Development and external validation of a nomogram to predict the need for electrical cardioversion during radiofrequency ablation in patients with atrial fibrillation

Abstract

Objectives

To identify predictors of intraoperative electrical cardioversion and develop a predictive model for patients undergoing radiofrequency ablation for atrial fibrillation (AF).

Methods

We retrospectively analyzed data from 1,348 patients with AF who underwent radiofrequency catheter ablation at Tongji Hospital between January 2018 and December 2023. Clinical, echocardiographic, and CT imaging data were collected. The Boruta algorithm and multivariable logistic regression were used to identify predictors and construct a nomogram. Model performance was assessed using the area under the ROC curve (AUC), calibration plots, and decision curve analysis (DCA). External validation was performed on 121 patients treated at Hubei No. 3 People’s Hospital of Jianghan University from June 2023 to February 2025.

Results

Patients were divided into training and validation sets (7:3 ratio). Five independent predictors were identified: AF type (OR = 13.63), valvular regurgitation (OR = 3.25), BMI (OR = 1.06), left atrial diameter (OR = 1.74), and systolic blood pressure (OR = 0.96). The nomogram showed excellent discriminative ability with AUCs of 0.881 (training), 0.879 (internal validation), and 0.866 (external validation). Calibration curves demonstrated good agreement between predicted and actual outcomes. DCA confirmed the model’s clinical utility.

Conclusions

The proposed nomogram accurately predicts the need for intraoperative electrical cardioversion during AF ablation and may aid in individualized procedural planning.

1. Introduction

Atrial fibrillation (AF) is one of the most common cardiac arrhythmias [1]. Currently, it affects approximately 59.7 million individuals worldwide, and is responsible for an estimated 315,000 deaths annually on a global scale [2]. AF is associated with various complications, including heart failure and stroke, and significantly impairs patients’ quality of life [3,4]. Catheter ablation, particularly pulmonary vein isolation (PVI) using various energy sources such as radiofrequency, cryoablation, or pulsed field ablation (PFA), has emerged as the most effective invasive treatment for AF, especially in patients who are intolerant of or unresponsive to antiarrhythmic medications [5,6]. As such, it has become the preferred therapeutic option in this patient population.

PVI and complex fractionated atrial electrogram (CFAE) ablation are two commonly employed strategies for targeting both the triggers and the arrhythmogenic substrate underlying AF [7,8]. In addition, adjunctive linear ablation—such as lines placed at the left atrial roof and mitral isthmus—may further modify the atrial substrate and improve procedural efficacy. Nevertheless, a substantial proportion of patients with persistent AF still require transthoracic ECV for rhythm restoration, even following extensive catheter ablation [9].

Although ECV is often regarded as a necessary adjunctive step when ablation alone fails to restore sinus rhythm, its clinical implications extend beyond acute rhythm conversion. Previous studies have suggested that intraoperative ECV may be associated with higher postoperative recurrence rates and an increased risk of rehospitalization [10,11]. The need for ECV also reflects the complexity of postoperative management, ultimately affecting long-term prognosis [12]. Therefore, understanding which patients are more likely to require intraoperative ECV is not only of procedural importance but also of prognostic relevance. To date, however, previous prediction models in the field have primarily concentrated on postoperative recurrence or long-term rhythm outcomes, with no validated models specifically developed to assess the likelihood of intraoperative ECV during catheter ablation. This underscores the novelty and clinical value of the present study.

A robust and clinically implementable prediction model is essential for effective risk stratification and informed clinical decision-making, as it enables the identification and tailored management of high-risk patients. However, the majority of recently developed risk models related to atrial fibrillation focus exclusively on predicting postoperative recurrence [13,14]. To date, no validated models are available for assessing the likelihood of intraoperative cardioversion, representing a significant gap in the current clinical toolkit.

The ability to predict the need for intraoperative ECV in advance could have several practical benefits: (1) optimizing patient preparation by informing patients about the likelihood of ECV and improving psychological readiness and informed consent; (2) guiding procedural planning by allowing operators to anticipate difficulties in achieving sinus rhythm with ablation alone and to adjust ablation strategies accordingly; and (3) enhancing peri-procedural resource allocation, such as ensuring anesthesia support and minimizing intraoperative delays. Together, these considerations highlight the strong clinical rationale for developing a prediction tool in this setting.

Therefore, this study aimed to identify risk factors associated with intraoperative electrical cardioversion during radiofrequency catheter ablation in patients with atrial fibrillation and to develop a predictive nomogram model. Importantly, beyond internal validation, we performed external validation using an independent set from a second medical center. This two-center design enhances the model’s generalizability and robustness, thereby improving its applicability in real-world clinical settings.

2. Patients and methods

2.1. Participants

This retrospective, two-center study included consecutive patients with AF who underwent radiofrequency catheter ablation. The development set consisted of patients treated at Tongji Hospital, Huazhong University of Science and Technology, between January 2018 and December 2023. The external validation set was derived from an independent group of AF patients treated at Hubei No. 3 People's Hospital between June 2023 and February 2025. Importantly, as the two sets were collected independently at different centers, the external validation period partially overlapped with and extended beyond the development set period, which was an intentional design to maximize validation sample size and ensure generalizability.

The inclusion and exclusion criteria were consistent across both centers to ensure data uniformity and enhance the model’s applicability. The inclusion criteria were: (1) age ≥ 18 years with a confirmed diagnosis of AF [15]; (2) availability of complete preoperative imaging, laboratory, and electrocardiogram (ECG) data; (3) provision of written informed consent and agreement to undergo standard treatment and follow-up; and (4) indication for radiofrequency catheter ablation according to the 2020 ESC Guidelines for the diagnosis and management of atrial fibrillation (developed in collaboration with EACTS), with the procedure successfully completed [16]. Exclusion criteria were: (1) presence of significant structural heart disease, including severe valvular abnormalities or cardiomyopathies; (2) detection of atrial or left atrial appendage thrombus prior to ablation; (3) history of congenital heart disease; (4) pre-existing persistent bradyarrhythmias or significant sinus node dysfunction; (5) incomplete medical records or missing intraoperative cardioversion data; (6) pregnancy or lactation; and (7) coexisting systemic conditions potentially affecting prognosis, such as active malignancies, severe infections, or end-stage hepatic or renal failure.

The study was approved by the Ethics Committees of Tongji Hospital, Huazhong University of Science and Technology (Approval No. TJ-IRB202404056), and Hubei No. 3 People's Hospital, Jianghan University (Approval No. LW2025011). The requirement for written informed consent for research was waived due to the retrospective design of the study. All surgical procedures were performed after obtaining written informed consent from the patients or their legally authorized representatives. All patient data were strictly anonymized prior to analysis to ensure confidentiality and protect patient privacy, and no identifiable personal information was accessible to the investigators. All procedures were conducted in accordance with the Declaration of Helsinki and relevant national and institutional ethical guidelines [17].

2.2. Data collection and variables

Clinical data were collected through a comprehensive review of electronic medical records, which documented detailed medical information for each patient. The extracted data included demographics (age, sex, and body mass index (BMI)), medical history (including heart failure, hypertension, coronary heart disease (CHD), diabetes mellitus, cerebral infarction or transient ischemic attack (TIA), cardiomyopathy, renal insufficiency, hyperlipidemia, chronic obstructive pulmonary disease (COPD), and obstructive sleep apnea syndrome), type and duration of AF, and history of smoking and alcohol use. Echocardiographic parameters included left atrial volume (LAV), left atrial diameter (LAD), left ventricular end-diastolic diameter (LVEDD), right atrial diameter (RAD), and right ventricular end-diastolic diameter (RVEDD), valvular regurgitation, and left ventricular ejection fraction (LVEF). The CHA₂DS₂-VASc score was also recorded. Laboratory and physiological parameters comprised high-sensitivity C-reactive protein (hs-CRP), systolic and diastolic blood pressure (DBP), pulse rate, alanine aminotransferase (ALT), aspartate aminotransferase (AST), troponin, cardiac troponin I (cTnI), N-terminal pro–B-type natriuretic peptide (NT-proBNP), estimated glomerular filtration rate (eGFR), serum creatinine (Cr), low-density lipoprotein cholesterol (LDL-C), hemoglobin, thyroid-stimulating hormone (TSH), free thyroxine (FT4), free triiodothyronine (FT3), and glycated hemoglobin (HbA1c). Medication records included the use of sodium-glucose co-transporter-2 (SGLT2) inhibitors, statins, β-blockers, amiodarone, propafenone, rivaroxaban, dabigatran, angiotensin-converting enzyme inhibitors (ACEIs), angiotensin II receptor blockers (ARBs), and angiotensin receptor–neprilysin inhibitors (ARNIs).

In this study, regurgitation graded as mild or greater in any of the four valves (mitral, aortic, tricuspid, or pulmonary) was classified as positive, and analyzed as a binary variable (present vs. absent), without distinguishing among valve types, to avoid excessive stratification and loss of statistical power.

Clinical comorbidities were documented at baseline based on patient history, medical records, and supporting laboratory or imaging findings. Standardized definitions were applied in accordance with contemporary clinical guidelines [16,[18], [19], [20], [21], [22], [23], [24]] to ensure consistency and reproducibility in data collection. Detailed definitions of all clinical variables are provided in Supplemental Table 1.

2.3. Radiofrequency catheter ablation procedure

In summary, all procedures were performed using radiofrequency energy. Patients with paroxysmal atrial fibrillation (PAF) typically underwent circumferential PVI alone. In contrast, those with persistent AF (PersAF) received additional linear ablations at the left atrial roof, mitral isthmus, and cavo-tricuspid isthmus [25]. Radiofrequency energy was delivered in power-controlled mode, with settings of 40–50 W on the anterior wall and up to 60 W on selected sites according to operator judgment. On the posterior wall, power was appropriately reduced or application time shortened (5–15 s) to minimize the risk of esophageal injury, in line with a high-power short-duration (HPSD) strategy. The maximum temperature limit was 43 °C, and the irrigation flow rate was 17–30 mL/min. Lesion duration was determined according to local impedance drop and catheter stability. Bidirectional conduction block across all ablation lines was confirmed prior to the completion of each procedure. The procedural objective was to achieve complete PVI in combination with durable bidirectional block of the additional linear lesions [26]. For endpoint verification, entrance and exit block of all pulmonary veins were tested after a 20-minute waiting period, and bidirectional conduction block was confirmed across each linear lesion using differential pacing maneuvers. In cases where atrial arrhythmias persisted following ablation, electrical cardioversion was performed to restore sinus rhythm. Electrical cardioversion was only performed when, in the operator’s judgment, further ablation was unlikely to terminate AF after these steps.

2.4. Statistical analysis

The study adhered to the TRIPOD standards [27] for transparent development and reporting of predictive models (refer to Supplementary TRIPOD Checklist). Statistical analyses were performed using R software version 4.4.2. In this study, the ‘comparegroups’, ‘Boruta’, ‘glm {stats}’, ‘forestmodel’, ‘ggplot’, ‘ggROC’, ‘ResourceSelection’, ‘dcurves’, ‘rmda’, ‘ROCR’ and ‘Dcurves’ packages were employed for data collation and visualization. For missing data management, variables with a missing rate of less than 30 % were included in the analysis, while parameters with more than 30 % missingness—specifically left atrial volume and right atrial diameter—were excluded, as illustrated in Supplementary Fig. 1. Multiple imputation was performed using the mice package to address incomplete data. We generated 5 imputed datasets (m = 5), each with 50 iterations (maxit = 50). The predictor matrix incorporated all available clinical and echocardiographic variables; predictive mean matching (pmm) was applied for continuous variables, while logistic regression and polytomous regression were used for binary and categorical variables, respectively.

For convergence assessment, we plotted and examined the trace plots of means and variances of the imputation chains for each partially observed variable, checking whether multiple chains showed good mixing without systematic trends and fluctuated around a stable level. Convergence was considered adequate when relative changes in chain means/variances were < 1 % during the final 20 iterations. We additionally compared the distributions of observed and imputed values using density and strip plots to evaluate the plausibility of imputations. As a robustness check, we increased the number of iterations to 100 and the number of imputations to 20; changes in key effect estimates and the fraction of missing information (FMI) were both < 0.02, and the study conclusions remained unchanged.

Baseline characteristics were summarized as follows: continuous variables were presented as mean ± standard deviation (SD) for normally distributed data, or as median with interquartile range (IQR) for skewed distributions; categorical variables were reported as frequencies and percentages. Between-group comparisons of continuous variables were conducted using the independent samples Student’s t-test for normally distributed data, or the Wilcoxon rank-sum test for non-normally distributed data. Categorical variables were compared using the chi-square test or Fisher’s exact test, depending on expected cell counts.

The dataset was stratified and split into training and validation sets based on outcome proportions. Feature selection was initially performed using the Boruta algorithm on the training dataset. As a robust wrapper method based on the random forest classifier, Boruta identifies and retains variables with statistically significant importance [28,29]. The selected features were subsequently incorporated into a multivariable logistic regression model using a backward stepwise selection procedure to determine the final set of predictors [30,31]. In multivariate analysis, it is generally recommended to include no more than one variable per 20 outcome events to ensure model reliability and reduce the risk of overfitting. Given that the training set comprises 251 positive outcome events, a maximum of 13 variables can be appropriately included in the model [32].

Model development was conducted using the rms package in R. Model discrimination was assessed using the receiver operating characteristic (ROC) curve, with the area under the curve (AUC) used as the primary performance metric (AUC > 0.70 was considered indicative of good discrimination). In addition, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated to comprehensively evaluate the model’s predictive performance, with detailed results provided in Figs. 4 and 5. Internal validation was performed using 500 bootstrap resamples. Calibration was evaluated via calibration curves, in which predicted probabilities were grouped into deciles and compared against the corresponding observed event rates. The “actual probability” was estimated using bootstrap resampling (500 repetitions) to obtain bias-corrected estimates. This provided both a visual and quantitative assessment of agreement between predicted and observed outcomes across different probability ranges.

Fig. 4.

A. AUC for the training set (n = 944): 0.881 (95 % CI = 0.858–0.903). The sensitivity, specificity, PPV and NPV for training set were 0.803, 0.870, 0.682 and 0.927, respectively. B. AUC for the validation set (n = 404): 0.879 (95 % CI = 0.841–0.916). The sensitivity, specificity, PPV and NPV for validation set were 0.791, 0.876, 0.692 and 0.923, respectively. C and D. Calibration curve of the nomogram model in the training and validation sets. E and F. Decision curve analysis of nomogram for intraoperative electrical cardioversion in the training and validation sets (x-axis represents threshold probability; y-axis indicates net benefit).

Fig. 5.

A. Receiver operating characteristic analysis for the external validation set (n = 121) showed an AUC of 0.866 (95 % CI: 0.794–0.938). Sensitivity, specificity, PPV, and NPV were 0.857, 0.791, 0.625, and 0.932, respectively. B. Calibration curve of the nomogram in the external validation set, indicating good agreement between predicted and observed outcomes. C. Decision curve analysis of the nomogram for predicting the need for intraoperative electrical cardioversion in the external validation set. D. Clinical impact curve of the nomogram in the external validation set, showing the number of patients classified as high risk (red line) and the number of true positives (blue line) across a range of threshold probabilities. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

External validation: The predictive model was applied to an independent external validation set. All selected predictor variables were fully available in this dataset. No recalibration was performed, and the model outputs (predicted probabilities) were interpreted in the same manner as in the training set. Model performance in the external set was assessed in terms of discrimination (ROC curve, AUC), calibration (calibration plots), and clinical utility (decision curve analysis, DCA). DCA was calculated across a range of threshold probabilities to evaluate the net benefit of applying the model in the external set.

To assess clinical utility, DCA was performed by calculating the net benefit across a range of threshold probabilities for the training/internal validation sets and the external validation set. All statistical tests were two-sided, and a P value ≤ 0.05 was considered statistically significant.

3. Results

3.1. Participants enrollment

Fig. 1 illustrates the overall workflow of the present study. A total of 1,348 patients were included, of whom 348 (25.81 %) underwent electrical cardioversion and 1,000 (74.19 %) did not. Significant differences in clinical characteristics were observed between the two groups (Table 1).

Fig. 1.

Workflow diagram: The dataset from Tongji Hospital was partitioned into a training set and an internal validation set in a 7:3 ratio according to the distribution of clinical outcomes. An independent dataset from Hubei No. 3 People's Hospital served as the external validation set. Feature selection was conducted using only the training data.

Table 1.

Baseline clinical and demographic characteristics by cardioversion outcomes.

Variables	Total (n = 1348)	No ECV (n = 1000)	ECV performed (n = 348)	P
Age (years)	62.58 ± 10.77	62.48 ± 11.02	62.86 ± 10.02	0.559
BMI (kg/m2)	24.41 (22.36, 26.26)	24.39 (22.27, 26.12)	24.61 (22.49, 26.44)	0.005
CHA2DS2-VASC	2.00 (1.00, 3.00)	2.00 (1.00, 3.00)	2.00 (1.00, 3.00)	0.180
CHADS-VA	1.00 (1.00, 3.00)	1.00 (1.00, 3.00)	1.00 (1.00, 3.00)	0.003
Course of AF (months)	12.00 (3.00, 36.00)	12.00 (2.00, 36.00)	24.00 (6.00, 60.00)	<0.001
Time of surgery (hours)	2.16 (2.00, 2.67)	2.10 (2.00, 2.50)	2.29 (2.00, 3.00)	<0.001
LAV (mL)	122.82 (122.82, 122.82)	122.82 (116.78, 122.82)	122.82 (122.82, 155.22)	<0.001
LAD (cm)	4.20 (3.80, 4.70)	4.15 (3.70, 4.60)	4.60 (4.20, 5.00)	<0.001
CRP (mg/L)	1.00 (0.50, 2.00)	1.00 (0.60, 2.00)	1.00 (0.40, 1.92)	0.060
SBP (mmHg)	129.00 (117.00, 140.00)	132.00 (119.00, 142.00)	122.00 (112.00, 132.00)	<0.001
DBP (mmHg)	82.00 (75.00, 88.00)	83.00 (76.00, 89.00)	82.00 (73.00, 88.00)	0.058
Pulse (beats/min)	75.00 (67.00, 88.00)	75.00 (67.00, 88.00)	79.00 (70.00, 92.00)	0.002
ALT (IU/L)	19.00 (15.00, 28.00)	19.00 (14.00, 27.00)	20.00 (16.00, 30.00)	0.009
AST (IU/L)	21.00 (18.00, 25.00)	21.00 (17.00, 25.00)	21.50 (18.00, 26.00)	0.138
CtnI (ng/mL)	3.40 (2.60, 5.70)	3.40 (2.70, 6.10)	3.40 (2.00, 3.70)	<0.001
NT-proBNP (pg/mL)	282.00 (139.00, 584.00)	282.00 (100.00, 336.00)	573.00 (305.00, 921.00)	<0.001
eGFR (mL/min/1.73 m2)	88.60 (73.80, 95.22)	88.60 (74.33, 94.90)	88.60 (72.50, 95.60)	0.926
Cr (mg/dl)	76.00 (66.00, 90.00)	76.00 (64.00, 89.00)	76.00 (70.00, 92.00)	0.001
LDL (mmol/L)	2.28 (2.03, 2.59)	2.28 (2.04, 2.52)	2.28 (2.03, 2.75)	0.081
Hb (g/L)	136.00 (123.00, 148.00)	133.00 (123.00, 147.00)	141.00 (128.75, 151.00)	<0.001
TSH (μIU/mL)	2.03 (1.41, 2.86)	2.03 (1.48, 3.03)	2.03 (1.38, 2.36)	0.063
FT4 (pmol/L)	1.20 (1.02, 15.80)	1.20 (1.02, 15.30)	1.20 (1.03, 16.90)	0.009
FT3 (pmol/L)	2.85 (2.52, 4.44)	2.85 (2.52, 4.44)	2.85 (2.57, 4.49)	0.067
HbA1c	5.70 (5.50, 6.00)	5.70 (5.40, 5.80)	5.80 (5.70, 6.30)	<0.001
LV (cm)	4.80 (4.50, 5.20)	4.80 (4.50, 5.20)	4.90 (4.60, 5.30)	0.003
RA (cm)	4.30 (4.30, 4.30)	4.30 (4.30, 4.30)	4.30 (4.10, 4.40)	0.122
EF (%)	61.00 (56.00, 65.00)	62.00 (57.00, 65.25)	59.00 (50.00, 63.00)	<0.001
Sex, n(%)				<0.001
Female	574 (42.58)	465 (46.50)	109 (31.32)
Male	774 (57.42)	535 (53.50)	239 (68.68)
HF, n(%)				0.148
No	1133 (84.05)	832 (83.20)	301 (86.49)
Yes	215 (15.95)	168 (16.80)	47 (13.51)
Hypertension, n(%)				0.039
No	711 (52.74)	544 (54.40)	167 (47.99)
Yes	637 (47.26)	456 (45.60)	181 (52.01)
Diabetes, n(%)				<0.001
No	1116 (82.79)	851 (85.10)	265 (76.15)
Yes	232 (17.21)	149 (14.90)	83 (23.85)
Cerebral infarction/TIA, n(%)				0.007
No	1141 (84.64)	862 (86.20)	279 (80.17)
Yes	207 (15.36)	138 (13.80)	69 (19.83)
CHD, n(%)				0.901
No	968 (71.81)	719 (71.90)	249 (71.55)
Yes	380 (28.19)	281 (28.10)	99 (28.45)
Type of AF, n(%)				<0.001
Paroxysmal	933 (69.21)	845 (84.50)	88 (25.29)
Non-paroxysmal	415 (30.79)	155 (15.50)	260 (74.71)
AF duration, n(%)				0.047
＜6 years	1105 (81.97)	832 (83.20)	273 (78.45)
≥6 years	243 (18.03)	168 (16.80)	75 (21.55)
Cardiomyopathy, n(%)				0.590
No	1198 (88.87)	886 (88.60)	312 (89.66)
Yes	150 (11.13)	114 (11.40)	36 (10.34)
Sleep apnea syndrome, n(%)				0.095
No	1268 (94.07)	947 (94.70)	321 (92.24)
Yes	80 (5.93)	53 (5.30)	27 (7.76)
Renal insufficiency, n(%)				0.006
No	1249 (92.66)	938 (93.80)	311 (89.37)
Yes	99 (7.34)	62 (6.20)	37 (10.63)
Hyperlipidemia, n(%)				0.167
No	1114 (82.64)	818 (81.80)	296 (85.06)
Yes	234 (17.36)	182 (18.20)	52 (14.94)
Valvular regurgitation, n(%)				<0.001
No	1176 (87.24)	901 (90.10)	275 (79.02)
Yes	172 (12.76)	99 (9.90)	73 (20.98)
Smoking history, n(%)				0.051
No	656 (48.66)	471 (47.10)	185 (53.16)
Yes	692 (51.34)	529 (52.90)	163 (46.84)
Drinking history, n(%)				0.487
No	1139 (84.50)	849 (84.90)	290 (83.33)
Yes	209 (15.50)	151 (15.10)	58 (16.67)
SGLT2i, n(%)				<0.001
No	1320 (97.92)	990 (99.00)	330 (94.83)
Yes	28 (2.08)	10 (1.00)	18 (5.17)
Statin, n(%)				0.933
No	696 (51.63)	517 (51.70)	179 (51.44)
Yes	652 (48.37)	483 (48.30)	169 (48.56)
Beta-blocker, n(%)				<0.001
No	782 (58.01)	621 (62.10)	161 (46.26)
Yes	566 (41.99)	379 (37.90)	187 (53.74)
Amiodarone, n(%)				0.002
No	97 (7.20)	85 (8.50)	12 (3.45)
Yes	1251 (92.80)	915 (91.50)	336 (96.55)
Propafenone, n(%)				0.902
No	1175 (87.17)	871 (87.10)	304 (87.36)
Yes	173 (12.83)	129 (12.90)	44 (12.64)
Rivaroxaban, n(%)				0.282
No	773 (57.34)	582 (58.20)	191 (54.89)
Yes	575 (42.66)	418 (41.80)	157 (45.11)
Dabigatran, n(%)				0.049
No	819 (60.76)	623 (62.30)	196 (56.32)
Yes	529 (39.24)	377 (37.70)	152 (43.68)
ACEI/ARB/ARNI, n(%)				0.003
No	684 (50.74)	531 (53.10)	153 (43.97)
Yes	664 (49.26)	469 (46.90)	195 (56.03)

BMI: body mass index, LAV: left atrial volume, LAD: left atrial diameter, CRP: C-reactive protein, SBP: systolic blood pressure, DBP: diastolic blood pressure, ALT: alanine aminotransferase, AST: aspartate aminotransferase, CtnI: cardiac troponin I, NT-proBNP: N-terminal pro-B-type natriuretic peptide, eGFR: estimated glomerular filtration rate, Cr: creatinine, LDL: low-density lipoprotein cholesterol, Hb: hemoglobin, TSH: thyroid-stimulating hormone, FT4: free thyroxine, FT3: free triiodothyronine, HbA1c: glycated hemoglobin, LV: left ventricular diameter, RA: right atrial diameter, EF: ejection fraction,HF: heart failure, TIA: transient ischemic attack, CHD: coronary heart disease, SGLT2i: sodium-glucose co-transporter 2 inhibitor use, ACEI/ARB/ARNI: angiotensin-converting enzyme inhibitor/angiotensin receptor blocker/angiotensin receptor neprilysin inhibitor use.

To facilitate model development and evaluation, the dataset was randomly divided into a training set and a validation set in a 7:3 ratio according to the proportion of outcome events. As a result, the training set comprised 944 patients, including 251 who underwent electrical cardioversion and 693 who did not, while the validation set comprised 404 patients, including 97 who underwent electrical cardioversion and 307 who did not. The detailed distribution is presented in Fig. 1 and Table 2.

Table 2.

Training set and Validation set variability analysis.

Variables	Total (n = 1348)	Training set (n = 944)	Validation set (n = 404)	P
Age (years)	64.00 (57.00, 70.00)	64.00 (57.00, 70.00)	65.00 (57.00, 70.00)	0.300
BMI (kg/m2)	24.41 (22.36, 26.26)	24.41 (22.36, 26.14)	24.50 (22.36, 26.40)	0.570
CHA2DS2-VASC	2.00 (1.00, 3.00)	2.00 (1.00, 3.00)	2.00 (1.00, 3.00)	0.721
CHADS-VA	1.00 (1.00, 3.00)	1.00 (1.00, 3.00)	2.00 (1.00, 3.00)	0.631
Course of AF (months)	12.00 (3.00, 36.00)	12.00 (3.00, 36.00)	12.00 (2.00, 36.00)	0.259
Time of surgery (hours)	2.16 (2.00, 2.67)	2.16 (2.00, 2.60)	2.20 (2.00, 2.67)	0.057
LAD (cm)	4.20 (3.80, 4.70)	4.20 (3.80, 4.70)	4.20 (3.80, 4.70)	0.933
CRP (mg/L)	1.00 (0.50, 2.00)	1.00 (0.50, 1.83)	1.00 (0.60, 2.30)	0.006
SBP (mmHg)	129.00 (117.00, 140.00)	129.00 (117.00, 140.00)	128.00 (116.00, 140.00)	0.579
DBP (mmHg)	82.00 (75.00, 88.00)	83.00 (75.00, 88.00)	82.00 (75.00, 88.00)	0.683
Pulse (beats/min)	75.00 (67.00, 88.00)	75.00 (67.00, 87.00)	76.00 (67.00, 88.00)	0.535
ALT (IU/L)	19.00 (15.00, 28.00)	19.00 (15.00, 28.00)	20.00 (15.00, 27.00)	0.617
AST (IU/L)	21.00 (18.00, 25.00)	21.00 (18.00, 25.00)	21.00 (18.00, 26.00)	0.385
CtnI (ng/mL)	3.40 (2.60, 5.70)	3.40 (2.40, 5.60)	3.40 (2.70, 6.03)	0.169
NT-proBNP (pg/mL)	282.00 (139.00, 584.00)	282.00 (139.00, 573.00)	282.00 (143.00, 610.00)	0.812
eGFR (mL/min/1.73 m2)	88.60 (73.80, 95.22)	88.60 (74.33, 95.60)	88.60 (73.55, 94.90)	0.573
Cr (mg/dl)	76.00 (66.00, 90.00)	76.00 (66.00, 90.00)	77.00 (66.00, 90.00)	0.946
LDL (mmol/L)	2.28 (2.03, 2.59)	2.28 (2.03, 2.69)	2.28 (2.04, 2.51)	0.342
Hb (g/L)	136.00 (123.00, 148.00)	136.00 (123.00, 148.00)	135.00 (123.00, 149.00)	0.831
TSH (μIU/mL)	2.03 (1.41, 2.86)	2.03 (1.42, 2.86)	2.03 (1.39, 2.75)	0.714
FT4 (pmol/L)	1.20 (1.02, 15.80)	1.20 (1.02, 15.80)	1.20 (1.04, 15.80)	0.561
FT3 (pmol/L)	2.85 (2.52, 4.44)	2.85 (2.52, 4.49)	2.85 (2.52, 4.42)	0.629
HbA1c	5.70 (5.50, 6.00)	5.70 (5.50, 6.00)	5.70 (5.50, 6.00)	0.339
LV (cm)	4.80 (4.50, 5.20)	4.80 (4.50, 5.20)	4.90 (4.60, 5.20)	0.123
RA (cm)	4.30 (4.30, 4.30)	4.30 (4.30, 4.30)	4.30 (4.30, 4.30)	0.509
EF (%)	61.00 (56.00, 65.00)	61.00 (56.00, 65.00)	61.00 (56.00, 65.00)	0.511
Sex, n(%)				0.813
Female	574 (42.58)	400 (42.37)	174 (43.07)
Male	774 (57.42)	544 (57.63)	230 (56.93)
HF, n(%)				0.459
No	1133 (84.05)	798 (84.53)	335 (82.92)
Yes	215 (15.95)	146 (15.47)	69 (17.08)
Hypertension, n(%)				0.559
No	711 (52.74)	493 (52.22)	218 (53.96)
Yes	637 (47.26)	451 (47.78)	186 (46.04)
Diabetes, n(%)				0.933
No	1116 (82.79)	781 (82.73)	335 (82.92)
Yes	232 (17.21)	163 (17.27)	69 (17.08)
Cerebral infarction/TIA, n(%)				0.185
No	1141 (84.64)	791 (83.79)	350 (86.63)
Yes	207 (15.36)	153 (16.21)	54 (13.37)
CHD, n(%)				0.001
No	968 (71.81)	702 (74.36)	266 (65.84)
Yes	380 (28.19)	242 (25.64)	138 (34.16)
Type of AF, n(%)				0.227
Paroxysmal	933 (69.21)	644 (68.22)	289 (71.53)
Non-paroxysmal	415 (30.79)	300 (31.78)	115 (28.47)
AF duration, n(%)				0.455
＜6 years	1105 (81.97)	769 (81.46)	336 (83.17)
≥6 years	243 (18.03)	175 (18.54)	68 (16.83)
Cardiomyopathy, n(%)				0.087
No	1198 (88.87)	848 (89.83)	350 (86.63)
Yes	150 (11.13)	96 (10.17)	54 (13.37)
Sleep apnea syndrome, n(%)				0.447
No	1268 (94.07)	891 (94.39)	377 (93.32)
Yes	80 (5.93)	53 (5.61)	27 (6.68)
Renal insufficiency, n(%)				0.448
No	1249 (92.66)	878 (93.01)	371 (91.83)
Yes	99 (7.34)	66 (6.99)	33 (8.17)
Hyperlipidemia, n(%)				0.652
No	1114 (82.64)	783 (82.94)	331 (81.93)
Yes	234 (17.36)	161 (17.06)	73 (18.07)
Valvular regurgitation, n(%)				0.128
No	1176 (87.24)	815 (86.33)	361 (89.36)
Yes	172 (12.76)	129 (13.67)	43 (10.64)
Smoking history, n(%)				0.167
No	656 (48.66)	471 (49.89)	185 (45.79)
Yes	692 (51.34)	473 (50.11)	219 (54.21)
Drinking history, n(%)				0.788
No	1139 (84.50)	796 (84.32)	343 (84.90)
Yes	209 (15.50)	148 (15.68)	61 (15.10)
SGLT2i, n(%)				0.870
No	1320 (97.92)	924 (97.88)	396 (98.02)
Yes	28 (2.08)	20 (2.12)	8 (1.98)
Statin, n(%)				0.758
No	696 (51.63)	490 (51.91)	206 (50.99)
Yes	652 (48.37)	454 (48.09)	198 (49.01)
Beta-blocker, n(%)				0.027
No	782 (58.01)	566 (59.96)	216 (53.47)
Yes	566 (41.99)	378 (40.04)	188 (46.53)
Amiodarone, n(%)				0.243
No	97 (7.20)	73 (7.73)	24 (5.94)
Yes	1251 (92.80)	871 (92.27)	380 (94.06)
Propafenone, n(%)				0.494
No	1175 (87.17)	819 (86.76)	356 (88.12)
Yes	173 (12.83)	125 (13.24)	48 (11.88)
Rivaroxaban, n(%)				0.078
No	773 (57.34)	556 (58.90)	217 (53.71)
Yes	575 (42.66)	388 (41.10)	187 (46.29)
Dabigatran, n(%)				0.199
No	819 (60.76)	563 (59.64)	256 (63.37)
Yes	529 (39.24)	381 (40.36)	148 (36.63)
ACEI/ARB/ARNI, n(%)				0.154
No	684 (50.74)	491 (52.01)	193 (47.77)
Yes	664 (49.26)	453 (47.99)	211 (52.23)

BMI: body mass index, LAV: left atrial volume, LAD: left atrial diameter, CRP: C-reactive protein, SBP: systolic blood pressure, DBP: diastolic blood pressure, ALT: alanine aminotransferase, AST: aspartate aminotransferase, CtnI: cardiac troponin I, NT-proBNP: N-terminal pro-B-type natriuretic peptide, eGFR: estimated glomerular filtration rate, Cr: creatinine, LDL: low-density lipoprotein cholesterol, Hb: hemoglobin, TSH: thyroid-stimulating hormone, FT4: free thyroxine, FT3: free triiodothyronine, HbA1c: glycated hemoglobin, LV: left ventricular diameter, RA: right atrial diameter, EF: ejection fraction, HF: heart failure, TIA: transient ischemic attack, CHD: coronary heart disease, SGLT2i: sodium-glucose co-transporter 2 inhibitor use, ACEI/ARB/ARNI: angiotensin-converting enzyme inhibitor/angiotensin receptor blocker/angiotensin receptor neprilysin inhibitor use.

Compared with the non-cardioversion group, patients in the cardioversion group had a higher BMI (24.61 vs. 24.39 kg/m², P = 0.005), longer AF duration (24.00 vs. 12.00 months, P < 0.001), and longer procedure time (2.29 vs. 2.10 h, P < 0.001). Additionally, the cardioversion group demonstrated significantly larger LAD (4.60 vs. 4.15 cm, P < 0.001), lower systolic blood pressure (122.00 vs. 132.00 mmHg, P < 0.001), and higher resting pulse rate (79.00 vs. 75.00 bpm, P = 0.002). Clinically, the elevated heart rate in the cardioversion group may suggest a longer duration of AF or inadequate ventricular rate control, both of which are associated with atrial electrical instability and a reduced likelihood of spontaneous sinus rhythm restoration after ablation. This could partly explain the higher requirement for cardioversion in this group. Furthermore, elevated ALT levels (20.00 vs. 19.00 IU/L, P = 0.009) and higher NT-proBNP concentrations (573.00 vs. 282.00 pg/mL, P < 0.001) were observed in the cardioversion group.

Additionally, patients in the cardioversion group exhibited lower cTnI levels (median 3.40 ng/mL with a narrower interquartile range, P < 0.001), higher hemoglobin concentrations (141.00 vs. 133.00 g/L, P < 0.001), and slightly elevated FT4 levels (P = 0.009). HbA1c (5.80 vs. 5.70 %, P < 0.001), LVEDD (4.90 vs. 4.80 cm, P = 0.003), and serum creatinine (76.00 vs. 76.00 μmol/L, with a higher upper range, P = 0.001) were also significantly higher. In contrast, LVEF was significantly lower in the cardioversion group (59.00 % vs. 62.00 %, P < 0.001).

Regarding categorical variables, the cardioversion group had a higher proportion of males (68.68 % vs. 53.50 %, P < 0.001), and a greater prevalence of diabetes mellitus (23.85 % vs. 14.90 %, P < 0.001), hypertension (52.01 % vs. 45.60 %, P = 0.039), cerebral infarction or TIA (19.83 % vs. 13.80 %, P = 0.007), and renal insufficiency (10.63 % vs. 6.20 %, P = 0.006). Non-paroxysmal atrial fibrillation (non-PAF) was significantly more common (74.71 % vs. 15.50 %, P < 0.001), as was an AF duration ≥ 6 years (21.55 % vs. 16.80 %, P = 0.047) and valvular insufficiency (20.98 % vs. 9.90 %, P < 0.001).

In terms of medication use, patients in the cardioversion group were more frequently treated with SGLT2 inhibitors (5.17 % vs. 1.00 %, P < 0.001), β-blockers (53.74 % vs. 37.90 %, P < 0.001), amiodarone (96.55 % vs. 91.50 %, P = 0.002), dabigatran (43.68 % vs. 37.70 %, P = 0.049), and ACEIs/ARBs/ARNIs (56.03 % vs. 46.90 %, P = 0.003).

No significant differences were observed between the two groups in age, CHA₂DS₂-VASc score, eGFR, LDL-C, AST, DBP, hs-CRP, smoking and alcohol history, or the use of statins and rivaroxaban.

Significant differences were identified in hs-CRP, CHD, and β-blocker use between the training and internal validation sets (P < 0.05). However, no significant differences were observed in the remaining variables (P > 0.05), indicating a generally balanced dataset and comparable sets (Table 2). Similarly, comparisons between the training and external validation sets showed no significant differences in the majority of variables (P > 0.05), as presented in Supplementary Table 2.

3.2. Variables selection

A total of 44 independent variables were included in this study (see Fig. 2). The dependent variable was defined as the use of electrical cardioversion during radiofrequency catheter ablation (RFCA) in patients with AF. Based on the occurrence of outcome events, patients were randomly assigned to a training set and a validation set in a 7:3 ratio. Feature selection was performed on the training set using the Boruta algorithm, a robust and widely recognized method built on the random forest classifier [28,29]. This algorithm identifies and retains variables that demonstrate significantly greater importance than their randomized counterparts (“shadow features”), thereby ensuring relevance to the predictive model.

Fig. 2.

Feature selection based on Boruta algorithm. Green represents acceptable variables. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Through Boruta, eight important variables were identified: LVEF, Cr, BMI, SBP, valvular insufficiency, LAD, NT-proBNP, and AF type (paroxysmal or non-paroxysmal) (Fig. 2).

These variables were then entered into a multivariable logistic regression model using a backward stepwise selection strategy based on the least Akaike Information Criterion (AIC). Five independent predictors of electrical cardioversion during RFCA were ultimately retained (Table 3): AF type (P < 0.001), valvular insufficiency (P < 0.001), BMI (P = 0.050), LAD (P < 0.001), and SBP (P < 0.001). Each was significantly associated with the likelihood of requiring electrical cardioversion during RFCA.

Table 3.

Multivariate analysis on variables for predicting the patient's risk of ECV.

Variable	β	Odds ratio (95 %CI)	P value
Type of AF	2.612	13.62 (9.202–20.52)	＜0.001
Valvular regurgitation	1.179	3.250 (2.006–5.290)	＜0.001
BMI	0.058	1.059 (1.000–1.123)	0.05
LAD	0.557	1.744 (1.292–2.370)	＜0.001
Systolic blood pressure	−0.039	0.961 (0.949–0.974)	＜0.001

To assess multicollinearity, the variance inflation factor (VIF) was calculated for all variables [33]. All VIF values were well below the conventional threshold of 5, indicating low collinearity and supporting the model’s robustness (Supplementary Table 3).

3.3. Construction and validation of nomogram

Based on the results of multivariable logistic regression, we constructed a nomogram incorporating five independent predictors: AF type, valvular insufficiency, BMI, LAD, and SBP, as shown in Fig. 3. A web-based dynamic nomogram was published on shinyapp (See Supplementary Fig. 4), which can be used through the link (https://af-ecv.shinyapps.io/dynnomapp/).

Fig. 3.

Nomogram for predicting intra-procedural electrical cardioversion in AF patients undergoing radiofrequency ablation. Type of AF: 0: Paroxysmal AF; 1: Persistent AF/ Long-standing persistent AF/ Permanent AF; Valvular regurgitation: 0: No; 1: Yes.

The nomogram exhibited excellent discriminative performance in the training set, with an area under the receiver operating characteristic curve (AUC) of 0.881 (95 % CI: 0.858–0.903), a sensitivity of 0.803, and a specificity of 0.870. In the internal validation set, the model maintained high performance, yielding an AUC of 0.879 (95 % CI: 0.841–0.916), with a sensitivity of 0.791 and specificity of 0.876 (Fig. 4A, B). To further confirm robustness, a 5-fold cross-validation was conducted within the training dataset. The ROC curves for each fold demonstrated consistently high discriminative ability, with mean AUC = 0.867 (95 % CI: 0.843–0.892), indicating stable performance across different resampling partitions (Supplementary Fig. 3).

Calibration plots based on 500 bootstrap resamples demonstrated good agreement between predicted probabilities and observed outcomes in both the training and validation sets (Fig. 4C, D). The Hosmer–Lemeshow test indicated no significant lack of fit (P = 0.08 for the training set and P = 0.06 for the validation set), supporting the model's calibration.

Decision curve analysis (DCA) further confirmed the clinical utility of the model, showing a favorable net benefit across a wide range of threshold probabilities (2 %–68 %) compared to both the “treat-all” and “treat-none” strategies (Fig. 4E, F). In this context, the threshold probability represents the risk level at which a clinician would consider intervening (e.g., performing cardioversion), while the net benefit reflects the balance between true positives and false positives across different thresholds. This framework facilitates evaluation of the model’s value in clinical decision-making.

In the external validation set, the nomogram demonstrated favorable performance, achieving an AUC of 0.866 (95 % CI: 0.794–0.938), with a sensitivity of 0.857 and specificity of 0.791 (Fig. 5A). Calibration analysis showed good agreement between predicted and observed outcomes, with a calibration intercept of − 0.00 (95 % CI: −0.49 to 0.49) and slope of 1.00 (95 % CI: 0.62 to 1.38) (Fig. 5B). Consistent with the internal results, DCA in the external validation set also suggested a net clinical benefit over a wide threshold range (Fig. 5C). Furthermore, the clinical impact curve demonstrated that the nomogram identified a substantial number of patients at high risk who truly experienced the outcome, supporting its potential clinical utility (Fig. 5D). Nevertheless, these findings should be interpreted with caution due to the limited number of events in the external validation cohort.

4. Discussion

In this study, we developed and externally validated a nomogram to predict the need for intraoperative ECV during radiofrequency ablation (RFA) in patients with AF. Our model incorporated five readily available clinical and echocardiographic predictors—AF type, valvular regurgitation, BMI, LAD, and systolic blood pressure—all of which were independently associated with the likelihood of requiring ECV during ablation. The nomogram demonstrated excellent discrimination and calibration in both the internal and external validation sets, suggesting strong generalizability and potential clinical applicability.

AF type emerged as the strongest predictor of intraoperative ECV. Patients with persistent AF were significantly more likely to require ECV compared to those with paroxysmal AF. This finding is consistent with electrophysiological evidence showing that persistent AF is associated with more extensive atrial remodeling, including fibrosis, conduction heterogeneity, and electrical refractoriness [34]. These changes reduce the likelihood of spontaneous termination of AF during ablation and necessitate additional interventions, such as ECV, to restore sinus rhythm [35]. Thus, the inclusion of AF type in the nomogram allows for early identification of patients with a more resistant arrhythmic substrate.

Valvular regurgitation, particularly involving the mitral and tricuspid valves, was another significant predictor. Regurgitant flow leads to volume overload and pressure changes in the left and right atria, promoting atrial dilation and structural remodelling [36,37]. These alterations create a pro-arrhythmic milieu, increasing the likelihood that AF will persist despite ablation. Intraoperatively, the presence of valvular disease may correlate with delayed termination of AF and a greater need for ECV. This association also underscores the importance of comprehensive echocardiographic assessment prior to ablation to evaluate for structural heart disease.

BMI, while a borderline predictor in our analysis (p = 0.05), adds meaningful context. Obesity has been widely linked to AF initiation and maintenance through mechanisms such as increased epicardial fat, systemic inflammation, and autonomic imbalance [38]. Elevated BMI is closely associated with atrial structural and electrical remodeling, which reduces the likelihood of spontaneous AF termination and increases the probability of requiring intraoperative ECV. In the procedural setting, higher BMI may contribute to suboptimal catheter contact or longer procedure duration, both of which can influence the effectiveness of ablation and rhythm conversion [39]. Furthermore, elevated BMI is frequently associated with comorbidities such as hypertension and obstructive sleep apnea, compounding the risk for AF persistence and resistance to ablation alone [40].

LAD is a well-established indicator of atrial remodelling [41]. In our study, a larger atrial diameter was strongly associated with the need for ECV. A dilated left atrium reflects chronic atrial stretch and structural degeneration, including myocyte hypertrophy and fibrosis, which impair conduction and promote reentry circuits [39]. These pathological changes diminish the chances of spontaneous AF termination, even after extensive ablation. Given its predictive power, left atrial size remains a cornerstone variable in procedural planning and outcome forecasting in AF management.

SBP was identified as an independent protective factor against intraoperative ECV in patients undergoing radiofrequency catheter ablation (OR = 0.96, 95 % CI: 0.95–0.97, p < 0.001). Lower pre-procedural SBP may indicate impaired cardiac reserve or subclinical heart failure, reflecting compromised atrial–ventricular electromechanical coupling and hemodynamic fragility. From a hemodynamic perspective, higher SBP before the procedure may reflect relatively preserved myocardial contractility and adequate stroke volume, which help maintain effective forward flow and stable atrial–ventricular electromechanical coupling. In contrast, patients with relatively higher SBP are more likely to have preserved left ventricular systolic function and adequate cardiac output, favoring spontaneous rhythm conversion without reliance on ECV. Therefore, pre-procedural SBP may serve as a surrogate marker of overall cardiac performance and provide useful prognostic information for procedural planning.

Our findings have important clinical implications. The ability to predict the need for intraoperative ECV may help electrophysiologists anticipate procedural challenges and tailor perioperative management strategies. For instance, patients with high nomogram scores might benefit from earlier or more aggressive rhythm control strategies, or closer post-procedural monitoring.

Preprocedural SBP was identified as an independent protective factor against the need for intraoperative electrical cardioversion (OR = 0.96, 95 % CI: 0.95–0.97, p < 0.001). The underlying mechanisms remain uncertain, and this association should not be interpreted as SBP being a direct marker of cardiac function. SBP is influenced by multiple factors, including vascular tone, neurohormonal activity, and autonomic regulation. Lower SBP may reflect impaired cardiovascular reserve or comorbid conditions predisposing to rhythm instability, whereas higher SBP could indicate more favorable baseline physiology. Importantly, this interpretation is hypothesis-generating, and further studies are warranted to clarify the causal pathways.

Compared to prior predictive tools focusing on post-ablation outcomes, our model is novel in its specific focus on the intraoperative requirement for ECV—a critical yet understudied aspect of AF ablation. The use of robust statistical methods including the Boruta algorithm for variable selection, as well as external validation in an independent set, adds methodological strength to our work.

In positioning our study within the broader literature, it is important to note that most prior research has focused on predicting either external cardioversion outcomes or long-term recurrence following AF ablation. For example, Antoun et al. [42] systematically reviewed predictors of external direct current cardioversion (DCCV) success, highlighting demographic, biochemical, imaging, and ECG-based factors as relevant determinants. Tourni et al. [43] applied electromechanical cycle length mapping (ECLM), a novel echocardiographic technique, to predict short- and mid-term outcomes of DCCV, demonstrating the value of functional imaging markers. Roney et al. [44] combined population-level data with patient-specific computational atrial models to improve prediction of AF recurrence after ablation, underscoring the role of advanced simulations and machine learning in this field. While these studies differ in outcome focus and methodological complexity, they collectively reinforce the importance of integrating structural, electrical, and functional parameters into prediction frameworks. By contrast, our study uniquely targets the intraoperative need for ECV—an outcome not directly addressed in prior literature—using readily available preprocedural clinical and echocardiographic variables. Future work may explore whether imaging-based or computational biomarkers, as highlighted in these studies, could further enhance the predictive accuracy and clinical utility of intraoperative ECV models.

Nevertheless, several limitations should be acknowledged. First, the retrospective design may introduce selection bias. Although we attempted to mitigate this through multicenter data and randomization, potential sources of bias remain, including differences in referral patterns between centers, exclusion of patients with incomplete clinical or follow-up data, and variability among operators in the timing and indications for electrical cardioversion. These factors may influence baseline characteristics and procedural strategies, thereby limiting the representativeness of our cohort. Second, although external validation was performed, the generalizability of our model to non-Chinese populations or centers with different ablation protocols remains to be tested. In particular, this study was conducted exclusively in the context of RFA. The findings cannot be directly extrapolated to PFA, where the procedural workflow is substantially different. This distinction should be considered when interpreting the clinical applicability of our model. Third, certain factors such as atrial fibrosis (e.g., via MRI or voltage mapping) were not included due to limited availability, and may enhance predictive power if incorporated in future models. In addition, details regarding the type and severity of valvular insufficiency were not systematically recorded. Most cases involved mitral regurgitation, but a precise stratification between mitral, tricuspid, and aortic insufficiency was not feasible in this retrospective dataset, which limits the interpretability of this factor. Fourth, our model did not incorporate detailed ECG parameters beyond the basic clinical classification of AF. Recent studies have demonstrated that non-invasive surface ECG analysis—including metrics such as P-wave duration and fibrillatory wave characteristics derived from signal processing techniques (e.g., frequency domain analysis)—can provide valuable information on atrial electrical remodeling, organization, and substrate complexity, and may improve prediction of ablation outcomes. The absence of these ECG markers represents a limitation of the current model, and incorporating them in future studies could further enhance predictive performance and clinical applicability. Fifth, procedural heterogeneity is another limitation. Patients with persistent AF routinely underwent additional linear ablations beyond PVI, whereas most patients with paroxysmal AF received PVI only. Importantly, for paroxysmal AF patients who did not achieve sinus rhythm after PVI, additional lesion ablations—including at the left atrial roof, mitral isthmus, and tricuspid isthmus—were performed prior to electrical cardioversion. This variability in ablation extent may influence the consistency and interpretability of the model’s outputs. Acknowledging this procedural heterogeneity improves methodological transparency and contextualizes the applicability of our model in clinical settings with more uniform ablation strategies.

In conclusion, we have developed a clinically relevant and statistically robust nomogram for predicting intraoperative ECV during RFA in AF patients. This tool has the potential to enhance personalized procedural planning and risk assessment, though further prospective validation and integration with imaging or electrophysiological biomarkers are warranted.

5. Conclusions

Our study constructed and validated a nomogram integrating AF type, valvular regurgitation, BMI, LAD, and systolic blood pressure to predict the likelihood of intraoperative electrical cardioversion in patients with AF undergoing radiofrequency ablation. The nomogram demonstrated favorable discrimination, calibration, and clinical applicability, which may facilitate individualized risk assessment and optimize perioperative management strategies in clinical practice.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to patient privacy and institutional restrictions, but de-identified data are available from the corresponding author on reasonable request. Data sharing will require approval by the institutional ethics committee, and access will be granted to qualified researchers for academic purposes only, in compliance with applicable regulations.

CRediT authorship contribution statement

Yahui Li: Writing – original draft, Software, Formal analysis, Data curation, Conceptualization. Xindi Yue: Data curation, Conceptualization. Yidan Chen: Data curation, Conceptualization. Xuhui Liu: Writing – original draft, Data curation. Xujie Wang: Resources, Data curation. Ru Sun: Data curation. Haojiang Li: Data curation. Qingqing Li: Data curation. Nianfang Luo: Data curation. Feng Wang: Writing – review & editing, Supervision, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Ling Zhou: Writing – review & editing, Supervision, Software, Resources, Project administration, Conceptualization. Chunxia Zhao: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Conceptualization.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committees of Tongji Hospital, Huazhong University of Science and Technology (Approval No. TJ-IRB202404056), and Hubei No. 3 People's Hospital, Jianghan University (Approval No. LW2025011). Written informed consent was obtained from all participants or their legally authorized representatives prior to enrollment. All methods were performed in accordance with the Declaration of Helsinki. The data used in this study were anonymized, and no individual patient information is disclosed in this manuscript. All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publication

Not applicable.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: