched set). The same calculations can be viewed as using the varyingly sized matched sets that full matching had generated to specify a standard population, then comparing VCD and non-VCD segments of the matched sam19 November 2013 Annals of Internal Medicine Volume 159 • Number 10 661
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure
ple in terms of weighted means and percentages, weighting both subsamples to reflect the same standard population (12). The Appendix Figure and Appendix Table 2 (both available at www.annals.org) report descriptive statistics for the subsamples that have been standardized in this way. A similar pattern of close agreement at baseline between matched VCD and non-VCD samples was observed for the alternate, simplified match. However, for brevity, we omitted those comparisons. Second, we assessed differences in an unrelated outcome (contrast-induced nephropathy [CIN]) that would not be expected to be influenced by VCD use (13). Presence of a differing outcome between the 2 groups in the absence of biological plausibility would suggest residual confounding. Outcome analyses used conditional logistic regression to model vascular complications, the need for subsequent transfusion, or in-hospital death as a parsimonious function of VCD use and selected variables preceding the VCD intervention; the 3 analyses each incorporated matching results using conditioning for matched sets (14). The additional independent variables (besides VCD use and the propensity score matches) included in these outcome regressions were glomerular filtration rate, ST-segment elevation myocardial infarction, prior congestive heart failure, congestive heart failure on admission, prior PCI, and the hospital in which the procedure was performed. Accordingly, VCD coefficients estimated with these models are interpretable as increments to the log odds of vascular complications, transfusion need, or death, respectively, associated with VCD use, after taking into account these clinical characteristics and the hospital in which the procedure was performed. Subgroup analyses considered elaborations of these models to include interactions of the VCD variables with exact matching variables; evidence of differences across a particular categorization is reported only if its interacting with VCD usage significantly improved the model’s overall fit. The association of VCDs with these complications was thus assessed in the entire population and within the prespecified subgroups. Subgroup analyses are reported only for outcomes with a sufficient number of events that permitted fitting of the models, including interactions. A sensitivity analysis was performed to assess the effect of an unmeasured confounder as previously described (15). Analysis was performed using R, version 2.15.1, and matching was performed using R package optmatch, version 7.3 (R Foundation for Statistical Computing, Vienna, Austria). Role of the Funding Source
The BMC2 registry is funded by Blue Cross Blue Shield of Michigan. Funding was also provided by the National Science Foundation. The funding source had no role in the analysis, study design, or decision to publish these results.
RESULTS Our study examined 85 048 PCI procedures, of which 28 528 (37%) used a VCD. The use of VCDs ranged from 0% to 83% across the participating institutions. The most commonly used device was the Angio-Seal (St. Jude Medical, Minneapolis, Minnesota), which was used in 56.5% of the VCD users, followed by Perclose (Abbott, Santa Clara, California), which was used in 9.7% of VCD users, and Starclose (Abbott, Santa Clara, California), which was used in 9.5% of VCD users. Supplements 5 and 6 (available at www.annals.org) depict an artist’s rendition of the 2 most commonly used devices. In unadjusted analysis, several significant differences between patients who were treated with manual closure and those treated with VCDs were found (Table 1 and Appendix Table 1). Patients treated with VCDs were generally younger and less likely to have comorbid conditions or primary PCI, but the matched population was wellbalanced (Table 1, Appendix Table 1, and Appendix Figure). Of the 85 048 patients, 68 874 were included in the matched analysis; of these, 1314 (1.9%) experienced vascular complications and 1893 (2.7%) required a postprocedure transfusion. Vascular closure devices were associated with a significant reduction in the odds of vascular complications (odds ratio [OR], 0.78 [95% CI, 0.67 to 0.90]; P ⫽ 0.001) and need for postprocedure transfusion (OR, 0.85 [CI, 0.74 to 0.96]; P ⫽ 0.011) (Table 2). The effect of VCDs on reduction in vascular complications was evident in all subgroups except those with a BMI less than 25 kg/m2 and those who received a platelet GP IIb/IIIa inhibitor (Figure). The reduction in vascular complications was especially marked in patients who received bivalirudin. When the specific subtypes of vascular complications were evaluated, the most common complication was hematoma, which occurred in 924 patients (1.3%), whereas pseudoaneurysms occurred in 325 (0.5%) and retroperitoneal bleeding developed in 239 (0.3%). Vascular closure devices were associated with a reduction in hematomas (OR, 0.69 [CI, 0.58 to 0.82]; P ⬍ 0.001) or pseudoaneurysms (OR, 0.57 [CI, 0.41 to 0.78]; P ⫽ 0.007), whereas the odds of retroperitoneal bleeding (OR, 1.57 [CI, 1.13 to 2.16]; P ⫽ 0.001) were higher. The other outcomes were too infrequent to provide reliable estimates of the comparative efficacy of VCDs and manual compression. These trends were preserved in all subgroups except patients with normal or lean BMI and those who received platelet GP IIb/IIIa inhibitors in whom no benefit of VCDs was noted. Among patients who received GP IIb/ IIIa inhibitors, despite an overall lack of benefit of VCDs with respect to vascular complications, pseudoaneurysms were significantly reduced (n ⫽ 143; OR, 0.53 [CI, 0.31 to 0.90]; P ⫽ 0.019), which was counterbalanced by significantly increased odds of retroperitoneal bleeding (n ⫽ 171; OR, 1.88 [CI, 1.25 to 2.83]; P ⫽ 0.003). There was
662 19 November 2013 Annals of Internal Medicine Volume 159 • Number 10
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
www.annals.org
Comparative Safety of Vascular Closure Devices and Manual Closure
Original Research
Table 1. Covariate Balance Before and After Matching for Select Variables* Variable
Age, y Female Weight, kg White Hypertension Type 1 diabetes Type 2 diabetes Congestive heart failure Peripheral vascular disease Preprocedure hemoglobin level, g/L GFR, mL/min/1.73 m2 Primary PCI for STEMI Bivalirudin Platelet GP IIb/IIIa inhibitor Ejection fraction, %
Before Matching
After Matching
No VCD Use
VCD Use
P Value
No VCD Use
VCD Use
P Value
64.27 0.33 89.43 0.87 0.85 0.13 0.23 0.15 0.14 136 78.89 0.12 0.35 0.42 52.18
63.87 0.33 89.86 0.84 0.85 0.13 0.24 0.16 0.15 135 78.97 0.10 0.30 0.44 51.99
0.00 0.88 0.00 0.00 0.01 0.06 0.14 0.00 0.00 0.00 0.75 0.00 0.00 0.00 0.03
64.18 0.35 89.00 0.86 0.85 0.13 0.23 0.16 0.17 135 78.41 0.10 0.32 0.42 52.16
64.34 0.35 88.81 0.86 0.85 0.13 0.23 0.16 0.17 135 78.14 0.10 0.32 0.42 52.02
0.15 1.00 0.19 0.95 0.71 0.83 0.95 0.67 1.00 0.28 0.36 0.46 1.00 0.40 0.18
GFR ⫽ glomerular filtration rate; GP ⫽ glycoprotein; PCI ⫽ percutaneous coronary intervention; STEMI ⫽ ST-segment elevation myocardial infarction; VCD ⫽ vascular closure device. * Distribution of selected baseline demographic variables, comorbid conditions, and preprocedural and procedural therapy in patients who were treated with VCDs compared with those treated with manual compression in unmatched and matched data.
no significant reduction in general hematomas (n ⫽ 572; OR, 0.82 [CI, 0.65 to 1.04]; P ⫽ 0.102). No difference in overall mortality was observed with VCDs (n ⫽ 112; OR, 0.82 [CI, 0.45 to 1.48]; P ⫽ 0.51), but the small number of events made this comparison imprecise. The findings were consistent across institutions for vascular complications and transfusion (data not shown). There was no association between VCD use and the risk for CIN (OR, 0.94 [CI, 0.86 to 1.02]; P ⫽ 0.14). Findings about vascular complications and transfusion were moderately robust to the presence of unmeasured confounding. When a hypothetical unmeasured confounder was posited to have a moderate effect size, the significance of the effect of VCD on vascular complications and transfusion remained unchanged in each scenario considered (the scenarios varied according to prevalence of the unmeasured confounder in patients with a VCD as opposed to patients without a VCD). Allowing a hypothetical omitted confounder to have a large effect size and to differ in its prevalence in the 2 comparison groups (VCD and no VCD) in a manner unlike most of the measured variables in the data led to a loss of significance of the effect of VCD on transfusion and, for a specific set of prevalence rates, vascular complications (Supplement 1). Further, the stability check of substituting the simplified, alternate match into the analysis left the estimation results largely unchanged with respect to directionality of treatment effect estimates, although differences in significance were noted (Appendix Table 3, available at www .annals.org). First, although no significant association for in-hospital death was observed in our results, with use of the alternate simplified match, this finding would nearly have reversed for in-hospital mortality (P ⫽ 0.056 based on the analysis using the alternate match). Second, CIN is www.annals.org
associated with nonuse of VCDs after either match but significantly so only under the simplified match (P ⫽ 0.14 vs. 0.005).
DISCUSSION The key finding of our study is that VCD use is associated with a significant reduction in vascular complications and need for transfusion in patients having transfemoral PCI. The benefits of VCDs were consistently observed in multiple subgroups except for patients with a lean or normal BMI or those who received platelet GP IIb/IIIa inhibitors, in whom the clinical efficacy of these devices was attenuated. The results of our study are important for many reasons. Vascular closure devices are better-tolerated and permit more rapid ambulation than manual closure. Although VCD use is not reimbursed by most third-party payers, their relatively low cost (approximate list price of $275 to $350 per device) combined with these factors probably account for the widely prevalent use of VCDs despite a relative paucity of high-quality data supporting their efficacy.
Table 2. Association of Vascular Closure Device Use With Key Clinical Outcomes Variable
Total Patients, n
Events, n
Odds Ratio (95% CI)
P Value
Vascular complications Transfusion needed In-hospital death
68 874 68 874 68 874
1314 1893 112
0.78 (0.67–0.90) 0.85 (0.74–0.96) 0.82 (0.45–1.48)
0.001 0.011 0.51
19 November 2013 Annals of Internal Medicine Volume 159 • Number 10 663
Downloaded From: http://annals.org/pdfaccess.ashx?url=/data/journals/aim/928838/ by a Fudan University User on 01/22/2017
Original Research
Comparative Safety of Vascular Closure Devices and Manual Closure
Figure. Propensity score–matched odds of vascular complications and transfusion in the overall cohort and subgroups of patients at varying risk for access site complications. Odds Ratio (95% CI) for Vascular Complications
P Value Events, n (for Interaction)
Women (n = 22 827)
775
Men (n = 46 047)
539
BMI
