There is increasing interest in performing comparative effectiveness analyses in large observational databases, yet these analyses must adjust for treatment selection issues.
To conduct comparative safety and efficacy analyses of prasugrel vs clopidogrel bisulfate after percutaneous coronary intervention and to evaluate inverse probability of treatment weighting (a propensity score method) and instrumental variable methods.
Design, Setting, and Participants
This study used data from the Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study. Included in the study were patients undergoing percutaneous coronary intervention for myocardial infarction, 26.0% of whom received prasugrel. The study dates were April 4, 2010, to October 31, 2012.
Choice of initial antiplatelet agent (prasugrel or clopidogrel).
Main Outcomes and Measures
Safety and efficacy outcomes included 1-year composite major adverse cardiovascular events, moderate to severe bleeding, and stent thrombosis. Hospitalizations for pneumonia, bone fractures, and planned percutaneous coronary intervention were used as the falsification end points.
The study cohort comprised 11 784 participants (mean [SD] age, 60.0 [11.6] years, and 28.0% were female). Using inverse probability of treatment weighting adjustment, prasugrel and clopidogrel had similar major adverse cardiovascular events (hazard ratio [HR], 0.98; 95% CI, 0.83-1.16) and bleeding outcomes (1.18; 0.77-1.80), but prasugrel had a lower rate of stent thrombosis (0.51; 0.31-0.85). Using instrumental variable methods, prasugrel use was associated with a lower rate of the major adverse cardiovascular event end point (HR, 0.68; 95% CI, 0.47-1.00) but nonsignificant differences in the rates of bleeding (0.95; 0.41-2.08) and stent thrombosis (0.67; 0.16-2.00). There was no significant treatment difference noted in any of the falsification end-point rates when analyses were performed using inverse probability of treatment weighting, although the bone fracture end point approached statistical significance. Nevertheless, a lower rate of pneumonia-related hospitalizations was noted in the prasugrel-treated patients when analyses were performed using instrumental variable methods.
Conclusions and Relevance
Conclusions regarding the safety and efficacy of antiplatelet therapy varied depending on analytic technique, and none were concordant with the results from randomized trials. In addition, both statistical strategies demonstrated concerning associations when tested in the falsification analyses. A high level of scrutiny and careful attention to assumptions and validity are required when interpreting complex analyses of observational data.
Observational comparative effectiveness of therapeutics provides several advantages over randomized clinical trials, including real-world populations that may be excluded from trials, practice patterns that may differ from protocol-driven treatments used in trials and potentially lower costs, and avoidance of ethical issues regarding randomization.1,2 However, treatments are not assigned at random in observational data, so the potential for bias is a persistent methodological challenge due to factors that confound the relationship between treatment and outcome.3 Several analytic approaches have been developed to account for treatment selection bias in observational treatment comparisons.4 Despite the underlying differences in the statistical assumptions, cross-methodological comparisons of study results remain uncommon.
The Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study was a prospectively designed observational study investigating the comparative safety and efficacy of prasugrel and clopidogrel bisulfate, which are 2 adenosine diphosphate (ADP) receptor inhibitor treatments. For the primary prespecified outcomes analysis, the TRANSLATE-ACS study used an inverse probability of treatment weighting (IPTW) Cox proportional hazards regression model, which is a type of propensity score (PS) analysis.5
Instrumental variable (IV) methods provide an alternative through which treatments may be compared. With an IV approach, one can adjust for measured confounding factors as well as potentially remaining unmeasured differences in treatment choice. These methods rely on the identification of valid instruments, which are factors that affect the treatment the patient receives. Yet after adjustment for observed factors, these instruments do not independently affect patient outcomes except through their effect on treatment. The identification of valid IVs remains a challenge, but promising approaches have been reported, including the use of differential distance between facilities with different treatment capabilities,6,7 temporal variation in procedure use,8-10 and small area variation in practice patterns.11-14
The objectives of our analyses were 2-fold. First, we compared the safety and efficacy of prasugrel vs clopidogrel using 2 separate analytic approaches (IPTW and IV). Second, we used falsification end-point analyses to determine the potential for either approach to identify likely false associations.
Box Section Ref ID
Question Are associations between the use of prasugrel rather than clopidogrel bisulfate for patients with acute coronary syndrome robust to analytic method?
Findings In this secondary analysis of data from the Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study, prasugrel use was associated with a reduction in major adverse cardiovascular events when analyzed by an instrumental variable approach but not by propensity score analysis via inverse probability of treatment weighting. However, the findings from falsification testing raise concerns about the validity of these results.
Meaning A high level of scrutiny and careful attention to assumptions and validity are required when interpreting complex analyses of observational studies.
Authorization for this study was provided by the Duke University Health System Institutional Review Board. A waiver of informed consent was obtained.
TRANSLATE-ACS Study Design
The TRANSLATE-ACS study design has been previously described in detail.5 In summary, the TRANSLATE-ACS study was a prospective cohort investigation enrolling patients with myocardial infarction who underwent percutaneous coronary intervention (PCI) and were treated with an ADP receptor inhibitor during the index hospitalization. The study enrolled 12 365 patients between April 4, 2010, and October 31, 2012. Of these patients, 11 969 received prasugrel or clopidogrel as their initial ADP receptor inhibitor and were included in the primary analysis. Follow-up was conducted by telephone interview at 6 weeks, 6 months, 12 months, and 15 months after the date of the index PCI procedure.
We evaluated prespecified TRANSLATE-ACS safety and efficacy end points, all of which were measured through 365 days after the patient’s index PCI. The prespecified primary analysis used an as-treated approach. Events occurring more than 7 days after the medication discontinuation or switch were censored in patients who discontinued or switched to a different ADP receptor inhibitor. The primary efficacy end point was a major adverse cardiovascular event (MACE), including all-cause death, myocardial infarction, stroke, or unplanned coronary revascularization. The secondary efficacy end point was stent thrombosis, judged to be definite using Academic Research Consortium criteria.15 The primary safety end point was moderate to severe bleeding events, defined using the Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries (GUSTO) criteria.16 All 3 of these end points were independently validated and adjudicated by study physicians.
In addition to the 3 safety and efficacy end points used in the primary outcomes analysis, 3 additional (not prespecified) falsification end points were constructed for this analysis. Falsification end points are end points for which the treatment is a priori believed to be unlikely to have an effect.17 If the treatment does have an effect on these outcomes, then this finding raises concern that there are confounding factors present, analogous to negative controls in laboratory science. The outcome selected for falsification must also be strongly correlated with potential confounding factors of concern.18 In the TRANSLATE-ACS study, 2 confounding factors that are potentially incompletely measured in the data are frailty and coronary disease severity. To evaluate frailty, we evaluated admissions in whichever of the first 2 diagnosis codes indicated pneumonia or orthopedic fracture, as specified by Clinical Classifications Software diagnosis categories 8.1.1 and 16.2, respectively.19 Coronary disease severity was assessed using a planned coronary revascularization end point as part of the TRANSLATE-ACS study follow-up process.
In this study, we considered the following 2 instruments: (1) the facility variation in prasugrel use11-14 and (2) the variation in prasugrel use occurring over time.8-10 The facility variation was operationalized by using the treatment choices of all patients treated at the facility, excluding the patient as the instrument (ie, a facility-level rate). Temporal variation was operationalized as the month and year of the patient’s index PCI procedure. We included both the facility variation and temporal instruments in the primary analyses. For the sensitivity analyses, we also performed IV analysis using the instruments separately and combined by calculating a patient’s IV value as the rolling average of the last 5 patients’ treatment choices (excluding the first 5 patients treated at each facility). Because the facility variation may be correlated with facility characteristics, we augmented covariates used in the original TRANSLATE-ACS study models by adding facility characteristics from the American Hospital Association Annual Survey Database.
For modeling outcome, an important analytic consideration is the high rate of discontinuation and switching of antiplatelet therapy observed in the TRANSLATE-ACS study. For this reason, we used a Cox proportional hazards regression model in which patients are censored at loss to follow-up, 12 months of follow-up, or 7 days after discontinuation or switching of therapy, which mirrors the primary analysis used in the TRANSLATE-ACS study. The IV component was implemented using 2-stage residual inclusion estimation, in which the first stage was a logistic regression predicting receipt of treatment (in this case, the use of prasugrel) as a function of instruments and covariates, and the second stage was the Cox proportional hazards regression model predicting outcome as a function of treatment and covariates (excluding the instruments), with addition of the residual term from the first-stage equation.20 The facility variation instrument introduces potential correlation among the observations. Consequently, when the facility variation instrument was used, standard errors were calculated by bootstrapping (1000 iterations), with observations clustered by facility. Percentile-based 95% CIs were reported.21
Inverse Probability of Treatment Weighting
Inverse probability of treatment weighting is a form of PS analysis in which probability weights are used to reduce imbalance in potential confounding factors between treated and control patients.22 As in the primary TRANSLATE-ACS analyses, we constructed a PS using logistic regression, and this model was refit in the study subset. We added the additional facility factors used in the IV analysis to the prespecified covariates included in the primary outcomes analysis so that the IPTW model and IV results would be comparable. Consistent with the primary study analysis, we then calculated stabilized average treatment effect weights, capped at 10 times the average weight value. Imbalance of covariates before and after IPTW adjustment was assessed using standardized differences,23 treating a difference of more than 10% as concerning.24 The weights were then used to analyze outcomes using cause-specific Cox proportional hazards regression models to account for the time-to-event nature of the data.
Wilcoxon rank sum, χ2, and log-rank tests were used to evaluate the statistical significance of unadjusted continuous, categorical, and time-to-event data, respectively. A 2-sided α = .05 was considered statistically significant for all analyses. Statistical analyses were performed using software (SAS, version 9.4; SAS Institute Inc and Stata Statistical Software, version 14.0; StataCorp LP).
The TRANSLATE-ACS study included 11 969 patients with ST-segment elevation myocardial infarction (STEMI) and non-STEMI who underwent PCI at 233 US hospitals between April 4, 2010, and October 31, 2012. Of these patients, 11 784 (98.5%) comprised the cohort for this study (excluding those facilities that enrolled <10 patients to ensure accurate measurement of the facility variation instrument). Their mean (SD) age was 60.0 (11.6) years. The included patients were similar to those excluded (eTable 1 in the Supplement). Among the 196 facilities that were included in the study, the median prasugrel use rate was 27.3% (25th to 75th percentiles, 12.7%-45.6%; range, 0.0%-85.7%) (Figure 1A). After adjustment for patient and facility characteristics, facility-level variation in the prasugrel use rate remained significant (estimated variance on constant, 1.17; 95% CI, 0.90-1.51) (Figure 1B). There was also significant temporal variation in prasugrel use (Figure 1C), which remained statistically significant after adjustment (P < .001) (Figure 1D).
Patient Demographic and Clinical Characteristics and Facility Characteristics
In general, prasugrel recipients had characteristics consistent with a lower risk of events than clopidogrel recipients. Prasugrel recipients were younger, had fewer diseased vessels, and had lower rates of prior myocardial infarction, revascularization, cerebrovascular disease, and most comorbid conditions (Table 1 and Figure 2). Regardless of how the IV approach was formulated (facility variation or calendar time), patient characteristics were better balanced across the values of the IVs than they were when comparing prasugrel recipients vs clopidogrel recipients. A notable exception was that, based on the facility variation instrument, patients were less likely to have been transferred for PCI at facilities with greater than the median rate of prasugrel use than they were at facilities with less than the median rate of prasugrel use. Nonetheless, there were distinct differences in facility characteristics across the values of the IVs: higher prasugrel use was associated with facilities that were less likely to have teaching status, less likely to be a member of a hospital system, less likely to have a cardiac intensive care unit, and more likely to have smaller bed size and fewer total admissions.
Unadjusted MACE and moderate to severe bleeding were both higher for clopidogrel vs prasugrel (17.2% vs 13.1%, P < .001 and 3.7% vs 2.6%, P = .01, respectively). Stent thrombosis rates did not differ by a statistically significant margin (1.2% vs 1.0%, P = .16). Among the falsification end points, the rates of pneumonia hospitalization were more than double for clopidogrel vs prasugrel (1.9% vs 0.9%, P = .001), but the rates of fracture hospitalization did not differ (0.8% vs 0.6% in both arms, P = .41). The rates of planned revascularization were lower for clopidogrel than for prasugrel (3.3% vs 4.5%, P = .005).
Inverse Probability of Treatment Weighting
Propensity scores ranged from 0.2% to 82.0% among clopidogrel recipients and from 0.9% to 81.6% among prasugrel recipients, with 99.4% of the sample in the region of common support (eFigure 1 in the Supplement). After adjustment using weights, the mean imbalance in standardized differences was reduced from 13.8% to 2.2%, and the maximum standardized difference was reduced from 46.6% to 6.0% (eFigure 2 in the Supplement).
Unadjusted and Adjusted Safety and Efficacy End Points
In unadjusted analyses, receipt of prasugrel was associated with lower rates of the composite MACE end point (hazard ratio [HR], 0.76; 95% CI, 0.67-0.85), bleeding (HR, 0.71; 95% CI, 0.55-0.93), and stent thrombosis (HR, 0.73; 95% CI, 0.47-1.14) (Table 2). After adjustment with IPTW, prasugrel use was associated with a nonsignificant difference in MACE (HR, 0.98; 95% CI, 0.83-1.16) and bleeding outcomes (HR, 1.18; 95% CI, 0.77-1.80) but a significant difference in the rate of stent thrombosis (HR, 0.51; 95% CI, 0.31-0.85). When using both facility variation and calendar time as IVs, prasugrel use was associated with a statistically significant reduction in MACE (HR, 0.68; 95% CI, 0.47-1.00) but nonsignificant differences in the rates of bleeding (HR, 0.95; 95% CI, 0.41-2.08) and stent thrombosis (HR, 0.67; 95% CI, 0.16-2.00). The results were consistent with the combined IV findings when using the IVs independently and parameterized instead of using lagged treatment (eTable 2 in the Supplement).
Unadjusted and Adjusted Falsification End Points
In unadjusted analyses, receipt of prasugrel was associated with lower rates of pneumonia (HR, 0.48; 95% CI, 0.30-0.76), orthopedic fracture (HR, 0.78; 95% CI, 0.44-1.40), and the composite of pneumonia or fracture (HR, 0.58; 95% CI, 0.39-0.80), as well as higher rates of planned revascularization (HR, 1.37; 95% CI, 1.09-1.70) (Table 3). After adjustment with IPTW, none of the differences in outcomes were statistically significant, although fracture approached significance (HR, 2.33; 95% CI, 0.99-5.33). In contrast, using IV methods, prasugrel use was associated with a significantly lower rate of pneumonia (HR, 0.22; 95% CI. 0.05-0.73) and the composite of pneumonia or fracture end point (HR, 0.26; 95% CI, 0.09-0.76).
Observational study validity is always threatened by the potential bias induced by treatment selection. Therefore, a variety of analytic techniques have been developed to adjust for confounding. We compare the results using 2 common analytic techniques, namely, IPTW and IV analyses (by means of several instruments). We found that the comparative safety and efficacy conclusions varied depending on whether an IPTW or IV approach was used. Also, the IV method found an unexpected and likely false association between prasugrel use and a significant reduction in hospitalizations for pneumonia with IVs, while IPTW methods also suggested a near-significant reduction for orthopedic fractures. Neither statistical method produced data that were concordant with the results from randomized clinical trials.
Several assumptions underlie both the IPTW and IV methods.25,26 First, both methods rely on the stable unit value treatment assumption, which is that (1) the treatment choices of one patient do not affect the effectiveness of treatment in other patients and (2) the nature of the treatment is the same for all patients. Given that the treatment is a pharmaceutical, the first component is straightforward, but the second component is more complicated because antiplatelet therapy is a longitudinal treatment, and differences in adherence may affect the results of either analysis. Second, there are 2 IV-specific assumptions, including (1) the IVs must strongly affect the patient’s treatment choices, which is a testable assumption met in this study, and (2) the IVs must affect all individuals in the same direction, also known as the monotonicity assumption. While untestable, the monotonicity assumption appears intuitively reasonable in our study: both greater facility use of prasugrel and a temporal trend toward increasing prasugrel use are unlikely to result in other patients receiving less prasugrel. Third, both methods rely on the ignorability assumption. For PS analyses, it is important that, after adjustment for observable confounders via the PS analysis, the relationship between treatment and outcome is not confounded. For IV analyses, the ignorability assumption has the following 2 components: (1) the relationship between the IVs and treatment is not confounded by unmeasured factors and (2) the relationship between the IVs and outcome is not confounded by unmeasured factors (often called the exclusion restriction). Given the findings of the falsification tests, the ignorability assumption may be faulty, particularly with respect to the facility variation instrument, because it appears likely that patients treated at facilities that use higher rates of prasugrel systematically differ from patients treated at facilities that use lower rates of prasugrel or that other differences in the processes of care associated with these facilities affect patient outcomes.27
The TRANSLATE-ACS study is part of a new generation of observational comparative effectiveness research studies and one of the first to rely on large-scale prospective data collection to address potential confounding factors between treatment and outcome. Because the study was prospectively designed with the intention of performing PS, it is perhaps not surprising that the IPTW models performed better than the IV models with respect to the falsification end points. Given current interest in performing IV methods as a part of observational comparative effectiveness research and often in concert with IPTW methods, it will improve the validity of these approaches to prespecify both analytic methods and to collect the data needed to remove potential confounding factors from both methods.
Our study has several limitations. First, the falsification end points were not prespecified in the TRANSLATE-ACS study, and the pneumonia and fracture hospitalization end points were obtained from diagnosis codes rather than from clinically adjudicated end points. Nevertheless, such end-point misclassification would likely similarly affect both treatments and both analytic perspectives. Second, the falsification end points are only indirect tests of validity. While helpful guides, the assumptions underlying both PS and IV methods are not directly testable. The IV and IPTW results apply to different elements of the overall study population. The IPTW results are intended to serve as an average effect of prasugrel treatment in the overall study population, whereas the IV results apply to those marginal patients whose treatment choice was affected by the IVs. Third, this comparison was performed in the context of a single (albeit large, prospectively designed) study, so the results may not be broadly generalizable. Our comparison focuses on IV as implemented by 2-stage residual inclusion and PS as implemented by IPTW; therefore, our study is not intended to be an exhaustive comparison of the various implementations of both methods.
Of note, the results of the IPTW and IV methods differ from each other and from the randomized clinical trials that led to the approval of prasugrel.28 In particular, the Trial to Assess Improvement in Therapeutic Outcomes by Optimizing Platelet Inhibition With Prasugrel–Thrombolysis in Myocardial Infarction 38 (TRITON-TIMI 38) demonstrated a decrease in the MACE end point but an increase in major bleeding.28 In this study, we found no difference in the MACE end point when the results were analyzed using the IPTW method, while with the IV method, there was no increased risk of bleeding. These differences may represent persistent bias in the comparison of the prasugrel and clopidogrel populations that both methods failed to completely control. Alternatively, it may also represent a difference in the performance of prasugrel when evaluated outside of a randomized clinical trial in real-world populations, and in this sense, randomized trial data are an imperfect criterion standard for comparison of the results from observational studies.
In our analysis of patients receiving prasugrel vs clopidogrel for the treatment of acute coronary syndrome, we found a MACE end-point reduction by IV analysis that had not been observed in the main IPTW analysis; however, falsification test results raise concerns regarding the validity of this finding. Observational studies in which both the IPTW and IV methods are contemplated can be prospectively designed to minimize validity threats to both methods by identifying and collecting potential confounding factors so that they may be adjusted for in comparative analyses. However, particularly in the case of observational comparative effectiveness research done in the absence of randomized clinical trial findings, ensuring unbiased comparisons of treatments remains a methodological challenge.
Accepted for Publication: May 8, 2016.
Corresponding Author: Tracy Y. Wang, MD, MHS, MSc, Duke Clinical Research Institute, Duke University Medical Center, 2400 Pratt St, Durham, NC 27705 (email@example.com).
Published Online: July 20, 2016. doi:10.1001/jamacardio.2016.1783
Author Contributions: Dr Federspiel had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Federspiel, Xian, Effron, Faries, Zettler, Mauri, Yeh, Peterson, Wang.
Acquisition, analysis, or interpretation of data: Federspiel, Anstrom, Xian, McCoy, Effron, Zettler, Peterson, Wang.
Drafting of the manuscript: Federspiel, McCoy, Zettler.
Critical revision of the manuscript for important intellectual content: Anstrom, Xian, Effron, Faries, Zettler, Mauri, Yeh, Peterson, Wang.
Statistical analysis: Federspiel, Anstrom, McCoy, Faries, Mauri, Peterson.
Obtained funding: Wang.
Administrative, technical, or material support: McCoy, Effron.
Study supervision: Effron, Peterson, Wang.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Anstrom reported receiving research support from AstraZeneca, Bristol-Myers Squibb, Lilly USA, LLC, Boehringer Ingelheim, the Pulmonary Fibrosis Foundation, and Medtronic; reported serving as a consultant for Abbott Vascular, AstraZeneca, Bristol-Myers Squibb, Gilead, Pfizer, and GlaxoSmithKline; reported serving on data monitoring committees for the National Institutes of Health, Forest, Pfizer, and GlaxoSmithKline; and reported having an equity interest in Biscardia. Dr Effron reported full-time employment and shareholding with Lilly USA, LLC. Dr Faries reported full-time employment and minor shareholding with Lilly USA, LLC. Dr Zettler reported shareholding with Lilly USA, LLC. Dr Peterson reported receiving institutional grant support from the American College of Cardiology, American Heart Association, Lilly USA, LLC, and Janssen and reported receiving consulting fees (including continuing medical education) from Merck & Co, Boehringer Ingelheim, Genentech, Janssen, and Sanofi. Dr Wang reported receiving research grant support from Lilly USA, LLC, Daiichi Sankyo Inc, AstraZeneca, Bristol-Myers Squibb, Boston Scientific, Gilead, GlaxoSmithKline, and Regeneron and reported providing consulting services for Lilly USA, LLC, AstraZeneca, and Premier. No other disclosures were reported.
Funding/Support: The Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study was funded by Daiichi Sankyo Inc and Lilly USA, LLC.
Role of the Funder/Sponsor: The sponsors were involved in the original design of the Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study and had the opportunity to review the final manuscript. The study authors conducted the study, including data collection, management, and analysis; interpretation of the data; preparation and revision of the manuscript; and decision to submit the manuscript for publication.
Group Information: The Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) investigators were A. Nasser Adjei, MD (Sparks Regional Medical Center), Agha Ahmed, MD (Galichia Heart Hospital, LLC), Bina Ahmed, MD (University of New Mexico Hospitals), Chowdhury Ahsan, MD (University Medical Center of Southern Nevada), Christopher Allen, MD (University of Pittsburgh Health System), Nishith Amin, MD (Arnot Ogden Medical Center), H. Vermon Anderson, MD (Memorial Hermann Hospital/University of Texas Health Science Center at Houston), Dominick Angiolillo, MD (Shands Jacksonville Medical Center/University of Florida), Mehrdad Ariani, MD (Northridge Hospital Medical Center), Imran Arif, MD (University of Cincinnati), Ghulam Awan, MD (University of South Alabama), Michael Azrin, MD (University of Connecticut Health Center), Richard Bach, MD (Washington University School of Medicine/Barnes-Jewish Hospital), Subhash Banerjee, MD (VA North Texas Health Care System), Kenneth Baran, MD (United Hospital), Stephen Battista, MD (Fairview Southdale Hospital), Hans Bauer, MD (Our Lady of Lourdes Medical Center), Anthony Bavry, MD (Malcom Randall VA Medical Center/North Florida/South Georgia Veterans Health System), Peter Berger, MD (Geisinger Wyoming Valley Medical Center), Charles Bethea, MD (INTEGRIS Baptist Medical Center, Inc), Jessica Birchem, MD (St John’s Hospital/St John’s Medical Research Institute, Inc), Richard Blankenbaker Jr, MD (Martin Memorial Medical Center), Paul Boffetti, MD (Lahey Cardiology/Southern New Hampshire Medical Center), Sorin Brener, MD (New York Methodist Hospital), David Brill, MD (Washington Adventist Hospital), Britta Brott, MD (University of Alabama at Birmingham Hospital), John Canto, MD (Lakeland Region Medical Center/Watson Clinic Center for Research, Inc), Charles Carey, MD (St Anthony’s Medical Center), Joel Carver, MD (Washington Regional Medical Center), Juan Chahin, MD (Westmoreland Regional Hospital), Jeffrey Chambers, MD (Metropolitan Cardiology Consultants at Mercy Hospital), Kollagunta Chandrasekhar, MD (Winter Haven Hospital), David Chang, MD (PinnacleHealth at Harrisburg Hospital), Michael Chang, MD (Mercy General Hospital), Adnan Chhatriwalla, MD (St Luke’s Hospital/Mid America Heart Institute), Pedro J. Colón Hernandez, MD (Transcatheter Medical Inc), Jeffrey Cook, MD (OSF Saint Anthony Medical Center/Rockford Cardiovascular Associates Research Foundation), Frank Corbally, MD (Lancaster General Hospital/Heart Specialists of Lancaster), William R. Craig, MD (St John’s Regional Medical Center/Mercy Health of Joplin), Linda Cuomo, MD (Westchester Medical Center/New York Medical College), Shukri David, MD (Providence Hospital), Michael Del Core, MD (The Cardiac Center of Creighton University Medical Center), Vishva Dev, MD (Los Robles Hospital & Medical Center), Lingareddy Devireddy, MD (St John Macomb Hospital), David Dobies, MD (Genesys Regional Medical Center), Sandhya Donepudi, MD (St Margaret Mercy Healthcare Centers), Joseph Doucette, MD (Overlake Hospital Medical Center), Anthony D’Souza, MD (Heart Specialists, PC of Southern Connecticut), Christopher Dyke, MD (Alaska Heart Institute), Melvin Echols, MD (Southeastern Regional Heart Center), Mehiar El Hamdani, MD (St Mary’s Medical Center), Bernard Erickson, MD (Saint Cloud Hospital/CentraCare Heart & Vascular Center), John Erwin III, MD (Scott and White Hospital and Clinic), Steven Falen, MD (Sutter Roseville Medical Center), Bobbie Farber, MD (St Francis Hospital), Robert Feldman, MD (MediQuest Research Group Inc at Munroe Regional Medical Center), Ronald Fields, MD (St Mary Medical Center), Robert Fishberg, MD (Overlook Hospital), Daniel Fisher, MD (Umass Memorial Healthcare), Jean Foucauld, MD (Cardiology Partners Clinical Research Institute, LLC), Anthony Frey, MD (Atlantic Cardiology Associates/Peninsula Regional Medical Center), George Gabriel, MD (Allegheny General Hospital/Allegheny Singer Research Institute), Neal Gaither, MD (Winchester Medical Center), Kirk Garratt, MD (Lenox Hill Hospital), Matthew Gibb, MD (Carle Physicians Group/Carle Foundation Hospital), Lee Giorgi, MD (St Luke’s Hospital Association of Duluth), Jeff Gladden, MD (Wise Regional Health System), Tyler Gluckman, MD (Providence St Vincent Medical Center), Rafael Gonzalez, MD (Scott and White Healthcare–Round Rock), Evelyne Goudreau, MD (Virginia Commonwealth University Medical Center), Bruce Graham, MD (Ball Memorial Hospital), Robert Greene, MD (Alta Bates Summit Medical Center), John Griffin, MD (Sentara Virginia Beach General Hospital/Cardiovascular Associates, Ltd), Paul Michael Grossman, MD (University of Michigan Healthcare System), Luis Gruberg, MD (Stony Brook University Medical Center), Darrel Gumm, MD (HeartCare Midwest/OSF Saint Francis Medical Center), Anjan Gupta, MD (Aurora St Luke’s Medical Center), Paul Gurbel, MD (Sinai Hospital of Baltimore/Platelet Thrombosis Research), Jason Hall, MD (Medicor Cardiology), Franklin Handel, MD (Kettering Medical Center/Kettering Health Network), Steven Hearne, MD (Delmarva Heart Research Foundation, Inc/Peninsula Regional Medical Center), Timothy Henry, MD (Minneapolis Heart Institute/Abbott Northwestern Hospital), Stuart Higano, MD (Missouri Baptist Medical Center), Anthony Hilliard, MD (Loma Linda University Medical Center), David Hoffman, MD (St Elizabeth Hospital Medical Center/Humility of Mary Health Partners), Paula Hollingsworth, MD (Central Baptist Hospital), Naseem Jaffrani, MD (Rapides Regional Medical Center/Alexandria Cardiology Clinic/Cambridge Medical Trials), Richard Jantz, MD (Parker Adventist Hospital), Werner Jauch, MD (Pasco Cardiology Center), Alonzo Jones, MD (Columbus Cardiology Associates), Wesley Kai, MD (Hawaii Pacific Health Research Institute/Straub Hospital), Brian Kaminsky, MD (Bon Secours Virginia Health System), Steven Karas, MD (Holmes Regional Medical Center), Barbara Karenko, MD (Metro Health Hospital), Jason Katz, MD (University of North Carolina at Chapel Hill), Kenneth Kent, MD (Western Maryland Health System Regional Medical Center), Michael Kesselbrenner, MD (The Valley Hospital), Suhail Khadra, MD (John H. Stroger Jr Hospital), Akshay Khandelwal, MD (Henry Ford Hospital/Henry Ford Heart & Vascular Institute), Dennis Killian, MD (Provena Saint Joseph Medical Center), Sung Sup Kim, MD (Swedish Covenant Hospital), Joseph Klag, MD (Scottsdale Healthcare Osborn and Scottsdale Healthcare Shea), Bruce Klugherz, MD (Abington Memorial Hospital/Abington Medical Specialists), Mark Koenig, MD (Saint Thomas Hospital), Devendra Koganti, MD (Southern Heart Research Insititute, LLC/Southern Regional Medical Center), Richard Konstance, MD (Carilion Roanoke Memorial Hospital), Mark Kozak, MD (Penn State University College of Medicine Milton S. Hershey Medical Center), Phillip Kraft, MD (Beaumont-Troy/William Beaumont Hospital), George Kramer, MD (WellStar Health System/WellStar Kennestone Hospital/WellStar Cobb Hospital), Rolf Kreutz, MD (Indiana University Health), Prasanna Kumar, MD (Memorial Hospital of Carbondale/Prairie Cardiovascular Consultants), Michael Kutcher, MD (Wake Forest University Baptist Medical Center/North Carolina Baptist Hospital), Gervasio Lamas, MD (Mount Sinai Medical Center), Charles Lambert, MD (University Community Hospital, Inc d/b/a Florida Hospital Tampa), Faisal Latif, MD (VA Medical Center, Oklahoma City), Joseph Lawton III, MD (Columbia Cardiology), Daniel Lee, MD (Bay Regional Medical Center), Kwan Lee, MD (University Physicians Healthcare Hospital/University of Arizona), Thomas LeGalley, MD (Marquette General Hospital/Upper Michigan Cardiovascular Associates, PC), Bernard Levi, MD (Menorah Medical Center/Midwest Cardiology Associates), Stephen Lewis, MD (Bethesda North Hospital/Hatton Institute for Research and Education), Mark Lurie, MD (Torrance Memorial Medical Center), Hugh MacIsaac, MD (St Elizabeth Medical Center), William Mackin, MD (Delnor Community Hospital), Calin Maniu, MD (Bon Secours Virginia Health System), Stephen Marshalko, MD (Bridgeport Hospital), Robert Marshall, MD (Meritus Medical Center), Scott Martin, MD (Covenant Medical Center), Marco Mazzella, MD (Kansas City Heart Foundation/St Joseph Medical Center), Raymond McKay, MD (Hartford Hospital), Ronald McKechnie, MD (Cardiovascular Associates Ltd/Chesapeake Regional Medical Center), Bruce McLellan, MD (St Charles Health System), John Messenger, MD (University of Colorado Denver), Paul Micale, MD (Riverside Hospital, Inc d/b/a Riverside Regional Medical Center), Jay Midwall, MD (Palm Beach Heart Research Institute, LLC at JFK Medical Center, Indian River Medical Center), Gary Miller, MD (Danville Regional Medical Center/Cardiology Consultants of Danville, Inc), Michael Montgomery, MD (North Kansas City Hospital), Mehesh Mulumudi, MD (Providence Regional Medical Center), Venkatesh Nadar, MD (Heritage Cardiology Associates/Holy Spirit Health System), Arthur Nazarian, MD (St Alexius Medical Center), Georges Nseir, MD (Chandler Regional Medical Center), Georges Nseir, MD (Mercy Gilbert Medical Center), Brian O’Murchu, MD (Temple University Hospital), John Ord, MD (Aurora Denver Cardiology Associates, PC), Martin O’Riordan, MD (Mercy Fitzgerald Hospital), Thomas Palmer, MD (Waukesha Memorial Hospital/ProHealth Care, Oconomowoc Memorial Hospital/ProHealth Care), John Paulowski, MD (Aultman Health Foundation), William Phillips, MD (Central Maine Medical Center), Richard Pish, MD (Pish Medical Associates), Thomas Pow, MD (Lakeland Hospital/Great Lakes Heart and Vascular Institute), Eric Powers, MD (Medical University of South Carolina), Antonis Pratsos, MD (Bryn Mawr Hospital), Peter Puleo, MD (St Luke’s Hospital and Health Network), Joshua Purow, MD (Jim Moran Heart and Vascular Research Institute, Holy Cross Hospital Inc), Edward Rachofsky, MD (Cardiology Associates of Somerset County, PA), Michael Ragosta, MD (University of Virginia Health System), Ganesh Raveendran, MD (University of Minnesota), Arthur Riba, MD (Oakwood Hospital & Medical Center, Dearborn), Wilfredo Rivera, MD (McLaren Regional Medical Center–Research Institute), David Roberts, MD (Sutter Heart and Vascular Institute), Roy Robertson, MD (Parkview Hospital, Inc), Scott Robertson, MD (Sentara Norfolk General Hospital), Arsenio Rodriguez, MD (Florida Hospital), Rolando Rodriguez, MD (Bay Area Cardiology Associates, PA), Michael Romanelli, MD (St John Hospital & Medical Center), Chanwit Roongsritong, MD (Renown Regional Medical Center), Eli Rosenthal, MD (Legacy Health), David Roth, MD (Kalispell Regional Medical Center, Inc d/b/a Glacier View Cardiology), Ahmed Sabe, MD (Mercy Medical Center), David Scherer, MD (Baylor Regional Medical Center at Grapevine/Baylor Research Institute), Jay Schlaifer, MD (St Elizabeth East), Gopi Shah, MD (Palmetto Health), Nicolas Shammas, MD (Midwest Cardiovascular Research Foundation), Fayez Shamoon, MD (Saint Michael’s Medical Center), Emanuel Shaoulian, MD (Hoag Memorial Hospital Presbyterian/Pacific Coast Cardiology), Trilok Sharma, MD (Southwest General Health Center), Thomas Shimshak, MD (Wheaton Franciscan Healthcare–All Saints, Inc), Arsalan Shirwany, MD (Baptist Memorial Hospital/Stern Cardiovascular Foundation, Inc), Paul Silverman, MD (Advocate Christ Medical Center), William Smith, MD (New Hanover Regional Medical Center), Dane Sobek, MD (Bozeman Deaconess Hospital), Ali Sonel, MD (VA Pittsburgh Healthcare System), Anthony Sonn, MD (Mercy Hospitals East Communities d/b/a Mercy Hospital St Louis), Nattapong Sricharoen, MD (University of Nebraska Medical Center), Nicholas Stamato, MD (UHS–United Health Services), Dwight Stapleton, MD (Donald Guthrie Foundation for Education and Research/Robert Packer Hospital), Robert Stenberg, MD (Conemaugh Valley Memorial Hospital d/b/a Memorial Medical Center), Brett Stoll, MD (AnMed Health), Damodhar Suresh, MD (St Elizabeth Medical Center/St Elizabeth’s Physicians Heart and Vascular Group), Peter Tadros, MD (University of Kansas Medical Center), Praveen Tamirisa, MD (The Toledo Hospital), Hoang Thai, MD (Southern Arizona VA Health Care System), Barry Uretsky, MD (Central Arkansas Veterans’ Healthcare System), Tudor Vagaonescu, MD (Cardiovascular Institute/Robert Wood Johnson Medical School), Imran Virk, MD (INTEGRIS Cardiology Southwest/INTEGRIS Southwest Medical Center), Kishor Vora, MD (Research Integrity, LLC), Ron Waksman, MD (Washington Hospital Center), Harry Wallner, MD (Trinity Medical Center), Jonathan Waltman, MD (Saint Joseph Hospital), Tracy Wang, MD (Duke University), Hal Wasserman, MD (Danbury Hospital), Richard Waters, MD (St Joseph’s Medical Center), Thomas Watson, MD (Santa Barbara Cottage Hospital), Larry Weathers, MD (Mercy Medical Center Northwest Arkansas), Richard Webel, MD (University of Missouri Health System), Bradley Weinberg, MD (Indiana Heart Hospital), Barry Weinstock, MD (Orlando Health, Inc), Sandra Weiss, MD (Christiana Care Health Services), James Welker, MD (Anne Arundel Medical Center), Donald Westerhausen Jr, MD (Elkhart General Hospital), James Wilson, MD (St Luke’s Episcopal Hospital), William Witmer, MD (Aurora BayCare Medical Center), Christopher Wolfram, MD (Bellin Memorial Hospital, Inc d/b/a Cardiology Associates of Bellin Health), Lambert Wu, MD (Cotton-O’Neil Clinical Research Center/Stormont-Vail), Mark Zainea, MD (McLaren Medical Center–Macomb), Patrik Zetterlund, MD (Salinas Valley Memorial Healthcare System), and James Zidar, MD (Rex Hospital).
et al. ACC 2009 Advocacy position statement: principles for comparative effectiveness research. J Am Coll Cardiol
. 2009;54(18):1744-1746.PubMedGoogle ScholarCrossref
T. Glossary for econometrics and epidemiology. J Epidemiol Community Health
. 2008;62(10):858-861.PubMedGoogle ScholarCrossref
U. Good research practices for comparative effectiveness research: analytic methods to improve causal inference from nonrandomized studies of treatment effects using secondary data sources: the ISPOR Good Research Practices for Retrospective Database Analysis Task Force Report, part III. Value Health
. 2009;12(8):1062-1073.PubMedGoogle ScholarCrossref
et al. Treatment With Adenosine Diphosphate Receptor Inhibitors–Longitudinal Assessment of Treatment Patterns and Events After Acute Coronary Syndrome (TRANSLATE-ACS) study design: expanding the paradigm of longitudinal observational research. Am Heart J
. 2011;162(5):844-851.PubMedGoogle ScholarCrossref
JP. Does more intensive treatment of acute myocardial infarction in the elderly reduce mortality? analysis using instrumental variables. JAMA
. 1994;272(11):859-866.PubMedGoogle ScholarCrossref
et al. Association between stroke center hospitalization for acute ischemic stroke and mortality. JAMA
. 2011;305(4):373-380.PubMedGoogle ScholarCrossref
et al. Evaluating the effectiveness of a rapidly adopted cardiovascular technology with administrative data: the case of drug-eluting stents for acute coronary syndromes. Am Heart J
. 2012;164(2):207-214.PubMedGoogle ScholarCrossref
et al; EVENT Registry Investigators. Survival benefit with drug-eluting stents in observational studies: fact or artifact? Circ Cardiovasc Qual Outcomes
. 2011;4(6):587-594.PubMedGoogle ScholarCrossref
R. Effect of highly active antiretroviral therapy on multiple AIDS-defining illnesses among male HIV seroconverters. Am J Epidemiol
. 2006;163(4):310-315.PubMedGoogle ScholarCrossref
MJ. Analysis of observational studies in the presence of treatment selection bias: effects of invasive cardiac management on AMI survival using propensity score and instrumental variable methods. JAMA
. 2007;297(3):278-285.PubMedGoogle ScholarCrossref
DE. Long-term outcomes of regional variations in intensity of invasive vs medical management of Medicare patients with acute myocardial infarction. JAMA
. 2005;293(11):1329-1337.PubMedGoogle ScholarCrossref
SC. Combining ecological and individual variables to reduce confounding by indication: case study: subarachnoid hemorrhage treatment. J Clin Epidemiol
. 2000;53(12):1236-1241.PubMedGoogle ScholarCrossref
S. Evaluating short-term drug effects using a physician-specific prescribing preference as an instrumental variable. Epidemiology
. 2006;17(3):268-275.PubMedGoogle ScholarCrossref
et al; Academic Research Consortium. Clinical end points in coronary stent trials: a case for standardized definitions. Circulation
. 2007;115(17):2344-2351.PubMedGoogle ScholarCrossref
GUSTO Investigators. An international randomized trial comparing four thrombolytic strategies for acute myocardial infarction. N Engl J Med
. 1993;329(10):673-682.PubMedGoogle ScholarCrossref
AB. Prespecified falsification end points: can they validate true observational associations? JAMA
. 2013;309(3):241-242.PubMedGoogle ScholarCrossref
PJ. Two-stage residual inclusion estimation: addressing endogeneity in health econometric modeling. J Health Econ
. 2008;27(3):531-543.PubMedGoogle ScholarCrossref
RJ. Advanced statistics: bootstrapping confidence intervals for statistics with “difficult” distributions. Acad Emerg Med
. 2005;12(4):360-365.PubMedGoogle ScholarCrossref
GW. Estimation of causal effects using propensity score weighting: an application to data on right heart catheterization. Health Serv Outcomes Res Methodol
. 2001;2(3):259-278.Google ScholarCrossref
PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med
. 2009;28(25):3083-3107.PubMedGoogle ScholarCrossref
DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Stat
. 1985;39(1):33-38.Google Scholar
DB. Identification of causal effects using instrumental variables. J Am Stat Assoc
. 1996;91(434):444-455.Google ScholarCrossref
SB. Potential bias of instrumental variable analyses for observational comparative effectiveness research. Ann Intern Med
. 2014;161(2):131-138.PubMedGoogle ScholarCrossref
et al; TRITON-TIMI 38 Investigators. Prasugrel versus clopidogrel in patients with acute coronary syndromes. N Engl J Med
. 2007;357(20):2001-2015.PubMedGoogle ScholarCrossref