Background
Transcatheter aortic valve replacement (TAVR) is a safe and effective treatment for patients with severe aortic stenosis at even low surgical risk [
1,
2]. Permanent pacemaker implantation (PPMI) remains a common complication after TAVR compared to surgical aortic valve replacement (SAVR) [
3,
4]. In a recent systematic review, 22.7% of patients developed permanent left bundle branch block (LBBB) after TAVR, and 5.9% to 32.0% required PPMI [
5].
PPMI after TAVR not only increases the occurrence of wire-related tricuspid regurgitation, implant infection, wire damage, pocket damage and hematoma, cardiac perforation and other complications [
6,
7], but also increases the economic burden of patients and the length of hospital stay [
8]. The results of a recent meta-analysis further confirmed a significant increase in mortality among patients requiring PPMI after TAVR [
5].
Previous studies suggested that prior right bundle branch block (RBBB) [
9,
10], short membranous septum length [
11,
12], calcification of device landing zone [
10,
13], intra-procedural atrioventricular block (AVB) [
4], prosthesis oversizing [
14], self-expanding valve [
15,
16], and lower implantation depth [
10,
13] were associated with PPMI after TAVR, but no single factor could predict PPMI accurately.
The majority of the existing prediction models for PPMI only used pre-procedural predictors [
17‐
19]. However, for post-procedural patients, there still lacks effective prediction models for PPMI risk assessment and timely guidance of pacemaker removal and patient discharge. In this study, we aim to develop a post-procedural prediction model for patients requiring PPMI after TAVR, on top of the pre-procedural risk prediction model.
Discussion
PPMI due to conduction block remains one of the major complications after TAVR, and there has been limited research on effective prediction models involving both pre- and post-procedural characteristics. In this study, we developed a post-procedural prediction model for PPMI via multivariate logistic regression and identified RBBB, pre-procedural AVA, post- to pre-procedural AVA ratio, and post-procedural AVA to PNA ratio as independent predictors for PPMI. Among them, having prior RBBB, large pre-procedural AVA and large post- to pre-procedural AVA ratio are risk factors, while large post-procedural AVA to PNA ratio is protective for PPMI. To our knowledge, AVA ratio and AVA–PNA ratio are identified as independent predictors for the first time and these two factors may related to the mechanical stress of the prosthesis. Total risk score can be calculated by adding up individual risk score associated with each predictor according to the post-procedural prediction model, and patients can then be classified into different risk groups based on their total risk score. The model has been validated internally via bootstrap and externally via an independent cohort of patients from the same hospital. Studies showed that incidence of delayed PPMI is increasing [
25]. Therefore, we investigated the effects of potential predictors on how soon PPMI could occur quantitatively as the secondary analysis.
RBBB has been reported as a risk factor for PPMI in many previous studies [
9,
10,
13] and was also confirmed in our analysis. The high incidence of new-onset LBBB after TAVR [
5] could explain why prior RBBB strongly associated with CHB or HAVB.
The effect of pre-procedural AVA on PPMI is likely related to the buffering force of the prosthetic valve holder to surrounding tissue due to the stenosis valve. Since the atrioventricular conduction system is anatomically close to the subaortic valve structure, the risk of pressure damage to the conduction system during valve implantation is difficult to avoid [
26]. Senile aortic stenosis is often associated with severe calcification, and a reduction in valve area often means an increase in surrounding calcified tissue, which provides good support for the implanted valve and may reduce the direct pressure of the prosthetic valve holder on the subvalve tissue. The contact point between bicuspid aortic valve (BAV) and prosthetic valve holder may be closer to the axis and further away from membranous septum, therefore, pre-procedural AVA is more likely to affect the Chinese population with a higher prevalence of BAV stenosis. This may also explain why pre-procedural AVA was not found to be associated with PPMI among European and American population [
11].
Post- to pre-procedural AVA ratio reflects the extent to which tissue is squeezed between the prosthetic valve holder and the aortic wall after prosthetic implantation, with higher ratio indicating more squeezing and higher pressure of the prosthetic valve holder on the aortic wall and left ventricular outflow tract. High prosthetic pressure on surrounding tissues, especially the vulnerable areas where the His bundle and the left bundle branches locate, is a risk factor for PPMI [
27,
28]. Therefore, large post- to pre-procedural AVA ratio increases the risk of damage to the conduction system and PPMI after TAVR. Although AVA ratio and AVA–PNA ratio did not achieve statistical significance (AVA ratio: HR = 1.50,
p = 0.053; AVA–PNA ratio: OR = 0.01;
p = 0.052) at the 0.05 significance level, we would still consider them as important predictors for the time-to-PPMI and PPMI, given the relatively small sample size in the study. AVA–PNA ratio is an estimate of the relative expansion of the prosthetic valve holder, with smaller AVA–PNA ratio indicating more compression (i.e., less expansion) of the prosthetic on surrounding tissues. In vitro experiments also confirmed that the radial expansion of the prosthetic valve was approximately linearly related to its compression [
29]. However, estimation of the valve pressure and tissue compression was mostly done using finite element analysis (FEA) in the simulation setting, and is difficult to generalize to clinical settings due to technical limitation. In this study, we used a simple and more assessable approach by calculating the post-procedural AVA to PNA ratio as an alternative measure of tissue compression, and confirmed high prosthetic pressure as the risk factor for PPMI after TAVR [
27,
28].
AVA ratio and AVA–PNA ratio were both used to evaluate the mechanical stress. AVA ratio estimates the expansion of the aortic valve tissue pushed by prosthesis, while AVA–PNA ratio reflects the deformation of prosthesis under compression. The compressed prosthesis generated radial force due to continuous expansion of the memory metal, while the expansion level of the aortic valve tissue around the prosthesis reflects the response of the surrounding tissue to the force, which shows the stiffness and compliance of the tissue. The occurrence of PPMI after TAVR is closely related to the pressure of the prosthesis on the membranous septum, and the pressure degree is closely related to the compression of the valve frame. However, for patients with the same annulus size and the same prosthesis, the actual compression ratio of the prosthesis may not be the same due to different situation of valve leaf calcification, adhesion and fusion. Therefore, AVA–PNA ratio is likely to better reflect the above situation. This would explain why AVA–PNR ratio, rather than oversizing rate, was significantly different between PPMI and non-PPMI patients.
Previous studies have shown that myocardial injury and elevated troponin after TAVR were associated with increased risk of PPMI [
30,
31]. In this study, we assessed the elevation of troponin using the difference between post- and pre-procedural troponin-T. To better reflect the myocardial injury specifically caused by TAVR, we excluded patients undergoing concurrent PCI, SMA, and electrical cardioversion. However, the post-relative to pre-procedural difference of TnT was not statistically associated with binary PPMI endpoint due to small sample size. In the secondary analysis involving time-to-PPMI, the post-relative to pre-procedural difference of TnT was identified as a significant risk factor and patients with higher elevated troponin were more likely to have PPMI sooner, suggesting that early occurrence of PPMI might be caused by acute myocardial injury due to the value pressure of TAVR. However, the specificity of troponin is limited due to the simultaneous presence of coronary embolism caused by small emboli detachment during valve release [
32], and the effect of intra-procedural tachyarrhythmia.
Evidence from previous studies showed that PPMI can occur beyond 30 days after TAVR [
33]. The medical records of patients with PPMI beyond 30 days after TAVR in the derivation set showed that most of these patients experienced amaurosis or syncope within 30 days after TAVR, but did not seek appropriate medical treatment in time. The latest PPMI in the derivation set was observed on day 58 after TAVR due to symptomatic atrial fibrillation with slow ventricular response. Therefore, we decided to focus on the PPMI cases within 60 days after TAVR in the primary analysis. In the secondary analysis, we investigated potential factors that might influence patients’ time-to-PPMI and found that in addition to acute myocardial injury, greater valve pressure as indicated by large pre-procedural AVA, post- to pre-procedural AVA ratio and small post-procedural AVA to PNA ratio were associated with sooner PPMI.
Delayed high-degree atrioventricular block after TAVR is increasing in recent years, which may due to the early-discharge practices and surveillance strategies after discharge [
34]. This may cause syncope and fall out of the hospital, which may fatal to the already elderly patients. The prediction of delayed PPMI occurrence is however difficult and the related risk factor is still unknown. Current thinking suggests that the persistent oppression caused by the self-expandable valves maybe one of the main reasons. The factors derived from our predictive model may reflect the oppression stress and showed relevance with the timing of PPMI which may useful for future research.
The prediction model achieved good discriminative power in both the internal and external validation. For example, Shivamurthy et al. included transfemoral approach, LBBB without bradycardia, sinus bradycardia without LBBB, RBBB, LBBB with sinus bradycardia, and second-degree AVB in a prediction model with AUROC 0.674 [
19,
35]. In another prediction model by Tsushima et al., hypertension, first-degree AVB, self-expanding valve use and RBBB were selected as potential predictors for PPMI and the AUROC was 0.693 [
18]. We verified these two models and obtained an AUROC of 0.57 and 0.58 using patient data in the derivation set, respectively (Additional file
1: Figure S4), which suggests that existing prediction models might not be appropriate for the Chinese population.
Previous studies have suggested other predictors for PPMI, such as the distribution and sizing of aortic valve calcification and membranous septum length [
11]. However, these predictors cannot be easily obtained in clinical settings since they require specific imaging protocols and experienced technicians/physicians for image interpretations. Therefore, we chose not to include them in the predictive models in this investigation. Artificial intelligence is growing rapidly in recent years. The previously reported machine learning-based prediction models demonstrated significantly high predictive accuracy [
36,
37]. Due to the restricted data volume, machine learning is not applicable to our study yet. By using this method, we need to screen the input variables first, because in restrict to the computer calculation speed, we cannot put in the whole database. Our study may give some insights about the variables choosing in future research, especially about the delayed PPMI after TAVR. The majority of the existing prediction models for PPMI included only the pre-procedural predictors and obtained patient susceptibility before undergoing TAVR. The prediction model developed in our study combined both pre- and post-procedural characteristics and incorporated patient susceptibility into the total risk score after TAVR, and is thus more clinically meaningful for patients with high risk of PPMI.
Study limitations
First, the sample size in our study is small. According to the 10 event-per-variable (EPV) rule-of-thumb that the number of predictors in the model should not exceed one-tenth of the total number of events, we included only 4 predictors in the post-procedural prediction model. As a result, variables that were associated with PPMI in previous findings, such as history of syncope [
17], hypertension [
18], valve type [
18] and oversizing rate [
17], were not included due to insufficient power. Second, the majority of patients in our study were implanted with self-expanding valve. Therefore, the prediction model might not be generalizable to patients with balloon-expanding valve. Third, patients in the derivation set and external validation set had notable differences in terms of cardiovascular disease history, TnT and post-procedural AVA, which might affect results for external validation. Fourth, this is a single-center study and might not generalize to patients from other medical centers, although quite a bit of the TAVR within mainland China were conducted in our department. Fifth, our study is retrospective in design without prospective validation, and might be subject to selection bias. However, predictors identified in our post-procedural prediction model are all physiologically and anatomically reasonable. Sixth, the electrocardiogram after TAVR in our center is made in paper version beside the bed, electronic information is lacking. And we found many of the thermal drawings of 4–5 years ago can’t be seen clearly right now, so we can’t use them. Thus, we can’t analyze factors like the prolongation of QRS and PR after TAVR. But LBBB, CHB, HAVB and other cases requiring PPMI have been recorded in the disease course in time, this is the source of our data. Seventh, for patients with sinus rest and sick sinus syndrome, pacemakers are usually implanted before TAVR, and such patients are excluded from this study. But the two patients in our study showed normal electrocardiogram before TAVR, so it’s difficult to determine if this is related to TAVR or if it just wasn't detected by a regular electrocardiogram before TAVR. This may have some impact on the results.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.