New insights from a computational model on the relation between pacing site and CRT response

Introduction

Cardiac resynchronization therapy (CRT) produces clinical benefits in chronic heart failure patients with reduced left ventricular (LV) ejection fraction and left bundle branch block (LBBB).¹ LBBB causes delayed activation of the LV free wall (LVfw) relative to the septum, and consequently, dyssynchronous myocardial contraction and diminished cardiac pump function. In CRT, biventricular (BiV) pacing is employed to resynchronize the electrical activation and thereby to restore a more synchronized contraction of the ventricles.

The position of the pacing site on the LVfw is considered one of the important determinants of CRT response.^2,3 Current guidelines recommend to place the LV electrode, whenever possible, in a specific anatomical region, i.e. the postero-lateral region and away from the apex.^4,5 However, some studies report no significant difference in response to CRT when pacing from various circumferential and longitudinal positions on the LVfw.^6,7 In addition, the region that resulted in the best response is inconsistent between studies.^2,3 Alternatively, it has been proposed to position the LV electrode in the latest activated region (LAR) during intrinsic activation. Indeed, a positive correlation was found between the activation time in the LAR and response to CRT when pacing from this site.^8,9 However, most of this evidence has been derived from comparing single pacing sites between patients, rather than comparing the effect of pacing sites in the same patients and if so, only a few sites were accessible.¹⁰

The aim of this study was to investigate the relation between electrical activation time and cardiac function in a finite element (FE) model of left and right ventricular electromechanics. This approach enabled systematic investigation of the relation between LV pacing site and cardiac function. We assessed cardiac function during normal electrical activation, LBBB and fascicular blocks, and BiV pacing with different LV electrode positions. Subsequently, we investigated which LV pacing site resulted in maximum improvement of cardiac function.

Methods

The computational model of cardiac electromechanics has been published previously.¹¹ Details on the constitutive relations of the myocardium and parameter values are provided in the Supplementary material (see Supplementary material online) and a summary is given below.

Model of cardiac mechanics

The human right and left ventricle (RV and LV, respectively) were approximated by two intersecting ellipsoids. RV and LV wall volumes were set to 40 ml and 160 ml, respectively. The geometrical parameters¹¹ were chosen such that the resulting geometry was representative for a normal human heart (Figure 1A).

Figure 1

(A) The ellipsoidally shaped biventricular finite element (FE) mesh geometry for mechanical simulations incorporated in a lumped parameter model of the circulation. The valves are modelled as ideal diodes. Blood flow and storage of blood in both pulmonary and systemic circulation is modelled using resistances and compliances. We refer to the see Supplementary material online for the parameter settings and details on the numerical implementation. (B) Activation times (AT) in the right and left ventricle (RV and LV, respectively). Distributions of AT are shown without (top row) and with (bottom row) epicardium. In the normal situation (NORM, left), activation starts in five exit points (white stars) at both the RV and LV endocardium. During LBBB (middle left), activation only started in the RV. Cardiac resynchronization therapy is simulated by point stimulation at the RV endocardial apex and simultaneous stimulation at one of the locations on the LV-free wall (LVfw) as indicated by the symbols (middle right). An example of the activation pattern during BiV pacing with the LV pacing site at the mid-level antero-lateral wall is shown on the right.

Open in new tabDownload slide

The large spread in experimental data on myofiber orientation makes it difficult to motivate the choice within the range. We previously showed that small variations within the experimental range can have large effects on cardiac function.¹¹ To overcome this difficulty, the myofiber orientation was created by local adaptation of the myofiber orientation in response to fiber cross-fiber shear.¹¹ First, a rule-based longitudinal component of the myofiber orientation was prescribed, which varied non-linearly from ∼65° at the endocardium to ∼−50° at the epicardium. Subsequently, local adaptive myofiber reorientation was simulated until changes in local and global cardiac function reached a steady state. In the adapted myofiber orientation, a transmural component in myofiber orientation developed whereas major characteristics of the longitudinal component remained the same. In this way, resulting myofiber orientations were still within the experimental range and a better agreement between modelled and measured deformation data was observed.

Myocardial mechanical tissue properties were described by a passive and an active component.¹² The passive myocardium was assumed non-linearly elastic, transversely isotropic and nearly incompressible. The magnitude of active stress depended on time elapsed since activation, sarcomere length and sarcomere-shortening velocity.

Pre-load and after-load were simulated by the interaction of the heart with a lumped parameter model representing a human pulmonary and systemic circulation (Figure 1A). Parameter values for the resistances and compliances in the circulation model are listed in Supplementary material online Table S2.

Model of electrical wave propagation

The spatial distribution of the electrical activation time (AT) in each nodal position was determined by solving the Eikonal-diffusion equation in the end-diastolic configuration.¹² Computed ATs were used to initiate development of active stress in the model of cardiac mechanics.

In the normal case (NORM), activation started (AT =0 ms) from five Purkinje exit points defined at both the RV and the LV endocardium (Figure 1B, left). These exit points were based on three-dimensional mapping studies in the human heart.¹³ From the exit points, propagation of the wave front was assumed anisotropic with a conduction velocity of 0.70 m/s along the myofiber direction (c_f,myo) and 0.30 m/s transverse to the myofiber direction (c_t,myo).¹⁴ In addition, conduction velocity was assumed higher at the endocardium where it was set to 2.08 m/s in myofiber direction (c_f,en) and to 1.59 m/s transverse to the myofiber direction (c_t,en).^15,16 We chose a linear transition from the fast conducting endocardium into the normal conducting myocardium over the inner 20% of the relative wall thickness. Normal myocardial conduction velocities are taken into account in the remaining 80% of the relative wall thickness.

Numerical implementation

Computations of biventricular mechanics were performed in a mesh that consisted of 684 27-noded hexahedral elements resulting in an average spatial resolution of 3 mm. The Eikonal-diffusion equations were solved in a denser mesh that consisted of 29 760 8-noded hexahedral elements with an average spatial resolution of less than 1 mm. ATs that were computed in the denser mesh were projected on the coarser mesh.

Simulations

Starting from a reference simulation representing the normal heart,¹¹ we simulated LBBB and fascicular blocks, and 16 cases of BiV pacing each with a different LV pacing site. The heart rate was fixed at 75 beats per minute.

LBBB was modelled by a global 20% reduction of conduction velocities with respect to NORM.¹⁷ In addition, transmural conduction in the septum was further decreased by 15% at the RV endocardial side, and by 60% in the mid-myocardium and LV endocardial side.¹⁸ Exit points were only located at the RV endocardium (Figure 1B, middle left). In addition, two cases of fascicular blocks were created with a more distal block in the left bundle branch. The latter two baseline situations represented a left anterior and posterior fascicular block by allowing one exit point at the posterior (LAFB) or anterior (LPFB) side of the LV endocardium, respectively.

CRT was simulated by point stimulation at the RV endocardial apex with simultaneous stimulation from one of 16 different locations that were distributed over the LV-free wall (LVfw) (Figure 1B, middle right). Thereby, we assumed that during BiV pacing the ventricular activation is dominated by the activation wave fronts originating from the two pacing sites, i.e. no fusion with intrinsic activation. An example activation pattern during BiV pacing is presented in the right panel of Figure 1B.

In a first step, we evaluated the electrical activation in terms of cumulative activation of the LVfw, because especially this part of the ventricles is likely to be influenced by the LV pacing site and this metric accurately expressed timing and temporal variation of activation of this wall. In addition, we evaluated total LV AT (TATLV) and total AT of both ventricles (TATBiV, i.e. metric for QRS duration).

Next, we investigated the relation between cardiac function and clinically used measures of electrical dyssynchrony.¹⁹ Therefore, dyssynchrony was defined as inter-wall dyssynchrony (difference in mean AT) between (1) LVfw and septum, and (2) LVfw and RVfw, and as intra-wall dyssynchrony (standard deviation (SD) of AT) in (3) the entire LV and (4) the LVfw only. Cardiac function was quantified by maximum rate of rise of LV pressure (LV dp/dt_max) and by LV stroke work (SW, area enclosed by the pressure–volume relation). After simulating 10 cycles to allow hemodynamic stabilization, mean and SD of hemodynamic data were computed from the following five subsequent cardiac cycles.

Results

Variation in electrical activation caused differences in pump function, as shown in Figure 2. During LBBB, maximum LV pressure, stroke volume and thus LV SW were lower than in the NORM simulation. BiV pacing with the LV electrode positioned at the mid-level antero-lateral wall partly restored LV SW to NORM. The right panels of Figure 2 show simulated M-mode echocardiographic wall motion patterns. The inward motion of LVfw and septum occurred simultaneously in NORM. During LBBB, the motion pattern showed early septal inward motion (‘septal flash’), which disappeared during CRT.

Figure 2

Simulated pressure-volume relations (left) and M-mode echocardiographic wall motion patterns (right). Left: LV (black) and RV (grey) pressure–volume relation from NORM (dotted), LBBB (dashed) and CRT with the LV pacing site at the mid-level antero-lateral wall (CRT, solid) are shown. Right: simulated M-mode echocardiographic wall motion pattern of septum (Sept) and LV-free wall (LVfw) at equatorial level. Blue dots indicate the moment of maximum inward motion. IC,  isovolumic contraction; EJECT,  ejection; IR,  isovolumic relaxation; FILL,  filling.

Open in new tabDownload slide

Taking advantage of knowing the calculated onset of contraction for all locations within the ventricles, cumulative activation curves were calculated. These curves resemble pharmacological dose–response curves and can be analyzed accordingly. The activation was characterized by the activation time of 50% of the LVfw myocardium t₅₀ and the speed of activation recruitment at this point m_t50 (Figure 3).

Figure 3

Cumulative activation of the LVfw as a function of time. The top panel shows results of the NORM (white circles), LBBB (grey circles) and a selection of the pacing simulations (non-circled symbols, see Figure 1B). The time needed to activate 50% of the LVfw tissue (t₅₀) and the speed of activation recruitment at t₅₀ (m_t50) are determined from these graphs. The middle panels show the results of t₅₀ (left) and m_t50 (right) for all simulations. LAR, latest activated region; LVa, LV apex; AN, anterior; AL, antero-lateral; PL, postero-lateral; PO, posterior. The bottom panels show the relation of both t₅₀ and m_t50 with LV dp/dt_max.

Open in new tabDownload slide

In NORM (Figure 3, white circles), the direct activation of the LVfw from the exit points of the fast conducting endocardium, resulted in immediate activation in the LVfw with fast increase (t = 30 ms, m = 3.1%/ms). In LBBB (Figure 3, grey circles), delayed activation of the LVfw caused a significantly larger t₅₀ (105 ms). The slow progression of activation within the LVfw due to decreased conduction velocities and absence of Purkinje-system exit points in the LV, caused a more than 50% reduction of m_t50 (1.4%/ms). The various modes of BiV pacing led to lower t₅₀ and higher m_t50 values, but the extent of these changes depended on the specific site of pacing the LV. Largest reductions in t₅₀ were observed when pacing the mid-level of the antero- and postero-lateral wall (AL and PL, respectively). Largest increases in m_t50 were observed when pacing the latest activated region, the mid-level AL wall, or the base of the PL wall. Moreover, t₅₀ showed a stronger correlation with LV dp/dt_max than with m_t50.

When relating LV dp/dt_max to clinically used terms of dyssynchrony, a strong correlation was observed between LV dp/dt_max and inter-LVfw-septum dyssynchrony (Figure 4, top left). LVfw–RVfw dyssynchrony was small during all BiV pacing modes (bottom left), because the RV pacing component generated asynchronous activation within the RVfw with late activated regions in the RV base (right panel in Figure 1B). As a consequence of the small LVfw–RVfw dyssynchrony during BiV pacing, a poor correlation existed between LV dp/dt_max and LVfw-RVfw dyssynchrony (Figure 4, bottom left). Fair correlations were found for the relation of intra-LV and intra-LVfw dyssynchrony and LV dp/dt_max (Figure 4, right column).

Figure 4

Relation between electrical dyssynchrony and LV dp/dt_max. Inter-wall dyssynchrony (left) was determined between LVfw and septum (top), and LVfw and RVfw (bottom). Intra wall dyssynchrony (right) was determined in the entire LV (top), and the LVfw (bottom).

Open in new tabDownload slide

Figure 5 presents the relation between LV pacing site and change in LV dp/dt_max, LV SW, and resynchronization, the latter expressed as decrease in TATLV and in TATBiV. The location of LV pacing that provided maximum hemodynamic effect was the mid-level LV lateral wall. Importantly, a beneficial effect on LV dp/dt_max was observed at many LV pacing sites. Sites providing better hemodynamic response corresponded with those increasing resynchronization of the LV, i.e. reducing TATLV, but not with those reducing TATBiV (metric for QRS duration).

Figure 5

Functional epicardial maps showing the hemodynamic effect (change in LV dp/dt_max and LV stroke work), and resynchronization of the LV (decrease of TATLV), and of both ventricles (decrease of TATBiV) as function of the LV pacing site. The complete contour map was created by interpolating the results from the 16 pacing sites indicated by the symbols in the left plot that were simulated. Each set shows two orientations: antero-lateral (top) and postero-lateral (bottom). The colour code reflects the change relative to LBBB. Baseline values: LV dp/dt_max =1680 mmHg/s, LV SW = 0.92 J, TATBiV = 174 ms, TATLV = 174 ms. Note the different scales for each function variables.

Open in new tabDownload slide

The hypothesis that LV pacing in the LAR is an important determinant for effect of CRT was tested by plotting the relation between change in LV dp/dt_max during BiV pacing and the activation delay of the LV pacing site (‘Q-LV’) at baseline (Figure 6). When using LBBB as baseline, a statistically significant but not perfect relation was found (R = 0.78, P < 0.001). It can be observed that at a given AT, the mid-level AL and PL pacing sites performed better than some later activated regions.

Figure 6

Relation between change in LV dp/dt_max and the activation delay of the LV pacing site during three baseline situations: LBBB (left), LAFB (middle) and LPFB (right). The activation pattern of each baseline situation is shown on top. LV dp/dt_max values (mean ± SD) were 1680 ± 15, 1863 ± 72 and 1923 ± 50 mmHg/s, respectively.

Open in new tabDownload slide

To further investigate the above hypothesis, we investigated the relation between change in LV dp/dt_max and the activation delay of the LV pacing site during the two other baseline situations of simulated anterior and posterior fascicular block. Under these baseline conditions, a very poor relation existed between AT during intrinsic activation and increase in LV dp/dt_max during BiV pacing. Moreover, the extent of increase was smaller than during complete LBBB (Figure 6, middle and left panel). The on an average smaller effect of BiV pacing was explained by the higher baseline LV dp/dt_max during incomplete LBBB. The relation between AT and hemodynamic response was not significant in case of LAFB (R = 0.11, P = 0.68) and LPFB (R = 0.28, P = 0.43). In this regard, it should be noted that during BiV pacing the ventricular activation is dominated by the activation wave fronts originating from the two pacing sites, which makes ventricular activation during BiV pacing independent of the baseline activation sequence.

To simulate LBBB with heart failure, ventricular contractility was globally decreased by 40%. CRT was applied to this ventricle and a summary of the results is presented in Figure 7. The left panel shows that due to a decrease in contractility, pump function decreased irrespective of the activation pattern: maximum LV pressure and stroke volume were decreased (black lines) compared to the simulations with normal contractility (grey lines). Pressure–volume relations were shifted to the right in LBBB with decreased contractility indicating ventricular dilatation. Like in the simulations with normal contractility, pump function decreased during LBBB and was partly restored during biventricular pacing. Similar to Figure 4, the right panel of Figure 7 shows the results for wall dyssynchrony. LV dp/dt_max was further decreased, but the relative differences between simulations remained. Consequently, decreasing contractility only resulted in a scaling effect. Therefore, also for this more severe type of heart failure, a strong correlation was observed between LV dp/dt_max and the inter LVfw-septum dyssynchrony.

Figure 7

Global pump function and wall dyssynchrony in simulations with decreased contractility. Left: pressure–volume relations of simulations NORM (dotted), LBBB (dashed) and CRT with the LV pacing site at the mid-level antero-lateral wall (solid). The grey lines represent the results from the left panel in Figure 2. The black lines are the results from the simulations with a 40% global decrease of contractility. Right: relation between electrical dyssynchrony and LV dp/dt_max in simulations with 40% decrease in ventricular contractility and with normal activation (dotted white circle), LBBB (dotted grey circle), and BiV pacing with the LV lead positioned at the antero- and postero-lateral midwall (dotted yellow and blue squares; note that the blue square is behind the yellow square). Results with normal contractility are made transparent.

Open in new tabDownload slide

Discussion

In this study, we investigated the relation between LV pacing site during biventricular pacing and cardiac function in a finite element (FE) model of left and right ventricular electromechanics. The model replicated general clinical observations such as the increase of LV dp/dt_max and stroke work and the disappearance of a septal flash during BiV pacing. In addition, values of model computed global cardiac function during normal activation, such as a maximum LV pressure of ∼120 mmHg and a stroke volume of ∼65 ml, are well within the human physiological range. Acute hemodynamic response, expressed as increase in LV dp/dt_max and LV stroke work, was maximal when pacing the LV at the mid-basal lateral wall. Pacing from this region resulted in best resynchronization of the LV through fast activation recruitment, and consequently, a low total LV electrical activation time. The LV pacing site that led to largest CRT response co-located with the latest activated region (LAR) for the simulation of typical LBBB, but not for the fascicular block simulations. These results indicate that it is the site of LV pacing that creates the largest reduction in total LV activation time during CRT, which is the primary determinant of CRT response and not the timing of baseline electrical activation (‘Q-LV’).

Maximum CRT response by pacing the mid-basal lateral wall

The observation in our computational study that the fastest recruitment of LV activation and the largest hemodynamic effect were achieved by mid-basal LV lateral wall pacing, is corroborated by data from animal experiments²⁰ and patient studies.²¹ The good performance of the mid-basal lateral wall as pacing site may be understood easiest by considering that from this site, the activation wave front can move in all directions. In other words, the average distance to the pacing site is smaller for mid-basal LV lateral wall regions than for basal, often later activated, regions. Maximum CRT response through fast activation recruitment confirms suggestions from other computational studies.²²

Role of anatomical location

Our observation that the pacing lead can be positioned in a relatively large area of the LV wall to obtain a substantial hemodynamic response to CRT was also corroborated by observations made in canine experiments.²⁰ This large region with at least reasonable haemodynamical effect might explain why larger clinical studies do not consistently find an effect of the anatomical region on (long-term) response to CRT.⁶

Role of latest activated region

Results from our LBBB simulation are in line with observations in clinical studies that showed that response to CRT was better when baseline AT at the LV pacing site was higher.^7,–9 However, the simulations also show that the mechanism for this better response is the optimal resynchronization during CRT rather than the degree of normalization of the baseline conduction. The latter is demonstrated by the results from the fascicular block simulations, where the latest activated region was different from that during LBBB but yet the optimal pacing site was the same as during LBBB. This demonstrates that the ventricular activation during BiV pacing is determined solely by the activation wave fronts originating from the two pacing sites, irrespective of the underlying conduction abnormality.

The results of clinical studies that advocate for the use of the LAR may partly be explained by the fact that the delay at this location provides information whether this site is later activated than its adjacent regions, but also by the fact that the entire LV is late activated (e.g. the presence of LBBB), the latter implying that there is a substrate amenable for resynchronization. The importance of the latter is illustrated from the results in the simulations with fascicular blocks in our study. In addition, these simulations showed that improvement in LV dp/dt_max can range from zero to 30%, which is in good agreement with observations from clinical studies.^23,24 Altogether, our results indicate that the association between LAR and optimal LV pacing site is based on coincidence rather than on electromechanical mechanisms.

Reduction of QRS duration

On the one hand, the present simulation data also provide a possible explanation why reduction of the QRS duration does not always predict a good response to CRT.²⁵ After all, for optimal LV function, it is predominantly the optimal LV resynchronization that determines the CRT response. On the other hand, the RV-pacing component of BiV pacing may increase activation time in the RVfw and thereby affect QRS duration. The present study indicates that largest reduction of TATBiV was obtained when the LV electrode was positioned in anterior regions, regions that showed poor increase of haemodynamic response (Figure 5), as is also known from clinical studies.²⁶

In this study, CRT was simulated as it is applied in current clinical practice, i.e. complete capture of the wave fronts originating from the pacing leads. Considering the increased activation time in the RVfw, LVfw pacing and fusion with intrinsic activation might even give a better response. This could be investigated in a next study.

Study limitations

The computational model is by definition a simplified representation of the human cardiovascular physiology. For example, in the absence of complete human data, some material laws and model parameter values of cardiac tissue properties were based on animal experiments. However, the observation that global LV function parameters in the NORM simulation were well within the human physiological range suggests that the simulation results may be translatable to the human situation.

Besides a distinction between endocardial and myocardial conduction velocities, no other regional differences in conduction (like scar) were taken into account, nor dilated cardiomyopathy. In addition, we assumed a linear transition from the faster conduction velocities at the endocardium to the conduction velocities at the mid-myocardium over a relative wall thickness of 20%, because experimental data on the real transition is not available.

The LBBB simulation mimicked the condition of a complete, proximal LBBB, without any change in other conduction parameters. The fact that the resultant electrical conduction pattern as well as the mechanical consequences, such as septal flash were reproduced by the model indicated that the model shown replicated the human situation very well. The two fascicular block cases are merely examples of non-typical LBBB conduction which is used to illustrate the independence of the relation between baseline conduction pattern with corresponding location of the latest activated region and CRT response.

By globally decreasing ventricular contractility, a more severe example of heart failure was simulated that resulted in a further decrease of pump function. The relative differences between simulations remained, because decreased contractility affected pump function and not the electrical activation pattern. Thus, best resynchronization of the LV, and consequently, maximum response was still obtained by pacing the mid-basal lateral wall. Nevertheless, the situation in heart failure patients may be even more complicated than the one we simulated because these patients might also have several ischemic or scarred regions and changes in passive myocardial behaviour. Finally, the results only concern the acute hemodynamic response, while reverse remodeling was not taken into account. Therefore, the data from the present study should serve as hypothesis generating and need confirmation in clinical studies.

Outlook

The use of computational models has several advantages. For example, all tissue properties are available for analysis and new insights can be gained from these analyses for generating new hypotheses. The simulation results presented in this study provide more insight into the possible mechanism for maximizing CRT response by LV pacing site. The present study provides several clues: (1) the anatomical position can be useful, especially aiming at mid-basal LV lateral wall; (2) measuring the delay in the latest activated region (LAR) can confirm the electrical substrate, but aiming at the LAR is not advisable; and (3) aiming at maximum QRS reduction is not advisable. More research is needed to test if the generated hypotheses hold for the real patient.

Another implication of the present study would be that for optimal application of CRT it may be useful to determine, during CRT implantation, the site that not only provides maximal resynchronization of the LV wall but also maximal resynchronization between LVfw and septum.

Conclusions

The present computational study provides novel insight in how maximum acute hemodynamic response by BiV pacing can be achieved. Most important factors are fast recruitment of LV activation, leading to maximum reduction in inter and intra LV dyssynchrony. This was achieved by pacing a central location on the LV lateral wall. Coincidentally, this region was near the LAR, but only in case of a typical LBBB activation.

Supplementary material

Supplementary material is available at Europace online.

Funding

This research was performed within the framework of CTMM, the Center for Translational Molecular Medicine (www.ctmm.nl), Project COHFAR (Grant 01C-203). F.W. Prinzen has received research grants from Medtronic (Minneapolis, USA; Maastricht, The Netherlands), EBR Systems (Sunnyvale, USA), MSD (Whitehouse Station, USA) Sorin (Paris, France), Biotronik (Berlin, Germany), St. Jude Medical (St. Paul, USA). J. Lumens has received a grant within the framework of the Dr. E. Dekker program of the Dutch Heart Foundation (NHS-2012T010).

References