Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Circ Cardiovasc Interv. Author manuscript; available in PMC 2017 April 1.
Published in final edited form as:
PMCID: PMC4832135

Effectiveness of Arterial Closure Devices for Preventing Complications With Percutaneous Coronary Intervention: An Instrumental Variable Analysis

Neil J. Wimmer, MD, MSc,^ Eric A Secemsky, MD, MSc,# Laura Mauri, MD, MSc,* Matthew T. Roe, MD, MHS, Paramita Saha-Chaudhuri, PhD, David Dai, PhD, James M. McCabe, MD,§ Frederic S. Resnic, MD, MSc,|| Hitinder S. Gurm, MBBS, and Robert W. Yeh, MD, MSc$



Bleeding is associated with poor outcomes after percutaneous coronary intervention (PCI). While arterial closure devices (ACDs) are widely used in clinical practice, whether they are effective in reducing bleeding complications during transfemoral PCI is uncertain. The objective of this study was to evaluate the effectiveness of ACDs for the prevention of vascular access site complications in patients undergoing transfemoral PCI using an instrumental variable approach.

Methods and Results

We performed a retrospective analysis of CathPCI Registry from 2009-2013 at 1,470 sites across United States. Variation in the proportion of ACDs used by each individual physician operator was used as an instrumental variable to address potential confounding. A two stage instrumental variable analysis was used as the primary approach. The main outcome measure was vascular access site complications, and non-access site bleeding was used as a “falsification endpoint” (negative control) to evaluate for potential confounding. A total of 1,053,155 ACDs were used during 2,056,585 PCIs during the study period. The vascular access site complication rate was 1.5%. In the instrumental variable analysis, the use of ACDs was associated with a 0.40% absolute risk reduction in vascular access site complications (95% confidence interval (95% CI):0.31%−0.42%, number needed to treat=250). Absolute differences in non-access site bleeding were negligible (risk difference 0.04%, 95% CI:0.01%−0.07%), suggesting acceptable control of confounding in the comparison.


ACDs are associated with a modest reduction in major bleeding after PCI. The number needed to treat with ACDs to prevent one major bleeding event is high.

Keywords: instrumental variable, closure device, percutaneous coronary intervention

Vascular complications related to arterial access are a common cause of morbidity, mortality, and cost in patients undergoing PCI.1, 2 Multiple strategies have been employed to decrease the risk of bleeding and vascular access site complications at the time of PCI.3 These strategies focus have focused on several areas in the care of patients at the time of PCI including the choice of arterial access site for the procedure, the choice of optimal pharmacologic therapy during PCI, and the decision to use ACDs for patients when a transfemoral route is chosen. While transradial PCI is likely the most effective mechanism for preventing access site complications,4 only a minority of patients in the US receive PCI via this route at present.5

The role of ACDs in patients having PCI is controversial. A large number of prior studies have evaluated the efficacy of ACDs in preventing complications in patients undergoing PCI. Some of the initial randomized trials were small, underpowered, and do not reflect current PCI practice. The largest meta-analysis on the subject raised concerns that ACDs may be associated with an increased risk in complications.6 The most recent randomized trial found non-inferiority between the use of ACDs and manual compression in patients undergoing diagnostic coronary angiography, some of whom also had PCI.7 Large observational studies have also evaluated the efficacy of ACDs,5, 8-13 and have generally found reductions in bleeding or vascular complications associated with their use. However, because the selection of ACD use at a patient level is often determined by factors such as the location and quality of the arterial puncture and the severity of atherosclerotic disease or calcification of the accessed femoral artery - characteristics that are rarely included in detailed clinical registries - there is reason to believe these studies may be confounded.

We therefore sought to determine whether ACDs are effective in preventing vascular access site complications with PCI using an instrumental variable approach, which may be less susceptible to confounding by unmeasured variables.14 Our hypothesis was that ACDs would not be associated with a clinically meaningful reduction in in-hospital vascular access complications after PCI.


Dataset and Patient Selection

The National Cardiovascular Data Registry’s CathPCI Registry, a partnership between the American College of Cardiology and the Society for Cardiovascular Angiography and Interventions, is a national quality improvement program that includes in-hospital data in patients undergoing cardiac catheterization and PCI procedures, and has been previously described.15, 16 The analyses here utilized version 4 of the CathPCI Registry, which contained data from 2,492,770 hospital admissions for PCI from July 1, 2009 until September 30, 2013. A full description of the data elements in the version 4 of the CathPCI Registry is available at Patients were excluded from the analysis if they underwent PCI via a non-femoral approach, if they underwent more than one PCI during the hospital admission, if the PCI procedure was classified as a salvage procedure, if the PCI was performed with an intra-aortic balloon pump or other form of mechanical circulatory support, or if the PCI was performed by a physician operator who had performed fewer than 25 PCIs in the CathPCI Registry during the study period. Waiver of written informed consent and authorization for this study was granted by Chesapeake Research Review Incorporated.

Study Outcomes

The primary outcome variable for the analysis was vascular access site complications which were defined as a composite of: bleeding at the access site with associated transfusion, hematoma > 5 cm at the access site, and retroperitoneal bleeding post-PCI. In addition, to ensure the absence of confounding in our main comparison, we pre-specified the use of a negative control, or falsification, hypothesis by assessing the effect of ACDs on non-access site bleeding.17 Non-access site bleeding was defined as gastrointestinal bleeding, genitourinary bleeding, cardiac tamponade, or hemorrhagic stroke post PCI. Such endpoints would not be expected to be influenced in a causal manner by the use of ACDs, such that any differences in these outcomes would likely be the result of residual confounding. We also analyzed secondary endpoints including the individual endpoints comprising the composite endpoint, NCDR major bleeding,18 in-hospital mortality, and post-procedure length of stay. NCDR major bleeding was defined as any of the following occurring before hospital discharge: arterial access site bleeding (defined as external bleeding at access site or a hematoma >10 cm for femoral access, >5 cm for brachial access, or >2 cm for radial access); retroperitoneal, gastrointestinal, or genitourinary bleeding; intracranial hemorrhage; cardiac tamponade; post-procedure hemoglobin decrease of 3 g/dl in patients with a pre-procedure hemoglobin level <=16 g/dl; or post-procedure non-bypass surgery–related blood transfusion for patients with a pre-procedure hemoglobin level >=8 g/dl.

Patient and Procedural Characteristics

We identified patient and procedural characteristics among patients treated with ACDs and those treated without ACDs at the time of the index PCI procedure. These variables included sociodemographic characteristics (age, sex, race, and region), medical history (diabetes mellitus, hypertension, hyperlipidemia, smoking status, previous PCI, previous MI, previous coronary artery bypass graft surgery, congestive heart failure, peripheral vascular disease, cerebrovascular disease, chronic lung disease, atrial fibrillation, history of cancer, history of gastrointestinal bleeding, chronic kidney disease, hemodialysis, history of lung disease), and presentation characteristics (STEMI versus NSTEMI versus other presentation, duration of acute coronary syndrome, presentation with shock). Peri-procedural anti-platelet use and anticoagulant use were also evaluated. Factors related to the physician operator, including whether they practiced at a teaching hospital or not and the annual operator volume of PCIs reported in the CathPCI dataset were also evaluated. These variables were used for statistical adjustment for all models subsequently described.

Statistical Analysis

We used an instrumental variable approach to attempt to account for selection bias and unmeasured confounding, as many PCI procedural variables and treatment decisions are influenced by characteristics that are poorly ascertained in observational analyses19. Valid instrumental variables, or instruments, are causally related to the exposure of interest but unrelated to the outcome of interest except through the pathway of the exposure itself. Thus, by exploiting the natural randomness in treatment assignment through an instrumental variable approach, we are able to reduce the influence of bias in the analysis. We used the proportion of ACD use of PCI-performing physicians as our instrumental variable. This approach, which takes advantage of physician practice variation, has been referred to as a “preference-based” instrumental variable, and has been previously described.20 Specifically, transfemoral PCIs performed by operators with less <20% ACD use (low) or >80% ACD use (high) were identified. We included only “low” and “high” ACD users in order to take advantage of operators that were very consistent in their choice to use or not use ACDs.

To perform the instrumental analysis, we utilized a standardized approach, the two-stage least squares linear regression method, which produce adjusted estimates of treatment effect on an absolute, rather than a relative scale21, 22. We pre-specified several analyses to confirm the adequacy of our selected instrumental variable prior to examining the primary outcomes. First, we examined the distribution in operator propensity to use ACDs, by constructing a histogram of the proportion of PCIs involving ACDs for each operator. After confirming adequate variation to support the approach, we then calculated standardized differences for a wide range of patient factors and procedural factors between those patients treated by low ACD use operators or high ACD use operators. Standardized differences were used to compare differences between groups instead of p-values as these measures are able to calculate the effect size between groups independent of sample size. Although standardized differences were initially developed in the context of comparing the mean of a continuous variable between two groups, this measure has also been used to compare binary and categorical variables.23 Standardized differences <10% were considered significant, as previously described.23

Two-stage least squares regression was then performed using the SAS procedure PROC SYSLIN24 to assess the adjusted absolute risk reduction in vascular access site complications with ACDs. This method relied on building two sequential linear regression models, with treatment of the dependent variable in the second-stage linear model as a binary variable22. In the first-stage model, we regressed the predicted use of an ACD on our instrument, the proportion of PCI cases with ACD use by the individual physician operator, adjusting for all covariates described above. In the second stage model, we regressed our outcome variable on the predicted value of receiving an ACD as estimated from the first stage model, in addition to adjusting for all of the covariates described above. The absolute differences between those treated with ACDs and those not treated with ACDs was estimated based on the coefficient of the instrumental variable in the second regression model. Robust standard errors (SEs) were estimated for all analyses and p-values <0.05 were considered significant. Analyses were performed using SAS v9.3 (Cary, NC).

Evaluation of the Instrumental Variable Assumptions

A valid instrument requires that several assumptions be justified.21 First, the instrument should strongly predict the exposure of interest. In this case, the assumption is that patients treated by high ACD use operators were more likely to receive an ACD during the index PCI. A second assumption is that the instrument affects the outcome only through its association with the primary predictor of interest, a fundamentally untestable assumption. A third assumption is that the instrumental variable should effectively “randomize” patients such that patients should be similar with respect to measured and unmeasured factors across levels of the instrument.


The patient population used for the analysis is shown in Figure 1. A total of 1,053,155 ACDs were used during 2,056,585 PCIs during the study period. Patients who did not receive an ACD had higher rates of peripheral arterial disease and were more likely to present with severer symptoms of heart failure. Otherwise, patients and procedural characteristics were overall similar between patients who received an ACD compared with those who did not (Table 1).

Figure 1
Flow diagram of procedures considered for the simple IV analysis and the full instrumental variable analysis.
Table 1
Baseline clinical and procedural characteristics by ACD use

Data from 1,121,714 patients with PCI performed by 4,331 physicians at 1,470 centers were analyzed when assessing those patients treated by high and low ACD-use operators. The distribution of the proportion of ACD use by physician operator is shown in Figure 2. There were 2,035 low ACD-use operators and 2,296 high ACD-use operators. These 4,331 operators represent 55.3% of the total number of physician operators in the study sample. The clinical and procedural characteristics of patients treated by high ACD-use operators and low ACD-use operators are shown in Tables 1 and and2.2. Among those treated by low ACD-use operators, 94.1% of patients received no ACD, 3.9% received a sealant device, 0.9% received a suture device, and 0.6% received a patch-based device. Among those patients treated by the high ACD-use operators, 9.6% of patients received no ACD, 60.5% of patients received a sealant device, 17.7% of patients received a suture-based device, and 5.8% of patients received a patch-based device.

Figure 2
Utilization of arterial closure devices by physician
Table 2
Baseline clinical and procedural characteristics by low vs high ACD operator

Vascular access site complications occurred in 1.5% of patients overall. CathPCI-defined major bleeding occurred in 4.6% of patients. Non-access site bleeding occurred in 0.4% of patients. Among those treated by high ACD use operators, vascular access site complications occurred in 1.6%, major bleeding occurred in 3.6%, and non-access site bleeding occurred in 0.4%. Among those treated by low ACD use operators vascular access site complications occurred in 1.3%, major bleeding occurred in 5.7%, and non-access site bleeding occurred in 0.3%.

Patient and procedural characteristics for PCIs performed by high versus low use ACD operators were similar. Standardized differences were less than <10% in 42 of 46 variables, and < 15% all 46 clinical, demographic, and procedural factors evaluated in the two populations. High ACD use physicians and low ACD use physicians had similar annual PCI volumes and were similarly likely to perform procedures in teaching hospitals (Table 3).

Table 3
Operator characteristics by performing physician procedure and practice characteristics.

In the instrumental variable analysis, physician operator was highly predictive of actual ACD use, as expected (Table 2). High ACD use operators used ACDs in 90.3% of PCIs while low ACD use operators used ACDs in 5.9% of PCIs. The use of ACDs was associated with a 0.36% absolute risk reduction in vascular access site complications (95% confidence interval (95% CI):0.31%−0.42%, number needed to treat=250) (Table 4). Absolute differences in non-access site bleeding were negligible (risk difference 0.04%, 95% CI: 0.01%−0.07%), suggesting acceptable control of confounding in the comparison. The use of ACDs was associated with a 0.73% absolute risk reduction in major bleeding (95% CI: 0.64%−0.82%, number needed to treat=137). ACD use was also associated with a small reduction in post-PCI length of stay (0.11 days, 95% CI: 0.10 days −0.12 days). ACD use was not associated with a reduction in post-PCI mortality (0.03%, 95% CI: −0.07%−0.04%).

Table 4
Results of the instrumental variable analysis.


We found that the use of ACDs is associated with a small, but significant reduction in vascular access site complications compared to manual compression in patients undergoing PCI. There was wide variation in the physician use of such devices, although this predilection did not appear to be related to measured patient or physician factors. These findings highlight the lack of strong evidence supporting the use of such devices, creating the preconditions for unexplained variation in care. Exploiting this variation through the instrumental variable approach allowed for the estimation of the impact of ACD use on important bleeding and vascular complications that was less susceptible to confounding by traditional regression approaches.

In this analysis, ACD use corresponded with an absolute 0.36% reduction in access-site bleeding. As the absolute difference in our falsification endpoint, non-access site bleeding, between treatment strategies was small and not clinically meaningful, we believe that the reduction in access-site bleeding in the primary analysis is most likely attributable to ACD use rather than residual confounding by unmeasured variables. While our findings suggest that ACDs are associated with some clinical benefit, we conclude that that benefit is relatively small in terms of preventing complications. We calculated that 250 patients need to be treated with ACDs in order to prevent one vascular access site complication. Given their relatively modest cost in the scheme of an overall PCI, patient and/or physician convenience may be enough to justify their use. However, we also acknowledge that given the low rates of bleeding and vascular complications, even interventions that are highly efficacious would still have a relatively large cost per complication prevented. A similar critique has been made of the recent ISAR-CLOSURE trial,7, 25 which compared ACDs against manual compression in patients undergoing coronary angiography without PCI.

Several prior authors in this field have called ACDs “bleeding avoidance strategies” that can be used systematically as a way to reduce bleeding or access site complications during transfemoral PCI.3, 5, 9, 26, 27 For example, Marso et al. found a 0.7% reduction in periprocedural bleeding with use of an ACD compared with manual compression following transfemoral PCI9. We found lower reductions in bleeding, perhaps due to differences in methodology (e.g. less susceptibility to confounding) and also differences in endpoint definitions and study time periods. Regardless, these data suggest that while ACDs can have a role in preventing complications, the role is likely modest when seen in the totality of the care of the patient with coronary artery disease undergoing PCI. Other interventions, including transradial PCI or choices in antithrombotic or antiplatelet agents, likely have a much larger impact in preventing bleeding or access site complications overall.

The analyses presented here are based on an instrumental variable approach to causal inference. Instrumental variable analyses exploit situations where some degree of randomness affects how patients are selected for a treatment.14 In this case, the variation in likelihood of using ACDs across physicians was used as the instrumental variable. The observation that patient characteristics among high versus low ACD use operators were very similar supports the validity of using this approach. We additionally employed a falsification endpoint in the analysis, namely non-access site bleeding, which would not be expected to be influenced by ACD use. We do note that the p-value for non-access site bleeding was nominally significant, secondary to the extremely large sample size and the inability of any observational study to completely eliminate potential confounding. However, we are reassured that the actual difference (0.04%) was unlikely to be clinically meaningful. In contrast, we have previously shown that the use of non-access site bleeding could effectively show that comparisons of radial and femoral access among PCI patients are susceptible to confounding17.

Beyond the validity of the chosen instrumental variable, our study is limited in several additional ways. Our study was not designed to detect differences in the effectiveness of individual ACDs, as the instrumental variable approach is not well suited to examine subgroups. While our results give an average overall effect of ACD use, the effectiveness of some ACDs may be different than others. We also did not take into account the learning curve with these devices.28 In addition, we are limited by the accuracy of the event ascertainment in our dataset. Endpoints in this dataset are not individually adjudicated and both under and over-reporting are possible. Finally, we are limited by the fact that the dataset does not capture all of the clinical factors that are used in routine practice to determine whether use of a particular ACD is appropriate, as we did not have access to femoral angiograms, or their interpretation, nor do we have access or to historical factors such as whether an ACD was used during a prior angiogram or PCI.


In summary, in this large, contemporary, real-world PCI registry, significant variation in use of ACDs existed among interventional cardiologists performing transfemoral PCIs. By exploiting this randomness of operator preference for ACD use during PCI through an instrumental variable approach that also employed a falsification endpoint, we demonstrated that use of these devices was associated with a modest but significant reduction in vascular access site complications compared with manual compression. A detailed cost-benefit analysis may be helpful in informing future treatment decisions.

What is Known

  • The effectiveness of arterial closure devices to reduce bleeding complications following transfemoral percutaneous coronary intervention is uncertain.
  • Randomized trials comparing arterial closure devices to manual compression in patients undergoing percutaneous coronary intervention have been inconclusive.

What the Study Adds

  • We used operator variation in preference of arterial closure devices after transfemoral percutaneous coronary intervention as an instrumental variable to assess the effectiveness of these devices at reducing arterial access site bleeding compared with manual compression.
  • We showed that use of arterial closure devices was associated with a modest but significant reduction in access-site bleeding.
  • This analysis highlights the use of an instrumental variable approach as a method to control for unmeasured confounding in non-randomized data.


Sources of Funding

This research was supported by the American College of Cardiology Foundation’s National Cardiovascular Data Registry and by the National Heart, Lung and Blood Institute Grant 1K23HL118138 (RWY).

Dr. Yeh serves on an advisory board for Abbott Vascular. He has received an educational honorarium from Gilead Sciences and has provided expert witness testimony for Merck. He also receives funding from the Hassenfeld Scholars Fund and National Heart, Blood and Lung Institute (1K23HL118138). Dr. Mauri receives grants to institution from Abbot, Boston Scientific, Eli Lilly, Daiichi Sankyo, Sanofi Aventis, and Bristol Myers Squibb. She provides consulting to St. Jude Medical, Biotronik, ReCor, DC Devices, Boeringher Ingelheim, and Eli Lilly. Dr. McCabe receives honoraria from Abbott Vascular. Dr. Resnic is a consultant for St. Jude Medical, Inc. Dr. Gurm received research funding from the National Institutes of Health and serves as a consultant for Osprey Medical, Inc. and Histiosonics, Inc. Dr. Roe has received research funding from Eli Lilly, Revalesio, Sanofi, American College of Cardiology, American Heart Association, and the Familial Hyperlipidemia Foundation; and has received consulting or honoraria from Eli Lilly, AstraZeneca, Sanofi, Janssen Pharmaceuticals, Merck, Regeneron, and Daiichi-Sankyo.


*This manuscript was presented as an oral presentation at the TCT 2014 national meeting in Washington, DC on September 14, 2014.


The other authors have nothing to disclose.


1. Manoukian SV, Feit F, Mehran R, Voeltz MD, Ebrahimi R, Hamon M, Dangas GD, Lincoff AM, White HD, Moses JW, King SB, 3rd, Ohman EM, Stone GW. Impact of major bleeding on 30-day mortality and clinical outcomes in patients with acute coronary syndromes: an analysis from the ACUITY Trial. J Am Coll Cardiol. 2007;49:1362–8. [PubMed]
2. Pinto DS, Stone GW, Shi C, Dunn ES, Reynolds MR, York M, Walczak J, Berezin RH, Mehran R, McLaurin BT, Cox DA, Ohman EM, Lincoff AM, Cohen DJ. Economic evaluation of bivalirudin with or without glycoprotein IIb/IIIa inhibition versus heparin with routine glycoprotein IIb/IIIa inhibition for early invasive management of acute coronary syndromes. J Am Coll Cardiol. 2008;52:1758–68. [PubMed]
3. Dauerman HL, Rao SV, Resnic FS, Applegate RJ. Bleeding avoidance strategies. Consensus and controversy. J Am Coll Cardiol. 2011;58:1–10. [PMC free article] [PubMed]
4. Jolly SS, Yusuf S, Cairns J, Niemela K, Xavier D, Widimsky P, Budaj A, Niemela M, Valentin V, Lewis BS, Avezum A, Steg PG, Rao SV, Gao P, Afzal R, Joyner CD, Chrolavicius S, Mehta SR. Radial versus femoral access for coronary angiography and intervention in patients with acute coronary syndromes (RIVAL): a randomised, parallel group, multicentre trial. Lancet. 2011;377:1409–20. [PubMed]
5. Daugherty SL, Thompson LE, Kim S, Rao SV, Subherwal S, Tsai TT, Messenger JC, Masoudi FA. Patterns of use and comparative effectiveness of bleeding avoidance strategies in men and women following percutaneous coronary interventions: an observational study from the National Cardiovascular Data Registry. J Am Coll Cardiol. 2013;61:2070–8. [PMC free article] [PubMed]
6. Koreny M, Riedmuller E, Nikfardjam M, Siostrzonek P, Mullner M. Arterial puncture closing devices compared with standard manual compression after cardiac catheterization: systematic review and meta-analysis. Jama. 2004;291:350–7. [PubMed]
7. Schulz-Schupke S, Helde S, Gewalt S, Ibrahim T, Linhardt M, Haas K, Hoppe K, Bottiger C, Groha P, Bradaric C, Schmidt R, Bott-Flugel L, Ott I, Goedel J, Byrne RA, Schneider S, Burgdorf C, Morath T, Kufner S, Joner M, Cassese S, Hoppmann P, Hengstenberg C, Pache J, Fusaro M, Massberg S, Mehilli J, Schunkert H, Laugwitz KL, Kastrati A. Comparison of vascular closure devices vs manual compression after femoral artery puncture: the ISAR-CLOSURE randomized clinical trial. Jama. 2014;312:1981–7. [PubMed]
8. Gurm HS, Hosman C, Share D, Moscucci M, Hansen BB. Comparative safety of vascular closure devices and manual closure among patients having percutaneous coronary intervention. Ann Intern Med. 2013;159:660–6. [PubMed]
9. Marso SP, Amin AP, House JA, Kennedy KF, Spertus JA, Rao SV, Cohen DJ, Messenger JC, Rumsfeld JS. Association between use of bleeding avoidance strategies and risk of periprocedural bleeding among patients undergoing percutaneous coronary intervention. Jama. 2010;303:2156–64. [PubMed]
10. Arora N, Matheny ME, Sepke C, Resnic FS. A propensity analysis of the risk of vascular complications after cardiac catheterization procedures with the use of vascular closure devices. Am Heart J. 2007;153:606–11. [PubMed]
11. Ahmed B, Piper WD, Malenka D, VerLee P, Robb J, Ryan T, Herne M, Phillips W, Dauerman HL. Significantly improved vascular complications among women undergoing percutaneous coronary intervention: a report from the Northern New England Percutaneous Coronary Intervention Registry. Circ Cardiovasc Interv. 2009;2:423–9. [PubMed]
12. Applegate RJ, Sacrinty MT, Kutcher MA, Kahl FR, Gandhi SK, Santos RM, Little WC. Trends in vascular complications after diagnostic cardiac catheterization and percutaneous coronary intervention via the femoral artery, 1998 to 2007. JACC Cardiovasc Interv. 2008;1:317–26. [PubMed]
13. Sanborn TA, Ebrahimi R, Manoukian SV, McLaurin BT, Cox DA, Feit F, Hamon M, Mehran R, Stone GW. Impact of femoral vascular closure devices and antithrombotic therapy on access site bleeding in acute coronary syndromes: The Acute Catheterization and Urgent Intervention Triage Strategy (ACUITY) trial. Circ Cardiovasc Interv. 2010;3:57–62. [PubMed]
14. Iwashyna TJ, Kennedy EH. Instrumental variable analyses. Exploiting natural randomness to understand causal mechanisms. Ann Am Thorac Soc. 2013;10:255–60. [PMC free article] [PubMed]
15. Brindis RG, Fitzgerald S, Anderson HV, Shaw RE, Weintraub WS, Williams JF. The American College of Cardiology-National Cardiovascular Data Registry (ACC-NCDR): building a national clinical data repository. J Am Coll Cardiol. 2001;37:2240–5. [PubMed]
16. Weintraub WS, McKay CR, Riner RN, Ellis SG, Frommer PL, Carmichael DB, Hammermeister KE, Effros MN, Bost JE, Bodycombe DP. The American College of Cardiology National Database: progress and challenges. American College of Cardiology Database Committee. J Am Coll Cardiol. 1997;29:459–65. [PubMed]
17. Wimmer NJ, Resnic FS, Mauri L, Matheny ME, Yeh RW. Comparison of transradial versus transfemoral percutaneous coronary intervention in routine practice: evidence for the importance of "falsification hypotheses" in observational studies of comparative effectiveness. J Am Coll Cardiol. 2013;62:2147–8. [PubMed]
18. Rao SV, McCoy LA, Spertus JA, Krone RJ, Singh M, Fitzgerald S, Peterson ED. An updated bleeding model to predict the risk of post-procedure bleeding among patients undergoing percutaneous coronary intervention: a report using an expanded bleeding definition from the National Cardiovascular Data Registry CathPCI Registry. JACC Cardiovasc Interv. 2013;6:897–904. [PubMed]
19. Imbens GW, Angrist JD. Identification and Estimation of Local Average Treatment Effects. Econometrica. 1994;62:467–475.
20. Rassen JA, Brookhart MA, Glynn RJ, Mittleman MA, Schneeweiss S. Instrumental variables II: instrumental variable application-in 25 variations, the physician prescribing preference generally was strong and reduced covariate imbalance. J Clin Epidemiol. 2009;62:1233–41. [PMC free article] [PubMed]
21. Brookhart MA, Rassen JA, Schneeweiss S. Instrumental variable methods in comparative safety and effectiveness research. Pharmacoepidemiol Drug Saf. 2010;19:537–54. [PMC free article] [PubMed]
22. Rassen JA, Schneeweiss S, Glynn RJ, Mittleman MA, Brookhart MA. Instrumental variable analysis for estimation of treatment effects with dichotomous outcomes. American journal of epidemiology. 2009;169:273–84. [PubMed]
23. Austin PC. Using the Standardized Difference to Compare the Prevalence of a Binary Variable Between Two Groups in Observational Research. Communications in Statistics - Simulation and Computation. 2009;38:1228–1234.
24. SAS/ETS(R) 9.2 User's Guide. Website: Accessed January 23, 2016.
25. Hoffer EK. Assessing the benefit of vascular closure devices after femoral artery puncture. Jama. 2015;313:855. [PubMed]
26. Subherwal S, Peterson ED, Dai D, Thomas L, Messenger JC, Xian Y, Brindis RG, Feldman DN, Senter S, Klein LW, Marso SP, Roe MT, Rao SV. Temporal trends in and factors associated with bleeding complications among patients undergoing percutaneous coronary intervention: a report from the National Cardiovascular Data CathPCI Registry. J Am Coll Cardiol. 2012;59:1861–9. [PMC free article] [PubMed]
27. Strauss CE, Porten BR, Chavez IJ, Garberich RF, Chambers JW, Baran KW, Poulose AK, Henry TD. Real-time decision support to guide percutaneous coronary intervention bleeding avoidance strategies effectively changes practice patterns. Circ Cardiovasc Qual Outcomes. 2014;7:960–7. [PubMed]
28. Resnic FS, Wang TY, Arora N, Vidi V, Dai D, Ou FS, Matheny ME. Quantifying the learning curve in the use of a novel vascular closure device: an analysis of the NCDR (National Cardiovascular Data Registry) CathPCI registry. JACC Cardiovasc Interv. 2012;5:82–9. [PubMed]