Cardiac MRI-based strain analysis for the assessment of left atrial dysfunction in patients with atrial fibrillation

Abstract

Atrial fibrillation (AF) involves changes in atrial myocardium and early left atrial (LA) dysfunction, which are difficult to identify using conventional static measurements. This study aimed to investigate the value of cardiac magnetic resonance feature tracking (CMR-FT) derived LA strain parameters for assessing LA dysfunction in AF patients and to identify sensitive indicators for early detection of impaired LA function.This retrospective study included 113 AF patients and 52 age- and sex-matched healthy controls. LA structural (LAD, LAVI) and functional parameters (LAEF, εs, εe, εa) were measured. AF patients were divided into AF1 (abnormal LAVI/LAEF, n = 92) and AF2 (normal LAVI/LAEF, n = 21). Group comparisons, correlation analysis, ROC analysis, and univariate and multivariate logistic regression analyses were performed.AF patients showed significantly increased LAD and LAVI, decreased LAEF, and impaired LA strain (all P < 0.001). Even in the AF2 subgroup with normal LAVI/LAEF, LA strain remained significantly impaired (all P < 0.001). LA strain was positively correlated with LVEF and LAEF, and negatively correlated with LAD and LAVI (all P < 0.001). εs showed the strongest correlation with total LAEF (r = 0.902). ROC analysis showed high efficacy for identifying LA dysfunction (AUC > 0.80), with εs achieving the highest AUC (0.917).CMR-FT enables non-invasive quantitative assessment of LA function. LA strain is more sensitive than conventional static parameters for detecting established LA dysfunction in AF patients, although patients already exhibited structural remodeling and reduced LVEF. The small AF2 subgroup limits the robustness of early-detection claims. εs is a promising imaging biomarker for assessing LA dysfunction.

Introduction

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia, characterized by progressive atrial myocardial remodeling including inflammation, fibrosis and structural changes, which leads to left atrial (LA) dysfunction and further aggravates the progression of AF [1, 2]. Accurate assessment of LA dysfunction is the key to early diagnosis and intervention of AF. Traditional evaluation of LA function relies on static structural parameters such as LA volume and ejection fraction, which can only reflect the late structural changes of the atrium and cannot detect the early functional impairment before LA enlargement [3, 4].

The left atrium maintains cardiovascular homeostasis through three functional phases (storage, conduit, and contractile), and impairment of any phase leads to abnormal left ventricular filling and cardiac hemodynamics [5, 6]. Due to the thin myocardial wall and complex anatomical location of the LA, it is difficult to accurately measure its myocardial deformation by traditional echocardiography [7]. Cardiac magnetic resonance feature tracking (CMR-FT) is an emerging non-invasive quantitative imaging technology with high spatial resolution and excellent tissue contrast, which can track myocardial contours throughout the cardiac cycle and generate accurate strain parameters to reflect the dynamic mechanical changes of the atrium [8, 9]. Previous studies confirmed that CMR-FT derived LA strain parameters can sensitively reflect atrial myocardial remodeling and are expected to be a new index for evaluating LA function [10, 11]. LA myocardial strain is divided into total strain (εs), passive strain (εe), and active strain (εa) based on the three functional phases of the LA, and εe is calculated as the difference between εs and εa, which can comprehensively reflect the mechanical state of the LA myocardium.

Cardiac rhythm during CMR acquisition strongly influences LA strain measurements, and rhythm status must be controlled or adjusted for in statistical models [19, 20]. In addition, large-cohort studies have validated the reproducibility and generalizability of atrial strain measurements, supporting their clinical translation [19, 20].

This study aimed to compare the differences in CMR-FT derived LA strain parameters between AF patients and healthy controls, analyze the correlation between LA strain and traditional LA structural/functional parameters, and evaluate the value of LA strain parameters for assessing left atrial dysfunction.

Patients and Methods

Patients

A total of 750 hospitalized patients who underwent CMR imaging at Shanghai East Hospital from 2020 to 2024 were retrospectively reviewed, among whom 209 had a confirmed history of paroxysmal or persistent AF. After strict exclusion (prior electrophysiological tests/catheter ablation for AF, severe valvular heart disease, poor CMR image quality), 113 AF patients were finally included. A control group of 52 age- and sex-matched healthy subjects without AF or other cardiac diseases was selected during the same period.

AF patients were divided into two subgroups based on LAVI and LAEF values: AF1 subgroup (n = 92, abnormal LAVI and/or LAEF beyond the healthy reference range) and AF2 subgroup (n = 21, normal LAVI and LAEF within the healthy reference range).

Inclusion criteria for AF group

(1) Underwent routine CMR imaging prior to catheter ablation or pulmonary vein isolation; (2) AF was confirmed by clinical history, 12-lead ECG, or 24-hour Holter monitoring.

Exclusion criteria for AF group

(1) Significant valvular heart disorder; (2) Previous experience of a heart attack; (3) Major cardiomyopathy conditions; Significant dysfunction of hepatic or renal systems.(5) Limitations for undergoing CMR scans.

Inclusion criteria for normal control (NC) group

(1) Aged ≥ 18 years; There are no heart or lung conditions, nor related risk factors.(3) The EKG indicates a stable cardiac rhythm and pace. (4) Normal cardiac chamber sizes and ventricular function on CMR; (5) Stable sinus rhythm and heart rate < 80 bpm during CMR scanning.

CMR Scanning Strategy

All CMR images were acquired using a 1.5T MRI machine from Philips Healthcare equipped with a 32-channel coil specifically designed for cardiac imaging. Cardiac rhythm during scanning was continuously monitored and recorded for all participants. AF patients received heart rate control treatment (oral β-receptor blockers) before scanning to maintain heart rate < 80 bpm and reduce rhythm‑related variability in strain analysis [19, 20].

Standard axial and sagittal black blood T2-weighted sequences were obtained using the half-Fourier single-shot turbo spin-echo technique. Left ventricular (LV) long-axis 2-chamber, 4-chamber, short-axis views and LA long-axis/short-axis views were captured with steady-state free precession (SSFP) cine sequences with the following parameters: slice thickness 8 mm, slice gap 0–4 mm, TR 2.9–3.4 ms, TE 1.5–1.7 ms, flip angle 45°–50°, FOV 320 × 320–380 × 380 mm, matrix 192 × 224–224 × 256. LA images ensured full coverage from the atrioventricular ring to the atrial floor.

Post-processing

All CMR images were analyzed by two experienced radiologists (8 and 10 years of cardiac CMR experience) using CVI 42 software (Circle Cardiovascular Imaging Inc., Version 5.14). Conflicts were settled through mutual agreement.

LA and LV myocardial contours were manually delineated on SSFP cine sequences, and the software automatically tracked the contours throughout the cardiac cycle to generate quantitative parameters: (1) LA structural parameters: LAD, LAVI (LAVImax, LAVIpre-ac, LAVImin); (2) Assessments of left atrium function: total, passive, and active left atrial emptying fractions.(3) LA strain metrics include: εs, εe, εa, global circumferential strain (GCS). (4) LV parameters: LVEF. All strain parameters were measured three times, and the average value was used for statistical analysis.

Formal reproducibility analysis was not performed in this study; however, excellent reproducibility of CMR-derived atrial strain has been well established in previous large validation studies [19, 20].

Statistical Analysis

SPSS version 29 was utilized for a comprehensive analysis of the dataset. The evaluation of normality was conducted using Q-Q plots along with the Shapiro-Wilk test.For variables exhibiting a normal distribution, data were presented as the mean with the standard deviation (x ± s). Variables not adhering to a normal distribution were shown as the median along with the interquartile range.Categorical variables were represented using frequencies and their corresponding percentages (n, %).

To examine differences in continuous variables across groups, independent sample t-tests were utilized, whereas Chi-square (χ²) tests were applied to assess categorical variables. Pearson correlation coefficients were used to analyze the linear relationship between LA strain and other cardiac parameters. To assess the efficacy of each parameter for left atrial dysfunction, ROC curves were developed, and the best threshold was identified using the Youden index. Univariate and multivariate logistic regression analyses were performed to identify factors associated with LA dysfunction. Multivariate models were adjusted for rhythm status during scanning, heart rate, age, BMI, diabetes, LVEF, and beta-blocker use. A p-value less than 0.05 was deemed to indicate statistical significance.

Results

Clinical Baseline Characteristics of AF and NC Groups

The average age within the AF group was significantly higher, recorded at 65 years(65.01 ± 11.58 vs. 59.12 ± 10.65 years, P = 0.002) and had a higher BMI (25.09 ± 3.33 vs. 23.90 ± 2.99 kg/m², P = 0.030) than the NC group. The prevalence of hypertension (59.29% vs. 21.15%) and diabetes (27.43% vs. 3.85%) in the AF group was notably greater compared to the NC group (all P < 0.01). There were no notable disparities between the two groups in terms of gender distribution, body surface area (BSA), blood pressure, fasting blood glucose, or heart rate, with all P-values exceeding 0.05.

Comparison of LA and LV Function/Strain between AF and NC Groups

Compared with the NC group, AF patients had significantly increased LAVI (LAVImax, LAVIpre-ac, LAVImin) and LAD, reduced all LAEF parameters, and impaired all LA strain parameters (all P < 0.001). Meanwhile, the AF group had significantly lower LVEF and impaired LV strain parameters (all P < 0.001), suggesting that AF patients were accompanied by obvious structural remodeling and functional impairment of the left atrium and ventricle (Table 1). (Fig. 1)

Table 1.

Analysis of left atrial (LA) and left ventricular (LV) function and strain in the AF group and NC group (x ± s)

Index	NC(n = 52)	AF(n = 113)	p
LAVI max (ml/m2)	32.55 ± 9.12	60.02 ± 28.85	< 0.001**
LAVI pre-ac (ml/m2)	18.16 ± 4.08	45.72 ± 27.85	< 0.001**
LAVI min (ml/m2)	13.54 ± 5.23	43.38 ± 27.62	< 0.001**
LAD(mm)	35.04 ± 3.72	44.96 ± 7.77	< 0.001**
LAEF total(%)	58.80 ± 8.47	32.31 ± 15.96	< 0.001**
LAEF passive(%)	42.75 ± 8.93	26.60 ± 15.68	< 0.001**
LAEF active(%)	27.19 ± 14.33	9.31 ± 9.25	< 0.001**
εs	35.64 ± 9.68	14.40 ± 10.79	< 0.001**
εe	19.08 ± 8.07	8.23 ± 6.45	< 0.001**
εa	16.56 ± 5.50	6.17 ± 5.11	< 0.001**
GCS	−18.51 ± 3.43	−9.00 ± 5.32	< 0.001**
LVEF(%)	60.72 ± 7.09	45.98 ± 15.78	< 0.001**

Note: LAVI = left atrial volume index; LAD = left atrial diameter; LAEF = left atrial ejection fraction; εs = LA total strain; εe = LA passive strain; εa = LA active strain; GCS = global circumferential strain; AV = atrioventricular; LVEF = left ventricular ejection fraction; **P < 0.01 vs. NC group

Fig. 1.

Comparison of key LA structural, functional and strain parameters between AF group and NC group. Note: Compared with the NC group, the AF group showed significantly increased LAVI min, and significantly decreased LAEF total and εs (all P < 0.001). LAVI min=left atrial minimum volume index; LAEF total=left atrial total ejection fraction; εs = LA total strain; NC=normal control; AF=atrial fibrillation

Strain Analysis of AF1 and AF2 Subgroups

AF1 subgroup (abnormal LAVI/LAEF) had significantly more severe impairment of all LA strain parameters than AF2 subgroup (normal LAVI/LAEF) (all P < 0.001). Notably, AF2 subgroup still had significantly impaired LA strain and LV function parameters compared with the NC group (all P < 0.001).Refer to Table 2 for details.(Fig. 2)

Table 2.

Strain analysis of the NC group, total AF group, AF1 subgroup, and AF2 subgroup (x ± s)

Index	NC (n = 52)	AF (n = 113)	AF1 (n = 92)	AF2 (n = 21)	P
εs(%)	35.64 ± 9.68	14.40 ± 10.79	10.68 ± 7.16	30.70 ± 8.72	< 0.001**
εe(%)	19.08 ± 8.07	8.23 ± 6.45	6.10 ± 4.13	17.58 ± 6.53	< 0.001**
εa(%)	16.56 ± 5.50	6.17 ± 5.11	4.58 ± 3.63	13.12 ± 4.91	< 0.001**
GCS(%)	−18.51 ± 3.43	−9.00 ± 5.32	−7.29 ± 4.04	−16.50 ± 3.46	< 0.001**
LVEF(%)	60.72 ± 7.09	45.98 ± 15.78	43.07 ± 15.57	58.72 ± 9.12	< 0.001**

Note: AF1 = AF with abnormal LAVI/LAEF; AF2 = AF with normal LAVI/LAEF; **P < 0.01 for intergroup comparison

Fig. 2.

Comparison of LA strain parameters among NC group, AF2 subgroup and AF1 subgroup. Note: Compared with the NC group, the AF2 subgroup showed significant reductions in εs, εe, εa, and GCS; compared with the AF2 subgroup, the AF1 subgroup had further impairments in all strain parameters (all P < 0.001). εs=left atrial total strain; εe=left atrial passive strain; εa=left atrial active strain; GCS=global circumferential strain; NC=normal control; AF1 = AF with abnormal LAVI/LAEF; AF2 = AF with normal LAVI/LAEF

Correlation between LA Strain and Cardiac Functional Parameters in AF Patients

The parameters of LA strain (εs, εe, εa) showed a strong positive correlation with LVEF and all LAEF parameters. Conversely, they were significantly negatively correlated with LAD and all LAVI parameters (all P < 0.001).Among the variables examined, εs exhibited the highest positive correlation with the total LAEF, with a correlation coefficient of r = 0.902, P < 0.001. Refer to Table 3 for additional details.(Fig. 3)

Table 3.

Pearson correlation analysis of LA strain parameters with cardiac functional parameters in the AF group

Index	εs		εe		εa
	r	P	r	P	r	P
LVEF(%)	0.599	<0.001**	0.623	<0.001**	0.480	<0.001**
LAD(mm)	−0.585	<0.001**	−0.566	<0.001**	−0.522	<0.001**
LAVI max(ml/m2)	−0.546	<0.001**	−0.504	<0.001**	−0.516	<0.001**
LAVIpre-ac(ml/m2)	−0.638	<0.001**	−0.589	<0.001**	−0.603	<0.001**
LAVI min(ml/m2)	−0.673	<0.001**	−0.621	<0.001**	−0.638	<0.001**
LAEF total(%)	0.902	<0.001**	0.848	<0.001**	0.835	<0.001**
LAEF passive(%)	0.723	<0.001**	0.678	<0.001**	0.671	<0.001**
LAEF active(%)	0.551	<0.001**	0.542	<0.001**	0.479	<0.001**

Note: r = Pearson correlation coefficient; **P<0.01

Fig. 3.

Scatter plot depicting the relationship between εs and total LAEF in patients with atrial fibrillation. Note: εs showed a strong positive correlation with the overall LAEF (with a correlation coefficient of r = 0.902, P < 0.001) in individuals with atrial fibrillation. The continuous line illustrates the optimal linear regression fit, while the shaded region indicates the range within 95% certainty. εs denotes the total deformation experienced by the left atrium. LAEF stands for the overall ejection fraction of the left atrium

Assessment Value of LA Parameters for Left Atrial Dysfunction (ROC Curve Analysis)

ROC curve analysis showed that all LA structural, functional and strain parameters had high efficacy for assessing left atrial dysfunction with AUC > 0.80. Among the various LAVI metrics, LAVImin exhibited the greatest AUC (0.913, 95%CI: 0.862–0.964) with an advised limit of 22.65 ml/m² (sensitivity 89.4%, specificity 82.7%). Among LAEF parameters, LAEF total had the highest AUC (0.908, 95%CI: 0.854–0.962) utilizing a best threshold limit of 45.23% (sensitivity 87.6%, specificity 84.6%). Among the left atrial strain parameters, εs exhibited the greatest area under the curve of AUC (0.917, 95%CI: 0.868–0.966) utilizing a threshold optimally determined to be 24.58% (sensitivity 90.3%, specificity 86.5%), supporting its role as a potential imaging biomarker for LA dysfunction.(Fig. 4)

Fig. 4.

ROC curves of LA metrics for assessing left atrial dysfunction Note: A: ROC curves of LAVI parameters (LAVI max, LAVI pre-ac, LAVI min); B: ROC curves for parameters of Left Atrial Expansion Fraction (LAEF total, LAEF passive, LAEF active); C: ROC curves of LA strain parameters (εs, εe, εa). εs had the highest AUC (0.917, 95% CI: 0.868–0.966) among all parameters, with a sensitivity of 90.3% and a specificity of 86.5% at the optimal cut - off value of 24.58%. AUC = area under the curve; AF = atrial fibrillation

Univariate and Multivariate Logistic Regression Analysis for Left Atrial Dysfunction

Univariate logistic regression showed that age, BMI, hypertension, diabetes, and all cardiac imaging parameters were significantly associated with LA dysfunction (all P < 0.05).

To avoid multicollinearity among left atrial parameters, three adjusted multivariate models were constructed incorporating hypertension and distinct LA markers:

Model 1: Hypertension + LAVI.

Model 2: Hypertension + LAEF.

Model 3: Hypertension + LA strain.

After adjustment for age, BMI, diabetes, heart rate, and rhythm status, hypertension showed a paradoxical inverse association (OR < 1) in all models, which is inconsistent with established clinical knowledge and most likely indicates a coding or modeling artifact rather than a true biological effect.LA strain, LAEF, and LAVI remained independently associated with LA dysfunction in their respective models.

The C-index of Model 1, 2, and 3 were 0.661, 0.637, and 0.655, respectively, indicating moderate discriminative performance.Notably, the high AUC values from ROC analysis contrasted with relatively modest C-index values in multivariable models, likely due to overestimation from the healthy control design. Internal validation (e.g., bootstrap resampling) was not performed in this study, which represents a methodological limitation.

Discussion

This study confirmed that CMR-FT derived LA strain parameters can sensitively reflect LA dysfunction in AF patients, and is a more effective index than traditional static parameters for the assessment of LA dysfunction. The main findings of this study are as follows: (1) AF patients have significant structural and functional remodeling of the left atrium and ventricle, manifested as increased LAD/LAVI, decreased LAEF and LVEF, and obvious impairment of LA strain parameters; (2) LA strain parameters are impaired even in AF patients with normal LAVI/LAEF, suggesting that strain parameters can detect early LA myocardial functional damage before the abnormality of traditional static parameters; (3) εs has the strongest correlation with LAEF total, which can comprehensively reflect the overall functional state of the left atrium; (4) εs has the highest value for assessing LA dysfunction among all LA parameters and represents a promising imaging biomarker.

Consistent with recent evidence [19, 20], cardiac rhythm during CMR acquisition significantly affects LA strain measurements. We therefore strictly controlled heart rate, documented rhythm status, and included it as a key covariate in regression models to ensure validity.Although formal reproducibility testing was not performed in this study, prior large-scale validation has confirmed excellent reliability of atrial strain measurements [19, 20].

Given the absence of external validation and direct head-to-head comparison with echocardiographic strain (the clinical standard of care), εs should be regarded as a promising potential imaging biomarker rather than an optimal diagnostic marker.

Traditional LA functional evaluation relies on static structural indices including LAVI and LAEF, which merely reflect left atrial morphological and volumetric alterations and fail to identify early myocardial mechanical dysfunction in patients with AF [3,4]. In the present study, significantly impaired LA strain metrics (εs, εe, εa) were observed in the AF2 subgroup with preserved LAVI and LAEF relative to healthy controls, while strain impairment was more severe in the AF1 subgroup with abnormal LAVI/LAEF. These data demonstrate that LA strain metrics can capture early left atrial myocardial mechanical deformation abnormalities, outperforming conventional static parameters for detecting incipient LA dysfunction among AF patients. The underlying mechanism may be that LA strain directly quantifies dynamic atrial myocardial deformation, enabling detection of early atrial electrical and structural remodeling in AF; by contrast, traditional static indices only shift remarkably once advanced atrial remodeling develops [12, 13]. This observation carries critical clinical value for early screening and timely intervention in AF cohorts, particularly early-stage paroxysmal AF patients without apparent left atrial enlargement.Correlation analyses revealed tight associations between LA strain metrics and conventional LA structural and functional parameters, among which peak systolic strain (εs) exhibited the strongest positive correlation with total LAEF (r=0.902). This result aligns with prior investigations [14, 15], verifying that εs can comprehensively characterize global left atrial functional performance. Global LA systolic strain (εs) integrates functional shifts across the three distinct atrial phases (storage, conduit, and contractile), mirroring the holistic mechanical deformation capacity of LA myocardium. By comparison, isolated passive strain (εe) and active contractile strain (εa) only reflect single-phase atrial function, explaining why εs correlates more strongly with total left atrial ejection fraction.ROC curve analyses revealed all LA-derived indices possessed satisfactory diagnostic performance for AF (all AUC > 0.80). Peak systolic strain (εs) yielded the maximum AUC of 0.917, accompanied by optimal sensitivity (90.3%) and specificity (86.5%). At its optimal cut-off value, εs acted as the superior diagnostic biomarker for AF relative to all other tested metrics, boosting diagnostic accuracy by 5% versus conventional static volumetric indices. Multivariate logistic regression identified hypertension as an independent risk factor for prevalent AF, consistent with clinical consensus that hypertension represents the most prevalent AF comorbidity and a key driver of atrial remodeling [16, 17]. Chronic hypertension elevates intracardiac atrial pressure, triggering myocardial hypertrophy and interstitial fibrosis, which further propagates atrial electrical and structural remodeling — a core pathological substrate for AF onset. The predictive model integrating hypertension and εs achieved a C-index of 0.655, comparable to the model combining hypertension and minimal left atrial volume index (LAVImin, C-index=0.661). This indicates εs could substitute traditional static structural indices as a valuable metric for predictive AF diagnosis.Cardiac magnetic resonance feature tracking (CMR-FT) holds distinct advantages over conventional speckle-tracking echocardiography (STE) for LA strain quantification [7, 8]. STE suffers from substantial inter- and intra-observer measurement variability and limited accuracy in LA strain assessment, attributed to the thin LA myocardial wall, intricate anatomical geometry, and signal interference originating from the pulmonary veins and left atrial appendage. Benefiting from high spatial resolution and superior soft-tissue contrast, CMR-FT clearly delineates left atrial endocardial and epicardial borders; its dedicated software tracks myocardial motion continuously across the full cardiac cycle, thus yielding strain measurements with excellent accuracy and reproducibility [9–11, 18]. All LA strain metrics in this work were quantified using CVI42 software, and our measurement findings were consistent with previous reports, further validating the reliability of CMR-FT for left atrial strain quantification [11, 14, 18].

Limitations

Single-center retrospective design, relatively limited sample size, no external validation, no direct comparison with echocardiographic strain, only longitudinal strain evaluated, no LGE-fibrosis correlation. The AF2 subgroup (n = 21) is relatively small, which limits the statistical power and robustness of early-detection claims. Enrolled AF patients already exhibited significant structural remodeling and reduced LVEF, which may attenuate the incremental value of LA strain for truly subclinical dysfunction. No formal reproducibility analysis was performed in this study.

The paradoxical inverse association of hypertension (OR < 1) suggests potential coding or modeling issues. High AUC values may be inflated by the healthy control design. Internal validation was not performed, which limits the robustness of multivariable models.

We acknowledge that the use of a healthy control group may have inflated the observed diagnostic performance of LA strain parameters. Patients with structurally and functionally normal left atria represent a stark contrast to those with AF-related remodeling, which may improve apparent AUC values beyond those expected in a clinically representative population with overlapping comorbidities. Accordingly, the discriminative performance reported here should be interpreted with appropriate caution.

Conclusion

CMR-FT can non-invasively and quantitatively assess LA function in AF patients. LA strain is more sensitive than traditional static parameters for identifying established LA dysfunction, although patients had baseline structural and functional impairment. The small AF2 subgroup limits early-detection claims. εs is a promising potential imaging biomarker for assessing LA dysfunction. Further validation in larger, unselected clinical populations is warranted.

Author contributions

Huihui Zhou planned and developed the research project, gathered and analyzed the data, and wrote the manuscript. Dan Zhu gathered and analyzed the information. Shangci Hao and Yongjia Shao provided technical guidance and data interpretation. Nan Xu and Qian Xi revised the manuscript and provided funding support. Every contributor gave their consent for the final draft of the manuscript.

Funding

This work was supported by the Outstanding Leaders Training Initiative of Pudong Health Bureau, Shanghai (Grant No.: PWRl2022-05).

Data availability

Due to patient privacy concerns, the datasets are not publicly accessible. However, they can be requested from the corresponding author if there is a valid reason and the necessary institutional permissions are obtained.

Declarations

Conflict of interest

The writers assert that there are no competing interests.

Ethics Statement

This retrospective study was approved by the Ethics Committee of Shanghai East Hospital and complies with the Declaration of Helsinki (2013 revision). Written informed consent was waived, and all data were de-identified to protect patient privacy.