Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Comparing Long-term Mortality After Carotid Endarterectomy vs Carotid Stenting Using a Novel Instrumental Variable Method for Risk Adjustment in Observational Time-to-Event Data

Comparing Long-term Mortality After Carotid Endarterectomy vs Carotid Stenting Using a Novel... Key Points Question Can a novel instrumental IMPORTANCE Choosing between competing treatment options is difficult for patients and clinicians variable method designed for time- when results from randomized and observational studies are discordant. Observational real-world dependent outcomes more accurately studies yield more generalizable evidence for decision making than randomized clinical trials, but determine the relative long-term unmeasured confounding, especially in time-to-event analyses, can limit validity. mortality after carotid endarterectomy vs carotid artery stenting? OBJECTIVES To compare long-term survival after carotid endarterectomy (CEA) and carotid artery Findings In this registry-based, stenting (CAS) in real-world practice using a novel instrumental variable method designed for time- multicenter cohort study of 86 017 to-event outcomes, and to compare the results with traditional risk-adjustment models used in patients, the hazard ratio of long-term observational research for survival analyses. mortality for carotid endarterectomy vs carotid artery stenting was 0.83 (95% DESIGN, SETTING, AND PARTICIPANTS A multicenter cohort study was performed. The Vascular CI, 0.70-0.98) using instrumental Quality Initiative, an observational quality improvement registry, was used to compare long-term variable analysis, compared with 0.69 mortality after CEA vs CAS. The study included 86 017 patients who underwent CEA (n = 73 312) or (95% CI, 0.65-0.74) using a traditional CAS (n = 12 705) between January 1, 2003, and December 31, 2016. Patients were followed up for Cox regression model. long-term mortality assessment by linking the registry data to Medicare claims. Medicare claims data were available through September 31, 2015. Meaning Results from this instrumental variable method show that the survival EXPOSURE Procedure type (CEA vs CAS). advantage conferred by carotid endarterectomy is more modest than MAIN OUTCOMES AND MEASURES The hazard ratios (HRs) of all-cause mortality using suggested by traditional adjustment unadjusted, adjusted, propensity-matched, and instrumental variable methods were examined. The methods, aligning with results from instrumental variable was the proportion of CEA among the total carotid procedures randomized clinical trials. (endarterectomy and stenting) performed at each hospital in the 12 months before each patient’s index operation and therefore varies over the study period. Invited Commentary RESULTS Participants who underwent CEA had a mean (SD) age of 70.3 (9.4) years compared with Supplemental content 69.1 (10.4) years for CAS, and most were men (44 191 [60.4%] for CEA and 8117 [63.9%] for CAS). The Author affiliations and article information are observed 5-year mortality was 12.8% (95% CI, 12.5%-13.2%) for CEA and 17.0% (95% CI, listed at the end of this article. 16.0%-18.1%) for CAS. The unadjusted HR of mortality for CEA vs CAS was 0.67 (95% CI, 0.64-0.71), and Cox-adjusted and propensity-matched HRs were similar (0.69; 95% CI, 0.65-0.74 and 0.71; 95% CI, 0.65-0.77, respectively). These findings are comparable with published observational studies of CEA vs CAS. However, the association between CEA and mortality was more modest when estimated by instrumental variable analysis (HR, 0.83; 95% CI, 0.70-0.98), a finding similar to data reported in randomized clinical trials. (continued) Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 1/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Abstract (continued) CONCLUSIONS AND RELEVANCE The study found a survival advantage associated with CEA over CAS in unadjusted and Cox-adjusted analyses. However, this finding was more modest when using an instrumental variable method designed for time-to-event outcomes for risk adjustment. The instrumental variable-based results were more similar to findings from randomized clinical trials, suggesting this method may provide less biased estimates of time-dependent outcomes in observational analyses. JAMA Network Open. 2018;1(5):e181676. Corrected on October 12, 2018. doi:10.1001/jamanetworkopen.2018.1676 Introduction Randomized clinical trials, which have internal validity and test efficacy under carefully designed 1-4 study conditions, produce findings that are often widely accepted. When results of observational studies are concordant with randomized clinical trials, clear messages emerge for patients, clinicians, 5-8 and payers to guide treatment decisions. Concordance of results is particularly important when assessing a long-term, time-to-event outcome such as mortality, as this suggests that the findings 9-11 seen in randomized clinical trials will be durable in clinical practice. Treatment decisions are more difficult, however, when the results of randomized clinical trials and observational studies are discordant. For example, for patients and clinicians considering carotid endarterectomy (CEA) or carotid artery stenting (CAS), 2 competing treatments to prevent stroke from carotid artery stenosis, long-term survival after the procedure remains a matter of debate. While randomized clinical trials have shown no statistically significant difference in mortality between the 2 procedures, observational evidence suggests survival following endarterectomy is 12-17 superior. Several potential explanations exist for this type of discordance. For example, treatment regimens and effects in randomized clinical trials may not reflect clinical practice, thereby limiting 19,20 generalizability. This limitation of randomized clinical trials as well as their high cost and complexity make observational studies an attractive alternative. However, risk adjustment for confounding in observational data remains challenging. While methods such as Cox proportional hazards regression and propensity score matching have been developed to adjust observational time-to-event data for measured confounding, the possibility that unmeasured or even unmeasurable confounding persists in observational analyses is an important concern faced by 11,21-24 patients and clinicians. Unmeasured confounding is of particular importance in patients with peripheral arterial disease considering invasive vs minimally invasive options, where surgeon selection bias and patient fitness for surgery have been shown to have an important association with clinical outcomes after aortic aneurysm repair. Selection bias and unmeasured confounding are likely to also occur in patients with carotid artery disease, where the decision to choose an invasive vs minimally invasive procedure is influenced by many factors. Instrumental variable analysis is a procedure unique in its ability to account for unmeasured confounding, and this method has been applied to linear and logistic regression models to evaluate 26,27 outcomes that are not time dependent. To date, adaptation of instrumental variable methods in areas of medicine such as cardiovascular disease that often examine time-to-event data using Cox 22,28-31 regression as the standard analytic tool has been limited. We recently developed an instrumental variable procedure for use with the Cox model and have shown that it outperforms the 24,32 traditional Cox model and 2-stage approaches that include the Cox model. We apply this procedure to adjust for suspected unmeasured confounding when comparing individuals’ long-term mortality between 2 competing treatments for carotid revascularization in a large observational data set. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 2/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Methods Data Sources Our analyses use data derived from the Vascular Quality Initiative registry, a national quality improvement registry that captures data on vascular procedures from more than 400 hospitals and practices across the United States and Canada. Patients and procedures entered in the registry 34,35 were linked to the Medicare Denominator File for mortality assessment. This database includes patient-level information on baseline demographics, comorbid conditions, presenting neurologic symptoms, operative management, and mortality on patients who underwent CEA and CAS. Data from the Vascular Quality Initiative were available from January 1, 2003, to December 31, 2016. Medicare data were available until September 2015. All data were collected under the auspices of an Agency for Healthcare Research and Quality–designated Patient Safety Organization and were deidentified. Our study was approved by the Center for the Protection of Human Subjects at Dartmouth; a waiver of participant consent was obtained. This study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Primary Exposure and Outcome The primary exposure of interest was procedure type (CEA vs CAS). These procedures represent the most common methods of carotid revascularization in current practice. Patients receiving more than 1 procedure type in the same day were assigned to the first procedure they received. Patients receiving repeated revascularization procedures during follow-up were assigned to the index procedure. The primary outcome was all-cause mortality. This was assessed for all patients in the registry using the Social Security Death Index. Patients were assessed from the time of their index procedure until death, and censored only at the end of their known follow-up period. Patients eligible for Medicare in the registry were linked to their respective Medicare claims file, analyzed through the end of September 2015. Successful linking was obtained in 92% and 91% of eligible patients who underwent CEA or CAS, respectively. Statistical Analysis We performed a sensitivity analysis with results stratified based on the presence or absence of focal neurologic symptoms (symptomatic vs asymptomatic) at the time of presentation. We used clinical variables from the Vascular Quality Initiative to group patients as symptomatic or asymptomatic. We defined patients as symptomatic if they had a documented stroke, transient ischemic attack, or other ischemic neurologic symptoms at the time of hospitalization for their index procedure. We reported unadjusted mortality as absolute and relative frequencies where appropriate. We calculated the hazard ratio (HR) of mortality for CEA vs CAS using 4 methods of estimation: unadjusted, Cox regression adjusted, propensity matched, and our instrumental variable procedure designed for time-to-event data. We applied these to the overall cohort, as well as in the sensitivity analysis based on the presence of focal neurologic symptoms at the time of revascularization. Statistical tests were 2-sided with P < .05 considered significant. All statistical analyses were performed with R, version 3.3.2 (R Foundation for Statistical Computing). Unadjusted and Adjusted Mortality We calculated unadjusted mortality rates using Kaplan-Meier estimation. We then used Cox regression to estimate the HR of postoperative mortality for CEA vs CAS to account for observed 24,38 confounding. Summary statistics for the confounding variables in the statistical models are noted in Table 1. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 3/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Propensity-Matched Analysis 23,39 We created a propensity-matched cohort balanced in baseline covariates. Using the observed covariates in Table 1, we created a logistic regression model in which the dependent variable was the treatment exposure (CEA vs CAS). Next, we calculated the fitted probability of CEA, known as the propensity score, for each patient. We then matched patients undergoing CEA to those undergoing CAS. We compared mortality between patients who underwent CEA vs CAS in the matched cohort. To account for the censoring, we applied Cox regression to our propensity-matched cohort to estimate the HR of mortality for CEA vs CAS. Instrumental Variable Analysis Our instrument was the proportion of CEA among the total number of carotid revascularization procedures (CEA and CAS) performed at each hospital in the 12 months prior to the index operation. We excluded hospitals not performing at least 10 revascularization procedures in the year prior to the index operation. In the presenting symptoms sensitivity analysis (patients presenting with focal neurologic symptoms vs not), we further excluded hospitals not performing at least 10 carotid revascularization procedures for each indication. For this reason, the number of patients included in the overall analysis slightly exceeds the total number of patients included in the sensitivity analysis. The instrumental variable procedure identifies patients who would have undergone CEA at some institutions and CAS at others based on the value of the instrument alone and not on patient characteristics. If patients choose hospitals based on convenience, or at least based only on Table 1. Cohort Characteristics All Patients (N = 86 017) Propensity Matched (n = 24 680) Variable CEA (n = 73 312) CAS (n = 12 705) P Value CEA (n = 12 340) CAS n = (12 340) P Value Demographics Age, mean (SD), y 70.3 (9.4) 69.1 (10.4) <.001 69.3 (9.7) 69.2 (10.4) .44 Male, No. (%) 44 191 (60.4) 8117 (63.9) <.001 7843 (63.6) 7867 (63.7) .76 Race, No. (%) White 67 768 (92.4) 11 524 (90.7) <.001 11 174 (90.6) 11 198 (90.8) .65 Black 3099 (4.2) 694 (5.5) <.001 681 (5.5) 672 (5.4) .82 Other 2445 (3.4) 487 (3.8) .005 485 (3.9) 470 (3.8) .64 Clinical factors, No. (%) Elective 64 022 (87.3) 10 252 (80.7) <.001 10 014 (81.2) 10 042 (81.4) .66 Symptomatic 28 836 (39.3) 6863 (54.0) <.001 6333 (51.3) 6519 (52.8) .02 TIA or amaurosis 14 200 (19.4) 3106 (24.4) <.001 2885 (23.4) 2970 (24.1) .21 Stroke 14 636 (20.0) 3757 (29.6) <.001 3448 (27.9) 3549 (28.8) .16 Hypertension 65 128 (88.8) 11 292 (88.9) .90 11 032 (89.4) 10 981 (89.0) .31 Smoking history 55 476 (75.7) 9643 (75.9) .59 9393 (76.1) 9361 (75.9) .64 Positive stress test 5937 (8.1) 988 (7.8) .23 977 (7.9) 973 (7.9) .94 Coronary disease 20 643 (28.2) 4150 (32.7) <.001 4153 (33.6) 4023 (32.6) .08 Heart failure 7512 (10.2) 1883 (14.8) <.001 1792 (14.5) 1782 (14.4) .87 Diabetes 25 637 (35.0) 4595 (36.2) <.001 4522 (36.6) 4451 (36.1) .88 COPD 16 261 (22.2) 3233 (25.4) <.001 3181 (25.8) 3107 (25.2) .29 Renal insufficiency 4101 (5.6) 722 (5.7) .70 722 (5.9) 702 (5.7) .60 Hemodialysis 937 (1.3) 25 (0.2) <.001 34 (0.3) 25 (0.2) .30 Prior CEA 10 132 (13.8) 4132 (32.5) <.001 3908 (31.2) 3803 (30.8) .15 Medications Antiplatelet therapy, No. (%) Aspirin 60 744 (82.9) 10 877 (85.6) <.001 10 363 (84.0) 10 545 (85.5) .001 P2Y12 inhibitor 21 163 (28.9) 9646 (75.9) <.001 9310 (75.4) 9281 (75.2) .68 β-Blocker 41 759 (57.0) 7003 (55.1) <.001 7005 (56.8) 6853 (55.5) .04 Statin 58 588 (79.9) 10 120 (79.7) .50 9833 (79.7) 9861 (79.9) .67 Abbreviations: CAS, carotid artery stenting; CEA, carotid endarterectomy; COPD, chronic obstructive pulmonary disease; TIA, transient ischemic attack. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 4/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method observed factors, then where a patient seeks treatment emulates a randomized encouragement design in which assignment to a hospital with a historical precedent for performing a high proportion of CEA randomly exposes the patient to a greater likelihood of undergoing CEA. The estimation of treatment effects using instrumental variables is well developed for linear or logistic regression 40,42,43 models. However, Cox proportional hazard models examining time-to-event outcomes selects on survivors over the course of follow-up, which is problematic for standard methods of 24,44-46 instrumental variable identification. Therefore, we used the new instrumental variable estimator for the Cox proportional hazards model to simultaneously deal with the problems of unmeasured confounding and censoring of the outcome. We used this new procedure to examine the HR for all-cause mortality after CEA vs CAS by including both the instrument and all known confounding variables described in Table 1. The mean (SD) value of the instrumental variable was 0.89 (0.12) for patients undergoing CEA and 0.65 (0.29) for patients undergoing CAS (P < .001). Further details on derivation of the instrumental variable and its distribution can be found in the eMethods and eFigure 1 in the Supplement. Instrument Assessment We measured the strength of our instrument by determining if increasing levels of the instrument were associated with changing levels of the exposure. This is reported using the F statistic, for which a value greater than 10 traditionally indicates acceptable strength. The F statistic assesses the instrument’s ability to show association with the exposure received beyond the effect of any covariates that are adjusted for the survival model. Results Cohort Characteristics We studied 86 017 patients who underwent carotid revascularization (CEA, n = 73 312; CAS, n = 12 705) from January 1, 2003, to December 31, 2016. Mean follow-up was 3.0 (SD, 2.4 years; range, <0.1-14.3 years; median [interquartile range] follow-up, 2.5 [1.3-4.0] years), yielding the equivalent of 259 700 person-years for analysis. Vital status was known for 75.0% of patients who were eligible (procedure date, 2011 or earlier). Compared with patients who underwent CEA, those who underwent CAS tended to be younger (mean [SD] age, 70.3 [9.4] vs 69.1 [10.4] years, respectively), were more likely to be male (44 191 [60.4%] vs 8117 [63.9%], respectively), and were more likey to have an urgent procedure (2453 [19.3%] vs 9290 [12.7%], respectively) (Table 1). More than 89% of patients were receiving some form of antiplatelet therapy, and more than 75% were receiving a statin. Characteristics of the sensitivity analysis cohorts (symptomatic and asymptomatic patients) were similar (eTables 1 and 2 in the Supplement). There were several clinically meaningful differences between patients who underwent CEA and those who underwent CAS. Approximately one-third of patients undergoing CEA underwent the procedure because of focal neurologic symptoms, compared with more than half of patients treated with CAS. Patients who underwent CAS were also more likely to have several chronic comorbid conditions, including coronary artery disease, heart failure, and pulmonary disease. Patients who underwent CAS were also more likely to have previously undergone carotid surgery. Given that several differences existed in the characteristics between patients treated with CEA and CAS, we created a propensity-matched cohort for analysis. The propensity-matched cohort consisted of 12 340 matched pairs of patients and was well balanced in baseline characteristics apart from a small difference in aspirin use (84.0% in the CEA group and 85.5% in the CAS group; P = .001), β-blocker prescription (56.8% in the CEA group and 55.5% in the CAS group; P = .04), and in the proportion of procedures performed for symptomatic stenosis (51.3% in the CEA group and 52.8% in the CAS group; P = .02). A graphical representation of the performance of the propensity score matching can be found in eFigure 2 in the Supplement. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 5/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Unadjusted, Cox-Adjusted, and Propensity-Matched Mortality by Procedure Type The unadjusted Kaplan-Meier estimate of all-cause mortality at 5 years for CEA was 12.8% (95% CI, 12.5%-13.2%) and for CAS was 17.0% (95% CI, 16.0%-18.1%; log rank, P < .001). At 10 years after the procedure, estimated mortality was 27.3% (95% CI, 26.3%-27.3%) for CEA and 27.4% (23.9%-30.7%) for CAS (log rank, P < .001; eFigure 3 in the Supplement). Sensitivity analysis by the presence of neurologic symptoms at the time of revascularization demonstrated similar findings (Figure 1). The unadjusted HR of all-cause mortality for CEA vs CAS was 0.67 (95% CI, 0.64-0.71) (Table 2). A Cox proportional hazards model adjusting for differences in patient characteristics showed a similar Figure 1. Kaplan-Meier Estimated Mortality, Overall and by Presenting Symptoms All patients CAS CEA Log rank P <.001 0 1 2 3 4 5 Time, y No. at risk CAS 12 705 10 000 6854 4466 2578 1185 CEA 73 312 60 901 44 974 30 440 19 139 11 362 B Symptomatic CAS Log rank P <.001 CEA 0 1 2 3 4 5 Time, y No. at risk CAS 6825 5210 3516 2138 1190 533 CEA 26 689 23 822 17 319 11 653 7306 4382 C Asymptomatic CAS CEA Log rank P <.001 0 1 2 3 4 5 Time, y No. at risk CAS indicates carotid artery stenting; CEA, carotid CAS 5809 4731 3289 2292 1363 635 CEA 44 395 36 884 27 481 18 655 11 740 6920 endarterectomy. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 6/14 Mortality, % Mortality, % Mortality, % JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method association (HR, 0.69; 95% CI, 0.65-0.74), further suggesting that CEA was associated with a survival advantage. The propensity-matched cohort also revealed a survival advantage associated with CEA (HR, 0.71; 95% CI, 0.65-0.77). Sensitivity analysis by the presence of neurologic symptoms before carotid revascularization continued to show a statistically significant association (Table 2 and 16,48 Figure 2). Instrumental Variable–Adjusted Mortality by Procedure Type The instrument, each individual hospital’s 12-month prior proportion of CEA procedures, demonstrated a very strong association with the type of carotid procedure performed (F = 18 631). Applying our instrumental variable procedure to all-cause mortality revealed that patients selected for CEA had a more modest survival advantage (HR, 0.83; 95% CI, 0.70-0.98) than was suggested by results of our other analytic methods. These results are similar to the findings of published randomized clinical trials (Figure 2). Similar results were obtained by our instrumental variable approach in a sensitivity analyses stratified by presenting symptoms, although the association was more pronounced in those who were symptomatic. The HRs for those with symptoms changed by an absolute 17% to 19% between traditional statistical methods and our instrumental variable model, compared with an absolute change of 11% to 14% in those who were asymptomatic (Table 2). Table 2. Mortality HRs for Carotid Endarterectomy vs Carotid Stenting HR (95% CI) No. of Mortality Events/Total No. Variable of Patients Crude Adjusted Propensity Matched Instrumental Variable Overall CEA 6600/73 312 0.67 (0.64-0.71) 0.69 (0.65-0.74) 0.71 (0.65-0.77) 0.83 (0.70-0.98) CAS 1405/12 705 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Symptomatic CEA 2559/28 689 0.61 (0.46-0.66) 0.61 (0.56-0.67) 0.59 (0.53-0.66) 0.78 (0.61-0.99) CAS 786/6825 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Asymptomatic CEA 4017/44 395 0.76 (0.70-0.83) 0.79 (0.72-0.87) 0.79 (0.71-0.90) 0.90 (0.70-1.14) CAS 607/5809 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Abbreviations: CAS, carotid artery stenting; CEA, carotid endarterectomy; HR, Event and total cohort numbers are different for the propensity-matched analysis. hazard ratio. Figure 2. Hazard Ratios (HRs) of Mortality for Carotid Endarterectomy vs Carotid Stenting A All patients B Symptomatic C Asymptomatic Favors Favors Favors Favors Favors Favors Source endarterectomy stenting endarterectomy stenting endarterectomy stenting Relevant RCT results Brott et al, 2016 (CREST; n = 2502) Rosenfield et al, 2016 (ACT 1; n = 1453) VQI-Medicare observational analyses Unadjusted model Adjusted model Propensity matched Instrumental variable 0.5 1.0 2 0.5 1.0 2 0.5 1.0 2 HR (95% CI) HR (95% CI) HR (95% CI) The CREST trial outcome represented is long-term stroke or periprocedural myocardial Revascularization Endarterectomy vs Stenting Trial; RCT, randomized clinical trial; VQI, infarction, stroke, or death. ACT 1 indicates Asymptomatic Carotid Trial; CREST, Carotid Vascular Quality Initiative. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 7/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Discussion In this observational study, unadjusted, adjusted, and propensity-matched models of long-term mortality all demonstrated that treatment with CEA was associated with a survival benefit relative to treatment with CAS. These results are comparable with published observational reports but conflict with randomized clinical trials, which suggest survival is similar following the 2 competing treatment 12-17,49,50 options. Our instrumental variable method designed for risk adjustment of time-to-event data estimated a more modest association with long-term mortality, a finding consistent with the results of randomized clinical trials. These findings were robust to a sensitivity analysis by the presence of focal neurologic symptoms. This method, which accounts for both measured and unmeasured confounding in observational time-to-event analyses, represents an advance for investigators evaluating long-term outcomes, especially when considering clinical questions where randomized clinical trials are not possible or would be prohibitively expensive or when use of real- world evidence would be advantageous. Discordance between randomized clinical trials and observational studies is neither new nor 52-57 uncommon. For example, differences in the efficacy of vitamin E and hormone replacement therapy for the prevention of heart disease as well as antioxidant therapy for cancer represent important examples where conflicting results from randomized clinical trials and observational 52-56 studies have affected evidence-based treatment decisions. Meta-analyses examining the relative findings of randomized and observational studies suggest that observational studies tend to generate a larger treatment effect, and these differences may be further potentiated when assessing 58-64 long-term outcomes such as mortality. However, this is not always the case; a recent Cochrane review estimated that treatment effects were similar between randomized clinical trials and observational studies (pooled ratio of odds ratios, 1.04; 95% CI, 0.89-1.21). These contradictory results highlight how challenging it can be for patients and clinicians to interpret observational study results, especially if the direction of bias cannot be foreseen. In the example of discordance used in our analysis, patients with CAS, the Asymptomatic Carotid Trial (ACT 1) reported no statistically significant difference in 5-year mortality after CEA vs CAS, and the Carotid Revascularization Endarterectomy vs Stenting Trial (CREST) reported no statistically significant difference at 10 years in the composite outcome of long-term risk of stroke or perioperative myocardial infarction, stroke, or death. Despite these randomized clinical trials, several large observational studies have documented inferior outcomes for stenting overall, especially in subgroup analyses of symptomatic patients, which may bias physicians and patients 12-15,50 away from choosing stenting as a procedural option. Our findings, which suggest that there is a modest association between survival and CEA, provide important granular detail to help inform this management decision. The design and execution of a randomized clinical trial that provides true, unbiased estimates is a very difficult task, as there are several threats to both the validity and generalizability of their results. In some cases, clinical trial participants may not be representative of the target 18-20 population. In others, heterogeneous treatment effects may impede the ability of physicians to 18-20 parse out which patients may benefit most from an intervention. In addition, intention-to-treat estimators used in randomized clinical trials may be biased toward the null if noncompliance is considerable. These limitations to many contemporary randomized clinical trials highlight the utility of observational studies where real-world evidence can be used, provided that adequate adjustment for confounding can be performed. An analytic technique capable of better risk adjustment for unmeasured confounding would improve the reliance that could be placed on results from observational studies. Such a technique would allow real-world observational data to more consistently reflect the true outcome of treatment independent of confounding. Our instrumental variable procedure was specifically designed to be used to analyze time-to-event outcomes. While determining a suitable instrument JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 8/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method may be difficult in some settings, there appears to be few disadvantages in applying this procedure 22,47,66,67 to observational questions with time-dependent outcomes such as mortality. While the findings of our instrumental variable analyses gave results that are similar to those in randomized clinical trials, this may not be the case when applied to other clinical scenarios. The importance of observational data is that it documents results from clinical practice, outside of the confines of randomized clinical trials. Observational studies often include a much broader patient population with treatment-effect heterogeneity than are found in randomized clinical trials. In these situations, results from instrumental variable analyses may be different than those found in randomized clinical trials and may better represent the results that can be expected when an intervention is incorporated into clinical practice. Application of instrumental variables to time-to- event data therefore represents an important step forward in the evaluation of interventions in contemporary practice. Limitations Our study had limitations. First, it is not possible to truly know whether our instrumental variable balanced all unmeasured confounding. However, our sensitivity analyses by the presence of neurologic symptoms are reassuring. One would anticipate that unmeasured confounding would have a greater impact on the symptomatic analysis as patients in this subgroup are frequently sicker and thus are at higher risk for clinician selection bias to play a substantial role in the treatment decision. An instrument that accounts for unmeasurable confounding would change the effect size to a greater extent in these patients, and this was noted in our analyses. Second, we did not examine stroke-free survival as our primary outcome because of the heterogeneity in stroke assessment methods across the sites in our observational registry, an issue not encountered when examining survival as an outcome. Third, while 5-year vital status was known in 75.0% of patients who were eligible, many patients were not eligible for this assessment because of the date of their procedure (after 2011). Changes over time in both practice patterns and procedural competency may have an impact on the HR of mortality between the 2 procedures. However, findings remained consistent among patients who had their operations in earlier years where the longest follow-up was possible. Therefore, we feel that our estimates reported herein are an accurate reflection of long-term mortality after CEA vs CAS. In addition, instrumental variables must satisfy three conditions: first, they must be associated with the treatment exposure; second, an instrument must have no relationship to the outcome except through the effect on the exposure; and third, there must be no variables that affect both the instrument and the outcome. Our F statistic demonstrated that our instrument was strongly associated with the treatment exposure, thereby satisfying the first condition. It is not possible to prove whether an instrument is unrelated to an outcome. However, we required that a center perform at least 10 CEA or CAS procedures in the prior year to have patients included in the instrumental variable analysis to limit the possibility that proportion of procedures 69,70 performed could be related to postoperative mortality. Conclusions Using a novel instrumental variable method designed for time-to-event data, we found only a modest difference in long-term mortality after CEA vs CAS, a result that is comparable with recent randomized clinical trials. These similarities provide evidence that results from our instrumental variable procedure are more closely aligned with the true relative long-term mortality between the 2 revascularization procedures than incumbent methods for analyzing observational data. This method, which allows instruments to be used for risk adjustment with the widely used Cox regression model, may improve the validity of results for time-dependent outcomes for clinical questions where randomized clinical trials are not possible or would be prohibitively expensive or when use of real-world evidence would be advantageous. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 9/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method ARTICLE INFORMATION Accepted for Publication: June 7, 2018. Published: September 7, 2018. doi:10.1001/jamanetworkopen.2018.1676 Correction: This article was corrected on October 12, 2018, to fix errors in the title and Figure 2. Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2018 Columbo JA et al. JAMA Network Open. Corresponding Author: A. James O’Malley, PhD, The Dartmouth Institute for Health Policy and Clinical Practice, One Medical Center Drive, Lebanon, NH 03756 (james.omalley@dartmouth.edu). Author Affiliations: The Dartmouth Institute for Health Policy and Clinical Practice, Lebanon, New Hampshire (Columbo, Martinez-Camblor, MacKenzie, Staiger, Kang, Goodney, O’Malley); Section of Vascular Surgery, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire (Columbo, Goodney); Department of Biomedical Data Science, Geisel School of Medicine, Lebanon, New Hampshire (MacKenzie, O’Malley); Department of Economics, Dartmouth College, Hanover, New Hampshire (Staiger). Author Contributions: Drs O’Malley and Martinez-Camblor had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: All authors. Acquisition, analysis, or interpretation of data: Columbo, Martinez-Camblor, Mackenzie, Staiger, Goodney, O’Malley. Drafting of the manuscript: Columbo, Goodney, O’Malley. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Martinez-Camblor, Mackenzie, Staiger, O’Malley. Obtained funding: O’Malley. Administrative, technical, or material support: Columbo, Kang, Goodney, O’Malley. Supervision: Columbo, O’Malley. Conflict of Interest Disclosures: Dr Staiger reported grants from the Patient-Centered Outcomes Research Institute during the conduct of the study; and personal fees, nonfinancial support, and grants from ArborMetrix, Inc outside the submitted work. Dr O’Malley reported receiving grants from the Patient-Centered Outcomes Research Institute and the National Institutes of Health during the conduct of the study. No other disclosures were reported. Funding/Support: This work was supported by Patient-Centered Outcomes Research Institute award ME-1503- 28261. Additional support was provided by the US Food and Drug Administration (U01-FD005478), the National Institute on Aging (PO1-AG019783), and the National Institutes of Health Common Fund (U01-AG046830). Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Disclaimer: All statements in this article, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute, its Board of Governors, or Methodology Committee. REFERENCES 1. Barnett HJM, Taylor DW, Haynes RB, et al; North American Symptomatic Carotid Endarterectomy Trial Collaborators. Beneficial effect of carotid endarterectomy in symptomatic patients with high-grade carotid stenosis. N Engl J Med. 1991;325(7):445-453. doi:10.1056/NEJM199108153250701 2. Endarterectomy for asymptomatic carotid artery stenosis: executive committee for the Asymptomatic Carotid Atherosclerosis Study. JAMA. 1995;273(18):1421-1428. doi:10.1001/jama.1995.03520420037035 3. Halliday A, Mansfield A, Marro J, et al; MRC Asymptomatic Carotid Surgery Trial Collaborative Group. Prevention of disabling and fatal strokes by successful carotid endarterectomy in patients without recent neurological symptoms: randomised controlled trial. Lancet. 2004;363(9420):1491-1502. doi:10.1016/S0140-6736(04) 16146-1 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 10/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 4. Brott TG, Halperin JL, Abbara S, et al; American College of Cardiology Foundation; American Stroke Association; American Association of Neurological Surgeons; American College of Radiology; American Society of Neuroradiology; Congress of Neurological Surgeons; Society of Atherosclerosis Imaging and Prevention; Society for Cardiovascular Angiography and Interventions; Society of Interventional Radiology; Society of NeuroInterventional Surgery; Society for Vascular Medicine; Society for Vascular Surgery. 2011 ASA/ACCF/AHA/ AANN/AANS/ACR/ASNR/CNS/SAIP/SCAI/SIR/SNIS/SVM/SVS guideline on the management of patients with extracranial carotid and vertebral artery disease: executive summary: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines, and the American Stroke Association, American Association of Neuroscience Nurses, American Association of Neurological Surgeons, American College of Radiology, American Society of Neuroradiology, Congress of Neurological Surgeons, Society of Atherosclerosis Imaging and Prevention, Society for Cardiovascular Angiography and Interventions, Society of Interventional Radiology, Society of NeuroInterventional Surgery, Society for Vascular Medicine, and Society for Vascular Surgery. Circulation. 2011;124(4):489-532. doi:10.1161/CIR.0b013e31820d8d78 5. Manson JE, Stampfer MJ, Colditz GA, et al. A prospective study of aspirin use and primary prevention of cardiovascular disease in women. JAMA. 1991;266(4):521-527. doi:10.1001/jama.1991.03470040085027 6. Kannel WB, Larson M. Long-term epidemiologic prediction of coronary disease: the Framingham experience. Cardiology. 1993;82(2-3):137-152. doi:10.1159/000175864 7. Baigent C, Blackwell L, Collins R, et al; Antithrombotic Trialists’ Collaboration. Aspirin in the primary and secondary prevention of vascular disease: collaborative meta-analysis of individual participant data from randomised trials. Lancet. 2009;373(9678):1849-1860. doi:10.1016/S0140-6736(09)60503-1 8. US Preventive Services Task Force. Aspirin for the prevention of cardiovascular disease: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med. 2009;150(6):396-404. doi:10.7326/0003- 4819-150-6-200903170-00008 9. Lederle FA, Freischlag JA, Kyriakides TC, et al; Open Versus Endovascular Repair Veterans Affairs Cooperative Study Group. Outcomes following endovascular vs open repair of abdominal aortic aneurysm: a randomized trial. JAMA. 2009;302(14):1535-1542. doi:10.1001/jama.2009.1426 10. De Bruin JL, Baas AF, Buth J, et al; DREAM Study Group. Long-term outcome of open or endovascular repair of abdominal aortic aneurysm. N Engl J Med. 2010;362(20):1881-1889. doi:10.1056/NEJMoa0909499 11. Schermerhorn ML, Buck DB, O’Malley AJ, et al. Long-Term outcomes of abdominal aortic aneurysm in the medicare population. N Engl J Med. 2015;373(4):328-338. doi:10.1056/NEJMoa1405778 12. Giles KA, Hamdan AD, Pomposelli FB, Wyers MC, Schermerhorn ML. Stroke and death after carotid endarterectomy and carotid artery stenting with and without high risk criteria. J Vasc Surg. 2010;52(6):1497-1504. doi:10.1016/j.jvs.2010.06.174 13. McPhee JT, Schanzer A, Messina LM, Eslami MH. Carotid artery stenting has increased rates of postprocedure stroke, death, and resource utilization than does carotid endarterectomy in the United States, 2005. J Vasc Surg. 2008;48(6):1442-1450, 1450e1. doi:10.1016/j.jvs.2008.07.017 14. Wang FW, Esterbrooks D, Kuo YF, Mooss A, Mohiuddin SM, Uretsky BF. Outcomes after carotid artery stenting and endarterectomy in the Medicare population. Stroke. 2011;42(7):2019-2025. doi:10.1161/STROKEAHA.110. 15. Nolan BW, De Martino RR, Goodney PP, et al; Vascular Study Group of New England. Comparison of carotid endarterectomy and stenting in real world practice using a regional quality improvement registry. J Vasc Surg. 2012;56(4):990-996. doi:10.1016/j.jvs.2012.03.009 16. Rosenfield K, Matsumura JS, Chaturvedi S, et al; ACT I Investigators. Randomized trial of stent versus surgery for asymptomatic carotid stenosis. N Engl J Med. 2016;374(11):1011-1020. doi:10.1056/NEJMoa1515706 17. Bonati LH, Dobson J, Featherstone RL, et al; International Carotid Stenting Study Investigators. Long-term outcomes after stenting versus endarterectomy for treatment of symptomatic carotid stenosis: the International Carotid Stenting Study (ICSS) randomised trial. Lancet. 2015;385(9967):529-538. doi:10.1016/S0140-6736(14) 61184-3 18. Anglemyer A, Horvath HT, Bero L. Healthcare outcomes assessed with observational study designs compared with those assessed in randomized trials. Cochrane Database Syst Rev. 2014;(4):MR000034. 19. Rothwell PM. External validity of randomised controlled trials: “to whom do the results of this trial apply?” Lancet. 2005;365(9453):82-93. doi:10.1016/S0140-6736(04)17670-8 20. Dhruva SS, Redberg RF. Variations between clinical trial participants and Medicare beneficiaries in evidence used for Medicare national coverage decisions. Arch Intern Med. 2008;168(2):136-140. doi:10.1001/ archinternmed.2007.56 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 11/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 21. Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet. 2002;359(9302): 248-252. doi:10.1016/S0140-6736(02)07451-2 22. Stukel TA, Fisher ES, Wennberg DE, Alter DA, Gottlieb DJ, Vermeulen 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. doi:10.1001/jama.297.3.278 23. D’Agostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17(19):2265-2281. doi:10.1002/(SICI)1097-0258(19981015)17: 19<2265::AID-SIM918>3.0.CO;2-B 24. Cox DR. Regression models and life-tables. J R Stat Soc B. 1972;34(2):187-220. http://www.jstor.org/ stable/2985181. Accessed September 1, 2017. 25. De Martino RR, Brooke BS, Robinson W, et al. Designation as “unfit for open repair” is associated with poor outcomes after endovascular aortic aneurysm repair. Circ Cardiovasc Qual Outcomes. 2013;6(5):575-581. doi:10. 1161/CIRCOUTCOMES.113.000303 26. Hernán MA, Robins JM. Instruments for causal inference: an epidemiologist’s dream? Epidemiology. 2006;17 (4):360-372. doi:10.1097/01.ede.0000222409.00878.37 27. Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation: addressing endogeneity in health econometric modeling. J Health Econ. 2008;27(3):531-543. doi:10.1016/j.jhealeco.2007.09.009 28. MacKenzie TA, Tosteson TD, Morden NE, Stukel TA, O’Malley AJ. Using instrumental variables to estimate a Cox’s proportional hazards regression subject to additive confounding. Health Serv Outcomes Res Methodol. 2014; 14(1-2):54-68. doi:10.1007/s10742-014-0117-x 29. Aalen OO, Cook RJ, Røysland K. Does Cox analysis of a randomized survival study yield a causal treatment effect? Lifetime Data Anal. 2015;21(4):579-593. doi:10.1007/s10985-015-9335-y 30. Li J, Fine J, Brookhart A. Instrumental variable additive hazards models. Biometrics. 2015;71(1):122-130. doi: 10.1111/biom.12244 31. Martinussen T, Nørbo Sørensen D, Vansteelandt S. Instrumental variables estimation under a structural Cox model. Biostatistics.2017. 32. Martínez-Camblor P, Mackenzie T, Staiger DO, Goodney PP, O’Malley AJ. Adjusting for bias introduced by instrumental variable estimation in the Cox proportional hazards model. Biostatistics. 2017. doi:10.1093/ biostatistics/kxx062 33. Vascular Quality Initiative. http://www.vascularqualityinitiative.org/. Accessed March 1, 2017. 34. Centers for Medicare and Medicaid Services. http://www.cms.gov. Accessed March 17, 2017. 35. Hoel AW, Faerber AE, Moore KO, et al. A pilot study for long-term outcome assessment after aortic aneurysm repair using vascular quality initiative data matched to Medicare claims. J Vasc Surg. 2017;66(3):751-759.e1. doi:10. 1016/j.jvs.2016.12.100 36. Vandenbroucke JP, von Elm E, Altman DG, et al; STROBE Initiative. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007;4(10):e297. doi: 10.1371/journal.pmed.0040297 37. Skerritt MR, Block RC, Pearson TA, Young KC. Carotid endarterectomy and carotid artery stenting utilization trends over time. BMC Neurol. 2012;12:17. doi:10.1186/1471-2377-12-17 38. Cortese G, Scheike TH, Martinussen T. Flexible survival regression modelling. Stat Methods Med Res. 2010;19 (1):5-28. doi:10.1177/0962280209105022 39. Rosenbaum PR, Rubin DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Stat. 1985;39(1):33-38. doi:10.2307/2683903 40. Stock J, Yogo M. Testing for weak instruments in linear IV regression. In: Andrews DWK, ed. Identification and Inference for Econometric Models. New York, NY: Cambridge University Press; 2005:80-108. doi:10.1017/ CBO9780511614491.006 41. West SG, Duan N, Pequegnat W, et al. Alternatives to the randomized controlled trial. Am J Public Health. 2008;98(8):1359-1366. doi:10.2105/AJPH.2007.124446 42. Staiger D, Stock JH. Instrumental variables regression with weak instruments. Econometrica. 1997;65(3): 557-586. doi:10.2307/2171753 43. Rassen JA, Schneeweiss S, Glynn RJ, Mittleman MA, Brookhart MA. Instrumental variable analysis for estimation of treatment effects with dichotomous outcomes. Am J Epidemiol. 2009;169(3):273-284. doi:10.1093/ aje/kwn299 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 12/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 44. Wan F, Small D, Bekelman JE, Mitra N. Bias in estimating the causal hazard ratio when using two-stage instrumental variable methods. Stat Med. 2015;34(14):2235-2265. doi:10.1002/sim.6470 45. Cai B, Small DS, Have TR. Two-stage instrumental variable methods for estimating the causal odds ratio: analysis of bias. Stat Med. 2011;30(15):1809-1824. doi:10.1002/sim.4241 46. MacKenzie TA, Brown JR, Likosky DS, Wu Y, Grunkemeier GL. Review of case-mix corrected survival curves. Ann Thorac Surg. 2012;93(5):1416-1425. doi:10.1016/j.athoracsur.2011.12.094 47. Tan HJ, Norton EC, Ye Z, Hafez KS, Gore JL, Miller DC. Long-term survival following partial vs radical nephrectomy among older patients with early-stage kidney cancer. JAMA. 2012;307(15):1629-1635. doi:10.1001/ jama.2012.475 48. Brott TG, Howard G, Roubin GS, et al; CREST Investigators. Long-term results of stenting versus endarterectomy for carotid-artery stenosis. N Engl J Med. 2016;374(11):1021-1031. doi:10.1056/NEJMoa1505215 49. Paraskevas KI, Kalmykov EL, Naylor AR. Stroke/death rates following carotid artery stenting and carotid endarterectomy in contemporary administrative dataset registries: a systematic review. Eur J Vasc Endovasc Surg. 2016;51(1):3-12. doi:10.1016/j.ejvs.2015.07.032 50. Timaran CH, Veith FJ, Rosero EB, Modrall JG, Valentine RJ, Clagett GP. Intracranial hemorrhage after carotid endarterectomy and carotid stenting in the United States in 2005. J Vasc Surg. 2009;49(3):623-628. doi:10.1016/ j.jvs.2008.09.064 51. US Department of Health and Human Services, US Food and Drug Administration. Use of real-world evidence to support regulatory decision-making for Medical devices. https://www.fda.gov/downloads/medicaldevices/ deviceregulationandguidance/guidancedocuments/ucm513027.pdf. Accessed September 1, 2017. 52. Menkes MS, Comstock GW, Vuilleumier JP, Helsing KJ, Rider AA, Brookmeyer R. Serum beta-carotene, vitamins A and E, selenium, and the risk of lung cancer. N Engl J Med. 1986;315(20):1250-1254. doi:10.1056/ NEJM198611133152003 53. Stampfer MJ, Colditz GA. Estrogen replacement therapy and coronary heart disease: a quantitative assessment of the epidemiologic evidence. Prev Med. 1991;20(1):47-63. doi:10.1016/0091-7435(91)90006-P 54. Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group. The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med. 1994;330(15):1029-1035. doi:10. 1056/NEJM199404143301501 55. Khaw KT, Bingham S, Welch A, et al. Relation between plasma ascorbic acid and mortality in men and women in EPIC-Norfolk prospective study: a prospective population study. European prospective investigation into cancer and nutrition. Lancet. 2001;357(9257):657-663. doi:10.1016/S0140-6736(00)04128-3 56. Beral V, Banks E, Reeves G. Evidence from randomised trials on the long-term effects of hormone replacement therapy. Lancet. 2002;360(9337):942-944. doi:10.1016/S0140-6736(02)11032-4 57. Heart Protection Study Collaborative Group. MRC/BHF Heart Protection Study of antioxidant vitamin supplementation in 20,536 high-risk individuals: a randomised placebo-controlled trial. Lancet. 2002;360 (9326):23-33. doi:10.1016/S0140-6736(02)09328-5 58. Deeks JJ, Dinnes J, D’Amico R, et al; International Stroke Trial Collaborative Group; European Carotid Surgery Trial Collaborative Group. Evaluating non-randomised intervention studies. Health Technol Assess. 2003;7(27):iii-x, 1-173. doi:10.3310/hta7270 59. Kunz R, Oxman AD. The unpredictability paradox: review of empirical comparisons of randomised and non-randomised clinical trials. BMJ. 1998;317(7167):1185-1190. doi:10.1136/bmj.317.7167.1185 60. Ioannidis JP, Haidich AB, Pappa M, et al. Comparison of evidence of treatment effects in randomized and nonrandomized studies. JAMA. 2001;286(7):821-830. doi:10.1001/jama.286.7.821 61. Brewer T, Colditz GA. Postmarketing surveillance and adverse drug reactions: current perspectives and future needs. JAMA. 1999;281(9):824-829. doi:10.1001/jama.281.9.824 62. Ross SD. Drug-related adverse events: a readers’ guide to assessing literature reviews and meta-analyses. Arch Intern Med. 2001;161(8):1041-1046. doi:10.1001/archinte.161.8.1041 63. Sutton AJ, Cooper NJ, Lambert PC, Jones DR, Abrams KR, Sweeting MJ. Meta-analysis of rare and adverse event data. Expert Rev Pharmacoecon Outcomes Res. 2002;2(4):367-379. doi:10.1586/14737167.2.4.367 64. Ioannidis JP, Mulrow CD, Goodman SN. Adverse events: the more you search, the more you find. Ann Intern Med. 2006;144(4):298-300. doi:10.7326/0003-4819-144-4-200602210-00013 65. Newcombe RG. Explanatory and pragmatic estimates of the treatment effect when deviations from allocated treatment occur. Stat Med. 1988;7(11):1179-1186. doi:10.1002/sim.4780071111 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 13/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 66. Hearst N, Newman TB, Hulley SB. Delayed effects of the military draft on mortality: a randomized natural experiment. N Engl J Med. 1986;314(10):620-624. doi:10.1056/NEJM198603063141005 67. Brooke BS, Goodney PP, Kraiss LW, Gottlieb DJ, Samore MH, Finlayson SRG. Readmission destination and risk of mortality after major surgery: an observational cohort study. Lancet. 2015;386(9996):884-895. doi:10.1016/ S0140-6736(15)60087-3 68. Garabedian LF, Chu P, Toh S, Zaslavsky AM, Soumerai SB. Potential bias of instrumental variable analyses for observational comparative effectiveness research. Ann Intern Med. 2014;161(2):131-138. doi:10.7326/M13-1887 69. Finks JF, Osborne NH, Birkmeyer JD. Trends in hospital volume and operative mortality for high-risk surgery. N Engl J Med. 2011;364(22):2128-2137. doi:10.1056/NEJMsa1010705 70. Urbach DR. Pledging to eliminate low-volume surgery. N Engl J Med. 2015;373(15):1388-1390. doi:10.1056/ NEJMp1508472 SUPPLEMENT. eMethods. Instrumental Variable Derivation eFigure 1. Distribution of the Instrumental Variable eFigure 2. Propensity Score Performance eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years eTable 1. Characteristics of the Sub-Analysis Cohort: Symptomatic (n=35 514) eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204) JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 14/14 Supplementary Online Content Columbo JA, Martinez-Camblor P, MacKenzie TA, et al. Comparing long-term mortality after carotid endarterectomy vs carotid stenting using a novel instrumental variable method for risk adjustment in observational time-to-event data. JAMA Netw Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 eMethod . Instrumental Variable Derivation eFigure 1. Distribution of the Instrumental Variable eFigure 2. Propensity Score Performance eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years eTable 1. Characteristics of the Sub-Analysis Cohort: Symptomatic (n=35 514) eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204) This supplementary material has been provided by the authors to give readers additional information about their work. © 2018 et al. JAMA Network Open. eMethod . Instrumental Variable Derivation We assume that the risk model has the form: ( | , , ) = ( )ex ( + + ) where X is the treatment, Z is a vector of measured confounders (see Table 1) and U is a ( ) potential unmeasured confounder. The target of inference is exp , the impact (in terms of a hazard ratio) of the treatment. In addition, we assume that the treatment assignment model has the form: = + W + Z + U + where W is the instrumental variable (IV) and is random error or white-noise. In our study, W is computed as the proportion of carotid endarterectomy (CEA) procedures out of the total CEA and carotid artery stenting (CAS) procedures performed over the past 12-months at the hospital where the patient underwent their procedure. Thus, = , where and denote the number of CEA and CAS procedures performed at the hospital in the 12-months prior to a given patient’s procedure. Because the IV is patient-specific, it has the appealing property of varying within a hospital across time and so can exploit variation in a hospital’s relative preference for CEA and CAS over time. To help validate the IV, the denominator of is included as a measured confounder in the Cox model. The two-stage residual inclusion frailty (2SRI-F) procedure estimates the treatment received by a patient via = + + . From this, we estimate the residuals, avoiding the estimation errors as R = X X = U + . © 2018 et al. JAMA Network Open. Finally, we add R as a covariate in the Cox model and account for the white noise from the estimated first-stage equation, the quantity , by also adding an individual (gaussian) frailty term. Assuming that all of the data are contained in a frame called DATA, the corresponding R code has the general form: # FIRST STAGE. MODEL ASSIGNMENT ESTIMATION RES<- lm(X~ W + age + gender + race + etnia + […], data=DATA)$residuals # SECOND STAGE. COX REGRESSION MODEL TSRF<- coxph(Surv(time,exitus)~ X + age + gender + race + etnia + […] + RES + frailty(1:nrow(DATA),dist="gaussian"), data=DATA) © 2018 et al. JAMA Network Open. eFigure 1. Distribution of the Instrumental Variable. CEA = Carotid Endarterectomy; CAS = Carotid Artery Stenting. © 2018 et al. JAMA Network Open. eFigure 2. Propensity Score Performance We used the traditional plot of the extent of the imbalance between the pre and post matched samples as a means to illustrate the performance of the propensity score matching procedure. PS = propensity score; TIA = transient ischemic attack; HTN = hypertension; CAD = coronary artery disease; CHF = congestive heart failure; COPD = chronic obstructive pulmonary disease; CEA = carotid endarterectomy. © 2018 et al. JAMA Network Open. eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years © 2018 et al. JAMA Network Open. eTable 1. Characteristics of the Sub-analysis Cohort: Symptomatic (n=35 514)* Variable All Patients p Propensity-Matched p CEA n=28,689 CAS n=6,825 CEA n=6,418 CAS n=6,418 Demographics Age mean (SD), 70.2 (9.4) 68.7 (10.8) <0.001 69.1 (9.8) 69.9 (10.7) 0.39 years Male (%) 17,302 (60.3) 4,4169 (64.7) <0.001 4,133 (64.4) 4,127 (64.3) 0.93 Race White 26,492 (92.3) 6,103 (89.4) <0.001 5,765 (89.7) 5,757 (89.7) 0.83 Black 1,231 (4.3) 430 (6.3) <0.001 375 (5.8) 389 (6.1) 0.63 Other 963 (3.4) 292 (4.3) <0.001 278 (4.3) 272 (4.2) 0.83 Clinical Factors Elective 25,061 (87.3) 4,704 (70.2) <0.001 4,805 (74.9) 4,725 (73.6) 0.11 Symptomatic 28,689 (100) 6,825 (100) NA 6,418 (100) 6,418 (100) NA TIA or amaurosis 14,122 (49.2) 3,086 (45.2) <0.001 2,980 (46.4) 2,947 (45.9) 0.57 Stroke 14,567 (50.8) 3,739 (54.8) <0.001 3,438 (53.6) 3,471 (44.1) 0.57 Hypertension 24,487 (88.7) 6,045 (88.6) 0.72 5,723 (89.2) 5,691 (88.7) 0.38 Smoking History 21,662 (75.5) 5,128 (75.1) 0.53 4,933 (75.3) 4,834 (75.3) 1 Positive Stress 2,379 (8.3) 466 (6.8) <0.001 472 (7.3) 446 (6.9) 0.39 Test Coronary Disease 8,086 (28.1) 2,071 (30.3) <0.001 2004 (31.2) 1,962 (30.6) 0.43 Heart Failure 2,864 (10.0) 1,012 (14.8) <0.001 905 (14.1) 916 (14.3) 0.8 Diabetes 10,044 (35.0) 2,491 (36.5) 0.02 2,350 (36.6) 2,334 (36.4) 0.34 COPD 6,199 (21.6) 1,714 (25.1) <0.001 1,573 (24.5) 1,601 (24.9) 0.58 Renal 1,439 (5.0) 358 (5.2) 0.46 349 (5.4) 338 (5.4) 0.7 Insufficiency HD 332 (1.2) 14 (0.2) <0.001 20 (0.3) 14 (0.2) 0.39 Prior CEA 4,013 (13.9) 1,935 (28.4) <0.001 1,720 (26.8) 1,608 (25.1) 0.03 Medications Antiplatelet therapy Aspirin 23,851 (83.1) 5,751 (84.3) 0.03 5,279 (82.2) 5,377 (83.8) 0.02 P2y12 inhibitor 8,286 (28.9) 5,100 (74.7) <0.001 4,824 (75.2) 4,693 (73.1) 0.009 Beta-blocker 16,339 (56.9) 3,544 (51.9) <0.001 3,362 (52.4) 3,446 (53.7) 0.14 Statin 23,006 (80.2) 5,865 (78.6) 0.004 5,643 (77.5) 5,076 (77.7) 0.49 *Centers not performing at least 10 procedures on symptomatic patients in the year prior to analysis were excluded. For this reason the number of patients in the sub-analyses less slightly less than the number of patients in the combined analysis. Legend: SD, standard deviation; CEA, carotid endarterectomy; CAS, carotid artery stenting; TIA, transient ischemic attack; NA, not applicable; COPD, chronic obstructive pulmonary disease; HD, hemodialysis. © 2018 et al. JAMA Network Open. eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204)* Variable All Patients p Propensity-Matched p CEA CAS n=5,809 CEA n=5,737 CAS n=5,737 n=44,395 Demographics Age mean (SD), 70.3 (9.5) 69.5 (9.9) <0.001 69.5 (9.5) 69.5 (9.9) 0.65 years Male (%) 26,844 (60.5) 3,658 (63.0) <0.001 3,603 (62.8) 3,632 (63.3) 0.59 Race White 41,064 (92.5) 5,356 (92.2) 0.42 5,311 (92.6) 5,290 (92.2) 0.48 Black 1,853 (4.2) 259 (4.5) 0.33 224 (3.9) 257 (4.5) 0.14 Other 1,473 (3.3) 194 (3.3) 0.96 202 (3.5) 190 (3.3) 0.57 Clinical Factors Elective 39,763 (87.3) 5,399 (92.9) <0.001 5,328 (92.9) 5,329 (92.9) 1 Symptomatic 0 (0) 0 (0) NA 0 (0) 0 (0) NA TIA or amaurosis 0 (0) 0 (0) NA 0 (0) 0 (0) NA Stroke 0 (0) 0 (0) NA 0 (0) 0 (0) NA Hypertension 30,464 (88.9) 5,181 (89.9) 0.51 5,116 (89.2) 5,122 (89.3) 0.88 Smoking History 33,641 (75.8) 4,464 (76.8) 0.08 4,449 (77.5) 4,414 (76.9) 0.45 Positive Stress 3,537 (8.0) 514 (8.8) 0.02 505 (8.8) 506 (8.8) 1 Test Coronary Disease 12,485 (28.1) 2,054 (35.4) <0.001 2,017 (35.1) 2,022 (35.2) 0.94 Heart Failure 4,625 (10.4) 862 (14.8) <0.001 783 (13.6) 828 (14.4) 0.24 Diabetes 15,512 (34.9) 2,684 (35.9) 0.17 2,060 (35.9) 2,063 (36.0) 0.97 COPD 10,015 (22.6) 1,501 (25.8) <0.001 1,460 (25.4) 1,474 (25.7) 0.71 Renal 2,651 (6.0) 360 (6.2) 0.54 351 (6.1) 352 (6.1) 1 Insufficiency HD 599 (1.3) 11 (0.2) <0.001 10 (0.2) 11 (0.2) 1 Prior CEA 6,098 (13.7) 2,178 (37.5) <0.001 2,112 (36.8) 2,106 (36.7) 0.92 Medications Antiplatelet therapy Aspirin 36,702 (82.7) 5,063 (87.2) <0.001 4,923 (85.8) 4,991 (87.0) 0.07 P2y12 inhibitor 12,800 (28.8) 4,491 (77.3) <0.001 4,435 (77.3) 4,419 (77.0) 0.74 Beta-blocker 25,294 (57.0) 3,422 (58.9) 0.005 3,389 (59.1) 3,382 (58.9) 0.91 Statin 35,398 (79.7) 4,700 (80.9) 0.04 4,657 (81.2) 4,657 (81.2) 1 *Centers not performing at least 10 procedures on symptomatic patients in the year prior to analysis were excluded. For this reason the number of patients in the sub-analyses less slightly less than the number of patients in the combined analysis. Legend: SD, standard deviation; CEA, carotid endarterectomy; CAS, carotid artery stenting; TIA, transient ischemic attack; NA, not applicable; COPD, chronic obstructive pulmonary disease; HD, hemodialysis. © 2018 et al. JAMA Network Open. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JAMA Network Open American Medical Association

Comparing Long-term Mortality After Carotid Endarterectomy vs Carotid Stenting Using a Novel Instrumental Variable Method for Risk Adjustment in Observational Time-to-Event Data

Download PDF
22 pages

Loading next page...
 
/lp/american-medical-association/comparing-long-term-mortality-after-carotid-endarterectomy-vs-carotid-O0NEiRWhBb

References (154)

Publisher
American Medical Association
Copyright
Copyright 2018 Columbo JA et al. JAMA Network Open.
eISSN
2574-3805
DOI
10.1001/jamanetworkopen.2018.1676
Publisher site
See Article on Publisher Site

Abstract

Key Points Question Can a novel instrumental IMPORTANCE Choosing between competing treatment options is difficult for patients and clinicians variable method designed for time- when results from randomized and observational studies are discordant. Observational real-world dependent outcomes more accurately studies yield more generalizable evidence for decision making than randomized clinical trials, but determine the relative long-term unmeasured confounding, especially in time-to-event analyses, can limit validity. mortality after carotid endarterectomy vs carotid artery stenting? OBJECTIVES To compare long-term survival after carotid endarterectomy (CEA) and carotid artery Findings In this registry-based, stenting (CAS) in real-world practice using a novel instrumental variable method designed for time- multicenter cohort study of 86 017 to-event outcomes, and to compare the results with traditional risk-adjustment models used in patients, the hazard ratio of long-term observational research for survival analyses. mortality for carotid endarterectomy vs carotid artery stenting was 0.83 (95% DESIGN, SETTING, AND PARTICIPANTS A multicenter cohort study was performed. The Vascular CI, 0.70-0.98) using instrumental Quality Initiative, an observational quality improvement registry, was used to compare long-term variable analysis, compared with 0.69 mortality after CEA vs CAS. The study included 86 017 patients who underwent CEA (n = 73 312) or (95% CI, 0.65-0.74) using a traditional CAS (n = 12 705) between January 1, 2003, and December 31, 2016. Patients were followed up for Cox regression model. long-term mortality assessment by linking the registry data to Medicare claims. Medicare claims data were available through September 31, 2015. Meaning Results from this instrumental variable method show that the survival EXPOSURE Procedure type (CEA vs CAS). advantage conferred by carotid endarterectomy is more modest than MAIN OUTCOMES AND MEASURES The hazard ratios (HRs) of all-cause mortality using suggested by traditional adjustment unadjusted, adjusted, propensity-matched, and instrumental variable methods were examined. The methods, aligning with results from instrumental variable was the proportion of CEA among the total carotid procedures randomized clinical trials. (endarterectomy and stenting) performed at each hospital in the 12 months before each patient’s index operation and therefore varies over the study period. Invited Commentary RESULTS Participants who underwent CEA had a mean (SD) age of 70.3 (9.4) years compared with Supplemental content 69.1 (10.4) years for CAS, and most were men (44 191 [60.4%] for CEA and 8117 [63.9%] for CAS). The Author affiliations and article information are observed 5-year mortality was 12.8% (95% CI, 12.5%-13.2%) for CEA and 17.0% (95% CI, listed at the end of this article. 16.0%-18.1%) for CAS. The unadjusted HR of mortality for CEA vs CAS was 0.67 (95% CI, 0.64-0.71), and Cox-adjusted and propensity-matched HRs were similar (0.69; 95% CI, 0.65-0.74 and 0.71; 95% CI, 0.65-0.77, respectively). These findings are comparable with published observational studies of CEA vs CAS. However, the association between CEA and mortality was more modest when estimated by instrumental variable analysis (HR, 0.83; 95% CI, 0.70-0.98), a finding similar to data reported in randomized clinical trials. (continued) Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 1/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Abstract (continued) CONCLUSIONS AND RELEVANCE The study found a survival advantage associated with CEA over CAS in unadjusted and Cox-adjusted analyses. However, this finding was more modest when using an instrumental variable method designed for time-to-event outcomes for risk adjustment. The instrumental variable-based results were more similar to findings from randomized clinical trials, suggesting this method may provide less biased estimates of time-dependent outcomes in observational analyses. JAMA Network Open. 2018;1(5):e181676. Corrected on October 12, 2018. doi:10.1001/jamanetworkopen.2018.1676 Introduction Randomized clinical trials, which have internal validity and test efficacy under carefully designed 1-4 study conditions, produce findings that are often widely accepted. When results of observational studies are concordant with randomized clinical trials, clear messages emerge for patients, clinicians, 5-8 and payers to guide treatment decisions. Concordance of results is particularly important when assessing a long-term, time-to-event outcome such as mortality, as this suggests that the findings 9-11 seen in randomized clinical trials will be durable in clinical practice. Treatment decisions are more difficult, however, when the results of randomized clinical trials and observational studies are discordant. For example, for patients and clinicians considering carotid endarterectomy (CEA) or carotid artery stenting (CAS), 2 competing treatments to prevent stroke from carotid artery stenosis, long-term survival after the procedure remains a matter of debate. While randomized clinical trials have shown no statistically significant difference in mortality between the 2 procedures, observational evidence suggests survival following endarterectomy is 12-17 superior. Several potential explanations exist for this type of discordance. For example, treatment regimens and effects in randomized clinical trials may not reflect clinical practice, thereby limiting 19,20 generalizability. This limitation of randomized clinical trials as well as their high cost and complexity make observational studies an attractive alternative. However, risk adjustment for confounding in observational data remains challenging. While methods such as Cox proportional hazards regression and propensity score matching have been developed to adjust observational time-to-event data for measured confounding, the possibility that unmeasured or even unmeasurable confounding persists in observational analyses is an important concern faced by 11,21-24 patients and clinicians. Unmeasured confounding is of particular importance in patients with peripheral arterial disease considering invasive vs minimally invasive options, where surgeon selection bias and patient fitness for surgery have been shown to have an important association with clinical outcomes after aortic aneurysm repair. Selection bias and unmeasured confounding are likely to also occur in patients with carotid artery disease, where the decision to choose an invasive vs minimally invasive procedure is influenced by many factors. Instrumental variable analysis is a procedure unique in its ability to account for unmeasured confounding, and this method has been applied to linear and logistic regression models to evaluate 26,27 outcomes that are not time dependent. To date, adaptation of instrumental variable methods in areas of medicine such as cardiovascular disease that often examine time-to-event data using Cox 22,28-31 regression as the standard analytic tool has been limited. We recently developed an instrumental variable procedure for use with the Cox model and have shown that it outperforms the 24,32 traditional Cox model and 2-stage approaches that include the Cox model. We apply this procedure to adjust for suspected unmeasured confounding when comparing individuals’ long-term mortality between 2 competing treatments for carotid revascularization in a large observational data set. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 2/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Methods Data Sources Our analyses use data derived from the Vascular Quality Initiative registry, a national quality improvement registry that captures data on vascular procedures from more than 400 hospitals and practices across the United States and Canada. Patients and procedures entered in the registry 34,35 were linked to the Medicare Denominator File for mortality assessment. This database includes patient-level information on baseline demographics, comorbid conditions, presenting neurologic symptoms, operative management, and mortality on patients who underwent CEA and CAS. Data from the Vascular Quality Initiative were available from January 1, 2003, to December 31, 2016. Medicare data were available until September 2015. All data were collected under the auspices of an Agency for Healthcare Research and Quality–designated Patient Safety Organization and were deidentified. Our study was approved by the Center for the Protection of Human Subjects at Dartmouth; a waiver of participant consent was obtained. This study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. Primary Exposure and Outcome The primary exposure of interest was procedure type (CEA vs CAS). These procedures represent the most common methods of carotid revascularization in current practice. Patients receiving more than 1 procedure type in the same day were assigned to the first procedure they received. Patients receiving repeated revascularization procedures during follow-up were assigned to the index procedure. The primary outcome was all-cause mortality. This was assessed for all patients in the registry using the Social Security Death Index. Patients were assessed from the time of their index procedure until death, and censored only at the end of their known follow-up period. Patients eligible for Medicare in the registry were linked to their respective Medicare claims file, analyzed through the end of September 2015. Successful linking was obtained in 92% and 91% of eligible patients who underwent CEA or CAS, respectively. Statistical Analysis We performed a sensitivity analysis with results stratified based on the presence or absence of focal neurologic symptoms (symptomatic vs asymptomatic) at the time of presentation. We used clinical variables from the Vascular Quality Initiative to group patients as symptomatic or asymptomatic. We defined patients as symptomatic if they had a documented stroke, transient ischemic attack, or other ischemic neurologic symptoms at the time of hospitalization for their index procedure. We reported unadjusted mortality as absolute and relative frequencies where appropriate. We calculated the hazard ratio (HR) of mortality for CEA vs CAS using 4 methods of estimation: unadjusted, Cox regression adjusted, propensity matched, and our instrumental variable procedure designed for time-to-event data. We applied these to the overall cohort, as well as in the sensitivity analysis based on the presence of focal neurologic symptoms at the time of revascularization. Statistical tests were 2-sided with P < .05 considered significant. All statistical analyses were performed with R, version 3.3.2 (R Foundation for Statistical Computing). Unadjusted and Adjusted Mortality We calculated unadjusted mortality rates using Kaplan-Meier estimation. We then used Cox regression to estimate the HR of postoperative mortality for CEA vs CAS to account for observed 24,38 confounding. Summary statistics for the confounding variables in the statistical models are noted in Table 1. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 3/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Propensity-Matched Analysis 23,39 We created a propensity-matched cohort balanced in baseline covariates. Using the observed covariates in Table 1, we created a logistic regression model in which the dependent variable was the treatment exposure (CEA vs CAS). Next, we calculated the fitted probability of CEA, known as the propensity score, for each patient. We then matched patients undergoing CEA to those undergoing CAS. We compared mortality between patients who underwent CEA vs CAS in the matched cohort. To account for the censoring, we applied Cox regression to our propensity-matched cohort to estimate the HR of mortality for CEA vs CAS. Instrumental Variable Analysis Our instrument was the proportion of CEA among the total number of carotid revascularization procedures (CEA and CAS) performed at each hospital in the 12 months prior to the index operation. We excluded hospitals not performing at least 10 revascularization procedures in the year prior to the index operation. In the presenting symptoms sensitivity analysis (patients presenting with focal neurologic symptoms vs not), we further excluded hospitals not performing at least 10 carotid revascularization procedures for each indication. For this reason, the number of patients included in the overall analysis slightly exceeds the total number of patients included in the sensitivity analysis. The instrumental variable procedure identifies patients who would have undergone CEA at some institutions and CAS at others based on the value of the instrument alone and not on patient characteristics. If patients choose hospitals based on convenience, or at least based only on Table 1. Cohort Characteristics All Patients (N = 86 017) Propensity Matched (n = 24 680) Variable CEA (n = 73 312) CAS (n = 12 705) P Value CEA (n = 12 340) CAS n = (12 340) P Value Demographics Age, mean (SD), y 70.3 (9.4) 69.1 (10.4) <.001 69.3 (9.7) 69.2 (10.4) .44 Male, No. (%) 44 191 (60.4) 8117 (63.9) <.001 7843 (63.6) 7867 (63.7) .76 Race, No. (%) White 67 768 (92.4) 11 524 (90.7) <.001 11 174 (90.6) 11 198 (90.8) .65 Black 3099 (4.2) 694 (5.5) <.001 681 (5.5) 672 (5.4) .82 Other 2445 (3.4) 487 (3.8) .005 485 (3.9) 470 (3.8) .64 Clinical factors, No. (%) Elective 64 022 (87.3) 10 252 (80.7) <.001 10 014 (81.2) 10 042 (81.4) .66 Symptomatic 28 836 (39.3) 6863 (54.0) <.001 6333 (51.3) 6519 (52.8) .02 TIA or amaurosis 14 200 (19.4) 3106 (24.4) <.001 2885 (23.4) 2970 (24.1) .21 Stroke 14 636 (20.0) 3757 (29.6) <.001 3448 (27.9) 3549 (28.8) .16 Hypertension 65 128 (88.8) 11 292 (88.9) .90 11 032 (89.4) 10 981 (89.0) .31 Smoking history 55 476 (75.7) 9643 (75.9) .59 9393 (76.1) 9361 (75.9) .64 Positive stress test 5937 (8.1) 988 (7.8) .23 977 (7.9) 973 (7.9) .94 Coronary disease 20 643 (28.2) 4150 (32.7) <.001 4153 (33.6) 4023 (32.6) .08 Heart failure 7512 (10.2) 1883 (14.8) <.001 1792 (14.5) 1782 (14.4) .87 Diabetes 25 637 (35.0) 4595 (36.2) <.001 4522 (36.6) 4451 (36.1) .88 COPD 16 261 (22.2) 3233 (25.4) <.001 3181 (25.8) 3107 (25.2) .29 Renal insufficiency 4101 (5.6) 722 (5.7) .70 722 (5.9) 702 (5.7) .60 Hemodialysis 937 (1.3) 25 (0.2) <.001 34 (0.3) 25 (0.2) .30 Prior CEA 10 132 (13.8) 4132 (32.5) <.001 3908 (31.2) 3803 (30.8) .15 Medications Antiplatelet therapy, No. (%) Aspirin 60 744 (82.9) 10 877 (85.6) <.001 10 363 (84.0) 10 545 (85.5) .001 P2Y12 inhibitor 21 163 (28.9) 9646 (75.9) <.001 9310 (75.4) 9281 (75.2) .68 β-Blocker 41 759 (57.0) 7003 (55.1) <.001 7005 (56.8) 6853 (55.5) .04 Statin 58 588 (79.9) 10 120 (79.7) .50 9833 (79.7) 9861 (79.9) .67 Abbreviations: CAS, carotid artery stenting; CEA, carotid endarterectomy; COPD, chronic obstructive pulmonary disease; TIA, transient ischemic attack. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 4/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method observed factors, then where a patient seeks treatment emulates a randomized encouragement design in which assignment to a hospital with a historical precedent for performing a high proportion of CEA randomly exposes the patient to a greater likelihood of undergoing CEA. The estimation of treatment effects using instrumental variables is well developed for linear or logistic regression 40,42,43 models. However, Cox proportional hazard models examining time-to-event outcomes selects on survivors over the course of follow-up, which is problematic for standard methods of 24,44-46 instrumental variable identification. Therefore, we used the new instrumental variable estimator for the Cox proportional hazards model to simultaneously deal with the problems of unmeasured confounding and censoring of the outcome. We used this new procedure to examine the HR for all-cause mortality after CEA vs CAS by including both the instrument and all known confounding variables described in Table 1. The mean (SD) value of the instrumental variable was 0.89 (0.12) for patients undergoing CEA and 0.65 (0.29) for patients undergoing CAS (P < .001). Further details on derivation of the instrumental variable and its distribution can be found in the eMethods and eFigure 1 in the Supplement. Instrument Assessment We measured the strength of our instrument by determining if increasing levels of the instrument were associated with changing levels of the exposure. This is reported using the F statistic, for which a value greater than 10 traditionally indicates acceptable strength. The F statistic assesses the instrument’s ability to show association with the exposure received beyond the effect of any covariates that are adjusted for the survival model. Results Cohort Characteristics We studied 86 017 patients who underwent carotid revascularization (CEA, n = 73 312; CAS, n = 12 705) from January 1, 2003, to December 31, 2016. Mean follow-up was 3.0 (SD, 2.4 years; range, <0.1-14.3 years; median [interquartile range] follow-up, 2.5 [1.3-4.0] years), yielding the equivalent of 259 700 person-years for analysis. Vital status was known for 75.0% of patients who were eligible (procedure date, 2011 or earlier). Compared with patients who underwent CEA, those who underwent CAS tended to be younger (mean [SD] age, 70.3 [9.4] vs 69.1 [10.4] years, respectively), were more likely to be male (44 191 [60.4%] vs 8117 [63.9%], respectively), and were more likey to have an urgent procedure (2453 [19.3%] vs 9290 [12.7%], respectively) (Table 1). More than 89% of patients were receiving some form of antiplatelet therapy, and more than 75% were receiving a statin. Characteristics of the sensitivity analysis cohorts (symptomatic and asymptomatic patients) were similar (eTables 1 and 2 in the Supplement). There were several clinically meaningful differences between patients who underwent CEA and those who underwent CAS. Approximately one-third of patients undergoing CEA underwent the procedure because of focal neurologic symptoms, compared with more than half of patients treated with CAS. Patients who underwent CAS were also more likely to have several chronic comorbid conditions, including coronary artery disease, heart failure, and pulmonary disease. Patients who underwent CAS were also more likely to have previously undergone carotid surgery. Given that several differences existed in the characteristics between patients treated with CEA and CAS, we created a propensity-matched cohort for analysis. The propensity-matched cohort consisted of 12 340 matched pairs of patients and was well balanced in baseline characteristics apart from a small difference in aspirin use (84.0% in the CEA group and 85.5% in the CAS group; P = .001), β-blocker prescription (56.8% in the CEA group and 55.5% in the CAS group; P = .04), and in the proportion of procedures performed for symptomatic stenosis (51.3% in the CEA group and 52.8% in the CAS group; P = .02). A graphical representation of the performance of the propensity score matching can be found in eFigure 2 in the Supplement. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 5/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Unadjusted, Cox-Adjusted, and Propensity-Matched Mortality by Procedure Type The unadjusted Kaplan-Meier estimate of all-cause mortality at 5 years for CEA was 12.8% (95% CI, 12.5%-13.2%) and for CAS was 17.0% (95% CI, 16.0%-18.1%; log rank, P < .001). At 10 years after the procedure, estimated mortality was 27.3% (95% CI, 26.3%-27.3%) for CEA and 27.4% (23.9%-30.7%) for CAS (log rank, P < .001; eFigure 3 in the Supplement). Sensitivity analysis by the presence of neurologic symptoms at the time of revascularization demonstrated similar findings (Figure 1). The unadjusted HR of all-cause mortality for CEA vs CAS was 0.67 (95% CI, 0.64-0.71) (Table 2). A Cox proportional hazards model adjusting for differences in patient characteristics showed a similar Figure 1. Kaplan-Meier Estimated Mortality, Overall and by Presenting Symptoms All patients CAS CEA Log rank P <.001 0 1 2 3 4 5 Time, y No. at risk CAS 12 705 10 000 6854 4466 2578 1185 CEA 73 312 60 901 44 974 30 440 19 139 11 362 B Symptomatic CAS Log rank P <.001 CEA 0 1 2 3 4 5 Time, y No. at risk CAS 6825 5210 3516 2138 1190 533 CEA 26 689 23 822 17 319 11 653 7306 4382 C Asymptomatic CAS CEA Log rank P <.001 0 1 2 3 4 5 Time, y No. at risk CAS indicates carotid artery stenting; CEA, carotid CAS 5809 4731 3289 2292 1363 635 CEA 44 395 36 884 27 481 18 655 11 740 6920 endarterectomy. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 6/14 Mortality, % Mortality, % Mortality, % JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method association (HR, 0.69; 95% CI, 0.65-0.74), further suggesting that CEA was associated with a survival advantage. The propensity-matched cohort also revealed a survival advantage associated with CEA (HR, 0.71; 95% CI, 0.65-0.77). Sensitivity analysis by the presence of neurologic symptoms before carotid revascularization continued to show a statistically significant association (Table 2 and 16,48 Figure 2). Instrumental Variable–Adjusted Mortality by Procedure Type The instrument, each individual hospital’s 12-month prior proportion of CEA procedures, demonstrated a very strong association with the type of carotid procedure performed (F = 18 631). Applying our instrumental variable procedure to all-cause mortality revealed that patients selected for CEA had a more modest survival advantage (HR, 0.83; 95% CI, 0.70-0.98) than was suggested by results of our other analytic methods. These results are similar to the findings of published randomized clinical trials (Figure 2). Similar results were obtained by our instrumental variable approach in a sensitivity analyses stratified by presenting symptoms, although the association was more pronounced in those who were symptomatic. The HRs for those with symptoms changed by an absolute 17% to 19% between traditional statistical methods and our instrumental variable model, compared with an absolute change of 11% to 14% in those who were asymptomatic (Table 2). Table 2. Mortality HRs for Carotid Endarterectomy vs Carotid Stenting HR (95% CI) No. of Mortality Events/Total No. Variable of Patients Crude Adjusted Propensity Matched Instrumental Variable Overall CEA 6600/73 312 0.67 (0.64-0.71) 0.69 (0.65-0.74) 0.71 (0.65-0.77) 0.83 (0.70-0.98) CAS 1405/12 705 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Symptomatic CEA 2559/28 689 0.61 (0.46-0.66) 0.61 (0.56-0.67) 0.59 (0.53-0.66) 0.78 (0.61-0.99) CAS 786/6825 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Asymptomatic CEA 4017/44 395 0.76 (0.70-0.83) 0.79 (0.72-0.87) 0.79 (0.71-0.90) 0.90 (0.70-1.14) CAS 607/5809 1 [Reference] 1 [Reference] 1 [Reference] 1 [Reference] Abbreviations: CAS, carotid artery stenting; CEA, carotid endarterectomy; HR, Event and total cohort numbers are different for the propensity-matched analysis. hazard ratio. Figure 2. Hazard Ratios (HRs) of Mortality for Carotid Endarterectomy vs Carotid Stenting A All patients B Symptomatic C Asymptomatic Favors Favors Favors Favors Favors Favors Source endarterectomy stenting endarterectomy stenting endarterectomy stenting Relevant RCT results Brott et al, 2016 (CREST; n = 2502) Rosenfield et al, 2016 (ACT 1; n = 1453) VQI-Medicare observational analyses Unadjusted model Adjusted model Propensity matched Instrumental variable 0.5 1.0 2 0.5 1.0 2 0.5 1.0 2 HR (95% CI) HR (95% CI) HR (95% CI) The CREST trial outcome represented is long-term stroke or periprocedural myocardial Revascularization Endarterectomy vs Stenting Trial; RCT, randomized clinical trial; VQI, infarction, stroke, or death. ACT 1 indicates Asymptomatic Carotid Trial; CREST, Carotid Vascular Quality Initiative. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 7/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method Discussion In this observational study, unadjusted, adjusted, and propensity-matched models of long-term mortality all demonstrated that treatment with CEA was associated with a survival benefit relative to treatment with CAS. These results are comparable with published observational reports but conflict with randomized clinical trials, which suggest survival is similar following the 2 competing treatment 12-17,49,50 options. Our instrumental variable method designed for risk adjustment of time-to-event data estimated a more modest association with long-term mortality, a finding consistent with the results of randomized clinical trials. These findings were robust to a sensitivity analysis by the presence of focal neurologic symptoms. This method, which accounts for both measured and unmeasured confounding in observational time-to-event analyses, represents an advance for investigators evaluating long-term outcomes, especially when considering clinical questions where randomized clinical trials are not possible or would be prohibitively expensive or when use of real- world evidence would be advantageous. Discordance between randomized clinical trials and observational studies is neither new nor 52-57 uncommon. For example, differences in the efficacy of vitamin E and hormone replacement therapy for the prevention of heart disease as well as antioxidant therapy for cancer represent important examples where conflicting results from randomized clinical trials and observational 52-56 studies have affected evidence-based treatment decisions. Meta-analyses examining the relative findings of randomized and observational studies suggest that observational studies tend to generate a larger treatment effect, and these differences may be further potentiated when assessing 58-64 long-term outcomes such as mortality. However, this is not always the case; a recent Cochrane review estimated that treatment effects were similar between randomized clinical trials and observational studies (pooled ratio of odds ratios, 1.04; 95% CI, 0.89-1.21). These contradictory results highlight how challenging it can be for patients and clinicians to interpret observational study results, especially if the direction of bias cannot be foreseen. In the example of discordance used in our analysis, patients with CAS, the Asymptomatic Carotid Trial (ACT 1) reported no statistically significant difference in 5-year mortality after CEA vs CAS, and the Carotid Revascularization Endarterectomy vs Stenting Trial (CREST) reported no statistically significant difference at 10 years in the composite outcome of long-term risk of stroke or perioperative myocardial infarction, stroke, or death. Despite these randomized clinical trials, several large observational studies have documented inferior outcomes for stenting overall, especially in subgroup analyses of symptomatic patients, which may bias physicians and patients 12-15,50 away from choosing stenting as a procedural option. Our findings, which suggest that there is a modest association between survival and CEA, provide important granular detail to help inform this management decision. The design and execution of a randomized clinical trial that provides true, unbiased estimates is a very difficult task, as there are several threats to both the validity and generalizability of their results. In some cases, clinical trial participants may not be representative of the target 18-20 population. In others, heterogeneous treatment effects may impede the ability of physicians to 18-20 parse out which patients may benefit most from an intervention. In addition, intention-to-treat estimators used in randomized clinical trials may be biased toward the null if noncompliance is considerable. These limitations to many contemporary randomized clinical trials highlight the utility of observational studies where real-world evidence can be used, provided that adequate adjustment for confounding can be performed. An analytic technique capable of better risk adjustment for unmeasured confounding would improve the reliance that could be placed on results from observational studies. Such a technique would allow real-world observational data to more consistently reflect the true outcome of treatment independent of confounding. Our instrumental variable procedure was specifically designed to be used to analyze time-to-event outcomes. While determining a suitable instrument JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 8/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method may be difficult in some settings, there appears to be few disadvantages in applying this procedure 22,47,66,67 to observational questions with time-dependent outcomes such as mortality. While the findings of our instrumental variable analyses gave results that are similar to those in randomized clinical trials, this may not be the case when applied to other clinical scenarios. The importance of observational data is that it documents results from clinical practice, outside of the confines of randomized clinical trials. Observational studies often include a much broader patient population with treatment-effect heterogeneity than are found in randomized clinical trials. In these situations, results from instrumental variable analyses may be different than those found in randomized clinical trials and may better represent the results that can be expected when an intervention is incorporated into clinical practice. Application of instrumental variables to time-to- event data therefore represents an important step forward in the evaluation of interventions in contemporary practice. Limitations Our study had limitations. First, it is not possible to truly know whether our instrumental variable balanced all unmeasured confounding. However, our sensitivity analyses by the presence of neurologic symptoms are reassuring. One would anticipate that unmeasured confounding would have a greater impact on the symptomatic analysis as patients in this subgroup are frequently sicker and thus are at higher risk for clinician selection bias to play a substantial role in the treatment decision. An instrument that accounts for unmeasurable confounding would change the effect size to a greater extent in these patients, and this was noted in our analyses. Second, we did not examine stroke-free survival as our primary outcome because of the heterogeneity in stroke assessment methods across the sites in our observational registry, an issue not encountered when examining survival as an outcome. Third, while 5-year vital status was known in 75.0% of patients who were eligible, many patients were not eligible for this assessment because of the date of their procedure (after 2011). Changes over time in both practice patterns and procedural competency may have an impact on the HR of mortality between the 2 procedures. However, findings remained consistent among patients who had their operations in earlier years where the longest follow-up was possible. Therefore, we feel that our estimates reported herein are an accurate reflection of long-term mortality after CEA vs CAS. In addition, instrumental variables must satisfy three conditions: first, they must be associated with the treatment exposure; second, an instrument must have no relationship to the outcome except through the effect on the exposure; and third, there must be no variables that affect both the instrument and the outcome. Our F statistic demonstrated that our instrument was strongly associated with the treatment exposure, thereby satisfying the first condition. It is not possible to prove whether an instrument is unrelated to an outcome. However, we required that a center perform at least 10 CEA or CAS procedures in the prior year to have patients included in the instrumental variable analysis to limit the possibility that proportion of procedures 69,70 performed could be related to postoperative mortality. Conclusions Using a novel instrumental variable method designed for time-to-event data, we found only a modest difference in long-term mortality after CEA vs CAS, a result that is comparable with recent randomized clinical trials. These similarities provide evidence that results from our instrumental variable procedure are more closely aligned with the true relative long-term mortality between the 2 revascularization procedures than incumbent methods for analyzing observational data. This method, which allows instruments to be used for risk adjustment with the widely used Cox regression model, may improve the validity of results for time-dependent outcomes for clinical questions where randomized clinical trials are not possible or would be prohibitively expensive or when use of real-world evidence would be advantageous. JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 9/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method ARTICLE INFORMATION Accepted for Publication: June 7, 2018. Published: September 7, 2018. doi:10.1001/jamanetworkopen.2018.1676 Correction: This article was corrected on October 12, 2018, to fix errors in the title and Figure 2. Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2018 Columbo JA et al. JAMA Network Open. Corresponding Author: A. James O’Malley, PhD, The Dartmouth Institute for Health Policy and Clinical Practice, One Medical Center Drive, Lebanon, NH 03756 (james.omalley@dartmouth.edu). Author Affiliations: The Dartmouth Institute for Health Policy and Clinical Practice, Lebanon, New Hampshire (Columbo, Martinez-Camblor, MacKenzie, Staiger, Kang, Goodney, O’Malley); Section of Vascular Surgery, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire (Columbo, Goodney); Department of Biomedical Data Science, Geisel School of Medicine, Lebanon, New Hampshire (MacKenzie, O’Malley); Department of Economics, Dartmouth College, Hanover, New Hampshire (Staiger). Author Contributions: Drs O’Malley and Martinez-Camblor had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: All authors. Acquisition, analysis, or interpretation of data: Columbo, Martinez-Camblor, Mackenzie, Staiger, Goodney, O’Malley. Drafting of the manuscript: Columbo, Goodney, O’Malley. Critical revision of the manuscript for important intellectual content: All authors. Statistical analysis: Martinez-Camblor, Mackenzie, Staiger, O’Malley. Obtained funding: O’Malley. Administrative, technical, or material support: Columbo, Kang, Goodney, O’Malley. Supervision: Columbo, O’Malley. Conflict of Interest Disclosures: Dr Staiger reported grants from the Patient-Centered Outcomes Research Institute during the conduct of the study; and personal fees, nonfinancial support, and grants from ArborMetrix, Inc outside the submitted work. Dr O’Malley reported receiving grants from the Patient-Centered Outcomes Research Institute and the National Institutes of Health during the conduct of the study. No other disclosures were reported. Funding/Support: This work was supported by Patient-Centered Outcomes Research Institute award ME-1503- 28261. Additional support was provided by the US Food and Drug Administration (U01-FD005478), the National Institute on Aging (PO1-AG019783), and the National Institutes of Health Common Fund (U01-AG046830). Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Disclaimer: All statements in this article, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute, its Board of Governors, or Methodology Committee. REFERENCES 1. Barnett HJM, Taylor DW, Haynes RB, et al; North American Symptomatic Carotid Endarterectomy Trial Collaborators. Beneficial effect of carotid endarterectomy in symptomatic patients with high-grade carotid stenosis. N Engl J Med. 1991;325(7):445-453. doi:10.1056/NEJM199108153250701 2. Endarterectomy for asymptomatic carotid artery stenosis: executive committee for the Asymptomatic Carotid Atherosclerosis Study. JAMA. 1995;273(18):1421-1428. doi:10.1001/jama.1995.03520420037035 3. Halliday A, Mansfield A, Marro J, et al; MRC Asymptomatic Carotid Surgery Trial Collaborative Group. Prevention of disabling and fatal strokes by successful carotid endarterectomy in patients without recent neurological symptoms: randomised controlled trial. Lancet. 2004;363(9420):1491-1502. doi:10.1016/S0140-6736(04) 16146-1 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 10/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 4. Brott TG, Halperin JL, Abbara S, et al; American College of Cardiology Foundation; American Stroke Association; American Association of Neurological Surgeons; American College of Radiology; American Society of Neuroradiology; Congress of Neurological Surgeons; Society of Atherosclerosis Imaging and Prevention; Society for Cardiovascular Angiography and Interventions; Society of Interventional Radiology; Society of NeuroInterventional Surgery; Society for Vascular Medicine; Society for Vascular Surgery. 2011 ASA/ACCF/AHA/ AANN/AANS/ACR/ASNR/CNS/SAIP/SCAI/SIR/SNIS/SVM/SVS guideline on the management of patients with extracranial carotid and vertebral artery disease: executive summary: a report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines, and the American Stroke Association, American Association of Neuroscience Nurses, American Association of Neurological Surgeons, American College of Radiology, American Society of Neuroradiology, Congress of Neurological Surgeons, Society of Atherosclerosis Imaging and Prevention, Society for Cardiovascular Angiography and Interventions, Society of Interventional Radiology, Society of NeuroInterventional Surgery, Society for Vascular Medicine, and Society for Vascular Surgery. Circulation. 2011;124(4):489-532. doi:10.1161/CIR.0b013e31820d8d78 5. Manson JE, Stampfer MJ, Colditz GA, et al. A prospective study of aspirin use and primary prevention of cardiovascular disease in women. JAMA. 1991;266(4):521-527. doi:10.1001/jama.1991.03470040085027 6. Kannel WB, Larson M. Long-term epidemiologic prediction of coronary disease: the Framingham experience. Cardiology. 1993;82(2-3):137-152. doi:10.1159/000175864 7. Baigent C, Blackwell L, Collins R, et al; Antithrombotic Trialists’ Collaboration. Aspirin in the primary and secondary prevention of vascular disease: collaborative meta-analysis of individual participant data from randomised trials. Lancet. 2009;373(9678):1849-1860. doi:10.1016/S0140-6736(09)60503-1 8. US Preventive Services Task Force. Aspirin for the prevention of cardiovascular disease: U.S. Preventive Services Task Force recommendation statement. Ann Intern Med. 2009;150(6):396-404. doi:10.7326/0003- 4819-150-6-200903170-00008 9. Lederle FA, Freischlag JA, Kyriakides TC, et al; Open Versus Endovascular Repair Veterans Affairs Cooperative Study Group. Outcomes following endovascular vs open repair of abdominal aortic aneurysm: a randomized trial. JAMA. 2009;302(14):1535-1542. doi:10.1001/jama.2009.1426 10. De Bruin JL, Baas AF, Buth J, et al; DREAM Study Group. Long-term outcome of open or endovascular repair of abdominal aortic aneurysm. N Engl J Med. 2010;362(20):1881-1889. doi:10.1056/NEJMoa0909499 11. Schermerhorn ML, Buck DB, O’Malley AJ, et al. Long-Term outcomes of abdominal aortic aneurysm in the medicare population. N Engl J Med. 2015;373(4):328-338. doi:10.1056/NEJMoa1405778 12. Giles KA, Hamdan AD, Pomposelli FB, Wyers MC, Schermerhorn ML. Stroke and death after carotid endarterectomy and carotid artery stenting with and without high risk criteria. J Vasc Surg. 2010;52(6):1497-1504. doi:10.1016/j.jvs.2010.06.174 13. McPhee JT, Schanzer A, Messina LM, Eslami MH. Carotid artery stenting has increased rates of postprocedure stroke, death, and resource utilization than does carotid endarterectomy in the United States, 2005. J Vasc Surg. 2008;48(6):1442-1450, 1450e1. doi:10.1016/j.jvs.2008.07.017 14. Wang FW, Esterbrooks D, Kuo YF, Mooss A, Mohiuddin SM, Uretsky BF. Outcomes after carotid artery stenting and endarterectomy in the Medicare population. Stroke. 2011;42(7):2019-2025. doi:10.1161/STROKEAHA.110. 15. Nolan BW, De Martino RR, Goodney PP, et al; Vascular Study Group of New England. Comparison of carotid endarterectomy and stenting in real world practice using a regional quality improvement registry. J Vasc Surg. 2012;56(4):990-996. doi:10.1016/j.jvs.2012.03.009 16. Rosenfield K, Matsumura JS, Chaturvedi S, et al; ACT I Investigators. Randomized trial of stent versus surgery for asymptomatic carotid stenosis. N Engl J Med. 2016;374(11):1011-1020. doi:10.1056/NEJMoa1515706 17. Bonati LH, Dobson J, Featherstone RL, et al; International Carotid Stenting Study Investigators. Long-term outcomes after stenting versus endarterectomy for treatment of symptomatic carotid stenosis: the International Carotid Stenting Study (ICSS) randomised trial. Lancet. 2015;385(9967):529-538. doi:10.1016/S0140-6736(14) 61184-3 18. Anglemyer A, Horvath HT, Bero L. Healthcare outcomes assessed with observational study designs compared with those assessed in randomized trials. Cochrane Database Syst Rev. 2014;(4):MR000034. 19. Rothwell PM. External validity of randomised controlled trials: “to whom do the results of this trial apply?” Lancet. 2005;365(9453):82-93. doi:10.1016/S0140-6736(04)17670-8 20. Dhruva SS, Redberg RF. Variations between clinical trial participants and Medicare beneficiaries in evidence used for Medicare national coverage decisions. Arch Intern Med. 2008;168(2):136-140. doi:10.1001/ archinternmed.2007.56 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 11/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 21. Grimes DA, Schulz KF. Bias and causal associations in observational research. Lancet. 2002;359(9302): 248-252. doi:10.1016/S0140-6736(02)07451-2 22. Stukel TA, Fisher ES, Wennberg DE, Alter DA, Gottlieb DJ, Vermeulen 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. doi:10.1001/jama.297.3.278 23. D’Agostino RB Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17(19):2265-2281. doi:10.1002/(SICI)1097-0258(19981015)17: 19<2265::AID-SIM918>3.0.CO;2-B 24. Cox DR. Regression models and life-tables. J R Stat Soc B. 1972;34(2):187-220. http://www.jstor.org/ stable/2985181. Accessed September 1, 2017. 25. De Martino RR, Brooke BS, Robinson W, et al. Designation as “unfit for open repair” is associated with poor outcomes after endovascular aortic aneurysm repair. Circ Cardiovasc Qual Outcomes. 2013;6(5):575-581. doi:10. 1161/CIRCOUTCOMES.113.000303 26. Hernán MA, Robins JM. Instruments for causal inference: an epidemiologist’s dream? Epidemiology. 2006;17 (4):360-372. doi:10.1097/01.ede.0000222409.00878.37 27. Terza JV, Basu A, Rathouz PJ. Two-stage residual inclusion estimation: addressing endogeneity in health econometric modeling. J Health Econ. 2008;27(3):531-543. doi:10.1016/j.jhealeco.2007.09.009 28. MacKenzie TA, Tosteson TD, Morden NE, Stukel TA, O’Malley AJ. Using instrumental variables to estimate a Cox’s proportional hazards regression subject to additive confounding. Health Serv Outcomes Res Methodol. 2014; 14(1-2):54-68. doi:10.1007/s10742-014-0117-x 29. Aalen OO, Cook RJ, Røysland K. Does Cox analysis of a randomized survival study yield a causal treatment effect? Lifetime Data Anal. 2015;21(4):579-593. doi:10.1007/s10985-015-9335-y 30. Li J, Fine J, Brookhart A. Instrumental variable additive hazards models. Biometrics. 2015;71(1):122-130. doi: 10.1111/biom.12244 31. Martinussen T, Nørbo Sørensen D, Vansteelandt S. Instrumental variables estimation under a structural Cox model. Biostatistics.2017. 32. Martínez-Camblor P, Mackenzie T, Staiger DO, Goodney PP, O’Malley AJ. Adjusting for bias introduced by instrumental variable estimation in the Cox proportional hazards model. Biostatistics. 2017. doi:10.1093/ biostatistics/kxx062 33. Vascular Quality Initiative. http://www.vascularqualityinitiative.org/. Accessed March 1, 2017. 34. Centers for Medicare and Medicaid Services. http://www.cms.gov. Accessed March 17, 2017. 35. Hoel AW, Faerber AE, Moore KO, et al. A pilot study for long-term outcome assessment after aortic aneurysm repair using vascular quality initiative data matched to Medicare claims. J Vasc Surg. 2017;66(3):751-759.e1. doi:10. 1016/j.jvs.2016.12.100 36. Vandenbroucke JP, von Elm E, Altman DG, et al; STROBE Initiative. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007;4(10):e297. doi: 10.1371/journal.pmed.0040297 37. Skerritt MR, Block RC, Pearson TA, Young KC. Carotid endarterectomy and carotid artery stenting utilization trends over time. BMC Neurol. 2012;12:17. doi:10.1186/1471-2377-12-17 38. Cortese G, Scheike TH, Martinussen T. Flexible survival regression modelling. Stat Methods Med Res. 2010;19 (1):5-28. doi:10.1177/0962280209105022 39. Rosenbaum PR, Rubin DB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am Stat. 1985;39(1):33-38. doi:10.2307/2683903 40. Stock J, Yogo M. Testing for weak instruments in linear IV regression. In: Andrews DWK, ed. Identification and Inference for Econometric Models. New York, NY: Cambridge University Press; 2005:80-108. doi:10.1017/ CBO9780511614491.006 41. West SG, Duan N, Pequegnat W, et al. Alternatives to the randomized controlled trial. Am J Public Health. 2008;98(8):1359-1366. doi:10.2105/AJPH.2007.124446 42. Staiger D, Stock JH. Instrumental variables regression with weak instruments. Econometrica. 1997;65(3): 557-586. doi:10.2307/2171753 43. Rassen JA, Schneeweiss S, Glynn RJ, Mittleman MA, Brookhart MA. Instrumental variable analysis for estimation of treatment effects with dichotomous outcomes. Am J Epidemiol. 2009;169(3):273-284. doi:10.1093/ aje/kwn299 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 12/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 44. Wan F, Small D, Bekelman JE, Mitra N. Bias in estimating the causal hazard ratio when using two-stage instrumental variable methods. Stat Med. 2015;34(14):2235-2265. doi:10.1002/sim.6470 45. Cai B, Small DS, Have TR. Two-stage instrumental variable methods for estimating the causal odds ratio: analysis of bias. Stat Med. 2011;30(15):1809-1824. doi:10.1002/sim.4241 46. MacKenzie TA, Brown JR, Likosky DS, Wu Y, Grunkemeier GL. Review of case-mix corrected survival curves. Ann Thorac Surg. 2012;93(5):1416-1425. doi:10.1016/j.athoracsur.2011.12.094 47. Tan HJ, Norton EC, Ye Z, Hafez KS, Gore JL, Miller DC. Long-term survival following partial vs radical nephrectomy among older patients with early-stage kidney cancer. JAMA. 2012;307(15):1629-1635. doi:10.1001/ jama.2012.475 48. Brott TG, Howard G, Roubin GS, et al; CREST Investigators. Long-term results of stenting versus endarterectomy for carotid-artery stenosis. N Engl J Med. 2016;374(11):1021-1031. doi:10.1056/NEJMoa1505215 49. Paraskevas KI, Kalmykov EL, Naylor AR. Stroke/death rates following carotid artery stenting and carotid endarterectomy in contemporary administrative dataset registries: a systematic review. Eur J Vasc Endovasc Surg. 2016;51(1):3-12. doi:10.1016/j.ejvs.2015.07.032 50. Timaran CH, Veith FJ, Rosero EB, Modrall JG, Valentine RJ, Clagett GP. Intracranial hemorrhage after carotid endarterectomy and carotid stenting in the United States in 2005. J Vasc Surg. 2009;49(3):623-628. doi:10.1016/ j.jvs.2008.09.064 51. US Department of Health and Human Services, US Food and Drug Administration. Use of real-world evidence to support regulatory decision-making for Medical devices. https://www.fda.gov/downloads/medicaldevices/ deviceregulationandguidance/guidancedocuments/ucm513027.pdf. Accessed September 1, 2017. 52. Menkes MS, Comstock GW, Vuilleumier JP, Helsing KJ, Rider AA, Brookmeyer R. Serum beta-carotene, vitamins A and E, selenium, and the risk of lung cancer. N Engl J Med. 1986;315(20):1250-1254. doi:10.1056/ NEJM198611133152003 53. Stampfer MJ, Colditz GA. Estrogen replacement therapy and coronary heart disease: a quantitative assessment of the epidemiologic evidence. Prev Med. 1991;20(1):47-63. doi:10.1016/0091-7435(91)90006-P 54. Alpha-Tocopherol, Beta Carotene Cancer Prevention Study Group. The effect of vitamin E and beta carotene on the incidence of lung cancer and other cancers in male smokers. N Engl J Med. 1994;330(15):1029-1035. doi:10. 1056/NEJM199404143301501 55. Khaw KT, Bingham S, Welch A, et al. Relation between plasma ascorbic acid and mortality in men and women in EPIC-Norfolk prospective study: a prospective population study. European prospective investigation into cancer and nutrition. Lancet. 2001;357(9257):657-663. doi:10.1016/S0140-6736(00)04128-3 56. Beral V, Banks E, Reeves G. Evidence from randomised trials on the long-term effects of hormone replacement therapy. Lancet. 2002;360(9337):942-944. doi:10.1016/S0140-6736(02)11032-4 57. Heart Protection Study Collaborative Group. MRC/BHF Heart Protection Study of antioxidant vitamin supplementation in 20,536 high-risk individuals: a randomised placebo-controlled trial. Lancet. 2002;360 (9326):23-33. doi:10.1016/S0140-6736(02)09328-5 58. Deeks JJ, Dinnes J, D’Amico R, et al; International Stroke Trial Collaborative Group; European Carotid Surgery Trial Collaborative Group. Evaluating non-randomised intervention studies. Health Technol Assess. 2003;7(27):iii-x, 1-173. doi:10.3310/hta7270 59. Kunz R, Oxman AD. The unpredictability paradox: review of empirical comparisons of randomised and non-randomised clinical trials. BMJ. 1998;317(7167):1185-1190. doi:10.1136/bmj.317.7167.1185 60. Ioannidis JP, Haidich AB, Pappa M, et al. Comparison of evidence of treatment effects in randomized and nonrandomized studies. JAMA. 2001;286(7):821-830. doi:10.1001/jama.286.7.821 61. Brewer T, Colditz GA. Postmarketing surveillance and adverse drug reactions: current perspectives and future needs. JAMA. 1999;281(9):824-829. doi:10.1001/jama.281.9.824 62. Ross SD. Drug-related adverse events: a readers’ guide to assessing literature reviews and meta-analyses. Arch Intern Med. 2001;161(8):1041-1046. doi:10.1001/archinte.161.8.1041 63. Sutton AJ, Cooper NJ, Lambert PC, Jones DR, Abrams KR, Sweeting MJ. Meta-analysis of rare and adverse event data. Expert Rev Pharmacoecon Outcomes Res. 2002;2(4):367-379. doi:10.1586/14737167.2.4.367 64. Ioannidis JP, Mulrow CD, Goodman SN. Adverse events: the more you search, the more you find. Ann Intern Med. 2006;144(4):298-300. doi:10.7326/0003-4819-144-4-200602210-00013 65. Newcombe RG. Explanatory and pragmatic estimates of the treatment effect when deviations from allocated treatment occur. Stat Med. 1988;7(11):1179-1186. doi:10.1002/sim.4780071111 JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 13/14 JAMA Network Open | Surgery Survival Analyses After Carotid Endarterectomy vs Carotid Stenting Using Instrumental Variable Method 66. Hearst N, Newman TB, Hulley SB. Delayed effects of the military draft on mortality: a randomized natural experiment. N Engl J Med. 1986;314(10):620-624. doi:10.1056/NEJM198603063141005 67. Brooke BS, Goodney PP, Kraiss LW, Gottlieb DJ, Samore MH, Finlayson SRG. Readmission destination and risk of mortality after major surgery: an observational cohort study. Lancet. 2015;386(9996):884-895. doi:10.1016/ S0140-6736(15)60087-3 68. Garabedian LF, Chu P, Toh S, Zaslavsky AM, Soumerai SB. Potential bias of instrumental variable analyses for observational comparative effectiveness research. Ann Intern Med. 2014;161(2):131-138. doi:10.7326/M13-1887 69. Finks JF, Osborne NH, Birkmeyer JD. Trends in hospital volume and operative mortality for high-risk surgery. N Engl J Med. 2011;364(22):2128-2137. doi:10.1056/NEJMsa1010705 70. Urbach DR. Pledging to eliminate low-volume surgery. N Engl J Med. 2015;373(15):1388-1390. doi:10.1056/ NEJMp1508472 SUPPLEMENT. eMethods. Instrumental Variable Derivation eFigure 1. Distribution of the Instrumental Variable eFigure 2. Propensity Score Performance eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years eTable 1. Characteristics of the Sub-Analysis Cohort: Symptomatic (n=35 514) eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204) JAMA Network Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 September 7, 2018 14/14 Supplementary Online Content Columbo JA, Martinez-Camblor P, MacKenzie TA, et al. Comparing long-term mortality after carotid endarterectomy vs carotid stenting using a novel instrumental variable method for risk adjustment in observational time-to-event data. JAMA Netw Open. 2018;1(5):e181676. doi:10.1001/jamanetworkopen.2018.1676 eMethod . Instrumental Variable Derivation eFigure 1. Distribution of the Instrumental Variable eFigure 2. Propensity Score Performance eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years eTable 1. Characteristics of the Sub-Analysis Cohort: Symptomatic (n=35 514) eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204) This supplementary material has been provided by the authors to give readers additional information about their work. © 2018 et al. JAMA Network Open. eMethod . Instrumental Variable Derivation We assume that the risk model has the form: ( | , , ) = ( )ex ( + + ) where X is the treatment, Z is a vector of measured confounders (see Table 1) and U is a ( ) potential unmeasured confounder. The target of inference is exp , the impact (in terms of a hazard ratio) of the treatment. In addition, we assume that the treatment assignment model has the form: = + W + Z + U + where W is the instrumental variable (IV) and is random error or white-noise. In our study, W is computed as the proportion of carotid endarterectomy (CEA) procedures out of the total CEA and carotid artery stenting (CAS) procedures performed over the past 12-months at the hospital where the patient underwent their procedure. Thus, = , where and denote the number of CEA and CAS procedures performed at the hospital in the 12-months prior to a given patient’s procedure. Because the IV is patient-specific, it has the appealing property of varying within a hospital across time and so can exploit variation in a hospital’s relative preference for CEA and CAS over time. To help validate the IV, the denominator of is included as a measured confounder in the Cox model. The two-stage residual inclusion frailty (2SRI-F) procedure estimates the treatment received by a patient via = + + . From this, we estimate the residuals, avoiding the estimation errors as R = X X = U + . © 2018 et al. JAMA Network Open. Finally, we add R as a covariate in the Cox model and account for the white noise from the estimated first-stage equation, the quantity , by also adding an individual (gaussian) frailty term. Assuming that all of the data are contained in a frame called DATA, the corresponding R code has the general form: # FIRST STAGE. MODEL ASSIGNMENT ESTIMATION RES<- lm(X~ W + age + gender + race + etnia + […], data=DATA)$residuals # SECOND STAGE. COX REGRESSION MODEL TSRF<- coxph(Surv(time,exitus)~ X + age + gender + race + etnia + […] + RES + frailty(1:nrow(DATA),dist="gaussian"), data=DATA) © 2018 et al. JAMA Network Open. eFigure 1. Distribution of the Instrumental Variable. CEA = Carotid Endarterectomy; CAS = Carotid Artery Stenting. © 2018 et al. JAMA Network Open. eFigure 2. Propensity Score Performance We used the traditional plot of the extent of the imbalance between the pre and post matched samples as a means to illustrate the performance of the propensity score matching procedure. PS = propensity score; TIA = transient ischemic attack; HTN = hypertension; CAD = coronary artery disease; CHF = congestive heart failure; COPD = chronic obstructive pulmonary disease; CEA = carotid endarterectomy. © 2018 et al. JAMA Network Open. eFigure 3. Kaplan-Meier Estimated Mortality Overall and by Presenting Symptoms: 10-Years © 2018 et al. JAMA Network Open. eTable 1. Characteristics of the Sub-analysis Cohort: Symptomatic (n=35 514)* Variable All Patients p Propensity-Matched p CEA n=28,689 CAS n=6,825 CEA n=6,418 CAS n=6,418 Demographics Age mean (SD), 70.2 (9.4) 68.7 (10.8) <0.001 69.1 (9.8) 69.9 (10.7) 0.39 years Male (%) 17,302 (60.3) 4,4169 (64.7) <0.001 4,133 (64.4) 4,127 (64.3) 0.93 Race White 26,492 (92.3) 6,103 (89.4) <0.001 5,765 (89.7) 5,757 (89.7) 0.83 Black 1,231 (4.3) 430 (6.3) <0.001 375 (5.8) 389 (6.1) 0.63 Other 963 (3.4) 292 (4.3) <0.001 278 (4.3) 272 (4.2) 0.83 Clinical Factors Elective 25,061 (87.3) 4,704 (70.2) <0.001 4,805 (74.9) 4,725 (73.6) 0.11 Symptomatic 28,689 (100) 6,825 (100) NA 6,418 (100) 6,418 (100) NA TIA or amaurosis 14,122 (49.2) 3,086 (45.2) <0.001 2,980 (46.4) 2,947 (45.9) 0.57 Stroke 14,567 (50.8) 3,739 (54.8) <0.001 3,438 (53.6) 3,471 (44.1) 0.57 Hypertension 24,487 (88.7) 6,045 (88.6) 0.72 5,723 (89.2) 5,691 (88.7) 0.38 Smoking History 21,662 (75.5) 5,128 (75.1) 0.53 4,933 (75.3) 4,834 (75.3) 1 Positive Stress 2,379 (8.3) 466 (6.8) <0.001 472 (7.3) 446 (6.9) 0.39 Test Coronary Disease 8,086 (28.1) 2,071 (30.3) <0.001 2004 (31.2) 1,962 (30.6) 0.43 Heart Failure 2,864 (10.0) 1,012 (14.8) <0.001 905 (14.1) 916 (14.3) 0.8 Diabetes 10,044 (35.0) 2,491 (36.5) 0.02 2,350 (36.6) 2,334 (36.4) 0.34 COPD 6,199 (21.6) 1,714 (25.1) <0.001 1,573 (24.5) 1,601 (24.9) 0.58 Renal 1,439 (5.0) 358 (5.2) 0.46 349 (5.4) 338 (5.4) 0.7 Insufficiency HD 332 (1.2) 14 (0.2) <0.001 20 (0.3) 14 (0.2) 0.39 Prior CEA 4,013 (13.9) 1,935 (28.4) <0.001 1,720 (26.8) 1,608 (25.1) 0.03 Medications Antiplatelet therapy Aspirin 23,851 (83.1) 5,751 (84.3) 0.03 5,279 (82.2) 5,377 (83.8) 0.02 P2y12 inhibitor 8,286 (28.9) 5,100 (74.7) <0.001 4,824 (75.2) 4,693 (73.1) 0.009 Beta-blocker 16,339 (56.9) 3,544 (51.9) <0.001 3,362 (52.4) 3,446 (53.7) 0.14 Statin 23,006 (80.2) 5,865 (78.6) 0.004 5,643 (77.5) 5,076 (77.7) 0.49 *Centers not performing at least 10 procedures on symptomatic patients in the year prior to analysis were excluded. For this reason the number of patients in the sub-analyses less slightly less than the number of patients in the combined analysis. Legend: SD, standard deviation; CEA, carotid endarterectomy; CAS, carotid artery stenting; TIA, transient ischemic attack; NA, not applicable; COPD, chronic obstructive pulmonary disease; HD, hemodialysis. © 2018 et al. JAMA Network Open. eTable 2. Characteristics of the Sub-Analysis Cohort: Asymptomatic (n=50 204)* Variable All Patients p Propensity-Matched p CEA CAS n=5,809 CEA n=5,737 CAS n=5,737 n=44,395 Demographics Age mean (SD), 70.3 (9.5) 69.5 (9.9) <0.001 69.5 (9.5) 69.5 (9.9) 0.65 years Male (%) 26,844 (60.5) 3,658 (63.0) <0.001 3,603 (62.8) 3,632 (63.3) 0.59 Race White 41,064 (92.5) 5,356 (92.2) 0.42 5,311 (92.6) 5,290 (92.2) 0.48 Black 1,853 (4.2) 259 (4.5) 0.33 224 (3.9) 257 (4.5) 0.14 Other 1,473 (3.3) 194 (3.3) 0.96 202 (3.5) 190 (3.3) 0.57 Clinical Factors Elective 39,763 (87.3) 5,399 (92.9) <0.001 5,328 (92.9) 5,329 (92.9) 1 Symptomatic 0 (0) 0 (0) NA 0 (0) 0 (0) NA TIA or amaurosis 0 (0) 0 (0) NA 0 (0) 0 (0) NA Stroke 0 (0) 0 (0) NA 0 (0) 0 (0) NA Hypertension 30,464 (88.9) 5,181 (89.9) 0.51 5,116 (89.2) 5,122 (89.3) 0.88 Smoking History 33,641 (75.8) 4,464 (76.8) 0.08 4,449 (77.5) 4,414 (76.9) 0.45 Positive Stress 3,537 (8.0) 514 (8.8) 0.02 505 (8.8) 506 (8.8) 1 Test Coronary Disease 12,485 (28.1) 2,054 (35.4) <0.001 2,017 (35.1) 2,022 (35.2) 0.94 Heart Failure 4,625 (10.4) 862 (14.8) <0.001 783 (13.6) 828 (14.4) 0.24 Diabetes 15,512 (34.9) 2,684 (35.9) 0.17 2,060 (35.9) 2,063 (36.0) 0.97 COPD 10,015 (22.6) 1,501 (25.8) <0.001 1,460 (25.4) 1,474 (25.7) 0.71 Renal 2,651 (6.0) 360 (6.2) 0.54 351 (6.1) 352 (6.1) 1 Insufficiency HD 599 (1.3) 11 (0.2) <0.001 10 (0.2) 11 (0.2) 1 Prior CEA 6,098 (13.7) 2,178 (37.5) <0.001 2,112 (36.8) 2,106 (36.7) 0.92 Medications Antiplatelet therapy Aspirin 36,702 (82.7) 5,063 (87.2) <0.001 4,923 (85.8) 4,991 (87.0) 0.07 P2y12 inhibitor 12,800 (28.8) 4,491 (77.3) <0.001 4,435 (77.3) 4,419 (77.0) 0.74 Beta-blocker 25,294 (57.0) 3,422 (58.9) 0.005 3,389 (59.1) 3,382 (58.9) 0.91 Statin 35,398 (79.7) 4,700 (80.9) 0.04 4,657 (81.2) 4,657 (81.2) 1 *Centers not performing at least 10 procedures on symptomatic patients in the year prior to analysis were excluded. For this reason the number of patients in the sub-analyses less slightly less than the number of patients in the combined analysis. Legend: SD, standard deviation; CEA, carotid endarterectomy; CAS, carotid artery stenting; TIA, transient ischemic attack; NA, not applicable; COPD, chronic obstructive pulmonary disease; HD, hemodialysis. © 2018 et al. JAMA Network Open.

Journal

JAMA Network Open – American Medical Association

Published: Sep 7, 2018

There are no references for this article.