Aprotinin—Are There Lessons Learned?

This past year has been a complicated one for aprotinin, an antifibrinolytic serine protease inhibitor used to lessen bleeding in patients undergoing cardiac surgery with cardiopulmonary bypass. Thirteen years after the initial approval of aprotinin by the US Food and Drug Administration,¹ the report by Mangano and colleagues² in this issue of JAMA will intensify the debate about the benefits and risks associated with use of this drug. Perhaps more important, this aprotinin story illuminates the larger issue of postmarket safety evaluation surrounding drugs and devices in the current medical environment.

In this article, the investigators report 5-year, all-cause mortality data from patients enrolled in an international registry of on-pump coronary artery bypass graft surgery. The authors conclude that use of aprotinin in routine and complex coronary artery bypass graft surgery is associated with an increased risk of late mortality as compared with use of no hemorrhage-sparing agent and with use of aminocaproic acid and tranexamic acid, 2 lysine analog antifibrinolytics that are less expensive and less toxic. The 3876 patients from 62 of 69 centers were observed (87% follow-up rate) for 5 years. The authors used extensive statistical analytical techniques to address bias inherent in observational analyses,^{3 - 4} and referenced select randomized clinical trial data to support their observational findings.⁵

This current study by Mangano et al² follows an earlier publication from 2006 addressing the perioperative risk of aprotinin therapy.⁶ The same registry was used for both studies and consisted of prospectively collected clinical data (7500 variables/patient) from 69 cardiac surgery sites in North and South America, Europe, and Asia. The earlier study⁶ analyzed 4374 patients, nonrandomly enrolled into the same 4 groups (ie, control, aminocaproic acid, tranexamic acid, aprotinin). Propensity-adjusted, multivariable logistic regression analysis demonstrated that as compared with control, aprotinin was associated with increased perioperative risk of death, acute renal failure, myocardial infarction, and stroke/encephalopathy. The lysine analogs were not associated with any of these adverse outcomes.

Importantly, the perioperative observational study findings⁶ seemed to conflict with findings from several dozen randomized trials, 2 meta-analyses, a Cochrane Collaboration summary on aprotinin use,⁷ and an evaluation by the US Food and Drug Administration.⁸ The challenges in translating clinical trial data into the “real world” are well known. Possible reasons⁹ that randomized trial results might not be available for assessment of adverse events with pharmacologic therapies include the fact that trials powered for efficacy are often too small to detect adverse events and too short to detect events requiring longer exposure. In addition, there are limited industry incentives to fund trials to quantify adverse events for drugs or devices, and head-to-head comparison of adverse events between different manufacturers may not be available. Moreover, observational-based studies (of the type reported by Mangano et al^{2 ,6} ) might be a model for the future in assessing adverse effects from drugs.⁹

A variety of observational data sets exist in cardiovascular disease, including large administrative databases (Medicare, other payer, and hospital- or network-based systems), industry-sponsored postmarket registries, and comprehensive databases and registries from professional organizations (the Society of Thoracic Surgeons, the American College of Cardiology, and the American Heart Association). If observational data sets and analyses specific to those data sets are potential candidates for use in the safety evaluation process for drugs and devices, 4 criteria will need to be met: (1) the data sets must be sources of information that cannot be acquired through other mechanisms; (2) the reason patients receive the drug/device therapy must be determinable; (3) integration of both periprocedural data and long-term data are necessary; and (4) observational database systems will need to be modified and prospectively integrated into the overall process of evaluation for safety and efficacy. All of these criteria impact this aprotinin story.

The studies by Mangano et al^{2 ,6} are important because they have generated new information not otherwise available about aprotinin use. However, it is not clear that the reasons patients received any specific agent (or no agent) are determinable. In the registry used by Mangano et al,^{2 ,6} the use of antifibrinolytics varied widely across surgeons, sites, and countries during the enrollment period, and was influenced by cost, case complexity, physician judgment, and other factors including differing patient selection and indication criteria. For example, the majority (57.5%) of control patients were from European centers, whereas nearly half (48.8%) of antifibrinolytic patients were from North American centers, which captured only 23.9% of control patients. Aprotinin use in cardiac surgery has never been uniformly standardized, but generally has been reserved for patients in whom the surgical team anticipated a higher risk for intraoperative blood loss. This anticipation was driven by the surgical team's perception of increased technical complexity, increased risk of adverse outcome, or both for the patient in question. Conversely, no antifibrinolytic drug or lysine analogs were routinely used in the remaining patients, specifically lower-risk coronary artery bypass graft patients. In this study, 33.3% of the antifibrinolytic patients underwent complex surgery (vs 25.0% of control patients), with a substantial predominance of these patients in the aprotinin group compared with the lysine analog groups. Importantly, these biases, at the level of the surgical team, were not captured in the extensive patient-level data collection process, nor in the analysis.

It is distinctive that this registry is not a comprehensive accounting of all patients under evaluation, as compared with a number of other comprehensive observational databases in cardiac surgery.² Rather, a sampling of patients was enrolled based on anticipated annual surgical volume at the site. This method, therefore, enrolled a greater proportion of patients from low-volume cardiac centers, which have been shown to have less optimal coronary artery bypass graft outcomes than higher-volume centers, particularly in higher-risk cases.¹⁰ As a reflection of the overall scope of the sampling process, only 18 patients per year per site would be needed to generate the study population.

Site-level confounding also is a characteristic of multi-center observational analyses, and is difficult to address with regression and propensity analysis techniques alone. Hierarchical regression techniques have been useful in accounting for site-level bias in other large observational analyses involving multiple sites,¹¹ but these were not conducted by the authors.²

The higher operative mortality rate in the 2006 registry study vs control persisted during the 5-year follow-up. The mechanism for this late mortality difference is not clear and causality cannot be inferred from this data set analysis.¹² Nevertheless, the importance of long-term data are emphasized, and distinguish this analysis from other published data on aprotinin use.

In addition, aprotinin was rapidly adopted for cardiac surgery after US Food and Drug Administration approval in 1993 for high-risk cardiac surgery patients. Subsets of patients with adverse outcomes were soon identified.^{13 - 14} It took another 3 to 4 years to fully understand and adopt how to modify anticoagulation dosage and monitoring for cardiopulmonary bypass in patients receiving aprotinin.¹⁵ A decade later these observational studies^{2 ,6} introduce new information about late safety. Perhaps it is these long time frames (mortality 5 years following drug administration, emergence of new safety data after 13 years of clinical experience) that in part make it somewhat difficult to put these registry findings into the proper context.

As the validation cycle for new drugs and devices, funded largely through industry-sponsored trials, continues to shorten, and the pressure for rapid adoption of new technology increases, the likelihood that late outcomes and safety data will raise new issues not addressed in the efficacy trials will increase. Cardiac morbidity related to cyclooxygenase 2 inhibitors, late stent thrombosis with drug-eluting coronary artery stents, and cardiac mortality related to anti-Parkinson drugs are just a few examples. Industry-sponsored registries are limited by inertia¹⁶ and by their small scale relative to the magnitude of rapid adoption. For instance, industry-sponsored postmarket drug-eluting stent registries have enrolled fewer than 15 000 real-world patients with multivessel disease receiving off-label stents^{17 - 18} compared with more than 6 million stents implanted worldwide,¹⁹ a majority of which are now used for multivessel disease.²⁰ More comprehensive, integrated, and cross-therapy observational data platforms will be necessary.

Thus, the controversy surrounding these studies on aprotinin risks^{2 ,6} centers on the degree to which the robust statistical analyses can account for the unmeasured biases in this registry. The concept that observational data sets and analyses can play a substantial role in postmarket surveillance and safety evaluation for drugs and devices has enormous merit and, many believe, feasibility. With regard to aprotinin use, the US Food and Drug Administration issued a relabeling of aprotinin on December 15, 2006, once again confining it to use in high-risk coronary artery bypass graft patients.²¹ While the admonition to “First, do no harm” certainly applies to ongoing use of aprotinin, an equally important consideration is when and where, at the drug/device safety evaluation level, the issue of first, do no harm actually could have been initially identified. Ultimately, going forward, the most important lesson learned from the aprotinin story is determining better ways to ensure drug safety and to eliminate and prevent these harms.