Pump thrombosis (PT) and stroke are serious complications of left ventricular assist device (LVAD) treatment. The occurrence of PT is an estimated 0.08 events per patient year. Thromboembolic event (TE)-related adverse events include ischemic stroke, transient ischemic attack, and peripheral thromboembolism, and their occurrences in HeartWare HVAD are 0.08, 0.07, and 0.1 events per patient year, respectively.1 PT is a medical emergency that prompts immediate diagnosis and treatment. The diagnosis of pump thrombosis is indirect and based on hemolysis parameters, pump energy consumption, and the clinical picture. An elevation of serum lactate dehydrogenase precedes clinically detectable TE/PT and suggests a gradual build-up.2 There is no diagnostic method for subclinical TE detection. Intermittent methods may miss the early signs of TE and PT. Thus, novel methods that enable continuous monitoring and accurate detection of these serious complications are warranted.
There has been an increasing interest in pump sound analysis for the detection of PT.3–5 Pump sound is the result of pump vibrations detected at the skin. Acoustic spectral analysis utilizes pump sounds recorded intermittently by a stethoscope. This approach enables quantification of the pump vibration frequencies. Distinct changes in the pump sound pattern in response to embolic events in an in vitro model using HeartMate II have been demonstrated.5 Yost et al.3 showed similar acoustic spectral changes during pump thrombosis using in vitro and in vivo measurements in HeartMate II. Kaufmann et al.4 have demonstrated that the frequency corresponding to the pump speed × 3 (third harmonic) is a sensitive and specific marker for PT in HVAD. The reason for this is that a thrombus mass attached to the impeller shifts its center of gravity and causes its displacement from the normal central position via the centrifugal force. When it approaches one of the three coils, it is drawn back toward the central position again. This occurs three times per revolution and results in an amplitude change of the third harmonic. The change depends on the mass of the thrombus and its position.
Our idea was to measure pump vibrations directly and continuously by mounting an accelerometer sensor on the LVAD casing. The aim of the study was to demonstrate the feasibility of accelerometer-based PT and TE detection. We hypothesized that the accelerometer sensor could detect changes in the third harmonic amplitude in response to TEs passing through the pump with high sensitivity and specificity. We also hypothesized that a proportion of TEs will lead to a persistent change in the third harmonic amplitude indicative of PT. We tested the accelerometer in an in vitro model with interventions that included TE, changes in pump speed, and pre- and afterload.
We used explanted LVAD pumps (HVAD, HeartWare, Framingham, MA) to build an in vitro model (Figure 1). Three in vitro experiments were performed with different pumps for each experiment. The LVAD was suspended in an elastic chamber to enable changing wall tension conditions. There was no pulsatile element in the model, resulting in a steady mock loop (i.e., does not mimic the beating heart). Pre- and post-pump limbs were fitted with adjustable clamps for pre- and afterload adjustments, respectively. The pre-pump pressure (“preload”) and post-pump pressure (“afterload”) were measured using a patient monitor (Siemens AG 7000C, Munich, Germany). The flow was measured in the post-pump limb using a System M3 ultrasonic detection system (Spectrum Medical, Cheltenham, England). A thrombus injection port joined the pre-pump limb immediately before the elastic chamber. A three-axis accelerometer (KXM52-1040, Kionix, Inc., NY, USA) was attached to the pump housing (Figure 2).6 The accelerometer functionality was tested against the gravitational field confirming 1 G in each axis. Low signal noise was visually confirmed before pump activation. The in vitro model was filled with heparinized whole blood (activated clotting time >400 seconds) collected from a pig.
Porcine blood (200 mL) was collected in a clot activator tube, centrifuged via 2500 revolutions per minute (rpm) for 15 minutes at 4°C, and left to coagulate for a minimum of 1 hour.
Our protocol included TE and control interventions (Figure 3). According to Kaufmann et al., thrombus residue in the pump leads to persistent signal changes; therefore, control interventions were performed before TE interventions. All interventions started and ended with baseline recordings under standardized settings: pump speed of 2800 rpm and open pre- and afterload clamps. Control interventions included pump speed change (n = 15), preload decrease (n = 15), afterload increase (n = 15), and saline bolus injections (n = 15). Each intervention included event series that matched the intervention type. Preload decrease (pre-pump clamping) and afterload increase (post-pump clamping) interventions included 20%, 40%, 60%, and 80% flow reductions from baseline at the beginning of the intervention, as measured by Spectrum Medical M3. We did not open the clamps between events. The clamps were opened at the end of the intervention (after the last event) to return to the baseline settings. The pump speed rates were 2400, 2600, 2800, 3000, and 3200 rpm. As the TE intervention included a saline 0.9% bolus injection, we performed similar standardized saline injections as part of the control interventions. Saline injection was performed via a 10-second-long injection using a pressure set pressurized to 300 mm Hg.
The TE intervention included five randomized repeated injections of thrombi (0.2–1 mL in 0.2 mL increments). Thrombi were flushed into the afferent limb through the thrombus injection system with a standardized saline bolus injection (Figure 1). The first experiment included one thromboembolic intervention of five repeated thrombi injections, whereas experiments 2 and 3 included 2 × 5 TE interventions (total n = 25).
For each event accelerometer, the hemodynamic and pump parameters were recorded at a steady state after a 2-minute “hands off” period. In addition, the maximal change in the pump energy consumption during the “hands off” period was registered. As HeartWare HVAD standard logging is performed with 15-minute intervals, it was considered insufficient for our purpose. The LVAD parameters were recorded as presented on the monitor, which corresponds to the best-case scenario in the clinical setting.
All pumps were disassembled at the end of the experiment for visual confirmation of PT.
Accelerometer Signal Analysis
The accelerometer signals (Gx, Gy, and Gz axes) were recorded on a dedicated computer (sampling rate 500 samples/s). Further signal analysis was performed in Matlab (R2013b, MathWorks Inc., Natick, MA).
Fast Fourier transform provides a signal in the time domain with an alternative representation in the frequency domain. The raw signal is “broken” down to a set of sinusoids, and summation reproduces the original signal. Each sinusoid is defined by its period, amplitude, and phase. In this article, we report the period (frequency) and amplitude.
Axes Gx, Gy, and Gz were analyzed using sliding window fast Fourier transform (Matlab, Signal Processing Toolbox, spectrogram function utilizing Hamming window). The third harmonic was extracted from the spectrogram.
All events were analyzed for both the third harmonic amplitude and third harmonic amplitude change. Each event was followed by 2-minute hands off. The value at the end of this 2-minute interval was considered to be the steady state value. We calculated the acute and steady state third harmonic amplitude change by subtracting the reference third harmonic value 30 seconds before the event from the values at maximal deflection and the steady state points, respectively. Acute temporary changes were defined as the signal returning to the reference value at the steady state 2 minutes after the event. This was performed by a semi-automatic Matlab procedure that identified the point of maximal deflection in the third harmonic amplitude within a 2-minute interval for each event.
The statistical analysis was conducted using SPSS v.21 (IBM, Armonk, NY). Data are reported as median and interquartile (IQR) range, and comparisons between TE and control interventions were performed using the Mann–Whitney U test. Receiver operating characteristics (ROC) analysis was used to estimate the sensitivity/specificity and threshold values for the methods (accelerometer third harmonic amplitude/third harmonic amplitude change and HVAD energy consumption/change in pump energy consumption) for the detection of TE.7 A p-value ≤0.05 was considered significant.
The protocol was completed in all experiments with accelerometer signals of high quality. All pumps had a massive thrombi mass on the impeller at the end of the experiment (Figure 4).
TE led to visible changes in the accelerometer raw signal that were temporary or prolonged (Figure 5 “Accelerometer Signal”). Alterations in the pre- and afterload also led to signal changes. Some saline bolus injections led to minor temporary signal changes during injection. In contrast to the TE interventions, the signal always returned to its original baseline after the standardized baseline settings were reestablished.
Although the accelerometer signal changes were clearly visible during most interventions, interpretation with respect to their etiology was difficult, and further signal analysis was fast Fourier transform based.
The spectrogram obtained from the accelerometer signals showed a similar pattern as described by Kaufmann et al. using acoustic spectral analysis, with prominent amplitudes at the pump speed frequency (first harmonic) and its second, third, and fourth harmonics (Figure 5 “Spectrogram”). The pattern was similar for all three accelerometer axes.
The spectrogram values showed little inter-pump variability (different experiments) at baseline, and further analysis was performed on pooled data from all three in vitro experiments.
Third harmonic analysis.
The third harmonic amplitude was considerably lower than the first, second, and fourth harmonics at baseline, around the amplitude of the nonharmonic frequencies. All control interventions had low third harmonic amplitudes in all accelerometer axes, regardless of the intervention type (Figure 5 and Figure 6A).
TE led to an acute increased amplitude in the third harmonic that could be temporary or prolonged (Figure 5A and Figure 6A). Temporary changes in the third harmonic were often part of a general increase in amplitude involving many adjacent nonharmonic frequencies (Figure 5A, TE1 and TE2). Prolonged changes typically exhibited prominent third harmonic amplitude increases, whereas changes in the nonharmonic frequencies subsided compared with the reference values (Figure 5A, TE3). Similar changes in the nonharmonic frequencies were not present in the control interventions.
After a prolonged third harmonic amplitude increase, the third harmonic amplitude remained high throughout the rest of the experiment. Further changes in the accelerometer signal as a result of new TEs led to an increase or decrease in the third harmonic amplitude.
Comparison of Thromboembolic and Control Interventions
For all accelerometer axes, the third harmonic amplitude and third harmonic amplitude change were significantly higher during TE than the control interventions (all p < 0.01) (Figure 6A). The ROC analyses for the different accelerometer axes are shown in Table 1. The third harmonic amplitude change, measured on the Gz axis, distinguished TE from the control intervention with a sensitivity of 0.92 and a specificity of 0.94 with an area under the curve of 0.966 (confidence interval 0.92–1, p < 0.001) (Figure 4C). Twenty-three of 25 TEs caused acute third harmonic amplitude changes above the ROC analysis threshold value; however, there was no correlation between the thrombus size and the level of third harmonic amplitude change (Figure 7). Sixteen of 25 thromboembolic interventions led to prolonged third harmonic amplitude changes.
Pump energy consumption.
In contrast to the accelerometer third harmonic, the HVAD energy consumption changes were minor for all interventions. The highest energy consumption values for the control and TE events were 5.8 and 5.0 W, with changes in energy consumption of 2.5 and 0.6 W, respectively. Furthermore, in contrast to the accelerometer measurements, the change in pump energy consumption was significantly higher in the control interventions than in the TE interventions: 0.55 W (IQR 1.0) and 0.2 W (IQR 0.3), respectively (p = 0.01) (Figure 6B, left panel). The maximal energy consumption change per experiment was 0.6, 0.7, and 1.2 W.
There was no significant difference in the HVAD energy consumption for the thromboembolic and control interventions: median 3.9 W (IQR 1.0) and 3.5 W (IQR 1.1), respectively (p = 0.153) (Figure 6B, right panel).
The ROC analysis of both HVAD energy consumption and energy consumption change showed a lower area under the curve and sensitivity/specificity than the accelerometer third harmonic (Table 1 and Figure 6C).
Both the control and TE interventions led to substantial hemodynamic changes (Figure 7), with little inter-pump variability for the different hemodynamic interventions.
This investigation is the first study to present data on TE and PT detection in LVAD using an accelerometer. The accelerometer showed excellent ability in diagnosing these conditions using the third harmonic amplitude change. Twenty-three of 25 thrombi caused acute third harmonic changes, and more than 60% of the thrombi caused a prolonged increased amplitude, which is indicative of PT. In contrast, the HVAD energy consumption was not sensitive or specific in the diagnosis of TE and PT.
It is unknown how many subclinical TE events precede a clinically significant stroke and are implicated in the development of a significant PT. To date, there is no method for monitoring TE events. The accelerometer detected TE events with very high sensitivity and specificity. This is an important finding, as it enables the early detection of TE and PT and provides an opportunity to intervene before irreversible function loss. All third harmonic value acquisitions were obtained by a semiautomatic procedure based on the event time and may be easily automated by scanning the signal for acute or gradual changes in the third harmonic amplitude. This creates the possibility for continuous real-time monitoring. An accelerometer is a low-cost, reliable technology currently used in many different applications. There are few risks concerning patient safety that hinder its integration into the pump as it does not interfere with the LVAD reliability, affect coagulability, or increase the risk of infection. Previous studies that employed acoustic spectral analysis used signal averaging and could not distinguish the third harmonic in the absence of PT.8 We did not average the signal, and the low-level third harmonic amplitude was visible in all pumps, as well as the control interventions. This suggests a very low signal–noise ratio and superior signal quality compared with an acoustic signal. An acoustic spectral analysis is intermittent, whereas an accelerometer signal is acquired continuously and analyzed in real time. With its considerable advantages and minor patient safety issues, it is logically the next step in LVAD monitoring technology.
Four of five thrombi sized 0.2 mL were detected, which indicates that the accelerometer is a sensitive method. Nevertheless, the minimal embolus size detectable by accelerometer is unknown and requires further investigation. We also determined that there was no correlation between the thrombus size and accelerometer signal change. This finding may be explained by the different routes a thrombus mass can pass through the pump (primary, secondary, or tertiary flow paths), thus leading to varying degrees of vibration disruptions.
The maximal pump energy consumption and change in pump energy consumption were only 5 and 1.2 W, respectively, which are below the threshold values recommended for the diagnosis of PT (10 and 2 W, respectively).7 This finding was obtained despite the fact that PT was verified by the end of all experiments. We identified a higher change in energy consumption in the control interventions than the TE interventions. This was partly a result of the vigorous hemodynamic changes and subsequent flow changes during the control interventions. Nevertheless, a high energy consumption change with a concomitant absence of change in the third harmonic during the control interventions demonstrates the inferiority of pump energy consumption compared with the accelerometer.
The limited change in pump energy consumption may be a result of concomitant flow obstruction and thrombus mass on the impeller, which affect the pump energy consumption in opposite directions. It is unclear how often these two mechanisms occur simultaneously; however, it is plausible in the setting of thrombus dislodgement from an intracardiac thrombus mass. These results raise doubts regarding the accuracy of energy consumption in the detection of pump thrombosis.
We propose that an accelerometer will be integrated into the pump. Accelerometer data transfer and power supply can follow the same route utilized by the pump. Currently, wired technology following the pump energy supply cable is sufficient. As wireless technology becomes available, the same solution may be implemented for the accelerometer.
Monitoring the pump for thromboembolic events depends on continuous monitoring as the signal changes are transient. We suggest a buffer-based analysis with a sliding window spectrogram. The numbers of events suspected to be thromboembolic events may be summated and logged with predetermined intervals. The analysis of these data and the combination with other information sources, such as gyrus information, may help to distinguish mechanical disturbance and thromboembolic events. It is plausible that patients with an increased thromboembolic event rate as diagnosed by the accelerometer are at an increased risk of ischemic stroke and would benefit from an adjustment of anticoagulation/antiplatelet therapy.
Thromboembolic event detection requires continuous monitoring. Pump thrombosis may be diagnosed by a single measurement using the third harmonic amplitude. This may be easily achieved by intermittent measurements. However, early detection requires frequent measurements. Continuous monitoring of the third harmonic and its rate of change may provide additional information regarding the time window available for intervention and the treatment effect. We believe that continuous monitoring with an accelerometer provides significant advantages compared with intermittent measurements.
When a problem is suspected, the logged information may be easily sent to the treating facility. Pump thrombosis is a medical emergency, and its presence mandates medical monitoring and treatment. The presence of an increased third harmonic amplitude should alarm the patient regarding a potential problem. An automatic data transfer (streaming) to the treating hospital is easily achieved and will further improve prehospital diagnosis and facilitate early treatment.
We employed a static model, which does not simulate the beating heart. Kaufmann et al.4 have demonstrated that the third harmonic is not affected by the beating heart. However, this finding was based on an averaged signal and is therefore inappropriate for the detection of short lasting signal changes. For this first experiment using an accelerometer, we chose to isolate the pump from other dynamic elements (among others the beating heart) to clearly define pump-related signal changes. The ability to identify and distinguish these changes in a dynamic system (beating heart) should be the subject of future investigations in an in vivo model.
The accelerometer third harmonic was only tested in HeartWare HVAD and most likely depends on the pump structure. However, acoustic spectral analysis has been suggested as a diagnostic method for PT in HeartMate II.4,5 It is therefore likely that an accelerometer may be used to detect similar changes. As a TE passes through the pump, it temporarily disrupts its vibration pattern, which will logically lead to phase shift regardless of the pump type. The place of phase shift in TE detection is currently being tested by our group. We used explanted pumps that could potentially be damaged. However, the in vitro model showed minor inter-pump variability between experiments and good reproducibility between measurements. The pump mass and location were confirmed only at the end of the experiments and not after each TE event. Therefore, it was impossible to determine the correlation between the third harmonic amplitude and the presence, size, and location of thrombus on the impeller.
It was difficult to test pump thrombus development de novo. This will require the reversal of heparinization or non-heparinized blood. A pilot project showed massive thrombosis in the reservoir and, to some degree, in the tubes of the model after reversing heparinization, which made it difficult to maintain a steady state. One can deduce from the paper by Kaufmanns et al.4 combined with the evidence provided in this study that pump thrombosis will be easily detected using the third harmonic amplitude change over time.
Future studies should concentrate on confirming these findings in large animal models and patients.
The results from this study demonstrate the feasibility of an accelerometer for the real time detection of TE and PT in HeartWare HVAD. The accelerometer was superior to the routine method pump energy consumption to detect TE and PT. This method creates the possibility for better continuous monitoring of LVAD without affecting patient safety.