Analysis of lead placement optimization metrics in cardiac resynchronization therapy with computational modelling

Abstract

Aims

The efficacy of cardiac resynchronization therapy (CRT) is known to vary considerably with pacing location, however the most effective set of metrics by which to select the optimal pacing site is not yet well understood. Computational modelling offers a powerful methodology to comprehensively test the effect of pacing location in silico and investigate how to best optimize therapy using clinically available metrics for the individual patient.

Methods and results

Personalized computational models of cardiac electromechanics were used to perform an in silico left ventricle (LV) pacing site optimization study as part of biventricular CRT in three patient cases. Maps of response to therapy according to changes in total activation time (ΔTAT) and acute haemodynamic response (AHR) were generated and compared with preclinical metrics of electrical function, strain, stress, and mechanical work to assess their suitability for selecting the optimal pacing site. In all three patients, response to therapy was highly sensitive to pacing location, with laterobasal locations being optimal. ΔTAT and AHR were found to be correlated (ρ < –0.80), as were AHR and the preclinical activation time at the pacing site (ρ ≥ 0.73), however pacing in the last activated site did not result in the optimal response to therapy in all cases.

Conclusion

This computational modelling study supports pacing in laterobasal locations, optimizing pacing site by minimizing paced QRS duration and pacing in regions activated late at sinus rhythm. Results demonstrate information content is redundant using multiple preclinical metrics. Of significance, the correlation of AHR with ΔTAT indicates that minimization of QRSd is a promising metric for optimization of lead placement.

What’s new?

• Using a detailed biophysical model of cardiac function, the location of the left ventricular pacing lead in cardiac resynchronization therapy (CRT) was optimized in silico in three patients.

• Laterobasal pacing locations were found to generally result in the best acute haemodynamic response (AHR).

• Changes in total activation time with pacing were found to be correlated with AHR, supporting optimization of CRT by minimizing QRS duration.

• Pacing in regions activated late at sinus rhythm was found to generally improve AHR.

Introduction

Cardiac resynchronization therapy (CRT) is an effective treatment for dyssynchronous heart failure,¹ however response rates to therapy remain suboptimal. Only two-thirds of patients selected using standard criteria benefit from CRT when compared to standard pharmacological therapy alone, while over a third of selected patients would improve even without receiving this additional invasive and expensive treatment.²

Clinical studies have demonstrated the potential for improving response to CRT through the optimization of pacing lead location.^3,4 In our recent paper, we demonstrated the importance of pacing lead placement for achieving optimal therapy outcomes.⁵ Targeting specific anatomical locations, pacing in late-activated regions and pacing away from the location of scar tissue have been proposed as methods for the optimization of the left ventricle (LV) pacing location, however, so far, none has been shown to achieve optimal results consistently.⁶

Acute, instantaneously measurable and clinically available data on cardiac function present an appealing methodology for optimization of CRT. In particular, the total activation time of the ventricles (TAT) and the acute haemodynamic response (AHR) have been hypothesized as potentially promising metrics that correlate with CRT outcome.⁷ Changes in TAT, usually measured by the QRS duration on electrocardiogram (ECG), are considered clinically important, as electrical resynchronization of the ventricles by CRT is expected to be measured by a decrease in TAT. An alternative measure is AHR, the fractional increase in the maximum rate of pressure development (dP/dt) in the LV on pacing, which has been used in the clinic as a measure of acute response to therapy and a predictor of long term remodelling.⁷ As these metrics are relatively easy to obtain, it is feasible to trial multiple pacing sites during the implant procedure and to select the optimal location. Indeed, the value of this approach has been demonstrated in the clinic,⁷ however due to practical constraints, the number of pacing locations can realistically be trialled in each patient is limited.

Computational modelling provides the opportunity to optimize CRT in silico by constructing a model of a patient’s cardiac function and applying it to exhaustively test for the best pacing configuration. In patients with sufficient data to enable model construction, this would allow inference of the best configuration prior to intervention. However, we can also use models constructed for a sample of patients to assess the value of clinically obtainable or novel model-derived metrics for CRT optimization, especially given the need to balance such metrics’ predictive value against any costs and/or risks in obtaining them.

In this study, we use three biophysically detailed, personalized models of cardiac electromechanics to demonstrate the potential for in silico optimization of the LV pacing lead in CRT, and analyse the effectiveness of 16 metrics of preclinical cardiac function for the selection of the optimal pacing location.

Methods

Patients and data

Patient data were acquired from a detailed study of CRT carried out at St. Thomas’ Hospital and Evelina Children’s Hospital (London, UK). Selected patients had two clinical procedures, an initial invasive electrophysiology mapping study and a separate device implantation. In addition to standard clinical measurements including echocardiography, magnetic resonance imaging (MRI) and ECG, invasive recordings were made of ventricular pressure and non-contact mapping (NCM) of LV electrophysiology was performed.

Three patients were selected for this modelling study, with demographics and baseline characteristics as shown in Table 1. All patients had non-ischaemic aetiology and New York Health Association (NYHA) class of III. Patients 1 and 3 responded to therapy, having had a reduction of ≥ 15% in end systolic volume at follow up, whereas Patient 2 was a non-responder.

Table 1.

Demographics and baseline clinical indices for the patients in this study

Case	Sex	Age	QRSd (ms)	EF (%)	EDV (mL)	NCB
1	M	63	188	23.5	310	No
2	F	81	139	24.7	172	Yes
3	M	77	171	19.6	331	Yes

QRSd, QRS duration; EF, ejection fraction; EDV, end diastolic volume; NCB, non-ischaemic conduction block.

Model development and personalization

A computational model of cardiac electromechanics was constructed from clinical data for each of the three patients in this study. The biophysically detailed, biventricular model was formed by coupling the monodomain tissue model of electrical activation⁸ and the ten Tusscher cellular electrophysiology model⁹ to a model of large deformation mechanics using the Guccione passive material law,¹⁰ a phenomenological model of myocardial active tension¹¹, and a three element Windkessel model of afterload.¹² The personalization and validation process is described briefly below. For a comprehensive description of the model, underlying computational methods and the process for model personalization and validation using clinical data, the interested reader is referred to the supplement to our recent paper.⁵

Cardiac anatomy and fibres

A finite element mesh describing the cardiac anatomy was fitted to segmented end diastolic 3D whole-heart MRI.¹³ Tissue orthotropy was determined from generic histological measurements and mapped to each patient-specific anatomical model,¹¹ and an additional finite element mesh specialized for the simulation of cardiac electrophysiology was generated at a high spatial resolution from the original mesh.

Electrical activation

The location and timing of initial activation in the ventricles were determined from ECG, NCM, angiography and the pacing protocol for both sinus rhythm (SR) and artificial pacing, supplemented by information from experimental studies performed on excised human hearts,¹⁴ where necessary. None of the patients modelled exhibited scar tissue, but Patients 2 and 3, which exhibited non-ischaemic conduction block in NCM data, had this block incorporated into the model by assigning a low conductivity to the tissue at the observed location. The electrophysiology model was validated by comparison of the predicted activation sequences with activation time maps determined from NCM and registered with the model.

Contraction

The full coupled model of cardiac electromechanics was simulated at both SR and following CRT. Two parameters from the passive material law, six parameters from the active tension model, and the three Windkessel model parameters were fitted to patient-specific pressure–volume loops at SR and to the observed AHR in the case of the active tension model. Mechanical contraction was validated by comparison of the predicted wall motion with short axis cine MRI at SR.

Software tools

Simulations of cardiac electrophysiology were run with the Cardiac Arrhythmia Research Package (http://carp.medunigraz.at/), developed at the Medical University of Graz (Graz, Austria) and the University of Bordeaux (Bordeaux, France). Starting at the first depolarization in the ventricle, 300 ms were simulated, covering one full depolarization sequence. Simulations were performed on ARCHER (http://www.archer.ac.uk/), the UK national high performance computing (HPC) resource, using 288 cores and requiring 2.5–4 h execution time per model. Large deformation mechanics was simulated using CMISS (http://www.cmiss.org/), developed at the University of Auckland (Auckland, New Zealand). One heart beat was simulated, for a duration of 1 s starting at the first depolarization in the ventricle. Mechanics simulations were executed on the HPC resource at the Department of Biomedical Engineering at King’s College London, using 4 cores and requiring 14–20 h execution time per model.

Pacing optimization study

The utility of the model for the optimization of LV pacing lead placement was evaluated in each of the three models. Maintaining the RV apex lead in the implanted location for each case, the LV lead was set to a range of different positions, and the model of cardiac electromechanics was simulated with simultaneous biventricular CRT pacing. Pacing sites were placed at regular intervals on the LV epicardium, as shown in Figure 1. It should be noted that most of these pacing locations will not be accessible via the coronary venous anatomy, however as this varies from patient to patient the full area of potential pacing sites was covered.

Figure 1.

Experimental design for the lead placement optimization study, shown on a bullseye plot. The letters A, S, I, and L correspond to the anterior, septal, inferior and lateral walls of the ventricle, respectively. Biventricular pacing was performed with RV apical pacing and epicardial LV pacing at each of the marked locations.

Simulations of cardiac electromechanics following CRT pacing were performed for each of the above LV pacing sites, resulting in a total of 168 simulations. The models were solved using the tools and resources as described in Section 2.2.4. The results from each simulation were used to evaluate the change in TAT (ΔTAT) and AHR for that pacing configuration.

Preclinical metrics for lead placement optimization

In order to better inform the selection of optimal lead position in a clinical setting, a number of metrics based on preclinical cardiac function were calculated using the computer model at SR. Each metric was calculated at the location of all 56 LV pacing sites in the optimization study, in each of the three patient cases in the study. The preclinical metrics calculated were:

• Electrical activation time, determined by the time of myocyte depolarization on the LV endocardium

• Mechanical activation time, determined by the time at which myocytes in the LV begin to shorten

additionally, for each of the following, metrics were calculated from both the maximum and the time at which the maximum occurs:

• Tissue strain

• Tissue strain rate

• Active stress (stress caused by tension generated in the sarcomeres)

• Passive stress (stress caused by the elastic behaviour of the tissue)

and finally the total work, maximum power and the time of the maximum power formed metrics for each of:

• Active work (work done by active stress).

• Passive work (work done by passive stress).

These metrics cover a range of aspects of cardiac electromechanical function, though it should be noted that not all are measurable with the same cost and risk in the clinic. Specifically, the acquisition of regional electrical activation times has generally required invasive catheterization, though newer technologies such as ECGI may alleviate this risk. Mechanical activation time can be determined from non-invasive imaging, though achieving a sufficiently high spatio-temporal resolution may pose a challenge. Strain metrics can already be determined from MRI tagging, which is already commonplace in preclinical assessment of heart failure patients. Stress and work may be difficult to determine in a clinical environment, and thus will likely remain restricted to patient-specific modelling studies. Some of the above metrics would also require additional data to be acquired, however the additional cost and risk may be warranted if a metric shows sufficient predictive power. In this instance, computational modelling provides a valuable tool to add value to existing clinical data.

Visualization and analysis

Maps of the response metrics (ΔTAT and AHR) and preclinical metrics were calculated by interpolating the values across the pacing region, and plotting isocontours on a bullseye projection of the ventricle. Correlations between metrics were evaluated using the Spearman’s rank correlation coefficient, taking value pairs at each of the 56 pacing sites in the optimization study. The correlations between ΔTAT and AHR and between AHR and each of the preclinical metrics were evaluated separately for each of the three cases in the study. Preclinical metrics were compared with AHR and not ΔTAT as AHR is the metric calculated by our model which is most directly linked with response to therapy.⁷

Results

The anatomical model, model parameters, initial conditions, and boundary conditions were personalized to clinical data. The generated anatomical models and comparisons with MRI are shown in Figure 2.

Figure 2.

For each of the three patient cases in this study, the fitted anatomical model (above) was generated from 3D whole-heart MRI at end diastole. A comparison of a single short axis slice with the corresponding contour of the model (green) is also shown (below).

Pacing optimization study

Bullseye plot maps of ΔTAT and AHR according to LV pacing site are shown in Figure 3. While all pacing sites resulted in a reduction of TAT relative to SR, results show that ΔTAT was sensitive to LV pacing lead location, with more lateral and basal pacing sites generally resulting in larger reductions in ΔTAT. A maximum AHR of 25.2, 15.7, and 43.2% was achieved in Cases 1, 2, and 3, respectively. The pressure evolution for these pacing locations and at SR is shown in Figure 4.

Figure 3.

Bullseye plot maps showing patient responses to therapy according to LV pacing lead location in each of the three cases in our in silico CRT optimization study. Responses are calculated as the change in total activation time (ΔTAT, A) and the acute haemodynamic response (AHR, B).

Figure 4.

Model-predicted systolic LV pressure in three cases at both sinus rhythm and with optimal pacing. It can be seen that there is a significant increase in peak dP/dt in all three cases.

Our results show that the pacing site used in the clinic (standard implantation in the posterolateral vein, exact location determined in the model from angiography) resulted in a near-optimal reduction in TAT for Cases 1 and 3, where the TAT with pacing at the clinical site was 117 and 94 ms, respectively compared with an optimal of 114 and 96 ms, respectively. However, in Case 2 the clinically recorded site was sub-optimal, with a TAT of 96 ms compared with the best possible model prediction of 68 ms.

Most pacing sites resulted in a positive AHR, reflecting an improvement in LV peak dP/dt, however, a small number of pacing sites next to the LV free wall—septum junction in Cases 1 and 3 resulted in a negative AHR. More lateral and basal pacing sites were again associated with a greater response to therapy, however the distribution was not identical to that of the improvements in TAT. As with ΔTAT, the clinical pacing sites of Cases 1 and 3 were shown to provide a near-optimal AHR of 23.8 and 37.9%, respectively, whereas the Case 2 clinical pacing site provided an AHR of only 8.08% compared with the maximum 15.7%.

Notably, the cases with near-optimal ΔTAT and AHR when pacing at the clinically recorded pacing site were also those which exhibited long term response to therapy. Case 2, the only non-responder at follow up, was also the only case to have suboptimal ΔTAT and AHR, suggesting that this patient could have experienced improved outcomes with an alternative pacing site. However, it is important to note in this small study that there are other factors that also distinguish Case 2 from Cases 1 and 2, notably, the sex of the patient and the degree of LV dilation.

Correlations

Figure 5 plots the fractional ΔTAT of the ventricles against AHR for each of the three models in this study. We can see that all three cases show a strong negative correlation, with a Spearman’s rank correlation coefficient of ρ = −0.914, −0.910 and −0.802 for Cases 1, 2, and 3, respectively, demonstrating that improved electrical synchrony is predicted to result in improved mechanical function.

Figure 5.

The fractional change in total activation time of the ventricles (ΔTAT) scattered against acute haemodynamic response (AHR) across a range of possible pacing sites for each of the three models in this study.

It can be seen in Figure 5 that Case 2 has a cluster of simulations of the same TAT that have a spread of AHR values. This occurs when the LV pacing lead is placed in a region of the anterolateral free wall for which the time of last activation is exclusively determined by the activation front caused by the RV pacing lead. For pacing sites in this region, the TAT metric is unchanged, however the specific activation pattern is changed and thus the mechanical contraction and therefore the AHR varies.

Preclinical metrics

Correlations between regional preclinical metrics evaluated at LV pacing sites and the AHR at those pacing sites were evaluated using Spearman’s rank and are shown in Figure 6. The metrics that showed a strong correlation with AHR across all three cases were local activation time determined by electrical depolarization (ρ ≥ 0.76) or myocyte shortening (ρ ≥ 0.73), the time of maximum active stress (ρ ≥ 0.82), and the peak active power (ρ ≥ 0.69). Stress and power are difficult to measure in vivo, however regional activation time based on electrical depolarization or mechanical deformation can be determined in the clinic.

Figure 6.

Correlations between acute haemodynamic response and a number of preclinical metrics measured at the location of the LV pacing lead. Correlation coefficients were calculated separately for each case in the study using Spearman’s rank.

To form a comparison with the response maps in Figure 3, bullseye plot maps of electrical and mechanical activation times in the ventricle at SR are shown in Figure 7. Markers indicating the pacing location of maximal response according to both ΔTAT and AHR show that there is indeed good overall correspondence between these outcomes and the location of latest activation in the model, however, the match is not exact. Pacing at the last activated location does not necessarily result in the maximal response. This is most noticeable in Case 3, where pacing in the last electrically activated segment (the anterobasal segment) would result in a suboptimal AHR.

Figure 7.

Activation time of the tissue according to (A) electrical activation, determined as the time of depolarization of cardiac myocytes on the endocardium, and (B) mechanical activation, determined as the time at which the myocardium begins to contract. Activation times are shown for each of the three cases in this study at sinus rhythm. Markers show the location of the optimal pacing site from the pacing optimization study according to both ΔTAT and AHR.

Discussion

Using three computational models of cardiac electromechanics personalized to the clinical data of three CRT patients, we were able to optimize the location of the LV pacing lead in silico. Our study revealed that there were significant variations in response to therapy depending on the location of the pacing lead location.

Optimal pacing region

Figure 3A showed that the greatest haemodynamic response was usually achieved by pacing in the more lateral and basal regions of the LV. The association of lateral pacing sites with AHR is consistent with both canine experiments¹⁵ and clinical studies,^16,17 however these results have not been reproduced in a larger multi-centre trial.³ This latter study did, however, show that apical pacing locations are suboptimal, consistent with our results. Furthermore, these authors suggested that apical pacing was suboptimal as the proximity to the RV apex pacing lead reduced the effectiveness of resynchronization. This is also supported by our study, in which pacing locations closest to the apex and RV were the least effective. While the proposition that there is a general optimal pacing region is disputed,⁶ our results support the conclusions of previous studies that laterobasal pacing will on average produce the strongest response to therapy.

Patient-specific optimization

In recent years, there is a growing consensus that patient-specific optimization of CRT is necessary to achieve the optimal response to therapy.⁶ Patients undergoing therapy can have varied scar, ischaemia, and non-ischaemic conduction block, which can significantly affect the suitability of pacing in different locations.

Optimization of therapy directly using AHR has been demonstrated in clinical research cases,⁷ however, to reduce costs, procedure times and risk of stroke due to thromboembolism it is desirable to determine the optimal site prior to pacing. Our modelling results show a high degree of correlation between AHR and the change in total activation time due to pacing (ρ ≤ –0.80 in all cases, see Figure 5), indicating that minimization of QRSd under pacing may offer a suitable surrogate for optimization of lead placement. Live optimization in the clinic using QRSd still requires a manual trial and error approach, however a computer model of cardiac electrophysiology alone, which is much faster to simulate and requires much less data for personalization than a full electromechanics model, could be used to simulate changes in activation time and help identify the optimal pacing location prior to the intervention.

This conclusion again mirrors results reported in both the clinic and animal laboratory, where greater reductions in TAT with pacing were observed to result in improved outcomes.^18,19 However, it is important to note that optimization through the minimization of QRSd has not been shown to be significantly more effective than existing lead placement strategies.²⁰ Indeed, it is clear from Figure 3 that the maps of ΔTAT and AHR are not in perfect agreement.

It is also interesting to note that in Figure 5, the two responder Cases, 1 and 2, have a very similar slope in their relationship between ΔTAT and AHR, whereas the gradient of this relationship for Case 2 is much smaller. This indicates that, for a similar reduction in TAT, Case 2 generally had a smaller AHR than Cases 1 and 2. This may suggest an electromechanical coupling disorder, however it was not possible to determine the specific mechanism in this study.

Optimization using preclinical metrics

In Figure 7, we compared 16 regional metrics of cardiac function at SR prior to the implantation of the CRT device to the efficacy of CRT pacing at the pacing sites in the optimization study. This revealed that pacing at locations with a late activation time, either in terms of electrical depolarization or the onset of contraction, the time of the maximum active stress, and the maximum active power were all strongly correlated with AHR in all three patient cases in this study. The time of maximum active stress is likely dependent on the activation time, and both it and the maximum active power would be difficult to measure clinically without significant advances in the integration of model-based prediction into the clinical context. For this reason we concentrate on the activation time metrics in the discussion below.

While our results show a strong link between preclinical activation time and AHR at given pacing locations, and we can see that there is overall good agreement between the activation time maps in Figure 7 and the AHR maps in Figure 3B, pacing the last activated region as a strategy for CRT optimization is controversial. Several studies support pacing the last activated region determined from electrophysiological mapping or echocardiography,²⁰ however this conclusion has also been disputed.⁶ Indeed, in our study, we can see that pacing in the last activated region would not necessarily result in the optimal response. This is particularly true in the situation where there are strong spatial gradients in response caused by structural conduction defects, for example in Case 3, where pacing the last activated site would result in an improved but still suboptimal response. It is, however, important to note that pacing targeted at the whole anterobasal segment where last activation occurs could in the worst case result in no response at all.

Using current pacing technologies, where the location of the LV lead is limited by the coronary venous anatomy, the optimal pacing location is often not accessible. However, new pacing technologies such as leadless endocardial pacing may create new opportunities for optimization of pacing lead location using these findings. Furthermore, by targeting the area of optimal pacing determined from preclinical assessment for local optimization during the intervention, additional improvements in therapy outcomes may be reached.

Limitations

Computational models can predict many measures of cardiac function, however the models in this study predict only acute response to therapy. Furthermore, only one heart beat is simulated, and any short term circulatory feedback mechanisms are therefore omitted. Key to any device optimization is either the direct prediction of long term response with the model, an exceptionally challenging problem, or the calculation of a clinical or model derived metric that accurately predicts long term response. In this study, we used the model to determine AHR, which represents the current best practice for prediction of CRT outcomes.⁷

In this study, in silico assessment of pacing location was performed in three patients. While the study is limited by the small number of subjects, it still represents significant progress in computational modelling in this field, where most studies to date have involved only a single patient case.¹¹ In addition, none of the patients in this case presented scar, which would need to be considered when extending these results to the wider population. Two of the three patients did, however, have non-ischaemic conduction block introducing similar complexities to the Electrophysiological (EP) activation, and this was incorporated into the models.

While we have seen that a personalized model can indeed be used for the planning of optimal therapy in advance of the procedure, it should be noted that the models used in this work were fitted to data recorded during the CRT implant procedure, including the AHR. It however may be possible that sufficient detail can be captured with only preclinical data to determine the relative responses of different pacing sites. While this would not allow an accurate prediction of the magnitude of response expected from a patient, however, it would nevertheless allow us to determine the optimal pacing region and therefore retain utility as a treatment planning tool. Furthermore, continuing advances in computer hardware and simulation software⁸ bring closer the prospect of parameterization of the model mid-procedure, whereby additional invasive measurements could be incorporated into the model and so used to update the prediction of optimal pacing site.

Conclusions

Personalized models of cardiac electromechanics were used to perform an in silico LV pacing lead location optimization study in three patient cases. Several metrics of regional cardiac function at SR were tested for their efficacy at predicting the optimal pacing site. Laterobasal pacing locations were found to maximize response of therapy, both in terms of the reduction of activation time of the ventricles and according to the acute haemodynamic response. Pacing locations with greater reductions in total activation time of the ventricles were found to result in a greater AHR. Pacing in regions activated later at SR, determined either by the time of electrical activation or of first shortening of the muscle, was found to generally result in better AHR, however pacing in the last activated site did not result in the optimal response to therapy in all cases.

Funding

The authors would like to acknowledge funding from the Engineering and Physical Sciences Research Council (EP/G0075727/2) and the Wellcome Trust Medical Engineering Centre at King’s College London (088641/Z/09/Z). This research was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at Guy’s and St. Thomas’ NHS Foundation Trust and King’s College London. PL holds a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (099973/Z/12/Z). G.P. is funded by the Austrian Science Fund (FWF) (F3210-N18) and the European Commission (CardioProof, 611232). S.A.N. receives funding from Boston Scientific. C.A.R. receives research funding and Honoraria from St Jude Medical, Medtronic and Boston Scientific.

Conflict of interest: none declared.