|
|
Background Aprotinin (Trasylol) is used to mitigate bleeding during coronary-artery bypass grafting (CABG). Accumulating evidence suggests that this practice increases mortality.
Methods Using electronic administrative records of the Premier Perspective Comparative Database, we studied hospitalized patients with operating-room charges for the use of aprotinin (33,517 patients) or aminocaproic acid (44,682 patients) on the day CABG was performed. We tabulated the numbers of patients with a hospital-discharge status of death and performed three types of analyses: a multivariable logistic-regression analysis (primary analysis); propensity-score matching in the highly selected subcohort of patients who received full amounts of the study drug, who underwent CABG by surgeons who performed 50 or more CABG surgeries during the study period, and for whom information on 10 additional covariates was available because the surgery occurred on hospital day 3 or later; and an instrumental-variable analysis of data from patients whose surgeons showed a strong preference for one of the two study drugs.
Results In all, 1512 of the 33,517 aprotinin recipients (4.5%) and 1101 of the 44,682 aminocaproic acid recipients (2.5%) died. After adjustment for 41 characteristics of patients and hospitals, the estimated risk of death was 64% higher in the aprotinin group than in the aminocaproic acid group (relative risk, 1.64; 95% confidence interval [CI], 1.50 to 1.78). In the first 7 days after surgery, the adjusted relative risk of in-hospital death in the aprotinin group was 1.78 (95% CI, 1.56 to 2.02). The relative risk in a propensity-score–matched analysis was 1.32 (95% CI, 1.08 to 1.63). In the instrumental-variable analysis, the use of aprotinin was found to be associated with an excess risk of death of 1.59 per 100 patients (95% CI, 0.14 to 3.04). Postoperative revascularization and dialysis were more frequent among recipients of aprotinin than among recipients of aminocaproic acid.
Conclusions Patients who received aprotinin alone on the day of CABG surgery had a higher mortality than patients who received aminocaproic acid alone. Characteristics of neither the patients nor the surgeons explain the difference, which persisted through several approaches to control confounding.
At the request of the manufacturer, we conducted a retrospective analysis of a large hospital inpatient database study in the United States to examine the association between aprotinin use and the risk of serious in-hospital events and to compare it with that for other antifibrinolytic agents. In this article, we focus on the outcome of death.
Methods
The original study plan defined a single primary analysis. We developed several secondary analyses to assess the potential for bias.
Data
The study cohort was drawn from the Premier Perspective Comparative Database, a repository of hospital administrative data that includes approximately one sixth of all hospitalizations in the United States. Premier provides data services to hospitals that include tabulation and benchmarking against the performance of other institutions. Service-level data that are recorded include charges for medications, procedures, and laboratory tests; characteristics of surgeons and hospitals are also available.9 The UB92 discharge form provides data on demographic characteristics, discharge diagnoses, and discharge status (including death, but not its cause).10 Premier data undergo verification, reconciliation, validation, checks that the use of supplies and other hospital resources was within an acceptable range, and manual and data warehouse audits (Craver C: personal communication). Premier does not verify the submitted data against the original medical records. The Premier database is used in the Centers for Medicare and Medicaid Services Hospital Quality Incentive Demonstration, which links reimbursement of hospitals to the quality of patient care and outcomes of care for selected procedures or conditions, including CABG, acute myocardial infarction, and congestive heart failure.11 The Food and Drug Administration (FDA) and a variety of pharmaceutical manufacturers and research organizations also use Premier data.
The full study period began on April 1, 2003, when all hospitals in the deidentified Premier data set were reporting services separately for each hospital day; some hospitals began reporting as early as January 1, 2003, and have been included from their start dates. The study period ended on March 31, 2006. The authors designed the study, including an a priori power calculation; obtained and analyzed the data; wrote all versions of the manuscript; and vouch for the accuracy of the analysis. The publication of findings was permitted, but not supported, under the original research agreement with the sponsor. This analysis was carried out using fully deidentified data, according to the 1996 Health Insurance Portability and Accountability Act.
Patients
We selected inpatients 18 years of age or older whose hospital records contained a code for CABG (code 36.1, or any subcode thereof, in the International Classification of Diseases, 9th Revision) and a charge for the use of intravenous aprotinin or aminocaproic acid on the day of the surgery. We excluded patients who received multiple antifibrinolytic agents on the day of surgery. This was the primary study cohort (Figure 1).
|
Classification of Drug Exposure and Study Outcome
The degree of drug exposure was classified as very low, low, or high, according to the amount of intravenous drug for which there was a charge on the day of surgery. Very low exposure was defined as the receipt of one vial, or less than 2 million KIU, of aprotinin or one vial, or less than 10 g, of aminocaproic acid. Low exposure was defined as the receipt of more than one vial, or 2 million to 4 million KIU, of aprotinin or more than one vial, or 10 g to 20 g, of aminocaproic acid. High exposure was defined as the receipt of more than 4 million KIU of aprotinin or more than 20 g of aminocaproic acid. Since the body weights of patients were not available, we could not conduct a dose–response analysis.
Follow-up began on the day of surgery. The study outcome was a discharge status of death, according to the routinely submitted UB92 hospital claims form.
Characteristics of the Study Patients and Hospitals
Four types of characteristics were extracted from the Perspective Comparative Database records. The sociodemographic factors studied were age, sex, race, income status (with low-income status defined as receipt of Medicaid or classification as indigent), whether or not patients were living with a domestic partner, smoking status, and year of admission. The markers of prognosis were the type of admission (emergency vs. elective); the day of hospitalization on which CABG was performed; the number of vessels involved in CABG; the presence or absence of repeat CABG, any additional surgery on the day of the index surgery, and percutaneous coronary procedure or thrombolysis before CABG surgery; and complex or noncomplex CABG.12,13,14 Complex CABG surgery was defined as emergency admission, repeat CABG, or additional cardiac surgery on the day of CABG.
Information on two types of coexisting conditions was considered. Chronic conditions noted in discharge diagnoses included diabetes, hypertension, liver disease, chronic obstructive pulmonary disease or asthma, cancer, previous myocardial infarction, previous stroke, endocarditis, peripheral-artery disease, chronic kidney disease, and hemostatic disorders (idiopathic thrombocytopenia, hemophilia, protein S deficiency, protein C deficiency, or leukemia).15,16,17,18,19 Coexisting conditions or markers of disease severity inferred from procedures and drug use before surgery were also considered. These included angina, renal failure, heart failure, arrhythmia, diabetes, cardiac arrest, and use of warfarin, fibrinolytic medications or direct thrombin inhibitors, clopidogrel or glycoprotein IIb/IIIa inhibitors, plasma expanders, or radiologic contrast medium.20,21,22 In the absence of a recorded diagnosis, the value of this covariate was set at 0.
Finally, we studied characteristics of hospitals and surgeons. These consisted of hospital teaching status (teaching or nonteaching hospital), hospital region (Midwest, Northeast, South, or West), hospital location (urban or rural), hospital size (number of beds), and the number of CABG procedures performed during the study period at each hospital and by each surgeon.
Statistical Analysis
We examined the risk of death during the entire hospitalization period and also within the first 7 days after CABG.
Primary Study Cohort
We calculated crude risks and risk ratios and the associated 95% confidence intervals, without correcting for multiple analyses. Using a logistic-regression analysis, we estimated the odds ratios adjusted for all 41 covariates for patients and hospitals, without further selection of variables. We used generalized estimating equations to account for clustering within a hospital.23 Because the odds ratio is an excellent approximation of the risk ratio in the case of rare outcomes, the results of the logistic-regression analysis are referred to as relative risks.24 To explore the utility of the covariates for the prediction of mortality, we calculated model-prediction c statistics separately for each drug group.25 We performed a conditional logistic-regression analysis with matching on the basis of the hospital.26 We repeated the analyses for mortality within 7 days after CABG, and examined the experiences of specific subcohorts: patients who underwent complex surgeries, patients who were treated by surgeons who had conducted 50 or more CABG surgeries during the study period, and patients with diabetes.
We performed sensitivity analyses to quantify the size of the association between an unmeasured confounding variable and aprotinin use or death that would be required to fully explain our study findings. We gathered data on additional risk factors not observed in the main study from the medical records of 98 study patients who underwent CABG in a single, urban teaching hospital of medium size (400 to 649 beds) at which a moderate number of CABG surgeries (100 to 500) were performed during the study period. This information was used to correct the relative-risk estimates for selected unobserved confounders (see the Supplementary Appendix, available with the full text of this article at www.nejm.org).27
Highly Selected Subcohort
We calculated the propensity for treatment with aprotinin on the basis of the 41 covariates used in the logistic-regression analysis, as well as 10 markers of coexisting conditions and disease severity measured before CABG surgery.28 We matched each patient in the aprotinin group to the patient in the aminocaproic acid group with the closest propensity score, using a standard greedy-matching algorithm after excluding patients who received very low amounts of either antifibrinolytic agent.29
Instrumental-Variable Subcohorts
When surgeons always or nearly always use one of the two antifibrinolytic agents, the choice is evidently independent of characteristics of the patient, and it is possible to use the surgeon's preferred agent as a substitute for the actual exposure (i.e., as an "instrumental variable") in analyses.30,31 We examined 47,334 patients who received low or high amounts of aprotinin or aminocaproic acid and who were treated by surgeons who performed 50 or more CABG surgeries during the study period (Figure 1). We classified surgeons who administered aprotinin to 90% or more of their patients as surgeons who preferred aprotinin and those who administered aprotinin to 10% or fewer of their patients as those who did not prefer the drug. Using this preference or lack thereof as an instrumental variable, we computed differences in the risk of the primary outcome between the aprotinin group and the aminocaproic acid group, using a two-stage linear regression analysis that also adjusted for all 41 measured covariates.32 We also ran the analysis using a stricter definition of surgeons' preference (administration of aprotinin to 100% vs. 0% of patients).
Results
Primary Study Cohort
A total of 78,199 patients were in the primary study cohort (Figure 1); 33,517 (42.9%) were given aprotinin. Characteristics of patients and markers of the severity of their disease were balanced in the aprotinin group and in the aminocaproic acid group, with instructive exceptions (Table 1). In particular, repeat CABG surgeries were recorded more often for aprotinin recipients than for aminocaproic acid recipients (4.0% vs. 1.7%), as were additional cardiac procedures (25.4% vs. 18.4%), mostly valve surgeries.
|
|
|
Sensitivity analyses showed that an unmeasured confounder present in 10% of patients would be required to elevate the risk of in-hospital death by a factor of 6 and would also have to have a prevalence among aprotinin recipients that would be six times that among aminocaproic acid recipients to explain a relative risk of 1.64 (Fig. A1 in the Supplementary Appendix). A history of CABG was strongly associated with aprotinin use, as compared with aminocaproic acid use, in the validation study (relative risk, 24.6), and this risk factor was frequently incompletely recorded in the Premier Perspective Comparative Database file. An increase in the risk of death by a factor of 2.6 in association with repeat CABG14 would produce a 25.8% overestimate in the adjusted relative risk of in-hospital death in the aprotinin group, reducing the relative risk from 1.64 to 1.51. The addition of data on history of percutaneous coronary intervention, history of congestive heart failure, hypertension, diabetes, previous use of clopidogrel or aspirin, and a long duration of cardiopulmonary bypass surgery (>120 minutes) from the validation study would reduce the relative risk to 1.47, assuming additivity and independence of these confounders (Table 4).
|
In all, 13,345 patients qualified for the highly selected subcohort, and of these, 9598 were successfully matched on the basis of the propensity score. There were no differences of any consequence between recipients of aprotinin and recipients of aminocaproic acid (Table 1). The risk of death was 32% higher in the aprotinin group than in the aminocaproic acid group (relative risk, 1.32; 95% CI, 1.08 to 1.63). Model prediction of mortality improved, but only slightly, after inclusion of the 10 additional covariates.
Instrumental-Variable Subcohorts
The multivariable analysis of surgeons' preference for aprotinin (the instrumental-variable analysis, in which preference was defined as administration of the drug to at least 90% of patients) or lack of preference (defined as administration to no more than 10% of patients) yielded an increased risk of 0.60 death per 100 patients receiving aprotinin rather than aminocaproic acid (95% CI, 0.00 to 1.21). The estimated increase in risk of in-hospital death was slightly higher with the stricter definition of surgeons' preference (100% vs. 0% of patients given aprotinin): 1.59 deaths per 100 patients (95% CI, 0.14 to 3.04). The corresponding estimates for the unadjusted instrumental-variable analyses and those adjusted for age and sex were 1.59 and 1.47, respectively.
Discussion
This large, hospital-based cohort study using administrative data showed meaningful increases in inpatient mortality among recipients of aprotinin during CABG surgery, as compared with recipients of aminocaproic acid, both within the overall cohort and in all predefined subcohorts. The increased mortality in a multicenter registry study,3,4 the recently suspended head-to-head randomized trial,7 and our finding need to be set against the solid evidence that the use of aprotinin reduces the number of blood transfusions during cardiac surgery. However, aprotinin shows little or no benefit above that from aminocaproic acid. A meta-analysis of head-to-head trials showed that the use of aprotinin resulted in 0.20 fewer unit of blood transfused per patient (95% CI, –0.49 to 0.10), with no reduction in mortality (relative risk of death, 1.27; 95% CI, 0.30 to 5.42).34
Concerns about the potential intravascular thrombosis due to aprotinin are not new,35 and a prolonged celite-based activated clotting time among patients receiving aprotinin has been documented.36 The greater frequency of revascularization and dialysis among aprotinin recipients than among aminocaproic acid recipients in our highly selected subcohort (Table 1) may point to more frequent hypercoagulable states, but the overall evidence of consequential hypercoagulability associated with aprotinin is weak.
Aprotinin rather than aminocaproic acid was used in sicker patients, and the modest reduction in the relative mortality estimates after the control of confounding by covariates is consistent with the hypothesis of confounding on the basis of indication.37 Multivariate analyses resulted in weaker associations between aprotinin and death than those reported in unadjusted analyses (unadjusted relative risk, 1.83; adjusted relative risk, 1.64). Matching according to propensity score permitted us to control for an additional 10 covariates in a highly selected cohort, which further reduced the relative-risk estimate.
Our analyses were adjusted for some, but not all, covariates typically included in risk-prediction scores for patients undergoing CABG.13,14,38 However, we adjusted for many covariates not typically included, and controlling for proxies of confounders results in control of the confounders themselves if the proxies capture the relations with the true confounding variable, exposure, and outcomes. Our joint adjustment for 41 characteristics before CABG was performed resulted in the prediction of in-hospital death that is as good as that from widely accepted clinical risk-prediction models for patients undergoing CABG.13,14 Prediction was almost identical for patients receiving aprotinin and for those receiving aminocaproic acid.
Drawing covariates from administrative data involves making inferences from procedures and patterns of drug use. Despite their excellent predictive value, the covariates may have residual error. For example, several studies have described preferential prescribing of aprotinin in patients undergoing repeat CABG surgery, a potent risk factor for in-hospital death3,5,39; failure to fully adjust for this confounder, as we saw in the sensitivity analysis, somewhat inflates the association between aprotinin use and mortality. A quantitative sensitivity analysis of residual confounding showed that implausible levels of unmeasured confounding would be required to explain the present observations. A small number of abstracted medical records of study patients, which we cannot claim to be representative or to contain all possible risk factors, nevertheless provides some insight into how little the difference in the primary effect estimates between the two drug groups would be diminished if residual confounding were captured. The estimated reduction of relative risk of in-hospital death in the aprotinin group from 1.64 to 1.47 probably overstates the role of residual confounding, because the component biases were summed as if they were entirely independent of each other and independent of the 41 adjusted covariates.27
We used some surgeons' preference for (or avoidance of) aprotinin to bypass the analytic problems posed when patients' risk profiles drive treatment selection, as they may do for surgeons who regularly use either of the study drugs.30 Instrumental-variable analysis has produced results similar to those from randomized trials but is less efficient and thus produces wider confidence intervals.40 The results of our instrumental-variable analysis were similar to those obtained from traditional regression models. The estimated excess risk of death of 1.6 percentage points in aprotinin recipients as compared with aminocaproic acid recipients, added to the overall risk of death among recipients of aminocaproic acid of 2.5 percentage points, would translate into a relative risk on the order of 1.6. An instrumental variable such as a strong treatment preference of surgeons should not be correlated with patients' risk factors after adjustment for the measured covariates. The possibility that clustering of sicker patients treated by surgeons who performed 50 or more CABG surgeries and who had a strong preference for aprotinin was greater than clustering captured by the use of the 41 measured covariates requires the unlikely scenario that patients choose their surgeon on the basis of the surgeon's preference for a specific antifibrinolytic agent.
On September 13, 2006, we transmitted a preliminary version of this report to the manufacturer of aprotinin, which passed it on to the FDA 2 weeks later. The revised and final analysis was presented on September 12, 2007, at an FDA Advisory Committee meeting. We have provided the study data to the manufacturer and to the FDA, which has independently evaluated the data using different methods and has reported substantially similar results.41 Representatives of and consultants for the manufacturer have disagreed with our methods and conclusions.42,43,44,45
Our analysis of hospital administrative data for patients undergoing CABG, involving more than 33,000 aprotinin recipients in comparison with some 45,000 aminocaproic acid recipients, supports the hypothesis that there is an increased risk of in-hospital death among aprotinin recipients. The findings are not readily attributable to chance or to distortions arising from any of the dozens of measured characteristics of patients, hospitals, and surgeons. Clinicians need to weigh this increased risk of death among aprotinin recipients, as compared with aminocaproic acid recipients, against the reduction in the need for transfusions during CABG.
Supported by a research contract between Bayer HealthCare and Ingenix, whose i3 Drug Safety unit has performed drug safety research under contract with many pharmaceutical firms and the Food and Drug Administration. Bayer HealthCare is the manufacturer of aprotinin (Trasylol).
Dr. Schneeweiss reports receiving consulting fees from i3 Drug Safety for this research and research grant support from Pfizer. Dr. Seeger and Ms. Landon report being employees of i3 Drug Safety, and Dr. Walker reports being an employee of i3 Drug Safety during the conduct of the study. No other potential conflict of interest relevant to this article was reported.
Source Information
From the Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School (S.S.); and the Department of Epidemiology, Harvard School of Public Health (J.D.S., A.M.W.) — all in Boston; and i3 Drug Safety, Waltham, MA (S.S., J.D.S., J.L.).
Address reprint requests to Dr. Schneeweiss at the Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, 1620 Tremont St. (Suite 3030), Boston, MA 02120, or at schneeweiss{at}post.harvard.edu.
References
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
