Mortality Associated With Aprotinin During 5 Years Following Coronary Artery Bypass Graft Surgery FREE

Aprotinin, a drug effective in limiting blood loss in patients undergoing surgery, was first approved in the United States in 1993 for use in high-risk patients needing coronary artery surgery. In 1998, its use was broadened to include any patient undergoing coronary artery surgery using cardiopulmonary bypass.¹ More than 4 million patients worldwide have received aprotinin since 1985, principally during cardiac surgery, with annual use increasing to 600 000 patients (2005) throughout the world, including 246 000 uses in the United States in the past year.²

However, during 2006, the safety of aprotinin came under scrutiny. A sponsor-independent, propensity-adjusted, observational study reported that aprotinin use was associated with a doubling to tripling of the risk of perioperative renal dysfunction and failure requiring dialysis in patients undergoing primary, repeat, or complex coronary artery surgery.¹ These findings were consistent with the original 1993 US Food and Drug Administration observation that in some patients involved in the initial 2 trials, aprotinin was associated with “ kidney toxicity.”³ The 2006 study also raised other concerns regarding the association of aprotinin with cardiovascular events (myocardial infarction, heart failure) and cerebrovascular events (encephalopathy) among patients undergoing primary surgery. Since then, several other reports have reached similar conclusions^{4 - 6} and a series of case studies indicated aprotinin's disproportionate association with in-hospital death due to acute pulmonary vascular thromboses.⁷

Despite this evidence, however, recent commentary and regulatory reviews have claimed that perioperative adverse events associated with aprotinin use were transitory and concluded that there should be little concern for longer-term sequelae.^{8 - 9} However, no comprehensive data exist that describe the long-term effects on survival of either serine protease inhibitors or lysine analogs. Moreover, considerable evidence exists regarding the association of aprotinin with acute coronary graft occlusion in vitro,^{10 - 11} in animals,^{12 - 13} and in humans^{14 - 17} —thrombotic events that can manifest clinically over months to years following arterial occlusion (long after hospital discharge).^{18 - 26}

Given the importance of documenting the long-term effects of blood-sparing therapies in general, and aprotinin and the lysine analogs in particular, we investigated this question hypothesizing that use of either serine protease inhibitors or lysine analogs in patients presenting for coronary artery surgery may be associated with higher long-term all-cause mortality.

Following institutional review board approval and written informed consent, we prospectively enrolled at 69 international institutions every nth patient with medically refractory coronary artery disease who was scheduled for coronary artery bypass surgery using cardiopulmonary bypass. The sampling rate was n = N ÷ 50 (to the closest integer), where N was the number of patients expected to undergo myocardial revascularization surgery over a 1-year period at that institution.

In-hospital data (more than 7500 fields per patient) were collected through discharge among 4374 patients, with race/ethnicity determined by the clinical investigator at each institution (Table 1).¹ Thereafter, 62 sites in North and South America, Europe, and Asia agreed to and completed the 5-year long-term study. Investigators prospectively assessed patients using a formal interview and validated instruments at each of 7 long-term time points: at 6 weeks, 6 months, and annually at 1, 2, 3, 4, and 5 years after hospital discharge, with all data fields adjudicated centrally (Ischemia Research and Education Foundation) by blinded investigators. National death registries, such as the Social Security Death Registry, were used to supplement death information. Enrollment commenced on November 11, 1996, the in-hospital phase was completed on June 30, 2000, and the long-term program 5-year data collection was completed on January 5, 2006, with the long-term database locked (for all-cause survival) on December 7, 2006.

Outcomes were prespecified and defined by protocol. Clinical decisions were not controlled by study protocol, and all patients participating in the long-term program, as well as those who had died in-hospital, were included (Figure 1).

The long-term mortality study was designed prospectively prior to completion of 5-year data collection, database locking, and study group unblinding. Baseline medical characteristics were compared between the control group and each drug group using the 2-tailed χ² test or Fisher exact test, the risk of 5-year mortality was compared with control using the 2-tailed χ² test or Fisher exact test, and 5-year mortality of the treated groups was compared with control using the 2-tailed χ² test or Fisher exact test. Odds ratios (ORs) with 95% confidence intervals (CIs) with P values are presented. Continuous variables were described as mean plus or minus 1 SD, for comparison of 2 groups we used Wilcoxon rank-sum, and for more than 2 groups we used Kruskal-Wallis. Three methods were used to assess drug associations with outcome: survival analysis, multivariable logistic regression, and propensity score adjustment.

Survival analysis was performed on all 4374 patients enrolled in the in-hospital study¹ using Cox regression, illustrated using covariate-adjusted survival with origin time set at the time and date of surgery, and including right censoring (date of last contact) applied to those patients completing the in-hospital phase but choosing not to participate thereafter, and to those lost to follow-up. The proportional hazards assumption was evaluated by including time-dependent covariates in the model, and found to be appropriate. Imputation for continuous variables, such as body surface area (4 missing values), was performed using mean values by sex; otherwise, a combined variable approach was used. Analyses using nonimputed data produced similar findings as that presented here for imputed data (data available from authors upon request, or from the Ischemia Research and Education Foundation Web site).²⁷ For covariate adjustment of survival curves, the corrected-group-prognosis method was used^{28 - 29} ; the average-of-covariates method produced similar results. Cumulative mortality was calculated as 1 minus adjusted survival.²⁷

We performed multivariable logistic regression using the 97 perioperative risk factors²⁷ to further evaluate the association of drug group with 5-year mortality among patients participating in and completing the 5-year follow-up program (Figure 1). Univariate associations significant at a nominal 2-tailed P ≤.20 were entered stepwise into multivariable logistic models. The final model assessed the association of each treatment (aprotinin, aminocaproic acid, or tranexamic acid vs no treatment) with 5-year mortality in the presence of the significant covariates, also including in-hospital outcome events. Additionally, post hoc, we expanded the list of predefined covariates²⁷ to challenge these findings, but found no material differences.

To assess selection bias not adequately controlled by standard multivariable approaches, we used propensity score adjustment methods.^{1 ,30 - 32} Using nonparsimonious logistic regression models, we developed propensity scores for any antifibrinolytic treatment (vs no treatment), including 45 treatment selection covariates.²⁷ Covariate interactions were assessed but were found unnecessary for balance of any baseline characteristic. The area under the receiver operating characteristic, referred to as the C-index, was used to assess the discriminate power of the propensity score. The derived propensity scores were then used for multivariable covariate adjustment together with the antifibrinolytic drug indicator variables. Additionally, we assessed the Cox model findings with and without propensity score adjustment. We also expanded (post hoc) the list of propensity covariates²⁷ to challenge these findings, but found no material differences.

In secondary analyses, we assessed the dose relationship¹ among aprotinin patients who received either a low-dose regimen (loading = 1 million KIU; total >2 million KIU), or a high-dose regimen (loading = 2 million KIU; total >4 million KIU), compared with control patients. We also repeated the analyses post hoc, forcing each of 3 sets of covariates into the multivariable and propensity models: (1) covariates for all variables in Table 1; (2) covariates for in-hospital and long-term mortality using 5 independent risk indices^{33 - 41} ; and (3) covariates for combined and individual in-hospital outcomes (cardiovascular, cerebrovascular, renal). These additional covariates are available from the authors upon request or from the Ischemia Research and Education Foundation Web site.²⁷

All statistical analyses were performed using SAS statistical software version 8.2 (SAS Institute Inc, Cary, NC). Statistical significance was set at P<.05 (2-tailed). Bonferroni adjustment was used, as appropriate, to adjust the significance limit for multiple comparisons.

Baseline characteristics according to treatment are summarized in Table 1. Acute and chronic vascular disease was prevalent among all patients, with several imbalances noted before propensity adjustment.

During 5 years, 223 deaths occurred among 1072 aprotinin-treated patients (20.8%), a death rate nearly two-thirds greater than control patients (128 deaths among 1009 patients, 12.7%; covariate adjusted hazard ratio [without propensity adjustment], 1.48; 95% CI, 1.19-1.85; P<.001). Rates were similar for aminocaproic acid patients (132 deaths among 834 patients, 15.8%; adjusted hazard ratio, 1.03; 95% CI, 0.80-1.33; P = .81) and for tranexamic acid patients (65 deaths among 442 patients, 14.7%; adjusted hazard ratio, 1.07; 95% CI, 0.80-1.45; P = .64). Covariate-adjusted survival analyses demonstrated significant association with death for aprotinin, but not for either aminocaproic or tranexamic acid (Figure 2)—the latter 2 biochemically similar lysine analogs having nearly indistinguishable mortality (survival) patterns despite disparate use and approval among countries and centers. Aprotinin's association with mortality persisted among patients who survived their index hospitalization (Figure 2). Proportional hazards analysis using multiple covariates confirmed the survival associations (Table 2, Table 3).

Multivariable logistic regression confirmed these findings, which indicates that aprotinin was an independent predictor of 5-year mortality (covariate adjusted OR, 1.51; 95% CI, 1.17-1.96) without propensity adjustment, or with adjustment (C-index = 0.70; OR, 1.48; 95% CI, 1.13-1.93), as well as among those surviving their index hospitalization.²⁷ In contrast, neither aminocaproic nor tranexamic acid was associated with increased 5-year mortality.²⁷

We investigated whether the occurrence of in-hospital ischemic events (cardiovascular, cerebrovascular, renal) affected our findings on mortality over the 5 years following discharge by including in-hospital nonfatal events as covariates. Findings were similar to those in Table 2, Table 3; the OR for aprotinin vs control was 1.52 (95% CI, 1.14-2.02), aminocaproic acid vs control, 1.06 (95% CI, 0.77-1.45), and tranexamic acid vs control, 1.10 (95% CI, 0.74-1.62), and with in-hospital complications significantly associated with postdischarge death, 1.41 (95% CI, 1.10-1.82).

In secondary descriptive analyses, the association of aprotinin with death was found among diverse patient risk profiles^{33 - 41} and surgical factors (Figure 3), and among high-risk patients identified using in-hospital and long-term risk indices (Figure 3).

Finally, the association between aproprotin dose and mortality was assessed using prospective dose definitions as previously reported.¹ We found differences in the point estimates for 5-year mortality (control, low-dose aprotinin vs high-dose aprotinin): 12.7%, 22.8%, 37.9% (P<.001); in the ORs (low-dose vs control, 1.58; 95% CI, 1.14-2.20 and high-dose vs control, 2.07; 95% CI, 1.08-3.95); and in the Kaplan-Meier survival differences (control, low-dose aprotinin vs high-dose aprotinin, P<.001 [log-rank]).²⁷

Our study, assessing the long-term safety of antifibrinolytic agents, demonstrated that aprotinin—but not aminocaproic acid or tranexamic acid—is associated with an increased risk of death during the first 5 years following surgery. Importantly, aprotinin's association with death sustained comprehensive covariate challenges, remaining significant when assessed among multiple subgroups with differing risk profiles and among patients surviving their index hospitalization. The minimal associations of the lysine analogs aminocaproic acid and tranexamic acid with long-term death, along with their previously reported efficacy and safety profiles,^{1 ,4 - 6} indicate that safe and inexpensive alternatives exist. Based on our data, we therefore believe that additional concern is now warranted regarding the long-term safety of aprotinin among patients undergoing coronary artery bypass surgery.

These observations suggest that the deleterious perioperative safety findings previously reported for the serine protease antagonist aprotinin are not self-limited, as has been suggested,^{8 9} but that their consequences likely continue over months to years following administration of aprotinin. To provide insight into this observation, we cite 3 related areas of research: (1) clinical experiences suggesting an association of aprotinin with thrombosis (arterial, venous, coronary graft); (2) the natural history of acute thrombosis as it relates to long-term outcome; and (3) prior experiences addressing long-term effects of acute, time-limited therapy.

Concerns of a possible relationship between aprotinin and thrombosis arose soon after approval,^{14 - 15} suggesting that an antagonist of serine protease activity (aprotinin) might be expected to promote thrombosis, given that tissue plasminogen activator, an agonist of serine protease activity, prevents coronary occlusion effects in similar patients.⁴² Several other studies, however, did not find an association of aprotinin with thrombosis.^{43 - 45}

The US Food and Drug Administration then conducted a retrospective review of 1307 placebo-treated and 2004 aprotinin-treated patients, finding a statistically significant association between aprotinin use and coronary graft closure (occluded grafts).⁴⁶ Based on those data, a prospective randomized trial (International Multicenter Aprotinin Graft Patency Experience [IMAGE]) was conducted, which found a statistically significant 41% increase in coronary graft occlusion occurring within 2 weeks of aprotinin use vs placebo (the primary end point).¹⁶ Given the US Food and Drug Administration's analysis and the prospective results of the IMAGE study, there exists reasonable evidence to suggest that graft occlusion may occur within days of aprotinin use in at least some at-risk patients.

The natural history of acute coronary thrombosis and occlusion, as well as the experiences in the settings of medical and surgical revascularization, also suggests that clinical sequelae may manifest after months to years following arterial occlusion,^{18 - 26} as also noted in the accompanying commentary to the IMAGE study.⁴⁷ Our observation of significant effects long after aprotinin administration is not surprising given the experiences with the serine protease agonist therapy, tissue plasminogen activator, the acute administration of which produces effects lasting for 5 years or longer.⁴⁸ Similar examples are provided by the long-term effects of fibrinolytics (streptokinase),^{49 - 50} as well as β-blockers (atenolol) in surgical patients.⁵¹ Other perioperative studies indicate that seemingly reversible perioperative adverse events, such as postoperative myocardial ischemia, can have fatal consequences that only are realized months to years after hospital discharge.^{52 - 53} Accordingly, the long-term adverse effects of aprotinin detected in our study appear consistent with reported clinical observations in similar patients, in whom the effects of an acutely administered therapy can manifest over multiple years.

We estimate that over the past year, aprotinin was prescribed worldwide to at least 200 000 cardiac surgery patients having a profile similar to patients in our study.² For such patients, our study found a 5% absolute increase in 5-year mortality (1% per year for 5 years) associated with aprotinin use, compared with either aminocaproic or tranexamic acid use. Thus, in 2006 alone, had aprotinin been replaced with either of these generic agents, we estimate that approximately 2000 deaths per year for the next 5 years (or 10 000 total deaths) might have been avoided.

Large-scale randomized controlled trials are accepted as the criterion standard for efficacy,⁵⁴ but for assessment of post-marketing safety, these are uncommonly performed even when mandated by regulatory agencies.⁵⁵ Further, even clinical trials are subject to the serious limitation of patient selection bias, given post-marketing imbedded practice—a consideration that has been recognized¹ and thoroughly addressed and is based on a clinician's hesitancy to either withhold a perceived effective therapy (placebo group assignment) or administer a perceived unsafe drug (drug group assignment).³² Despite the well-known limitations of observational studies, we note that our conclusions survived multiple primary and secondary covariate challenges and were consistent across multiple risk groups (Figure 3). Despite differences in use patterns by center and by country for the biochemically similar lysine analogs, aminocaproic acid and tranexamic acid, their long-term survival patterns were virtually identical (Figure 2). This likely reflected that drug effect predominated over any selection pattern biases that may have existed, findings confirmed by the negligible effects of propensity adjustment on their relative survival patterns, even when a nonparsimonious approach was taken.

Regarding the concern that patients receiving aprotinin may have had more complex or advanced preoperative disease (Table 1), we believe that such differences had little effect on our 5-year mortality findings for a number of reasons: (1) all variables included in Table 1 were assessed in proportional hazard and logistic regression analyses and found to have little effect on the association between aprotinin and mortality with or without propensity correction; (2) none of the imbalances (bleeding agent vs no bleeding agent groups) shown in Table 1 remained significant following propensity adjustment; (3) among the patients with risk factors for in-hospital renal, cardiovascular, and cerebrovascular events (Figure 3), the associations for aprotinin—and lack of association for aminocaproic and tranexamic acid—persisted; (4) these associations prevailed for at-risk patients as characterized by risk indices for in-hospital and long-term events^{33 - 41} (Cleveland Clinic, Society of Thoracic Surgeons, additive EuroSCORE, logistic EuroSCORE, Veterans Affairs indices) (Figure 3) and although these indices were developed primarily for in-hospital events, we believe they are useful in this study for descriptive purposes; and (5) regarding survival, our findings using proportional hazard techniques (Table 2, Table 3) and adjusted mortality (survival) curves (Figure 2) indicate that the association between aprotinin and mortality remained significant when adjusted for pre-existing disease. Thus, we believe that our findings are substantive for effect size and consistency among risk groups, as well as for their conformity with the earlier IMAGE trial and US Food and Drug Administration review findings regarding aprotinin's association with acute occlusion of newly placed coronary vein grafts.

We acknowledge that 7 of the original 69 centers did not participate in our long-term program, reducing our cohort by 11%; however, we believe this reduction had little effect because the in-hospital mortality by study group was similarly distributed among the 62 centers participating in this study vs the 7 centers that did not (data available from authors upon request). Additionally, although 87% of participating centers had complete 5-year survival data, 13% of patients were lost to follow-up between 6 weeks and 5 years and were censored in survival analyses. The distribution of these patients among the study groups, however, was similar.

Regarding secondary dose-relationship analyses, we suggest cautious interpretation. The relationships, which contrasted 3 groups (control-no drug, low-dose, high-dose) were based on dose criteria that—although defined prospectively—were made intentionally rigid, thereby minimizing clinician subjectivity in dose choice, use of comingled (low-high) doses, or inadequate- or excessive-dose regimens. While we believe our findings are reliable for the predefined subgroup, we recognize that a sizeable number of patients not meeting the criteria were not included in the analysis. Thus, the dose/relationship findings only are suggestive and should be interpreted as such.

The association between aprotinin and long-term mortality indicates that serious safety concerns extend beyond the perioperative period. Therefore, continued use of aprotinin in this population does not appear prudent, given that safer alternatives—aminocaproic acid and tranexamic acid—are available.