Stroke and pump thrombosis are serious complications for patients under mechanical circulatory support with left ventricular assist device (LVAD).1–4 Stroke can be the result of a thromboembolic event originating either from the left heart chambers or from the pump itself.5 Thrombus dislodgement from the left atrium or ventricle will pass through the pump or the aortic valve but may not necessarily manifest itself as a clinically significant event, such as a stroke.6 As for today, there is no method to diagnose subclinical thromboembolic events, and it is unknown how many such events precede a clinically significant stroke.
Pump thrombosis is a medical emergency. Its presentation, however, varies widely, ranging from asymptomatic slowly progressing process to acute heart failure. The diagnosis of pump thrombosis is challenging, and the only continuous parameter currently in use is the pump power. Elevation in serum lactate dehydrogenase and free hemoglobin can indicate a thrombus buildup, but this is a nonspecific and intermittent marker.2
It has been demonstrated, using acoustic spectral analysis in both centrifugal and axial VAD, that pump vibration pattern changes when intrapump thrombosis is present.7–9 Kaufmann et al. showed that the presence of vibration in the frequency corresponding to the third harmonic (pump speed × 3) in HeartWare HVAD is a specific and sensitive marker for intrapump thrombosis.
We have recently demonstrated, in an in vitro model, an alternative method to monitor pump vibration using an accelerometer sensor attached directly to the HeartWare HVAD housing.10 The accelerometer detected intrapump thrombosis with much greater sensitivity than the HVAD power. It also enabled the detection of thromboembolic events (“pass-through”) with no apparent change in pump power.
The main aim of this study was to test the ability of an accelerometer to detect thromboembolic events and intrapump thrombosis in vivo.
We hypothesized that 1) the accelerometer third harmonic is a more sensitive parameter for detection of pump thrombosis and thromboembolic events than pump energy consumption, 2) thrombi passing through the pump would lead to an immediate increase in the third harmonic, and 3) a portion of thromboembolic events would lead to persistent change in the third harmonic, a marker of intrapump thrombosis.
We used LVAD pumps (HVAD, HeartWare, Framingham, MA) explanted from patients undergoing heart transplantation. No patient had prior clinical evidence of pump thrombosis. Before pumps were explanted from the patients, their power cord was cut inside the thoracic cavity and pulled through the skin (to avoid risk of infection). The pump was flushed using a water jet through the inflow cannula, in order to remove any blood remnants. The cord of the pump was then reattached and insulated. The pump was immersed in a 5% glucose solution until the experiment day. Before the experiment, an accelerometer was attached to the pump, and both pump and accelerometer function was confirmed using an in vitro model. Low third harmonic amplitude was confirmed by visual inspection of the spectrogram. The same accelerometer was later used for the in vivo experiment.
Different pumps were used for each experiment. One pump was excluded because of high third harmonic amplitude measured by in vitro circuit.
Institutional approval was obtained (13/5,551). We used seven Noroc pigs of either gender (90 ± 9 kg). The animals fasted overnight but had free access to water. Before transportation to the operating theater, the animal was sedated using intramuscular injection of ketamine (30 mg/kg), atropine (0.02 mg/kg), and azaperone (4 mg/kg). Anesthesia was induced with intravenous bolus injection of pentobarbital (1–3 mg/kg) and morphine (0.5 mg/kg). The airway was secured with surgical tracheostomy. The animal was mechanically ventilated using LEON plus ventilator (Heinen + Löwenstein GmbH, Bad Ems, Germany). The initial ventilator settings (FiO2 0.4, tidal volume 10 ml/kg, frequency 15/min, positive end-expiratory pressure 5) were adjusted to achieve arterial pO2 > 10 kPa, and pCO2 4.5–5.8 kPa. Anesthesia was maintained with continuous intravenous infusion of pentobarbital (4 mg/kg/h), morphine (2 mg/kg/h), and midazolam (0.15 mg/kg/h). Maintenance fluid therapy was given at 20 ml/kg/h.
In order to reduce pulmonary hypertension known to develop during cardiopulmonary bypass in pigs, the animal received methylprednisolone (1 g intravenously) and indomethacin (5 mg/kg intravenously).11 Amiodarone was given as an intravenous bolus injection of 300 mg followed by infusion of 60 mg/h together with magnesium sulfate 20 mmol intravenously to reduce cardiac arrhythmias.
Three hundred milliliters of blood was drawn for thrombus material before the animal was anticoagulated with heparin 2 mg/kg. Activated clotting time was targeted at >350 s. The animal received 300 ml colloid infusion (Voluven, Fresenium Kabi AG, Else-Kroner Straße 1 Bad Homburg, Germany) over 10 min to reestablish euvolemia.
The experimental set-up is shown in Figure 1. The external jugular vein was cannulated with a 28-Fr single stage venous cannula (Medtronic Inc, Minneapolis, MN), and the right femoral artery with an 18-Fr arterial cannula (Medtronic Inc), for cardiopulmonary bypass. HVAD was implanted during cardiopulmonary bypass after an orientating echocardiographic assessment ensuring optimal position and direction of the inflow cannula. The pump graft (15 cm) to the ascending aorta was extended with a 5 cm PVC tube with side port to enable measurements of graft flow and pressure. A 20-Fr cannula was placed in the left atrium for thrombus introduction, and its correct position is confirmed by echocardiography.
Left heart failure was established by occlusion of diagonal branches of the left anterior descending coronary artery (LAD) aiming at a graft flow to cardiac output ratio above 0.5.
The animal was weaned from cardiopulmonary bypass before LVAD was started at a pump speed of 1,800 rpm.
Central venous and arterial pressures were measured by catheters in the right internal jugular vein and right internal carotid artery. A pulmonary artery catheter (Swan-Ganz CCOmbo; Edwards Lifesciences, Irvine, CA) was used to measure pulmonary artery pressures and for cardiac output calculation (the average of triplicate thermodilutions). Left ventricular assist device graft flow was measured using System M3 ultrasonic detection system (Spectrum Medical, Cheltenham, England). A three-axis accelerometer (KXM52-1040; Kionix, Inc, Ithaca, NY) was attached to the pump casing as shown in Figure 1. The accelerometer signal was recorded continuously throughout the experiment by customized software on a dedicated computer.
The experiment consisted of both thromboembolic interventions and control interventions (Figure 2). As thromboembolic interventions may lead to persistent changes in the vibration pattern of the pump, the control interventions were performed first.
Control interventions included pump speed change, graft obstruction, and saline bolus injections. Each intervention was composed of serial changes. Pump speed was stepwise increased and included speeds of 1,800–2,200–2,600 rpm. Graft obstruction was performed by stepwise external compression of the graft to achieve a preset target graft flow of 50% and 25% of the baseline flow with open graft at 2,600 rpm. In the thromboembolic intervention, introducing a thrombus into the left atrium was achieved by flushing it from the thrombus injection system using saline injection, and hence we included similar saline injections without thrombus material as one of the control interventions.
The thromboembolic intervention included a series of up to 10 thrombus injections (events) per experiment. The criterion used to stop thrombus injection was arrhythmias with subsequent hemodynamic instability. Thromboembolic events were initiated by flushing thrombus mass (thrombus mass volume: 0.3–0.4 ml measured using a 2.5 ml syringe) into the left atrium through the thrombus injection system (Figure 1). To minimize signal change while injecting thrombus material, we used two different techniques, either bolus delivered by infusion pump or bolus delivered by pressurized infusion set. Using infusion pump (10 ml injection delivered over 1 min by infusion pump) was considered a more gentle method and was attempted first. In the cases the thrombus mass did not advance into the atrium, a saline injection delivered by pressurized set (10 s injection with infusion set pressurized to 300 mm Hg) was used. We tested, in the control interventions, both forms of saline injection to establish whether accelerometer signal changes during thromboembolic event could be the result of the saline injection itself.
During all interventions, the following parameters were recorded: pump speed, pump flow, pump power, graft flow, arterial pressure, and pulmonary artery pressure. Cardiac output was measured after each event in the rpm change and graft occlusion interventions, and at the start and the end of the saline bolus and thromboembolic interventions. The HVAD graft flow was recorded continuously throughout all interventions.
We assumed that thrombus material would follow the blood flow. The in vivo model allowed thrombi to pass both through the pump and through the aortic valve. Thus, we expected that the number of thrombi passing through the pump leading to accelerometer signal changes would be proportional to the graft flow to cardiac output ratio ±10% to account for uncertainty.
The accelerometer signal was recorded continuously during all interventions. Signal analysis has been described in detail previously.10 In brief, the signal of each of the three accelerometer axes was analyzed using sliding window Fast Fourier Transform (Signal Processing Toolbox, spectrogram function utilizing Hamming window, Matlab (R2013b, MathWorks Inc., Natick, MA)). The third harmonic frequency was then extracted from the spectrogram (Figure 3). An automatic program received each event time, and a reference third harmonic value 30 seconds before the event time was extracted. It was used to calculate the value and timing of the maximal third harmonic amplitude change within a 2 min period after the event, and the third harmonic amplitude change at the end of the 2 min period which was considered to be a steady-state value.
The detection of thromboembolic event was based on third harmonic amplitude change in the maximal deflection point within two min period after each event, whereas the detection of intrapump thrombosis was based on persistently high third harmonic value at steady state. The same analysis was performed in all control interventions to test the specificity of the method.
The first, second, and fourth harmonic were also extracted from the spectrogram, and the steady-state value at each event time was recorded. All pumps were opened at the end of the experiment for visual inspection of intrapump thrombosis.
SPSS v.21 (IBM, Armonk, NY) was used for statistical analysis. The accelerometer third harmonic and pump power data did not show normal distribution, and thus these results are reported as median (interquartile range [IQR]). The comparisons were made using Mann–Whitney U test. A p value of ≤0.05 was considered significant.
Receiver operating characteristic (ROC) curve was constructed with calculations of sensitivity, specificity, and cutoff values for the detection of thromboembolic events for third harmonic amplitude change, pump power, and pump power change.
Seven animals (90 ± 9 kg) were included in the study, and the protocol was completed in four of them. Three animals were excluded because of ventricular fibrillation during the surgical preparation.
One of the included animals did not demonstrate any signal change during the first five thrombus injections. Echocardiography showed malpositioning of the thrombus injection cannula tip against the atrial appendage wall, precluding thrombi from entering the atrium. The injection cannula was repositioned under echocardiographic guidance, and the thrombus injections were repeated. The five thrombus injections with malpositioned injection cannula were excluded from the statistical analysis.
The averaged graft flow to cardiac output ratio for all experiments was 0.66 (Table 1). The graft flow to cardiac output ratio, the expected thromboembolic event number, and the actual number of third harmonic amplitude change for each experiment are shown in Table 1.
Thromboembolic Event Detection
In total, 35 thromboembolic events were included. Based on the graft flow to cardiac output ratio (±10%), 21–25 of these thrombi were expected to pass through the pump and thereby potentially leading to signal changes. Thromboembolic events led to an immediate increase (within 5 s) in third harmonic amplitude in 26 of 35 thrombi injections and thus corresponded well to the number of events expected (Table 1).
The increase in third harmonic amplitude as response to thromboembolic events was accompanied with increased amplitude in the surrounding frequencies (Figure 4). The changes in response to thromboembolic events were either short lasting, returning to the level before the event, or followed by persistent elevation in third harmonic (Figure 4).
All the accelerometer axes (Gx, Gy, and Gz, Figure 1) showed similar pattern in response to the thromboembolic events. However, based on all thromboembolic events including thrombi assumed to pass through the aortic valve (n = 35), the Z axis measuring acceleration (vibration) perpendicular to rotational axis of the pump impeller was the most sensitive and specific, with the largest area under the curve (AUC) of 0.954 in the ROC analysis (Figure 5, A). The sensitivity and specificity using a cutoff value of 21.56 were 0.74 and 1.00, respectively (Table 2).
The median accelerometer Gz-axis third harmonic amplitude for all thromboembolic events (n = 35) and the control interventions (n = 47) was 64.5 (IQR: 18.8–107.1) and 5.45 (IQR: 4.2–6.6), respectively (P < 0.01; Figure 5, B).
Pump Thrombosis Detection
Persistent increase in third harmonic amplitude was present in 14 of the 26 thrombus injections that lead to immediate signal change (53%) and was accompanied by elevated amplitude of the first and second harmonics (Figure 6).7
All pumps demonstrated a vibration pattern with persistent elevated third harmonic amplitude at end of the experiments. Intrapump thrombosis with thrombus material on the impeller was verified in all pumps by visual inspection at the end of the experiment.
The control interventions, including pump speed change, graft obstruction, and saline injection, did not cause any marked increase in the third harmonic amplitude (Figure 7).
Pump power showed little or no change during thrombus injections (Figure 3, B). Median pump power was 3.0 W (IQR: 2.9–3.3) in the thromboembolic interventions and 2.8 W (IQR: 2.4–2.9) in the control interventions (P < 0.01; Figure 5, B). The median value of pump power change did not differ significantly between thromboembolic interventions and control interventions, 0.1 W (IQR: 0–0.2) and 0.1 W (IQR: 0–0.5), respectively (P = 0.083; Figure 5, B, midpanel). Pump power demonstrated lower sensitivity and specificity than the accelerometer third harmonic (Figure 5, A), with AUC of 0.759 and sensitivity and specificity of 0.40 and 1.00, respectively, using a cutoff value of 3.15 W. The AUC for the pump power change was 0.39, <0.5 as indicated by the diagonal reference line in Figure 5, A. Thus, sensitivity and specificity for this variable was not calculated.
Pump power values for each pump, at the end of the thromboembolic intervention, were 2.7/3.4/2.4/3.2 W, compared with 3.0/2.9/3.1/3.4 W before the thromboembolic intervention.
The hemodynamic variables remained relatively constant during the experiment (Table 3). Left ventricular assist device graft flow was increased by pump speed increase in the “rpm change” intervention, but cardiac output by the pulmonary artery catheter decreased in response to this. Graft flow decreased as consequence of graft obstruction, but cardiac output remained relatively constant.
To our knowledge, there is no available method for detection of subclinical thromboembolic events. This in vivo experimental study confirms the feasibility of accelerometer signal–based diagnosis of both thromboembolic events and intrapump thrombosis using HeartWare HVAD.
Thromboembolic Event Detection
Our results demonstrated that an accelerometer is a promising new real-time monitoring modality to detect thromboembolic events in HeartWare HVAD. As treatment adjustments pose a risk for the patient, any diagnostic test aiming at thromboembolic event detection should have high specificity. In our ROC analysis, we therefore chose a high cutoff value that assured specificity of 1, with lower sensitivity as a consequence. The somewhat reduced sensitivity (0.74) of the accelerometer, compared with previous in vitro measurements, can be partly attributed to model limitation and inability to detect thrombi that bypass the pump and leaving the ventricle through the aortic valve. When taking flow considerations, i.e., graft flow to cardiac output ratio into account, the estimated and measured data corresponded closely and is in agreement with previous findings using an in vitro model.10
Intrapump Thrombosis Detection
The relationship between thromboembolic events and pump thrombosis is unclear. Persistent elevation in third harmonic amplitude was evident at the end of all experiments, and importantly, intrapump thrombosis was confirmed in all pumps by visual inspection. Acute changes in third harmonic caused persistent third harmonic amplitude change in 53% of the cases. This study suggests that thromboembolic event passing through the pump may, but not necessarily, leave a residual mass on the impeller.
The initiating event in the development of pump thrombosis is unclear. It could be a de novo buildup, or thromboembolism leading to residual mass on the impeller, or a combination of both. Once a persistent high third harmonic is present, it may be difficult to differentiate between them. Our results indicate that an acute thrombus leads to an abrupt occurrence in third harmonic amplitude. It is likely that a gradual buildup of an organized thrombus will lead to continuing increase over time. Analyzing trend in third harmonic over time may be used to differentiate these two situations. Interestingly, in our study all pumps had persistent high third harmonic after the thromboembolic events, and fibrin residual on the impeller was observed in all pumps after the experiments. No high third harmonic was present during the control interventions which were performed before the thromboembolic intervention. This result indicates that an organized thrombus may develop in response to thromboembolism. Further studies are needed to answer these questions, preferably using a survival in vivo model with continuous accelerometer monitoring over time with thromboembolism or placebo as interventions.
Despite the fact that pump power was higher in the thromboembolic interventions, the median amplitude difference was only 0.2 W per thromboembolic event and under 1.0 W for the whole intervention. These values are lower than the values used today clinically for diagnosing pump thrombosis.12,13 The minor changes in pump power and decline in pump power by the end of the experiments might be partly attributed to concomitant flow obstruction caused by massive pump thrombosis. The minimal change in pump power despite considerable changes in the third harmonic amplitude demonstrates that pump power is less sensitive and probably is a relatively late manifestation of pump thrombosis.
Our finding regarding intrapump thrombosis detection using accelerometer agrees with previous findings using acoustic spectral analysis. However, an important difference is that unlike the acoustic signal analysis, the accelerometer signal does not require signal averaging. Signal averaging leads to temporal information loss and prevents detection of short lasting events, such as thromboembolism. The fact that changes in third harmonic amplitude were easily detected in the nonaveraged accelerometer signal demonstrates the superiority of this technology to acoustic spectral analysis. A study comparing the two methods should be performed.
We did not test other pump types in this experiment; however, investigations of other pumps using acoustic spectral analysis for the detection of pump thrombosis have showed promising results.8,9 Because the acoustic and accelerometer signals originates from the same physical phenomenon, we believe that an accelerometer-based method will be usable for all pumps where acoustic methods have been shown to detect pump thrombosis.
It is probable that accelerometer signal analysis can be further refined including temporal qualitative morphological changes, as well as adopting other methods as signal phase shift analysis.
Despite efforts to improve pump technology, the combination of bleeding disorders and thrombotic complications coexist in the LVAD patient population.3,14,15 Detecting thromboembolic events before pump thrombosis occurs can allow anticoagulation adjustment before pump thrombosis. It is likely that a patient suffering from silent thromboembolism is at higher risk of developing clinically significant stroke or other end-organ failure. Early diagnosis of subclinical thromboembolism can initiate earlier diagnostic investigations such as computed tomography and echocardiography. In case of thrombus, for instant in the atrium, ventricle or around the HVAD inflow cannula, adjustment in anticoagulation could be made earlier. Another possibility with accelerometer monitoring is guidance of effect of anticoagulation for the individual patient. The potential goal is to reduce the number of clinical thromboembolic complications related to HVAD treatment. Continuous monitoring of third harmonic amplitude by accelerometer can also provide valuable information on the speed of development of a thromboembolic process. We believe that integrating the sensor into the pump housing and transmitting the data to a log file in real time is the way to address these important clinical challenges. This data can in turn be transferred automatically to the treating facility allowing for early evaluation and intervention. Such practice should be investigated in clinical trials before conclusive recommendations can be made.
Currently, the pump uses wired technology for energy supply. This makes using wired technology for accelerometer data transfer unproblematic, as it does not increase infection risk. As wireless solutions for LVAD becomes available the accelerometer can utilize the same technology.
We used explanted pumps. Low third harmonic amplitude near noise level was present during the in vitro testing of the pumps before the experiment and during control interventions (Figure 4, lower panels). This was most probably the result of high accelerometer signal quality with low noise level and not because of pump thrombosis. This has to be confirmed in pumps that have not been implanted.
In the porcine heart, the left ventricular cavity is small with high contractility. To deal with this, we 1) used large animals to give room for the LVAD cannula (90 kg pig), 2) occluded several diagonal branches of the LAD coronary artery to reduce contractility. By these precautions, suction of the left ventricular cannula was avoided.
We could not determine precisely how many thrombi that were actually passing through the pump or through the aortic valve, and, therefore, in the statistical analysis, we included all thromboembolic interventions regardless of the assumed path. A model with reliable method to determine thrombus course is a desirable improvement of the current model. This could include labeling of the thrombus and using imaging to detect its path in the system.
Our model did not induce a global left ventricular failure but regional dysfunction with intact loading response in the remaining contracting myocardium. The decrease in cardiac output during the pump speed change intervention, and the maintenance of cardiac output during the graft obstruction intervention, suggests active Frank–Starling mechanism with compensatory response to preload conditions. Despite the maintenance of contractility, we managed to achieve a high ratio of graft flow to cardiac output with a good correspondence between the estimated and measured number of thromboembolic events passing through the pump.
Our results demonstrated the feasibility of accelerometer-based in vivo detection of thromboembolic events and pump thrombosis during HeartWare HVAD treatment. The method opens a new possibility for continuous real-time monitoring of thromboembolic events and pump thrombosis and holds the potential to monitor the effects of treatment.
The authors greatly acknowledge the Department of Cardiothoracic Surgery for providing equipment and knowledge, in particular VAD coordinator Gro Sørensen, and the staff at the Intervention Centre and Dr Hilde Karlsen for excellent assistance during the experiments.
1. Willey JZ, Gavalas MV, Trinh PN, et al. Outcomes after stroke complicating left ventricular assist device. J Heart Lung Transplant 2016.35: 1003–1009
2. Starling RC, Moazami N, Silvestry SC, et al. Unexpected abrupt increase in left ventricular assist device thrombosis. N Engl J Med 2014.370: 33–40
3. Boehme AK, Pamboukian SV, George JF, et al. Anticoagulation control in patients with ventricular assist devices. ASAIO J 2017.63: 759–765
4. Loyaga-Rendon RY, Jani M, Fermin D, McDermott JK, Vancamp D, Lee S. Prevention and treatment of thrombotic and hemorrhagic complications in patients supported by continuous-flow left ventricular assist devices. Curr Heart Fail Rep 2017.14: 465–477
5. Potapov EV, Krabatsch T, Buz S, Falk V, Kempfert J. Cerebral protection system applied during washout of thrombus occluding inflow cannula of HeartWare HVAD left ventricular assist device. J Heart Lung Transplant 2015.34: 1640–1641
6. Das RR, Seshadri S, Beiser AS, et al. Prevalence and correlates of silent cerebral infarcts in the Framingham offspring study. Stroke 2008.39: 2929–2935
7. Kaufmann F, Hörmandinger C, Stepanenko A, et al. Acoustic spectral analysis for determining pump thrombosis in rotary blood pumps. ASAIO J 2014.60: 502–507
8. Yost GL, Royston TJ, Bhat G, Tatooles AJ. Acoustic characterization of axial flow left ventricular assist device operation in vitro
and in vivo
. ASAIO J 2016.62: 46–55
9. Hubbert L, Sundbom P, Loebe M, Peterzén B, Granfeldt H, Ahn H. Acoustic analysis of a mechanical circulatory support. Artif Organs 2014.38: 593–598
10. Schalit I, Espinoza A, Pettersen FJ, et al. Accelerometer detects pump thrombosis and thromboembolic events in an in vitro
HVAD circuit. ASAIO J 2018.64: 601–609
11. Hobbhahn J, Conzen PF, Zenker B, Goetz AE, Peter K, Brendel W. Beneficial effect of cyclooxygenase inhibition on adverse hemodynamic responses after protamine. Anesth Analg 1988.67: 253–260
12. Goldstein DJ, John R, Salerno C, et al. Algorithm for the diagnosis and management of suspected pump thrombus. J Heart Lung Transplant 2013.32: 667–670
13. Scandroglio AM, Kaufmann F, Pieri M, et al. Diagnosis and treatment algorithm for blood flow obstructions in patients with left ventricular assist device. J Am Coll Cardiol 2016.67: 2758–2768
14. Morrison KA, Jorde UP, Garan AR, Takayama H, Naka Y, Uriel N. Acquired von Willebrand disease during CentriMag support is associated with high prevalence of bleeding during support and after transition to heart replacement therapy. ASAIO J 2014.60: 241–242
15. Uriel N, Pak SW, Jorde UP, et al. Acquired von Willebrand syndrome after continuous-flow mechanical device support contributes to a high prevalence of bleeding during long-term support and at the time of transplantation. J Am Coll Cardiol 2010.56: 1207–1213