Share this article on:

Gastrointestinal Bleeding in Patients with Ventricular Assist Devices Is Highest Immediately After Implantation

French, Joshua B.*; Pamboukian, Salpy V.; George, James F.; Smallfield, George B.§; Tallaj, Jose A.; Brown, Robert N.; Smallfield, Melissa C.; Kirklin, James K.; Holman, William L.; Peter, Shajan§

doi: 10.1097/MAT.0b013e3182a4b434
Adult Circulatory Support

Ventricular assist device implantation is associated with gastrointestinal bleeding (GIB); however, outcomes in terms of initial and repeat GIB risk, severity, location of lesions, and endoscopic interventions need to be better defined. Consecutive patients from a database of adult patients with ventricular assist devices (VADs) implanted between January 1, 2000, and December 31, 2010, at a single center were reviewed and followed through May 31, 2011, in a retrospective manner. The GIB events were further classified by severity, lesion location, and lesion type. Hazard analysis models were calculated for the time to GIB events. Of 166 patients with a VAD, 38 patients experienced 84 GIB events. Seventeen patients experienced ≥2 GIB events. Maximal hazard for the first bleeding event was 2.23 events/patient-year at 21 days and declined to the constant hazard by 71 days postimplantation. The hazard for recurrent GIB was greatest immediately after the first GIB event. When considering all GIB events, most lesions (68%) were located in the proximal bowel. Angiodysplasia was the most common lesion type (17.5%) seen on endoscopy when all GIB events were considered, whereas ulcers were the most common type (13.8%) seen in initial GIB events. The actuarial risk of initial GIB events peaks in the first 3 months after VAD implantation followed by a stable lower risk of bleeding. The hazard for recurrent GIB events is substantially increased immediately after the initial GIB.

From the *Graduate Medical Education, Department of Medicine, Internal Medicine Residency Program, Department of Medicine, Division of Cardiovascular Disease, Department of Surgery, Cardiovascular/Thoracic Surgery, and §Department of Medicine, Division of Gastroenterology and Hepatology, The University of Alabama at Birmingham, Birmingham, Alabama.

Submitted for consideration March 2013; accepted for publication in revised form July 2013.

Disclosure: The authors have no conflicts of interest to report.

Reprint Requests: Joshua B. French, MD, Graduate Medical Education, Department of Medicine, The University of Alabama of Birmingham, Internal Medicine Residency Training Program, BDB 327, 1720 2nd Ave South, Birmingham, AL 35294-0012. Email:

Heart failure is a systemic and debilitating disease that currently affects 5.8 million adults in the United States.1 In recent years, ventricular assist devices (VADs) have been increasingly used as a therapy for refractory heart failure as a bridge to transplant and more recently as destination therapy, especially as technology has improved the duration of support, correlating with transition from devices based on pulsatile flow to those using continuous-flow principles.2–5 As with any implantable device, VADs are not without risks and complications.

Although gastrointestinal bleeding (GIB) is a known possible consequence of VAD implantation, there are few reports characterizing these bleeding lesions.6–8 Various authors have described a multitude of lesion types and endoscopic procedures used to identify these lesions. Colonoscopy, esophagogastroduodenoscopy (EGD), and capsule endoscopy have been used in attempts to identify these lesions.9 In addition, other investigators have found red blood cell scintigraphy as a viable means to localize these bleeding lesions for selective angiography and therapeutic treatment with embolization.10 Overall, however, there are limited data available characterizing these gastrointestinal lesions and their associated propensity to hemorrhage.

Most studies identifying bleeding lesions in the gastrointestinal tracts of patients with VADs are relatively small. Only recently have larger patient populations been analyzed.11 Strategies to manage GIB include holding or lowering anticoagulation and decreasing the pump speed of continuous-flow devices as a means of decreasing shear stress.12,13 To better characterize these bleeding lesions and the endoscopic management, we investigated a relatively large patient population using VADs to calculate hazard data for initial and subsequent GIB events. We also compared endoscopic findings regarding lesion locations, types of lesions, and endoscopic therapy modalities.

Back to Top | Article Outline


We reviewed the records of patients with a VAD implanted at the University of Alabama at Birmingham University Hospital between January 1, 2000, and December 31, 2010. All patients with a VAD who were 19 years of age or older were included in the study. We reviewed patient medical records for evidence of GIB events that required hospitalization or occurred during hospitalization in the time period in which their VAD was in place. The closing date for inclusion of a GIB event was set as May 31, 2011. A GIB event was defined as a patient having one or more of the following characteristics: melena, hematochezia, hematemesis, guaiac-positive stool, or a bleeding lesion seen on endoscopy or radiography. A recurrent GIB event was defined by the same criteria and occurred at least 7 days after a previous GIB event, with evidence of laboratory stabilization of hemoglobin and hematocrit in the interim. Also included, as recurrent events, were events that occurred anytime after an endoscopic intervention was used to successfully cease visible bleeding of the GIB event. Both inpatient events and the outpatient events requiring hospitalization were included. Pulsatile and continuous left and right ventricular assist devices were included in the study. Gastrointestinal bleeding events that took place within the first 10 days after VAD placement were excluded as a result of the proximity to surgery and other potential sources of blood loss in the immediate postoperative period.

Based on the number of units of packed red blood cells (pRBCs) transfused during hospitalization, the events were classified by severity as either minor (≤4 U pRBCs) or major (>4 U pRBCs). Generally, our institutional practice is to avoid transfusion unless hematocrit is <21%. However, the decision to transfuse for a hematocrit higher than this is left to the physician and based on several clinical factors, such as hemodynamic stability and rate of bleeding. The parametric hazards, with respect to time, were calculated for the time to initial GIB as well as time of freedom from subsequent GIB events based on severity of initial bleeding.

Endoscopic data were collected for those patients who underwent endoscopy during their hospitalization. Data regarding the endoscopic procedure performed, number of lesions identified, lesion locations, and endoscopic intervention that was performed on the identified lesions were collected. Endoscopic procedures included were EGD, flexible sigmoidoscopy, colonoscopy, and push enteroscopy. Lesions were classified according to their appearance in the proximal (esophagus and extending to the ligament of Treitz), middle (ligament to Treitz and extending to the ileocecal valve), or distal (cecum to the rectum) gastrointestinal tract. Nomenclature describing various lesions was not standardized; therefore, lesion types were categorized based on the description provided in the endoscopy report. Data on endoscopic interventions were captured and categorized into the following broad categories: cauterization (Argon plasma coagulation, heater probe, bipolar coagulation, hot biopsy, hot snare, erbium YAG laser, and radiofrequency ablation), mechanical (clipping, banding, snare removal, and lesion removal), and chemical (epinephrine injection).

A lesion was said to be the primary source of bleeding if active bleeding was seen from the lesion, there was stigmata associated with recent bleeding at the lesion site, or it was deemed by the endoscopist to be the most likely cause of bleeding. Secondary lesions or multiple lesions noted at the time of an initial GIB event but deemed not related were not excluded from future data collection if a subsequent GIB event occurred. Data for continuous variables were presented as descriptive statistics with mean and range. Binary variables were presented as totals with percentages. The mean freedom from GIB events was assessed nonparametrically using the Kaplan-Meier method. To determine time-related risk factors, we used a fully parametric multiphase hazard model.14 This methodology allows the decomposition of the instantaneous risk of GIB into independent phases, each of which can be modulated by distinct risk factors.

Back to Top | Article Outline


Of the charts reviewed, 166 patients met inclusion criteria. Selected patient demographics can be found in Table 1. Of the 166 patients, 38 patients experienced at least one GIB event. Seventeen patients experienced two or more GIB events. A total of 84 GIB events were recorded in the 38 patients. There were one to nine separate GIB events per person. The interquartile range for the number of events per person was 1.00–2.25 events. Based on the number of units of pRBCs transfused, of the 38 initial GIB events, 20 (52.6%) were characterized as minor, 14 (36.8%) were major, and four (10.5%) could not be characterized because of incomplete outside records. Of the total 84 GIB events, 51 (60.7%) were classified as minor, 28 (33.3%) were major, and five (6.0%) could not be characterized because of incomplete records (Table 2).

The maximum hazard for initial GIB events was 2.23 events/patient-year at 21 days after initial VAD implantation. After this initial sharp rise in hazard at 21 days, the hazard quickly declined to a constant hazard by 71 days postimplantation (Figure 1). For recurrent GIB events, the hazard for a subsequent bleeding event was maximal immediately after the initial event. It then sharply declined followed by a more gradual trend toward a constant hazard (Figure 2). Similarly, the overall hazard for a major GIB event was highest shortly after VAD implantation as well, regardless of whether it was an initial major GIB event or a major GIB event after a minor or uncharacterized GIB event (Figure 3). To better demonstrate the bleeding incidence between a pulsatile and a continuous VAD, a direct comparison of the bleeding incidence in patients using a HeartMate II and HeartMate XVE was also investigated (Figure 4). These devices were chosen because the HeartMate II and HeartMate XVE were the most implanted devices among the pulsatile and continuous devices. No statistically significant differences were found between these two groups in our analysis.

The regimen of anticoagulation used by patients was also investigated. Most patients were anticoagulated using a regimen that included warfarin. Of the 84 total GIB events, 45 (54%) were on warfarin at the time of the event. Of the initial GIB events among those on warfarin, the average international normalized ratio (INR) was 3.04 (range, 1.67–5.56), and among the total number of GIB events, the average INR was 3.31 (range, 1.26–12). Aspirin and clopidogrel were used for 32 (84%) patients and four (11%) patients, respectively, at the time of the initial GIB event. For all GIB events, both primary and subsequent events, 51 (61%) patients were on aspirin and 13 (15%) were on clopidogrel either as monotherapy or in combination with other agents (Table 3). There were no statistical differences in bleeding events among different regimens of anticoagulation.

Endoscopy was implemented in 92% of initial bleeding events and 93% of all GIB events. Esophagogastroduodenoscopy was the most common endoscopic procedure performed. In most cases, more than one lesion was present on endoscopy, but these other lesions did not always contribute to the active bleeding (Table 4). Of the 84 GIB events, most lesions, 76% of the initial GIB events and 68% of the total events, were located in the proximal bowel. The second most common location of lesions was in the distal bowel accounting for 24% of all GIB events (Table 5). Interestingly, ulcers were the most common cause of initial GIB events (13.8%), but angiodysplasias were the predominant cause when considering all GIB events (17.5%; Table 6). Endoscopic interventions, including mechanical, cauterization, and chemical techniques, were investigated to determine whether one particular intervention provided a longer freedom time from subsequent GIB events. Ultimately, the type of intervention performed on the initial bleeding lesions did not impact recurrent GIB episodes.

Back to Top | Article Outline


There are limited published data in patients with ventricular assist devices regarding the hazard for bleeding as a function of time, lesion location, type of lesion, and endoscopic therapy modalities. The pathophysiology behind the time-related risk for GIB events is poorly understood. There are several postulated explanations for increased rates of GIB in this setting. There is a baseline requirement for anticoagulation and antiplatelet therapy with these devices, which in itself increase bleeding diathesis. Increased shear stress and loss of pulsatility of blood flowing through the VAD lead to loss of high-molecular-weight multimers of von Willebrand factor, resulting in acquired von Willebrand disease. This phenomenon has been previously described in patients with aortic stenosis, known as Heyde’s syndrome. Reduced pulsatility and relative hypoxia of the gastrointestinal mucosa may lead to vascular dilatation and angiodysplasia.7–9,11,15–19 In most VAD patients, GIB is likely multifactorial from a combination of these, and possibly other, mechanisms.

Genovese et al.20 specifically investigated adverse events associated with VADs and found that within the first 60 days of VAD implantation, GIB was one of the most common adverse events. Furthermore, Elmunzer et al.21 recently described GIB from vascular lesions, with an average time from VAD placement to initial GIB event of 32 days. Similarly, in our study, the hazard for an initial GIB event was maximal at 21 days after implantation and then declined to a constant rate within the first 3 months. Furthermore, of these initial episodes, one third were categorized as major GIB events. The hazard for a second GIB event was at its maximum immediately after the initial bleeding event. Because one third of the initial GIB events were categorized as major, it seems clear that these patients require very close monitoring after VAD implantation and immediately after a GIB event. Although the bleeding hazard is highest in the first several weeks after implantation, risk declines over time to a constant baseline hazard. One explanation may be that early bleeding is related to a combination of general postoperative factors in addition to VAD-related factors. With time, as postoperative factors resolve, VAD-related factors remain accounting for the ongoing baseline hazard. A recent study of a large VAD population using the HeartMate II demonstrated a 19% GIB rate. The lesions identified included gastric arteriovenous malformations (AVMs), jejunal AVMs, Mallory–Weiss tears, hemorrhagic gastritis, colon polyps, and ischemic colitis.11 Similarly, we found multiple different lesion types in our population. Angiodysplasias were the most frequent cause of the bleeding, when considering all GIB events. Similarly, a recent study by Elmunzer et al.21 found that vascular lesions were most common. This supports prior studies demonstrating a propensity for VAD patients to form AVMs.8,11 Although angiodysplasias were the most common lesions overall, there were significant numbers of other lesion types that resulted in GIB events. Ulcerative lesions were the most common single lesions seen on endoscopic evaluation of initial GIB events supporting the hypothesis that perioperative factors influence GIB events soon after implantation. Further investigation will be needed to determine whether some early lesions are preventable, for example, with the extended use of high-dose proton pump inhibitors.

In this study, we compared the HeartMate II to the HeartMate XVE to better investigate the GIB incidence between pulsatile and continuous-flow devices. Although increased rates of GIB have been reported with continuous-flow compared with pulsatile devices, we did not observe a difference in bleeding rates between these two groups. This may have been because of the relatively small numbers of patients in each group in our analysis. Furthermore, our study included both short-term and chronic implantable VADs, and a specific analysis comparing short-term and long-term devices was not undertaken. Pulsatile index data were not collected in this study, which also could have influenced GIB event rates in the continuous-flow VAD patients.

In addition to limited published data regarding lesion type and locations, there are limited data evaluating endoscopic interventions for GIB in VAD patients. In our study, >90% of the patients with GIB events underwent endoscopy. Among those, 80% of endoscopy procedures revealed either a culprit lesion or a diffuse bleeding. Therefore, only a minority of procedures failed to visualize a bleeding source. Our report also demonstrated that most lesions visualized during endoscopy were found in the proximal bowel.

In this study, capsule endoscopy, radiographic studies, small bowel enteroscopy, and other methods of mid-bowel visualization were used on a very limited basis leading to the possibility that mid-bowel lesions were under-recognized and therefore under-reported. However, in many cases, mid-bowel visualization was not clinically necessary as a bleeding lesion was seen on EGD or colonoscopy. However, there were a number of EGDs and colonoscopies that did not visualize lesions, suggesting that mid-bowel lesions may have been the source. Our data highlight the role of early endoscopy (especially EGD) as a high-yield initial evaluation strategy in VAD patients with GIB. One type of intervention was not found to be superior over another in terms of preventing future GIB events. Future studies will need to compare the efficacy of various endoscopic therapeutic modalities in a prospective, randomized trial design.

A number of different anticoagulation regimens were used in the study population. The most common anticoagulant regimen consisted of aspirin and warfarin. Our data showed that the average INR was >3 at the time of onset of the GIB event. In our institution, target INR is generally 2–3 for durable devices and 2.5–3.5 for short-term devices. We observed that the average INR for all GIB events was 3.3 at the time of bleed, which is at the upper end of the therapeutic target even for short-term devices. This implies that less bleeding may be observed with lower INRs. However, this was not specifically examined in our retrospective review. Ultimately, balancing the risk of bleeding and the risk of thromboembolism associated with the VAD becomes a clinical decision to be decided by the physician based on patient factors. Morgan et al.22 reported similar findings leading to implementation of lower INR goals in some of their patients with left ventricular assist devices, and Suarez et al.23 recently described an institutional approach and algorithm for anticoagulation in their VAD population. However, randomized trials are necessary to determine whether lower INR goals can provide a morbidity and mortality benefit, balancing bleeding risk versus clotting risk including stroke and device thrombosis.

Our study has several limitations. First, it is a retrospective study at a single institution. Therefore, there may be institutional biases regarding the management of VADs with GIB events. Only the endoscopic data obtained at our institution were reviewed. This improves the consistency of the results by maintaining similar equipment, endoscopists, and follow-up schedule. In our community, endoscopic intervention in VAD patients is routinely referred to our center, so its unlikely events were treated at an outside facility and were missed. Although increased monitoring during the postoperative period might have led to increased detection rates shortly after implantation, in an effort to mitigate this bias, all GIB events occurring within 10 days of VAD implantation were excluded. When evaluating GIB events, all types of VADs were included in the study. Some of these devices were not designed for long-term utilization. Although it is possible that including devices intended only for short-term use might cause increased result variability, ultimately, the inclusion of these devices provides a more realistic clinical cross-section of patients in which these events occur.

Back to Top | Article Outline


Although GIB can be a significant source of morbidity after VAD implantation, it has not been fully investigated. Our data suggest that the highest hazard of GIB associated with VADs occurs relatively early postimplantation. Increased hazard for a second GIB event immediately after the initial event stresses the importance of close follow-up in the immediate postimplantation period and immediately after an initial GIB event. Although risk decreases with time, there is a constant baseline hazard in the chronic phase. Our data support the role of early endoscopy that identified lesions in over 80% of patients. The most common overall lesions seen were vascular in nature and most lesions occurred in the proximal bowel. Prospective trials of endoscopic interventions and anticoagulation strategies are needed to determine optimal approaches.

Back to Top | Article Outline


1. American. Heart Association. . Heart disease and stroke statistics. 2010 update. Circulation. 2010;121:e46–e215
2. Frazier OH, Rose EA, Oz MC, et al.HeartMate LVAS Investigators. Left ventricular assist system: Multicenter clinical evaluation of the HeartMate vented electric left ventricular assist system in patients awaiting heart transplantation. J Thorac Cardiovasc Surg. 2001;122:1186–1195
3. Rose EA, Gelijns AC, Moskowitz AJ, et al.Randomized Evaluation of Mechanical Assistance for the Treatment of Congestive Heart Failure (REMATCH) Study Group. Long-term use of a left ventricular assist device for end-stage heart failure. N Engl J Med. 2001;345:1435–1443
4. Miller LW, Pagani FD, Russell SD, et al. Use of a continuous-flow device in patients awaiting heart trans-plantation. N Engl J Med. 2007;357:885–896
5. Slaughter MS, Rogers JG, Milano CA, et al.HeartMate II Investigators. Advanced heart failure treated with continuous-flow left ventricular assist device. N Engl J Med. 2009;361:2241–2251
6. Letsou GV, Shah N, Gregoric ID, Myers TJ, Delgado R, Frazier OH. Gastrointestinal bleeding from arteriovenous malformations in patients supported by the Jarvik 2000 axial-flow left ventricular assist device. J Heart Lung Transplant. 2005;24:105–109
7. Crow S, John R, Boyle A, et al. Gastrointestinal bleeding rates in recipients of nonpulsatile and pulsatile left ventricular assist devices. J Thorac Cardiovasc Surg. 2009;137:208–215
8. Stern DR, Kazam J, Edwards P, et al. Increased incidence of gastrointestinal bleeding following implantation of the HeartMate II LVAD. J Card Surg. 2010;25:352–356
9. Hayes HM, Dembo LG, Larbalestier R, et al. Management options to treat bleeding in patients supported on rotary left ventricular assist devices: A single-center experience. Artificial Organs. 2010;34:703–706
10. Tulchinsky M. Lower gastrointestinal bleeding diagnosed by red blood cell scintigraphy in a patient with a left ventricular assist device. Clin Nucl Med. 2008;33:856–858
11. Demirozu ZT, Radovancevic R, Hochman LF, et al. Arteriovenous malformation and gastrointestinal bleeding in patients with the HeartMate II left ventricular assist device. J Heart Lung Transplant. 2011;30:849–853
12. John R, Lee S. The biological basis of thrombosis and bleeding in patients with ventricular assist devices. J Cardiovasc Transl Res. 2009;2:63–70
13. Slaughter MS. Hematologic effects of continuous flow left ventricular assist devices. J Cardiovasc Transl Res. 2010;3:618–624
14. Blackstone EH, Naftel DC, Turner ME. The decomposition of time-varying hazard into phases, each incorporating a separate stream of concomitant information. J Am Stat Assoc. 1986;81:615–624
15. Heyde EC. Gastrointestinal bleeding in aortic stenosis. N Engl J Med. 1958;259:196
16. Shoenfeld Y, Eldar M, Bedazovsky B, Levy MJ, Pinkhas J. Aortic stenosis associated with gastrointestinal bleeding. A survey of 612 patients. Am Heart J. 1980;100:179–182
17. Pate GE, Mulligan A. An epidemiological study of Heyde’s syndrome: An association between aortic stenosis and gastrointestinal bleeding. J Heart Valve Dis. 2004;13:713–716
18. Warkentin TE, Moore JC, Morgan DG. Aortic stenosis and bleeding gastrointestinal angiodysplasia: Is acquired von Willebrand syndrome the link? Lancet. 1992;240:35–37
19. Geisen U, Heilmann C, Beyersdorf F, et al. Non-surgical bleeding in patients with ventricular assist devices could be explained by acquired von Willebrand disease. Eur J Cardiothorac Surg. 2008;33:679–684
20. Genovese EA, Dew MA, Teuteberg JJ, et al. Incidence and patterns of adverse event onset during the first 60 days after ventricular assist device implantation. Ann Thorac Surg. 2009;88:1162–1170
21. Elmunzer V, Padhya KT, Lewis JL, et al. Endoscopic findings and clinical outcomes in ventricular assist device recipients with gastrointestinal bleeding. Dig Dis Sci. 2011;56:3241–3246
22. Morgan JA, Paone G, Nemeh HW, et al. Gastrointestinal bleeding with the HeartMate II left ventricular assist device. J Heart Lung Transplant. 2012;31:715–718
23. Suarez J, Patel CB, Felker GM, Becker R, Hernandez AF, Rogers JG. Mechanisms of bleeding and approach to patients with axial-flow left ventricular assist devices. Circ Heart Fail. 2011;4:779–784

gastrointestinal bleeding; ventricular assist device; hazard

Copyright © 2013 by the American Society for Artificial Internal Organs