An echocardiographic model for predicting the recurrence of paroxysmal atrial fibrillation after circumferential pulmonary vein ablation

Abstract Background Atrial fibrillation (AF) is a highly prevalent arrhythmia, with substantial associated morbidity and mortality. Circumferential pulmonary vein ablation (CPVA) is an effective rhythm control strategy, however, recurrence is an important factor influencing treatment decisions. Hypothesis To develop a predictive model based on left atrial (LA) structure and function, and evaluate its efficiency in predicting the recurrence of AF after CPVA. Methods Patients with paroxysmal AF who underwent CPVA were enrolled in this study and randomly divided into a development set and a validation set. The clinical and echocardiographic data of each patient were collected. In the development set, a least absolute shrinkage and selection operator (LASSO) regression was used to establish a LA ultrasound feature. By combining that LA ultrasound feature with independent clinical risk factors, we established an echocardiographic model using multivariate logistic regression and plotted the corresponding nomogram. Results The LA ultrasound feature established by LASSO regression included nine echocardiographic indicators related to LA structure and function. It also exhibited good predictive ability in both the development set and the validation set (AUC:0.944, 95%CI: 0.910–0.978; AUC:0.878, 95%CI: 0.816–0.942). Logistic regression analysis indicated that LA ultrasound feature and AF duration were independent predictors for AF recurrence. The combined model including LA ultrasound feature and AF duration also showed good discriminability in both the development set (AUC: 0.950, 95% CI:0.914–0.985) and the validation set (AUC: 0.890, 95% CI: 0.831–0.949). The calibration curve showed good agreement between the predicted value and observed value. Conclusions Our model that is based on LA structure and function measured by echocardiography is a useful non‐invasive preoperative tool, which exhibits good accuracy in predicting the recurrence of AF after CPVA.

0.831-0.949). The calibration curve showed good agreement between the predicted value and observed value.
Conclusions: Our model that is based on LA structure and function measured by echocardiography is a useful non-invasive preoperative tool, which exhibits good accuracy in predicting the recurrence of AF after CPVA.

| INTRODUCTION
Atrial fibrillation (AF) is the most common clinical arrhythmia, with a population incidence of 1%-2%. There are about 30-100 million AF patients in the world. 1 AF is also a progressive disease with an annual progression rate of 7%-15%, 2 and more than 50% of paroxysmal atrial fibrillation (PAF) cases will develop into persistent atrial fibrillation (PeAF) within 10 years. 3 Compared with PAF patients, the AF in patients with PeAF and more risk factors of cardiovascular disease usually progress faster to a more permanent form. 4,5 Therefore, positive rhythm control can slow down the progression of AF, as shown in a study where the progression rate of AF in the rhythm control set was 3.2 times lower than that in the ventricular rate control set. 5 Circumferential pulmonary vein ablation (CPVA) is an effective rhythm control strategy. Multi-center randomized clinical trials have found that CPVA works better than medications for treating PAF, providing a basis for using CPVA as a first-line treatment for AF. 6 The European Society of Cardiology 2020 recommends that AF catheter ablation for CPVA might be considered as a first-line rhythm control therapy for improving symptoms in selected AF patients. 7 However, the recurrence rate after CPVA is still high (up to 20%-40%). 8 Therefore, it is important to establish a model to predict recurrence after CPVA in PAF patients.
Moreover, the study of risk factors based on the left atrial (LA) structure and function is an important part of the comprehensive management of AF. In AF patients, the original electrical and structural characteristics of the left atrium are remodeled, and this remodeling is the structural basis for recurrence after CPVA. 9 Real-time three-dimensional echocardiography (RT3DE) is a new method for evaluating LA remodeling, and is more sensitive in determining LA dysfunction than other imaging modalities. The maximum LA volume (LAVmax) measured by 3-dimensional (3DE) has independent and improved prognostic value over the LAVmax measured by 2-dimensional echocardiography (2DE). [10][11][12] The LA structure and function can be used to identify patients with a high risk of AF recurrence after CPVA. [13][14][15] However, the LA indicators collected and calculated by echocardiography have multi-dimensional collinearity, and the clinically significant factors may be screened out during the traditional logistics regression process. Therefore, the least absolute shrinkage and selection operator (LASSO) regression, a type of linear regression that uses shrinkage, was used to screen and establish algorithms based on LA indicators while achieving comprehensive variable selection and parameter estimation.
In our study, we used RT3DE technology to evaluate LA structural remodeling, and then performed a LASSO regression to obtain a LA ultrasound feature that represents various indicators of the left atrium.
Furthermore, by combining the LA ultrasound feature with clinical risk factors, we established and verified a predictive model for AF recurrence after CPVA in PAF patients. Through this study, we hope to provide useful information for optimizing the rhythm control strategy of PAF patients.

| Objects
This is a single center, retrospective cohort study. The consecutive PAF patients who were hospitalized in The First People's Hospital of Changzhou City from October 2015 to December 2018 and underwent CPVA treatment for the first time were enrolled in our study. Inclusion criteria: (1) patients with AF confirmed by conventional electrocardiogram and/or 24-hour dynamic electrocardiogram, and sinus rhythm could be restored spontaneously or after intervention within 7 days of onset; (2) at least one type I or III antiarrhythmic drug treatment was ineffective, or patient who was unable to tolerate the side effects of drugs and were willing to receive CPVA treatment. Exclusion criteria: (1) the image quality of 2DE was poor and cannot be further analyzed and processed; (2) AF shown by echocardiography; (3) left ventricular ejection fraction (LVEF) lower than 50%; (4) patients younger than 18 years old or older than 80 years old; (5) patients with structural heart disease; (6) transesophageal echocardiography (TEE) showed LA appendage (LAA) emptying velocity lower than 40 cm/s; (7) patients with acute or prior myocardial infarction; (8) patients with a history of thoracotomy; (9) patients with moderate to severe valvular dysfunction; (10) patients with chronic obstructive pulmonary disease and were treated with betaagonists; (11) patients with hemorrhagic constitution or intolerant to heparin and anticoagulant drugs. This study complied with the principles of the Declaration of Helsinki. All patients signed the written informed consent. This study has been approved by the Scientific Ethics Committee.

| General information
By querying the medical record system, we obtained information on each patient including age, gender, body mass index (BMI), serum creatinine concentration (SCR), AF duration (the time from the first recorded electrocardiogram of AF to CPVA), smoking and drinking his-

| Image acquisition of echocardiography
Transthoracic echocardiography (TTE) and TEE were performed on the day before CPVA. All patients' images were collected during sinus rhythm. The inspection equipment was the Philips EPIQ 7C color

| Image analysis of echocardiography
Philips QLAB software (10.5, Philips Healthcare Royal Philips Electronics, Amsterdam, The Netherlands software pack) was used to analyze and obtain the 3-dimensional image of the left atrium, as well as its volume and function parameters. On the end-diastolic image of the apical 4-chamber, the 3DQ-A mode was selected, and the sampling point was placed on the LA wall. After adjusting the shape of the left atrium, we divided it into three parts, manually outlined the volume of each LA part, and calculated the minimum LA volume (LAVmin). Similarly, the end-systolic image of the apical four-chamber was used to calculate the LAVmax. The electrocardiographic image before the start of the P wave was used to calculate the LA pre-systolic volume (LAVpreA). From these volumes, total, passive, and active emptying volumes and fractions were derived. 16

| Postoperative treatment and follow-up
All patients converted to sinus rhythm during surgery and completed their 3-6 months follow-up. All patients took the previous anticoagulant for 2 months after the procedure, and then the medication was decided based on their stroke risk score. All patients continued to use an antiarrhythmic drug treatment for 3 months after surgery (amiodarone, 200 mg, Qd; propafenone 150 mg, Tid or morerazine 150 mg, Tid; sotalol 80 mg, Bid).Patients received a 12-lead electrocardiographic examination in each week or when arrhythmia symptoms occurred, Moreover, patients returned to the outpatient clinic every month to record medical history, obtain physical examinations, and receive the 12-lead electrocardiogram and 24-hour dynamic electrocardiogram. All follow-ups were conducted by personnel who had no information about their treatment plan. Any rapid atrial arrhythmia with ECG that occurred 3 months after CPVA surgery and lasted for more than 30s were defined as AF recurrence, which included AF, atrial flutter, or atrial premature contraction.

| Statistical analysis
Normally distributed data were expressed as mean ± SD, and nonnormally distributed data were expressed as median and interquartile range. Comparisons between groups were performed using independent sample t-test or Mann-Whitney U nonparametric test. Classification data were expressed as frequency or rate (%), and comparison between groups was performed by Pearson χ 2 test or Fisher exact probability method. The LASSO regression was used for analysis, which is suitable for the regression of high-dimensional data. 17  (206 cases) and a validation set (197 cases) by the split-sample function of the R software ( Figure 1).

| Comparison of general information between the development set and validation set
There was no statistical difference in general data between the two sets, including gender, age, LVEF, BMI, HR, SCR, AF duration, CHA 2 DS 2 -VASc score, conditions of hypertension, DM, HTG, CHF, smoking history, drinking history, history of stroke, or TIA, vascular disease, and so forth. There was also no significant difference in LA structure and function indexes LAmax, LAmin, LApreA, EI, DEI, AE, AEI, PE, and PEI between the two sets (all p ≥ .05) ( Table 1).

| Comparison between recurrence group and sinus rhythm group in the development set
In the development set, the patients were divided into a recurrence group (39 cases) and a sinus rhythm group (167 cases higher in the recurrence group, and the EI, DEI, PE, and PEI were lower in the recurrence group. The differences were statistically significant (p < .001). The AF duration of the recurrence group was significantly longer than the sinus rhythm group (p = .003). The difference in other indicators was not statistically significant (Table 2).

| Establishment of The LA ultrasound feature
Nine LA size-related features with nonzero coefficients were screened using a LASSO logistic regression model in the development set (Figure 2A-C) Figure 2D, Table S1). The difference of AUC between the two sets was not significant (z = 1.846, p = .065).

| Model construction based on LA ultrasound feature established by LASSO regression and clinical features
The LA ultrasound feature and AF duration were identified as independent predictors of AF recurrence in PAF patients by a multivariate logistic regression model (Table S2). Therefore, we constructed a stepwise (stepAIC) selected model from the observed data:

| DISCUSSION
Ablation is a well-established treatment for symptomatic drugrefractory paroxysmal AF, supported by a consistent body of evidence. 7 However, radiofrequency energy can cause LA cell damage and necrosis, which can lead to LA scar tissue contraction. The postoperative AF recurrence not only accelerates the progress of AF, but also increases the economic and social burden of the patients. Previous studies have found that the ampliative left atrium is a good predictor for AF recurrence. 19 A meta-analysis based on observational studies also indicated that the LAVi in patients with AF recurrence after ablation was significantly higher than the patients without recurrence. 20 LAVImax measured by 2DE has become as an important biomarker for adverse cardiac events under a variety of cardiovascular conditions. 21 In part because of the LA complex geometry and  The duration of AF was considered as an important prognostic factor for the outcome of cardiac rhythm after CPVA. The patients with a disease course of less than 2 years had a lower AF recurrence rate at 1 year after surgery than those with a disease course of more than 2 years, 26 and the PeAF patients with a disease course of more than 3 years were more likely to relapse after CPVA than the patients with disease course of less than 3 years. 27  To provide clinicians with an easy-to-use tool, we constructed a nomogram based on a multivariate logistic regression model, which showed good calibration and recognition capabilities in both the development and validation sets. In practice, clinicians can calculate the total score using the scores corresponding to the variables of the LA ultrasound feature and AF duration in the nomogram, and draw a straight line down to the total score to obtain the corresponding recurrence probability. When categorized into low-and high-risk groups based on the cutoff values of the risk score derived from the nomogram, the high-risk group had a significantly greater probability of AF recurrence. Therefore, our nomogram can serve as an accurate and reliable predictive tool for AF recurrence after CPVA in PAF patients.
The limitations of this study mainly include the lack of external validation of the model, and the requirement for multi-center validation with a larger sample size to obtain high-level clinical application evidence. In addition, serological indicators were not included in our model. According to previous studies, the baseline BNP concentration in the serum of PAF patient might be a predictor of recurrence after successful electrical cardioversion. 28 Our echocardiographic model is a non-invasive predictive tool that combines the LA ultrasound feature and the AF duration. It showed good predictive accuracy of AF recurrence after CPVA in PAF patients. As a convenient and fast non-invasive examination method, echocardiography can provide strong support to the comprehensive management of AF patients and the design of personalized treatment plans.

The Changzhou City Technology Bureau and Changzhou City Health
Bureau has no role in this study.

CONFLICT OF INTEREST
All authors declare that: (1) they have not received any support from any organization having an interest in the submitted works, whether financial or otherwise; (2) there are no other relationships or activities that may affect the submitted works.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.