Author + information
- Received February 20, 2013
- Revision received April 7, 2013
- Accepted May 28, 2013
- Published online December 10, 2013.
- Matthew C. Bunte, MD∗,
- Eugene H. Blackstone, MD†,‡,
- Lucy Thuita, MS‡,
- Jeff Fowler, DO§,
- Lee Joseph, MD§,
- Aska Ozaki, DO§,
- Randall C. Starling, MD, MPH∗,
- Nicholas G. Smedira, MD† and
- Maria M. Mountis, DO∗∗ ()
- ∗Department of Cardiovascular Medicine, Cleveland Clinic, Cleveland, Ohio
- †Department of Thoracic and Cardiovascular Surgery, Cleveland Clinic, Cleveland, Ohio
- ‡Department of Quantitative Health Sciences, Research Institute, Cleveland Clinic, Cleveland, Ohio
- §Department of Internal Medicine, Cleveland Clinic, Cleveland, Ohio
- ↵∗Reprint requests and correspondence:
Dr. Maria M. Mountis, Robert and Suzanne Tomsich Department of Cardiovascular Medicine, J3-4, Heart and Vascular Institute, Cleveland Clinic, 9500 Euclid Avenue, Cleveland, Ohio 44195.
Objectives The aim of this study was to characterize a single-center experience of major bleeding complications during HeartMate II (HMII) (Thoratec Corp., Pleasanton, California) left ventricular assist device support, with focus on the subtypes and temporal patterns of post-operative bleeding.
Background Bleeding complications are the most common post-operative adverse events after HMII implantation. The timing of bleeding events, relationship to coagulation status, and effect on post-operative survival are incompletely understood.
Methods From October 2004 to June 2010, 139 HMII recipients at the Cleveland Clinic received 145 devices as a bridge to transplant or destination therapy for advanced heart failure. Major bleeding was defined using Interagency Registry for Mechanically Assisted Circulatory Support criteria, with an additional category created to maximize sensitivity for events. Pre-operative variables, coagulation status, and bleeding recurrence were assessed for correlation to primary events using modulated renewal within a multivariable analysis.
Results The cumulative occurrence of major bleeding was 58% during 171 patient-years of follow-up. There were 1.14 major bleeds per patient-year, with 44% occurring as repeat bleeding events. A first bleed did not predict subsequent bleeding. The greatest risk of bleeding was noted within 2 weeks post-implantation. The international normalized ratio profile correlated poorly with the risk of bleeding. Bleeding early after surgery was associated with reduced survival while on HMII support.
Conclusions The risk of bleeding peaks early after HMII implantation. Bleeding of thoracic and gastrointestinal sources dominates these events, although many patients undergo transfusions for anemia without an apparent source of hemolysis or bleeding.
Despite advances in mechanical circulatory support, bleeding remains the most common adverse event after implantation of the HeartMate II (HMII) (Thoratec Corp., Pleasanton, California) ventricular assist device (1–4). Post-implantation bleeding complications diminish the overall effectiveness of HMII in treating advanced heart failure as a bridge to transplant (BTT) (5). Transfusion-related bleeding represents a decidedly ominous problem; the deleterious effect of red blood cell transfusion on HMII recipient survival has been well described (6,7). With growing use of the HMII as destination therapy, bleeding rates among this population are comparatively higher (3) relative to cohorts who receive an implant as a BTT (2). Therefore, efforts to reduce the rate of bleeding are imperative and urgently needed.
Bleeding complications are complex and dynamic phenomena that are likely dependent on many pre-operative and post-operative factors (5,6,8). From our clinical experience, some HMII recipients experience numerous bleeding events, whereas many others have none at all. Available reports that characterize HMII-related bleeding are often limited by variable bleeding definitions across studies, restricted characterization of bleeding subtypes, and heterogeneous inclusion of non-HMII devices. Accordingly, broad conclusions about post-HMII bleeding have been challenged by variability in post-operative management over time and across centers. Currently available HMII registries offer insufficient detail to account for features of bleeding complications. Whether the temporal patterns of bleeding, sources of blood loss, and impact of recurrent bleeding contribute to future events is not well understood. Therefore, we performed a detailed examination of post-HMII bleeding to address these limitations while providing novel insights that might direct future HMII care.
The aims of this study were to: 1) characterize the timing and subtypes of post-operative major bleeding after HMII implantation using standardized Interagency Registry for Mechanically Assisted Circulatory Support (INTERMACS) definitions for these events while using time-sensitive modeling techniques; 2) determine the influence of coagulation status on subsequent risks for bleeding; and 3) assess the association of bleeding events on post-surgical mortality.
From October 2004 to June 2010, 145 consecutive patients older than 18 years of age received a HMII device at the Cleveland Clinic. To limit potential confounding, patients who received other forms of pre- or post-operative mechanical circulatory support (e.g., concomitant right ventricular support device, prior pulsatile ventricular assist device, or prior biventricular support device) (n = 6) were not included in this analysis. Those receiving pre-operative intra-aortic balloon counterpulsation or extracorporeal membrane oxygenation were not excluded. The resulting 139 patients received 145 HMII implants, including 6 HMII reimplants. After institutional review board approval, patient characteristics, details of HMII implantation, and post-operative data were retrieved from prospectively populated institutional databases approved for research with patient consent waived.
Patients were followed up until transplantation, death, or a study completion date that allowed for at least 3 months of follow-up. Primary bleeding events were defined using INTERMACS adverse event criteria whenever possible (9). Because we encountered numerous major anemia events without a determined source of bleeding or hemolysis, a unique category of bleeding was created to maximize sensitivity for all events leading to transfusion (see “primary event definitions”). All primary bleeding outcomes were adjudicated against medical records applying these definitions.
Values for pre-operative variables were recorded nearest to the date of surgery and within 14 days of HMII implantation. Details of bleeding complications and medication use were retrospectively collected from chart review and input into an institutionally-licensed Research Electronic Data Capture online database (Vanderbilt University, Nashville, Tennessee) (10) hosted at the Cleveland Clinic and then merged with existing prospectively populated information from an institutional database for final analysis.
Institutional anticoagulation protocol
Like many centers, our initial HMII experience began with comparatively aggressive post-operative anticoagulation held over from an era of implanting pulsatile mechanical circulatory support devices. From our first HMII implant in 2004 to early 2008, most patients with implants received an anticoagluation regimen of aspirin 325 mg daily plus upfront heparin infusion with a bridge to warfarin. Warfarin was administered to a target international normalized ratio (INR) of 2.0 to 3.0 with subsequent discontinuation of heparin. High rates of post-operative bleeding coupled with emerging evidence of a low risk of early post-operative pump thrombosis (11) led to a more conservative anticoagulation protocol, essentially eliminating early post-operative heparin use. By mid-2008, heparin was no longer used routinely to bridge to a therapeutic INR. Initiation of warfarin was generally delayed until 1 to 2 weeks after surgery and administered to a more conservative goal INR of 1.7 to 2.5. The dose of aspirin was also adjusted to include doses of 81 to 325 mg daily. Figure 1 illustrates the timing of bleeding complications relative to time of surgery, heparin use, and initiation of warfarin among our 139 patients.
Primary event definitions
We sought high sensitivity for bleeding events that resulted in transfusion. INTERMACS adverse event definitions were used for primary events, including transfusion criteria of ≥4 U of red blood cells in the first post-operative week or any transfusion associated with bleeding thereafter (Online Appendix).
All bleeding was subtyped while satisfying time-sensitive INTERMACS transfusion criteria based on temporal distance from surgery. Thoracic and mediastinal bleeding was defined as bleeding that met transfusion requirements and required return to the operating room for surgical exploration or was associated with a clinically confirmed source of thoracic bleeding not managed surgically. Gastrointestinal (GI) bleeding was defined as bleeding that met transfusion requirements and was associated with a confirmed upper or lower source of bleeding. Hemorrhagic stroke was considered a major bleeding event, although it is technically listed as a central nervous system event under INTERMACS adverse event criteria. Other bleeding subtypes and cases of hemolysis were categorized separately.
When a clinically significant nonhemolytic anemia met INTERMACS transfusion criteria but a bleeding source could not be identified despite invasive or noninvasive testing, the event was classified as anemia of undetermined source (AUS). We considered AUS events important to capture given the reported association between transfusion and post-operative mortality, although they are not accounted for by INTERMACS adverse event criteria (6,7).
All data were analyzed using SAS software (SAS Institute, Inc., Cary, North Carolina). Continuous data in tables are summarized as mean ± SD. Continuous data in figures are reported as mean with corresponding 68% confidence intervals equivalent to 1 SE. Categorical data are summarized as percentages. Nelson’s cumulative event function provided nonparametric estimates of bleeding prevalence (12). Parametric estimates were obtained by multivariable hazard phase methodology to estimate the instantaneous risk of bleeding relative to temporal distance from HMII implantation (13). See the Online Appendix for a full list of demographic and clinical variables used.
Bleeding events after HMII implantation were modeled as repeated occurrences to determine an effect of prior bleeding on future events. Modulated renewal was used in a multivariable analysis such that primary bleeding outcome events were modeled as time-varying covariates, similar to techniques previously described (13). Applying this methodology, a patient who experienced a primary event was restarted in the model at time zero and followed up for occurrence of subsequent events (14). To this end, our model not only considered pre-operative covariates but also the timing and nonlinear effects of transient risk factors, such as a prior bleed, to predict subsequent events.
Risk factors for post-implantation bleeding were assessed using temporal decomposition hazard methodology to fit data by nonlinear mixed regression (SAS PROC NLMIXED). This technique permitted the assessment of specific risk factors resolved within 2 hazard phases: an early peaking (e.g., initial hospitalization) phase and late increasing (e.g., post-discharge) phase. Variable coefficients were then simultaneously modeled in early and late hazard phases using a log-linear function. Variables with more than 25% of values missing were excluded. For suitable variables with missing data, values were imputed 5-fold using the SAS PROC MI procedure and Markov Chain Monte Carlo technique (15,16). Bootstrap bagging was used to test variables for reliability and then select those for multivariable models (17). Variables were retained if they demonstrated 50% reliability at an alpha level ≤0.07 among 500 bootstrapped samples with automated stepwise selection.
Repeated post-operative INR measurements were analyzed longitudinally and modeled as time-varying covariates for effect on primary bleeding outcomes. Univariate and multivariable regression analyses were performed using the last INR measurement that preceded a bleed. See the Online Appendix for additional details on available INR records and how INR was used in this subanalysis.
The effect of post-surgical bleeding on mortality was analyzed using Nelson’s cumulative event function to obtain nonparametric survival estimates. Parametric estimates were obtained from a multiphase hazard function model of time-related events. Bootstrap bagging was again used to select variables for multivariable models (17). Post-operative bleeding was treated as a time-varying covariate, creating a modulated renewal process to determine the instantaneous risk of death after surgery. Survival analysis among patients treated with BTT was performed with the Kaplan-Meier estimator. Destination therapy patients were excluded from survival analysis to avoid confounding of bleeding effect among this inherently higher risk cohort.
A complete list of baseline characteristics is shown in the Online Appendix. For comparison, several baseline characteristics from our cohort are compared with those of other pivotal HMII studies, along with respective bleeding definitions and event rates (Table 1). Forty percent of our cohort received HMII support as BTT, 38% as bridge to candidacy, and 22% as destination therapy. Most patients had a pre-operative INTERMACS score of 1 (33%) or 2 (22%). The median follow-up time on HMII support was 489 days (interquartile range: 350 to 715 days) for a total of 170.9 patient-years of follow-up. Ten percent of survivors had follow-up >949 days (2.6 years). Among our cohort, 42% had no major bleeding events.
Occurrence and risk of bleeding events
Among the 139 HMII recipients, 81 patients (58%) had 145 major bleeding events (Table 2). The cumulative incidence of bleeding peaked within the first 3 post-operative months (Fig. 2). Most patients with bleeding experienced a single event (56%). A minority of patients experienced repeated bleeding occurrences, which included second, third, fourth, and a single fifth major bleed (Table 2). After adjusting for other factors, a first bleed did not predict subsequent bleeding events in a multivariable model. Moreover, after a first bleed, the probability of subsequent bleeding in the early hazard phase decreased (Table 3). When stratified into eras of implantation, including early-term (2004 to 2006), midterm (2007 to 2008), and late-term (2009 to 2010), the cumulative incidence of major bleeding decreased with each subsequent era (Fig. 3).
Incremental risk factors for overall bleeding were resolved into 2 hazard phases. In the early (i.e., index hospitalization) hazard phase, low values of pre-operative total bilirubin, increased cardiopulmonary bypass time, and early HMII implant year were associated with bleeding (Table 3). In the late (i.e., post-discharge) hazard phase, elevated pre-operative total bilirubin value and increased pre-operative pulmonary artery systolic pressure were independently predictive of bleeding events. Having a prior bleed was associated with a lower risk of future bleeds in the early hazard phase. Although incident major bleeding occurred most frequently among patients with the highest pre-operative INTERMACS acuity score, overall the INTERMACS profile was not a consistent predictor of bleeding (Online Appendix).
Thoracic and mediastinal bleeding predominated all subtypes of bleeding (Fig. 4) and drove the high overall risk of early post-operative bleeding within the first 2 post-operative weeks (Fig. 5). Thoracic and mediastinal bleeding developed at a median of 1 day from surgery (range: 0 to 89 days) and accounted for 42% of all major bleeds. The next most common subtype of bleeding was of GI sources, accounting for 21% of events. For analysis of GI bleeding, upper and lower bleed events were combined into a single category. GI bleeds occurred at a median of 33 days from surgery (range: 1 to 530 days), with the greatest risk within the first post-operative month (Fig. 5). Recurrent GI bleeds were common; one-third were repeat events.
AUS events occurred early and were also common, accounting for 20% of all bleeds. AUS events mirrored the temporal pattern of incident thoracic bleeds, with the majority of events occurring close to surgery (Fig. 4). INTERMACS-defined hemolytic events (plasma-free hemoglobin level >40 mg/dl in association with clinical signs associated with hemolysis occurring in the first 72 h post-implantation) were not observed among those categorized as AUS. A single patient in our cohort experienced INTERMACS-defined hemolysis that required transfusion; this anemia event was not included in the analysis of AUS bleeding.
Coagulation status and post-implantation hemostasis
We considered the last measurement of INR before the event for analysis. In unadjusted models, higher INR levels predicted combined bleeding events as well as thoracic and GI bleeding subtypes. After adjusting for other variables, INR was no longer predictive of a first or recurrent bleed by multivariable analysis (Online Appendix). Over the first year, the mean INR for first-time bleeding events was not significantly different from that of patients without these complications (Fig. 6).
Mortality associated with post-implantation bleeding
Parametric survival estimates among BTT and bridge-to-candidacy HMII recipients at 1, 6, 12, and 24 months of HMII support were 95%, 85%, 74%, and 54%, respectively (Online Appendix). The risk of death in the overall cohort was examined in a multiphase multivariable model of bootstrapped covariates, including time-related bleeding variables. The number of bleeding events increased the risk of death in the early hazard phase (Table 4). In the late hazard phase, elevated creatinine level and prolonged duration from surgery to bleeding were reliable, independent risk factors for death (Table 4). Although a reliable variable for modeling, the number of recurrent bleeding events did not provide independent, incremental prediction of late-term mortality. However, when analyzed over time, those who experienced a second bleed demonstrated a rapid early and sustained reduction in survival relative to those without bleeding or a first-time bleed (Fig. 7).
This study highlights the complex interplay of bleeding complications early after HMII implantation, predominated by early thoracic and mediastinal, GI, and AUS events. We assessed the influence of pre-operative variables on bleeding events as have others, but we also incorporated the dynamic factors of timing, type of bleeding, and anticoagulation status on subsequent events. To our knowledge, this is the first study to resolve the risk of post-HMII bleeding into 2 clinically relevant phases. In doing so, we uniquely demonstrate how time-sensitive factors relate to overall bleeding risk. Overall, these results provide a practical account of major bleeding requiring transfusion after HMII implantation.
A detailed classification of bleeding by subtype demonstrates important relations, including that of early thoracic, GI, and AUS events as well as late GI and central nervous system bleeds (Fig. 4). AUS events likely highlight limitations of endoscopic, noninvasive, and invasive diagnostic testing for post-HMII anemia as well as our retrospective analysis of bleeding events. Although inclusion of AUS events adds complexity to an already heterogeneous literature on bleeding, AUS importantly accounts for a significant portion of transfusions when a definitive source of blood loss is not apparent. Future prospective studies specifically targeting AUS may provide valuable insights into reducing transfusion events, regardless of their cause.
As anticoagulation protocols have evolved and technical challenges of implantation have been overcome, the incidence of bleeding during HMII support has decreased substantially. The risk of bleeding events declined over implantation eras, with the cumulative incidence of bleeding reduced by one-half between the 2007 to 2008 and 2009 to 2010 cohorts (Fig. 3). Although no single factor clearly accounted for these changes, the declining trend in bleeding complications is rather likely a reflection of several concurrent improvements to patient selection, operative techniques, and early post-operative anticoagulation protocols, among other factors. Nevertheless, the pattern of early post-operative bleeding remained consistent in our cohort over implantation eras, characterized by a prompt early rise and precipitous fall shortly after HMII implantation.
Having a history of post-HMII bleeding seemed to reduce the risk of future early-phase but not necessarily late-phase bleeds (Table 3). The mean INR among those who had prior bleeds tended to trend lower than among those with no or first-time bleeds (Fig. 6), possibly representing increased clinical vigilance and reduction in anticoagulation in response to bleeding. Importantly, early, but not late, bleeding complications tended to increase the risk of death in our cohort (Table 4). With the association of transfusion and mortality established (6), our results emphasize the detrimental role of bleeding. Perhaps through an association with early transfusion, these results also focus that risk in the days after surgery. Further prospective analysis of this association between transfusions and outcomes among HMII recipients may prompt strategies to avoid anemia and post-operative transfusions.
Multivariable predictors of bleeding in our cohort included low total bilirubin level and severe pulmonary hypertension. Our finding of a low bilirubin level associated with early bleeding risk seems contradictory to other reports (18–20). Regardless, additional studies are needed to further characterize links between post-HMII bleeding risk, hepatic dysfunction (18), and right ventricular failure (19,21), as has been suggested elsewhere. Considerable interest has focused on syndromes of platelet dysfunction potentially unique to the HMII device, including an acquired type IIA von Willebrand syndrome (22–26). Additional study of blood product use, platelet aggregometry, and thromboelastography may prove useful in perioperative bleeding risk stratification. Similar to scores used to predict post-HMII survival (27,28), an individualized score to predict bleeding could offer tailored post-operative anticoagulation that may limit bleeding.
Thoracic and mediastinal bleeding predominated within the first post-operative week and overall accounted for 42% of all bleeding events. Reported rates of thoracic bleeding in this analysis are considerably higher than the results of previously reported studies, including a 31% rate of reoperation for bleeding in an early experience (25). This noteworthy difference is a result of our goal of high sensitivity for bleeding or anemias that led to transfusion. We accounted for any source of thoracic bleeding, regardless of whether the patient returned to the operating room, as long as a thoracic source could be identified and transfusion criteria were met. Certainly, the wide scope of our definition of thoracic bleeding loses specificity when compared with other studies. All the same, with the use of an inclusive, transfusion-sensitive definition, we found that bleeding requiring transfusion was very common shortly after surgery.
The results of this observational analysis complement existing data with additional detail on post-HMII bleeding complications, although important limitations exist. This study spanned several years of HMII implantation at a single center, and so the ability to generalize these results may be limited. During 7 years of HMII implantation, many changes in the clinical management of device recipients occurred, which provide heterogeneity to our analysis. Furthermore, we did not perform subgroup analysis of INTERMACS score by implantation era. Such an analysis may have confirmed suppositions of reduced bleeding among lower acuity recipients while other era-related factors, such as anticoagulation protocols or implantation techniques, were also changing. Although we attempted to use comprehensive statistical modeling, there are likely influential factors for bleeding for which our analysis did not account. We attempted to use INTERMACS definitions for adverse events with limited manipulation. A retrospective analysis of bleeding events likely contributed to increased AUS categorization.
Bleeding complications peak in the early post-operative course after HMII, demonstrating a complex interplay of bleeding subtypes shortly after surgery. Although the observed decline in bleeding event rates over time is encouraging, bleeding and anemia requiring transfusion are common and associated with increased risk of post-surgical mortality. The additional influence of transfusions and alterations of platelet function during HMII support, particularly early after surgery, merit additional investigation. To this end, axial flow ventricular assist device technology will more safely and effectively fulfill its role among a growing number of patients with advanced heart failure.
For expanded information on Methods and Results, please see the online version of this article.
Dr. Starling has accepted honoraria from Thoratec Corporation, although all funds have been directed to a research fund of the Cleveland Clinic. All other authors have reported that they have no relationships relevant to the content of this paper to disclose.
- Abbreviations and Acronyms
- anemia of undetermined source
- bridge to transplant
- HeartMate II
- international normalized ratio
- Received February 20, 2013.
- Revision received April 7, 2013.
- Accepted May 28, 2013.
- American College of Cardiology Foundation
- Pagani F.D.,
- Miller L.W.,
- Russell S.D.,
- et al.
- Starling R.C.,
- Naka Y.,
- Boyle A.J.,
- et al.
- Eckman P.M.,
- John R.
- ↵INTERMACS Working Group (2008, Oct 30). Manual of Operations for the Interagency Registry for Mechanically Assisted Circulatory Support: Adverse Events Criteria (Version 2.3) [Word document]. Available at: http://www.uab.edu/ctsresearch/intermacs/Appendices_092107.htm. Accessed July 6, 2010.
- Nelson W.
- Kalbfleisch J.D.,
- Prentice R.L.
- Rubin D.B.
- Schafer J.L.
- ↵Kransdorf E, Moriguchi J, Kittleson M, et al. The use of the Model for End-Stage Liver Disease score predicts 90-day mortality in patients undergoing Heartmate II implantation. Paper presented at: American Heart Association Scientific Sessions; November 5, 2012; Los Angeles, CA.
- Matthews J.C.,
- Koelling T.M.,
- Pagani F.D.,
- Aaronson K.D.
- Matthews J.C.,
- Pagani F.D.,
- Haft J.W.,
- Koelling T.M.,
- Naftel D.C.,
- Aaronson K.D.
- Geisen U.,
- Heilmann C.,
- Beyersdorf F.,
- et al.
- Klovaite J.,
- Gustafsson F.,
- Mortensen S.A.,
- Sander K.,
- Nielsen L.B.
- Meyer A.L.,
- Malehsa D.,
- Bara C.,
- et al.
- Uriel N.,
- Pak S.W.,
- Jorde U.P.,
- et al.
- Cowger J.,
- Sundareswaran K.,
- Rogers J.G.,
- et al.