Leadless biventricular left bundle and endocardial lateral wall pacing versus left bundle only pacing in left bundle branch block patients

Biventricular endocardial (BIV-endo) pacing and left bundle pacing (LBP) are novel delivery methods for cardiac resynchronization therapy (CRT). Both pacing methods can be delivered through leadless pacing, to avoid risks associated with endocardial or transvenous leads. We used computational modelling to quantify synchrony induced by BIV-endo pacing and LBP through a leadless pacing system, and to investigate how the right-left ventricle (RV-LV) delay, RV lead location and type of left bundle capture affect response. We simulated ventricular activation on twenty-four four-chamber heart meshes inclusive of His-Purkinje networks with left bundle branch block (LBBB). Leadless biventricular (BIV) pacing was simulated by adding an RV apical stimulus and an LV lateral wall stimulus (BIV-endo lateral) or targeting the left bundle (BIV-LBP), with an RV-LV delay set to 5 ms. To test effect of prolonged RV-LV delays and RV pacing location, the RV-LV delay was increased to 35 ms and/or the RV stimulus was moved to the RV septum. BIV-endo lateral pacing was less sensitive to increased RV-LV delays, while RV septal pacing worsened response compared to RV apical pacing, especially for long RV-LV delays. To investigate how left bundle capture affects response, we computed 90% BIV activation times (BIVAT-90) during BIV-LBP with selective and non-selective capture, and left bundle branch area pacing (LBBAP), simulated by pacing 1 cm below the left bundle. Non-selective LBP was comparable to selective LBP. LBBAP was worse than selective LBP (BIVAT-90: 54.2 ± 5.7 ms vs. 62.7 ± 6.5, p < 0.01), but it still significantly reduced activation times from baseline. Finally, we compared leadless LBP with RV pacing against optimal LBP delivery through a standard lead system by simulating BIV-LBP and selective LBP alone with and without optimized atrioventricular delay (AVD). Although LBP alone with optimized AVD was better than BIV-LBP, when AVD optimization was not possible BIV-LBP outperformed LBP alone, because the RV pacing stimulus shortened RV activation (BIVAT-90: 54.2 ± 5.7 ms vs. 66.9 ± 5.1 ms, p < 0.01). BIV-endo lateral pacing or LBP delivered through a leadless system could potentially become an alternative to standard CRT. RV-LV delay, RV lead location and type of left bundle capture affect leadless pacing efficacy and should be considered in future trial designs.

His-Purkinje network generation pipeline. The first row shows the inputs for the His-Purkinje Network generation (from left to right): a biventricular mesh, the left and right ventricular endocardial surfaces the network is grown on, an apico-basal universal ventricular coordinate (UVC), ranging between 0 at the apex and 1 at the base, a transmural UVC, varying from 0 to 1 from the endocardium to the epicardium and a rotational UVC from - to + around the ventricles. The second row shows the root points for the five fascicles included in the modes. The third and rows represent all Purkinje-myocardium junctions (PMJs) and the final PMJs, remaining after deactivating the redundant PMJs. On the right, a picture of the His and fascicles, showing the root points as colored spheres.
The model accounts for three LV fascicles (colors refer to Figure 1, second to fourth row): anterior (purple), septal (orange) and posterior (light-blue), and for two RV fascicles: septal (green) and moderator band (pink).
The location of the root points (second row, Figure 1) was provided in terms of UVCs and was based on early activated areas in the Durrer maps. 3 The root points were then used to grow five independent networks that were joined to the His as shown in Figure 1, bottom-right. The His bundle is formed by filaments bundled together and insulated within a common cable. 4 These filaments are predestined to either the left bundle or the right bundle. To represent this anatomical property of the His bundle, we duplicated the His bundle segments in our Purkinje networks. One strand continues into the left bundle and the other continues into the right bundle.
The trees associated with the fascicles are grown independently from each other and can therefore overlap. This would not make a difference in sinus rhythm simulations, where the stimulus spreads from the His, fascicles, Purkinje and finally to the myocardium through the Purkinje-myocardium junctions (PMJs), e.g. the terminal points of the His-Purkinje network, as each node can activate only once. On the other hand, during pacing, additional PMJs can affect the simulation as the stimulus can enter the wrong fascicle. To prevent this, we deactivated redundant PMJs. First, the PMJs associated with each fascicle were found ( Figure 1, third row). We then identified the portions of LV and RV endocardial surfaces that were covered by each fascicle tree, and we identified areas where the trees overlapped. An eikonal simulation was run on the tree associated with each fascicle by stimulating the root point to find when each terminal point would activate. In areas where the fascicle trees overlapped, only the PMJs with the shortest activation time were connected with the surrounding myocardium and were therefore activated. The PMJs remaining active after this procedure are shown in Figure 1, bottom row. The image shows that there is no overlap between active PMJs belonging to different fascicles.
The timing of the first activations from the fascicles to ventricular myocardium was based on the Durrer maps. 3 The first LV activation occurs at the three simultaneous fascicle locations (anterior, septal and posterior). Therefore, we computed the conduction velocity (CV) of each LV fascicle separately to guarantee simultaneous activation of the root points of the fascicles. According to the Durrer maps, the RV fascicles activate about 10 ms later than the LV fascicles. Therefore, the CV of the RV fascicles was computed to achieve activation of the RV fascicles root points 10 ms later than the LV. This network generation pipeline was applied to all twenty-four patient-specific meshes. Proximal LBBB was simulated by disconnecting the left bundle branch from the LV Purkinje network along the His.

Model validation
We validated the baseline model using electrocardiographic imaging (ECGi) during left bundle branch block (LBBB) acquired from 8 LBBB patients as part of two ECGi studies. 5,6 Below, we describe the available clinical data and compare simulation results against them.

Electrocardiographic Imaging data
ECGi data were collected as part of two clinical studies. 5,6 The first study 6 (trial registration number NCT01831518) recruited 11 patients undergoing CRT implantation, including LBBB (N=7) and non-LBBB (N=4) baseline rhythm. For the purpose of this study, we compared our LBBB model against data from 7 LBBB patients. The details about data collection and analysis have been published previously 6 7 , computed as the difference between mean LV activation times and mean right ventricular (RV) activation times. The metrics computed from ECGi data accounted only for the ventricular epicardial surface, while the metrics computed from simulation results accounted for the whole ventricular myocardium. In both cases, the base of the ventricles, including the outflow tracts, was excluded from the mesh. We analyzed 3 baseline beats for each patient, leading to a total of 21 beats. We considered one additional LBBB patient from a second study 5 (trial registration number NCT 04322877) who also underwent ECGi. 3 baseline LBBB beats were collected and analyzed.  Figure 2 shows the comparison between the LBBB baseline metrics measured from the 24 beats extracted from the ECGi data and the metrics predicted by the model for all 24 patients. All metrics are within physiological ranges, showing that the model can reproduce features of LBBB activation pattern. We also show that the model replicates local LBBB activation by comparing an anterior and a posterior view of the epicardial maps for one patient at baseline with the epicardial activation times predicted by the model for all patients (Figure 3 and 4).

LBBB Baseline Validation
The model validation provided in this section shows that the baseline LBBB model reproduces standard LBBB activation patterns with large delays between the RV and the LV activations. We also show that metrics extracted from the model agree with the metrics computed from ECGi data. Therefore, we can conclude that the model can be used to make considerations about response to pacing.

The effect of septal scar on simulation results
We mapped a patient-specific septal scar and border zone geometries onto all twenty-four meshes to simulate the effect of septal scar on the results presented in the manuscript. We simulated baseline LBBB, selective LBP with and without optimized atrioventricular (AV) delay, leadless left bundle pacing (BIV-LBP) and leadless pacing with the LV lead placed in the free wall. The activation metrics were then computed as described in the methods of the manuscript and compared.  LBP performed with either a lead-based or a leadless system is ineffective in patients with septal scar, as the LBP stimulus does not capture healthy myocardium. On the other hand, BIV-endo lateral wall pacing efficacy is not affected by the presence of septal scar.