A machine learning model for predicting outcomes of MitraClip therapy

A machine learning model for predicting outcomes of MitraClip therapy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article A machine learning model for predicting outcomes of MitraClip therapy Hui Li, Ying Guo, Junsong Gong, Yiran Hu, Hongxia Qi, Fengwen Zhang, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5370589/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Severe mitral regurgitation (MR) is a life-threatening mitral valve disease. The MitraClip intervention offers a percutaneous solution for patients who are unsuitable for surgery. However, limited information is available on outcomes post-MitraClip intervention. This study aims to develop an approach for predicting MR outcomes after MitraClip intervention using machine learning-enhanced echocardiography. Methods: We enrolled 164 patients with MR ≥ 3 + degree who underwent MitraClip intervention at our institution between 2021 and 2024. Patients were monitored for approximately three years. The analysis included clinical data and echocardiographic parameters. Study endpoints were the recurrence of MR (2 + or above) and major adverse events during follow-up. A total of 147 patients were randomly divided into training (80%) and testing (90%) sets. An additional 17 patients comprised the validation cohort. Results: The best-performing model for predicting clinical outcomes utilized 81 features in a logistic regression classifier. Using all 81 features in the logistic regression model, specificity increased to approximately 0.797 (95% confidence interval: 0.739 ~ 0.854) and sensitivity to about 0.459 (0.370 ~ 0.549), resulting in an overall accuracy of 0.688 (0.632 ~ 0.745) for the validation dataset. The best-performing model achieved a receiver operating characteristic area under the curve value of 0.773 in both the test and validation groups. Conclusions: Our machine learning model, leveraging echocardiographic characteristics, demonstrated superior predictive performance. This model effectively forecasts patient outcomes following MitraClip intervention, proving beneficial within a clinical setting. Health sciences/Cardiology Physical sciences/Mathematics and computing mitral regurgitation MitraClip intervention machine learning model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Background Mitral valve regurgitation (MR) is the most common valvular heart condition, often caused by degenerative changes in the aging population and linked to increased mortality rates [ 1 , 2 ]. While surgery is a treatment option, about half of the patients are ineligible due to factors like frailty, comorbidities, or left ventricular dysfunction [ 3 ]. A viable alternative is transcatheter edge-to-edge mitral valve repair (TEER) using the MitraClip delivery system (Abbott Vascular, Abbott Park, IL, USA). The effectiveness of this approach has been evaluated in clinical trials [ 4 , 5 ]. The ability to predict outcomes after TEER can enhance candidate selection and provide patients with clearer expectations for their short- and long-term prognosis. Currently, the success of MitraClip intervention largely depends on the cardiologist's expertise. Several potential predictors for MR recurrence and major adverse events include clinical, echocardiographic, and procedural factors such as age, preoperative ventricular function, and procedural complexity [ 6 ]. However, the discussion around echocardiographic and procedural indicators associated with MR recurrence continues, highlighting the need for a deeper understanding of the underlying causes of MR recurrence [ 7 , 8 ]. Recently, the use of machine learning algorithms, a subset of artificial intelligence, has gained prominence in handling the extensive diversity of clinical data and has shown promising results across various medical fields [ 9 ]. Machine learning techniques are powerful analytical tools capable of recognizing complex data patterns for categorization. Many machine learning models have demonstrated superior ability to accurately identify high-risk patients for mortality, outperforming traditional statistical methods in classification and prognosis in general healthcare settings [ 10 , 11 ]. However, previous studies have not explored the potential of these methodologies to understand the factors contributing to unsuccessful MitraClip interventions. We hypothesized that machine learning approaches could identify new potential predictors and improve the comprehensive prediction of MR recurrence and major adverse events. The primary objective of this study was to construct a machine learning based prognostic model that enhances the interpretability of results and provides a deeper understanding of the underlying dataset. 2 Methods 2.1 Patient Cohorts We enrolled 164 patients (aged 18 years or older) with symptomatic moderate-to-severe or severe MR who met the echocardiographic inclusion criteria and were deemed inoperable or at high surgical risk by the heart team, with a life expectancy exceeding one year [ 12 , 13 ]. Exclusion criteria included patients with: (a) mitral stenosis requiring surgical intervention due to conditions such as rheumatic heart disease, endocarditis, or congenital causes; (b) simultaneous aortic valve replacement or repair, tricuspid valve replacement, or previous mitral, aortic, or tricuspid valve surgery; (c) concomitant congenital conditions (except for closure of patent foramen ovale) and bypass for obstructive coronary disease; (d) severe comorbid organ dysfunction. Patients requiring tricuspid valve repair for functional tricuspid regurgitation were not excluded from the study and met the same inclusion and exclusion criteria as the training cohort. After evaluation by the Heart Team, all 164 patients underwent TEER using the MitraClip system, with procedural details previously described [ 14 ]. 2.2 Echocardiography Skilled cardiologists conducted thorough transthoracic echocardiography evaluations on each patient before the procedure, immediately after the procedure, and during follow-up. These assessments were performed using either the EPIQ 7C or EPIQ CVX ultrasound platforms (Philips, Bothell, WA, USA). Echocardiographic imaging followed established guidelines for assessing valvular heart disease [ 15 ]. Left ventricular end-diastolic and end-systolic volumes, left ventricular ejection fraction, and left atrial volume were calculated using the biplane Simpson's method[ 16 ]. The severity of MR was graded into four categories (mild = 1+, moderate = 2+, moderate to severe = 3+, severe = 4+) and quantified by the effective regurgitant orifice area, regurgitant volume, and regurgitant fraction according to guidelines [ 15 , 17 ]. To determine the etiology and mechanism of regurgitation (primary or secondary, using Carpentier nomenclature), we assessed the anatomical characteristics of the mitral valve to evaluate suitability for TEER. This included measurements of the mitral valve area, mitral annulus diameter, coaptation length and depth, as well as leaflet thickness and length. Additionally, the function of the aortic and tricuspid valves was evaluated, and systolic pulmonary arterial pressure (PAPS) was calculated. 2.3 Prognostic follow-up Up to February 2024, follow-up information was collected through medical assessments, echocardiographic examinations, and telephone inquiries. The mean follow-up duration was 281 days (ranging from 30 to 872 days), with a 100% follow-up completion rate. The primary endpoint was the recurrence of residual MR (graded as 2 + or higher) during the follow-up period. Early-onset mitral regurgitation was defined as occurring within the first 3 days post-procedure, while late-onset mitral regurgitation was identified as occurring after the initial follow-up period. Grading of residual mitral regurgitation followed the guidelines set by the American Society of Echocardiography [ 18 ]. Secondary clinical endpoints included major adverse events such as mitral valve stenosis (defined as a mean pressure difference > 5 mmHg), overall mortality, and failed MitraClip intervention necessitating surgical repair or replacement of the mitral valve. Additionally, causes of death were determined through a detailed review of medical records. 2.4 Data process and model construction The study divided the 164 patients who underwent MitraClip intervention into two groups: 147 patients for model training and testing, and 17 patients for validation of the predictive model's efficiency. An endpoint prediction model was constructed using imaging and clinical features through machine learning methods. A total of 81 features from the 147 patients were used to establish the machine learning model. A 10-fold cross-validation test was applied during model training. The MR patient dataset was randomly divided into an 80% training dataset and a 20% testing dataset, and this process was repeated 10 times. Then, various machine learning classifiers and feature combinations were tested for each arrangement of the training and testing datasets. The performance consistency between the testing and validation sets was evaluated to select the best models. The workflow was shown in Fig. 1 . 2.5 Machine learning classifiers In this study, we employed a comprehensive range of machine learning classifiers to identify the best-performing algorithm. These classifiers are based on distinct mathematical principles and logical frameworks and can be categorized into various types: linear regression algorithms, tree-based algorithms, kernel-based learning algorithms, et al. After rigorous testing and evaluation, we employed 14 diverse machine learning classifiers, including: tree-based algorithms (such as adaBoost, catBoost, extra trees, gradient boosting, lightGBM, random forest, and XGBoost), linear regression algorithms (such as logistic regression, ridge regression, and softmax regression), kernel-based support vector machine (SVM), instance-based K-nearest neighbors (KNN), and probabilistic models (naive Bayes and quadratic discriminant analysis). 2.5.1 Tree-based algorithms Firstly, we focused on a selection of sophisticated tree-based algorithms. AdaBoost enhances model performance by focusing on difficult-to-classify samples. CatBoost excels in handling categorical data without extensive preprocessing. Extra trees algorithm introduces more randomness in selecting split points, offering enhanced robustness. Gradient boosting sequentially corrects errors from previous models, reducing bias and variance. LightGBM optimizes for large-scale data with its efficient histogram-based algorithm. Random Forest aggregates multiple decision trees to mitigate overfitting. XGBoost stands out for its performance and speed across diverse datasets. These models collectively are powerful tools for addressing complex prediction tasks with high accuracy and efficiency. 2.5.2 Linear regression models Next, we turned to linear models to predict outcomes by establishing linear relationships between features and target variables. Logistic regression excels at binary classification, providing probabilities for two distinct classes. To address model complexity and prevent overfitting, we employed ridge regression, which introduces L2 regularization into the linear regression framework, enhancing stability and performance. For multi-class classification, we employed softmax regression, an extension of logistic regression that efficiently assigns probabilities across multiple classes. This suite of linear models demonstrates their simplicity and adaptability, offering a solid foundation for both binary and multi-class prediction tasks with precision. 2.5.3 Kernel-based methods We also used a kernel-based method, the SVM, which excels at finding optimal boundaries between different categories. SVM is particularly notable for its ability to perform classification in higher-dimensional spaces by applying kernel tricks. This method effectively manages complex datasets where linear separability is not feasible, making it a powerful tool for nuanced classification tasks. 2.5.4 Instance-based learning For instance-based learning, we adopted the KNN algorithm due to its simplicity and effectiveness. KNN predicts the category of new samples based on the closest training examples in the feature space, bypassing the need for an explicit model training phase. This method excels in its straightforwardness and the intuitive principle that similar samples are likely to belong to the same class, making it a valuable component of our classification toolkit. 2.5.5 Probabilistic models Lastly, we explored probabilistic models to leverage probability theories for classification and prediction. Naive Bayes, a simplistic yet efficient probabilistic classifier, operates on the principle of feature independence as grounded in Bayes' theorem. Additionally, quadratic discriminant analysis incorporates quadratic terms to fit data distributions, making it suitable for scenarios where linear separability assumptions are inadequate. These probabilistic models provide a robust framework for understanding and predicting outcomes based on underlying probabilities, enriching our analytical toolkit. 2.6 Probabilistic models The dataset of 147 samples was divided into 10 folds for training and testing, with an 8:2 split for each fold. During model training, the folds were rotated 10 times between the training and testing sets to account for variations in model accuracy. This rotation led to fluctuations in model performance. Our goal was to identify the optimal combination of training and testing sets that consistently achieved high accuracy in both the testing and validation datasets. 2.7 Feature selection Eighty-one clinical and echocardiographic features were used to train and optimize the prediction models. The importance of the 81 variables was assessed using the XGboost classifier, while the mean decrease accuracy (MDA) was evaluated using random forest. Both importance and MDA values were co-evaluated for each variable. The top-ranked features were incrementally added to the model based on their priority. The model was then trained with these evolving features, and classification accuracy was monitored to observe the accuracy curve and ensure model convergence. Finally, an optimal combination of features was identified to maximize classification efficiency. 2.8 Statistical analysis Normally distributed variables were compared between groups using an independent-sample t test, while non-normally distributed variables were assessed with the Mann-Whitney u test. Categorical data were analyzed using the chi-square test or Fischer’s exact test. Statistical significance was defined as P < 0.05. All statistical analyses were performed using SPSS version 23.0 (IBM Corp, Armonk, NY). 3 Results 3.1 Clinical and echocardiographic information of patients A total of 164 patients (mean age 68.4 ± 10.5 years; 58 women, 35.4%) with ≥ 3 + MR (3+: 28.7%, 4+: 71.3%) were enrolled in this study. Post-MitraClip intervention, all patients had their MR reduced to ≤ 2+. Follow-up was completed by 100% of the participants. Baseline and follow-up characteristics are detailed in Table 1 . Common comorbidities included hypertension (46.3%), hyperlipemia (22.0%), atrial fibrillation (28.7%), diabetes mellitus (17.7%), coronary artery disease (17.7%) and so on. The mean plasma NT-proBNP value was 2328.5 ± 4227.9 pg/mL. During follow-up, significant reductions were observed in systolic blood pressure, diastolic blood pressure, plasma hemoglobin, low-density lipoprotein cholesterol, and NT-proBNP levels (Fig. 2 , P < 0.05). Table 1 Summary of clinical and echocardiographic characteristics of the subjects. Variables Pre-operation (n = 164) Follow-up (n = 164) P value Significance General information Gender-Female, n (%) 58 (35.4) Age-year, mean ± SD (min-max), Median 68.4 ± 10.5 (28–92), 69 Body surface area (BSA)-m 2 , mean ± SD (min-max), Median 1.73 ± 0.19 (1.34–2.15), 1.74 Body mass index (BMI)-kg/m 2 , mean ± SD (min-max), Median 24.0 ± 3.6 (13.8–33.2), 24.0 Clinical information Hypertension, n (%) 76 (46.3) Smoking, n (%) 39 (23.8) Hyperlipemia, n (%) 36 (22.0) Diabetes mellitus, n (%) 29 (17.7) Coronary artery disease, n (%) 29 (17.7) Coronary Artery Bypass Grafting, n (%) 8 (4.9) Valve surgery, n (%) 6 (3.7) Atrial fibrillation, n (%) 47 (28.7) 45 (27.4) 0.273 Systolic blood pressure-mmHg, mean ± SD (min-max), Median 121.4 ± 18.0 (79–172), 120.5 113.2 ± 12.6 (70–143), 112 2.54×10 − 6 ** Diastolic blood pressure-mmHg, mean ± SD (min-max), Median 71.9 ± 12.6 (46–143), 71.0 62.5 ± 9.1 (44–111), 62 7.72×10 − 13 ** Haemoglobin-g/L, mean ± SD (min-max) Median 134.0 ± 20.3 (79–177), 135 119.5 ± 21.8 (60–173), 120 1.38×10 − 9 ** Creatinine-umol/L, mean ± SD (min-max), Median 99.2 ± 54.5 (37.3-419.26), 86.7 108.2 ± 71.7 (36.4-589.89), 88.32 0.2 Uric acid-umol/L, mean ± SD (min-max), Median 436.1 ± 159.7 (134.14-1372.3), 408.22 406.6 ± 159.1 (84.58.05-1217.6), 390.37 0.095 Homocysteine-umol/L, mean ± SD (min-max), Median 18.5 ± 7.7 (6.56–51.21), 16.9 21.4 ± 8.3 (1.42–50.34), 20.91 0.001 ** Low-density lipoprotein cholesterol (LDL-C)-mmol/L, mean ± SD (min-max), Median 2.55 ± 0.95 (0.87–5.52), 2.28 2.28 ± 0.92 (0.52–5.96), 2.20 0.011 * NT proBNP-pg/mL, mean ± SD (min-max), Median 2328.5 ± 4227.9 (12.6-36763), 836 685.6 ± 926.9 (6.6–5465), 334.5 2.55×10 − 6 ** Conventional Echocardiographic parameters Left atrial anteroposterior diameter (LAD)-mm, mean ± SD (min-max), Median 48.7 ± 9.2 (33–89), 47 44.3 ± 8.7 (27–76), 43 1.72×10 − 5 ** Left atrial volume (LAV)-mL, mean ± SD (min-max), Median 119.1 ± 60.5 (42–413), 102.5 94.8 ± 58.3 (28–366), 76 2.39×10 − 4 ** Left ventricular end diastolic diameter (LVDd)-mm, mean ± SD (min-max), Median 57.6 ± 7.4 (37–81), 57 53.1 ± 7.8 (35–79), 52 1.84×10 − 7 ** Left ventricular end systolic diameter (LVDs)-mm, mean ± SD (min-max), Median 38.8 ± 9.4 (23–73), 37 37.2 ± 9.4 (20–72), 34 0.127 Left ventricular ejection fraction (LVEF)-%, mean ± SD (min-max), Median 59.3 ± 12.6 (28–81), 63 56.4 ± 12.7 (9–79), 60 0.04 * Right ventricular enddiastolic diameter (RVEDD)-mm, mean ± SD (min-max), Median 25.9 ± 4.3 (17–44), 26 25.6 ± 5.1 (2–60), 25 0.665 Mitral E wave, mean ± SD (min-max), Median 1.32 ± 0.35 (0.5–2.1), 1.3 1.40 ± 0.44 (0.5–3.2), 1.4 0.072 Tricuspid E wave, mean ± SD (min-max), Median 0.532 ± 0.138 (0.3–1.2), 0.5 0.570 ± 0.151 (0.3–1.3), 0.5 0.03 * Systolic pulmonary artery pressure (SPAP)-mmHg, mean ± SD (min-max), Median 42.2 ± 14.2 (18–90), 41 35.4 ± 12.8 (6–92), 32 6.81×10 − 6 ** Tricuspid regurgitation velocity-m/s, mean ± SD (min-max), Median 2.96 ± 0.54 (1.8–4.5) 2.95 2.67 ± 0.54 (0.6–4.5), 2.6 2.13×10 − 6 ** Mitral valve etiology Mitral valve area (MVA)-cm 2 , mean ± SD (min-max), Median 5.98 ± 1.47 (3.2–14.3), 5.8 Anterior mitral valve leaflet (AML)-mm, mean ± SD (min-max), Median 27.2 ± 4.3 (11–38), 27 Posterior mitral valve leaflet (PML)-mm, mean ± SD (min-max), Median 15.9 ± 4.2 (8–36), 15 Mitral valve leaflet thickening, n (%) 111 (67.7) Mitral annulus calcification (MAC), n (%) 13 (7.7) Antero-posterior mitral annulus diameter (MAD)-mm, mean ± SD (min-max), Median 33.1 ± 5.0 (21–49), 33 Medio-lateral mitral annulus diameter-mm, mean ± SD (min-max), Median 33.2 ± 4.6 (15–46), 33.2 Vena contracta-mm, mean ± SD (min-max), Median 6.71 ± 1.66 (2.3–13), 6.85 Proximal isovelocity surface area (PISA)-cm, mean ± SD (min-max), Median 0.946 ± 0.325 (0.4-3), 0.9 Effective regurgitant orifice area-cm 2 , mean ± SD (min-max), Median 0.484 ± 0.340 (0.04–2.49), 0.41 Regurgitant volume-mL/beat, mean ± SD (min-max), Median 66.7 ± 40.3 (7-259), 58 Degree of regurgitant, n (%) 0 - 38 (23.2) 4.00×10 − 58 ** 1 - 85 (51.8) 2 - 28 (17.1) 3 47 (28.7) 8 (4.9) 4 117 (71.3) 5 (3.0) Etiology, n (%) Degenerative mitral regurgitant (DMR) 108 (65.9) Functional mitral regurgitant (FMR) 47 (28.7) Both 9 (5.5) Mean pressure gradient-mmHg, mean ± SD (min-max), Median 2.16 ± 1.09 (1–6), 2 4.02 ± 2.72 (1–23), 3 3.26×10 − 14 ** 3.2 Building the MR recurrence risk prediction model We developed and optimized a high-performance predictive model at three levels: (a) selecting an appropriate combination of features to avoid underfitting nor overfitting; (b) identifying the best-performing classifier by testing various machine learning classifiers; (c) establishing the most representative training set to achieve high predictive accuracy in both testing and validating datasets. Initially, we tested the appropriate combination of features by incrementally adding the most important features to the model. The overall set of 81 features comprised clinical variables obtained before and during the MitraClip intervention, aimed at predicting prognosis. These features were ranked based on MDA and importance values (Fig. 3 ). The sequences of MDA and importance were simultaneously considered when adding features to the predictive model. All 14 machine learning classifiers were tested by incrementally adding features from 1 to 81, using 10-fold cross-validation between training (80%) and testing (20%) datasets. Predictive parameters, including precision, recall, f1 score for both positive and negative patients, and model accuracy, were analyzed and presented in Fig. 4 . Both mean and maximum limits were considered for selecting superior classifiers. Positive recall, being the lowest, reflects the model’s performance. LightGBM, logistic regression, ridge regression, softmax, SVM, and XGBoost demonstrated higher positive recall values compared to other classifiers in the testing dataset. To build the high-performance model, logistic regression, softmax, and SVM were selected from these six classifiers for their superior performance in the validation dataset, as their maximum parameter limits were higher (Fig. 4 ). The best-performance combination of features for each of the three selected classifiers, logistic regression, softmax and SVM, was presented in Table 2 and Fig. 5 A. For example, using all 81 features in the logistic regression model (logistic regression-81) with 10-fold cross-validation on an 80% training and 20% testing dataset, specificity increased to approximately 0.797 (0.739, 0.854) and sensitivity to about 0.459 (0.370, 0.549), resulting in an overall accuracy of 0.688 (0.632, 0.745) for the validation dataset. In one of the 10 cross-validation runs, a particular arrangement of the training (80%) and testing (20%) dataset (the best model of logistic regression-81) achieved a specificity of 0.900 and a sensitivity of 0.571, resulting in an accuracy of 0.765 (Fig. 5 B). The overall ROC AUC for the best models of logistic regression-81, softmax-71, and SVM-25 using all 164 patients is shown in Fig. 5 C. Among the three classifiers, logistic regression demonstrated the highest diagnostic efficacy (overall ROC AUC: 0.909; test and validation dataset ROC AUC: 0.774). Ultimately, the logistic regression-81 model was selected for its superior performance. This model's parameters improved and converged as more features were added, achieving the best and most balanced performance for both the testing and validation datasets when using all 81 features (Fig. 5 D). Table 2 Performance of three superior classifiers, each with a specific combination of features. Classifiers-No. of features, mean (95% CI); min-max Statistics Logistic regression-81 Softmax-71 Support vector machine-25 Testing set Accuracy 0.683 (0.639, 0.727); 0.567–0.767 0.713 (0.664, 0.763); 0.600–0.800 0.553 (0.418, 0.689); 0.267-0.800 0_precision (Specificity) 0.772 (0.731, 0.813); 0.682–0.857 0.780 (0.728, 0.833); 0.667–0.905 0.684 (0.577, 0.791); 0.455-0.900 0_recall 0.765 (0.692, 0.837); 0.565–0.938 0.805 (0.735, 0.875); 0.609–0.905 0.537 (0.333, 0.741); 0.095–0.833 0_f1 0.763 (0.725, 0.802); 0.667–0.821 0.788 (0.742, 0.835); 0.686–0.864 0.580 (0.405, 0.756); 0.160–0.857 1_precision (Sensitivity) 0.516 (0.395, 0.636); 0.286–0.875 0.571 (0.480, 0.662); 0.308–0.778 0.410 (0.301, 0.518); 0.158–0.625 1_recall 0.518 (0.441, 0.595); 0.333-0.700 0.527 (0.447, 0.606); 0.333–0.714 0.568 (0.447, 0.689); 0.286–0.778 1_f1 0.504 (0.427, 0.582); 0.353–0.667 0.535 (0.473, 0.598); 0.400-0.667 0.457 (0.364, 0.551); 0.214–0.667 Validation set Accuracy 0.688 (0.632, 0.745); 0.588–0.824 0.588 (0.522, 0.654); 0.471–0.765 0.547 (0.430, 0.664); 0.353–0.765 0_precision (Specificity) 0.797 (0.739, 0.854); 0.692–0.909 0.718 (0.658, 0.777); 0.615-0.900 0.718 (0.646, 0.790); 0.571–0.833 0_recall 0.758 (0.724, 0.792); 0.667–0.833 0.700 (0.636, 0.764); 0.583–0.917 0.558 (0.385, 0.732); 0.250–0.833 0_f1 0.775 (0.738, 0.813); 0.696–0.870 0.705 (0.658, 0.752); 0.636–0.818 0.611 (0.476, 0.745); 0.353–0.833 1_precision (Sensitivity) 0.459 (0.370, 0.549); 0.250–0.667 0.262 (0.129, 0.396); 0.000-0.571 0.368 (0.271, 0.466); 0.200–0.600 1_recall 0.520 (0.366, 0.674); 0.200–0.800 0.320 (0.139, 0.501); 0.000-0.800 0.520 (0.420, 0.620); 0.400–0.800 1_f1 0.484 (0.368, 0.599); 0.222–0.727 0.287 (0.134, 0.440); 0.000-0.667 0.415 (0.338, 0.491); 0.267-0.600 3.3 Clinical outcomes Our model demonstrated high specificity and sensitivity in predicting events after MitraClip intervention in MR patients. Follow-up completeness was 100%. During the follow-up period, the composite endpoint occurred in 53 patients (32.3%). These endpoints included the recurrence of moderate MR (2 + or higher), mitral valve stenosis (mean pressure difference > 5 mmHg), all-cause mortality, and TEER failure requiring subsequent mitral valve surgical repair or replacement. In this study, there were 4 deaths (3 cardiac and 1 non-cardiac). Causes of death included sudden cardiac death (1 patient), malignant arrhythmia (2 patients), and other (1 patient). Recurrence rates of severe MR (3 + or higher) were 3.0% at 30 days and 4.9% at three years. 4 Discussion The recurrence of MR following mitral valve intervention has been linked to a notable increase in long-term mortality rates and a higher risk of hospital readmission due to heart failure during the follow-up period [ 19 ]. In our study, severe recurrent MR after MitraClip intervention occurred in 7.9% of patients. In comparison, other studies reported recurrence rates of 18% in Everest 2 and 21% in ACESS-EU at one year [ 20 , 21 ]. Recently, various factors with prognostic significance following mitral valve interventions have been increasingly recognized. These factors encompass demographic, clinical, anatomic, and procedural features. The complexity of assessing recurrent MR is heightened by the large number of variables involved, making it difficult for clinicians to assess the risks for individual patients comprehensively. Despite this, predictors of MR recurrence remain poorly defined [ 19 ]. Previous studies on predicting MR recurrence after MitraClip intervention mainly relied on classic statistical modeling techniques constrained by assumptions such as distribution normality, non-informative or random censoring, and hazard risk linearity [ 19 , 22 ]. However, these traditional methods usually focus one or a few clinical features and overlook the potential effects of complex and hidden interactions among several weaker predictors. Machine learning algorithms, a subfield of artificial intelligence, can overcome these limitations by capturing high-dimensional nonlinear relationships among many clinical features [ 23 ]. Recently, several machine learning models have been introduced to handle the significant variability in clinical data and have demonstrated efficacy in various medical applications for cardiovascular diseases [ 11 , 24 ]. In our study, we employed logistic regression classifiers due to their state-of-the-art accuracy and interpretability. We used softmax regression because it is a popular method for handling high-dimensional predictors. Logistic regression was particularly useful for building risk models to address binary classification problems. Therefore, we utilized these methods for variable selection and risk model construction, ultimately choosing the best-performing model. Logistic regression is particularly suited for small datasets due to its computational efficiency. Compared to decision tree models, logistic regression offers several advantages, including faster runtime, better handling of outliers, greater flexibility in data processing, and superior performance. These benefits make logistic regression preferable, especially when dealing with high model complexity relative to dataset size. This efficiency and reduced generalization error are crucial in such scenarios. This study is likely the first to develop a predictive model for MitraClip intervention using echocardiography integrated with machine learning. The superior prediction model demonstrated an ROC AUC value of 0.909, with a sensitivity of 0.900 and specificity of 0.571 in the test group, and an ROC AUC value of 0.773 in the testing and validation dataset. Logistic regression exhibited excellent discriminatory performance in predicting the recurrence of MR and major adverse cardiovascular events. These findings highlight the potential of machine learning in assessing prognostic risk for MR patients undergoing MitraClip intervention. Accurately predicting the timeline for significant recurrent MR and major adverse cardiovascular events based on preoperative clinical and echocardiographic parameters is complex. Our results support expanding indications and refining patient selection criteria for those undergoing MitraClip intervention. In our machine learning model, we found various clinical, anatomic, and procedural factors linked to recurrent MR following MitraClip intervention. Conversely, traditional cardiovascular risk factors (such as hypertension, diabetes mellitus, and chronic obstructive pulmonary disease) and specific parameters used to evaluate MR severity (such as proximal isovelocity surface area, effective regurgitation area, and regurgitant volume) did not emerge as the most predictive factors for the endpoint in this specific patient population. First, a history of surgery is a predictor of recurrent MR after MitraClip intervention. Similar findings from the TRAMI registry and Boerlage-van Dijk indicate that patients with previous valve surgery have a poor prognosis following transcatheter edge-to-edge mitral valve repair [ 25 , 26 ]. Second, certain serum biomarkers reflecting pathophysiological states are related to clinical outcomes. Our machine learning model identified serum creatinine, which reflects renal function, as associated with worse outcomes after MitraClip intervention. Chronic kidney insufficiency can cause myocardial damage due to changes in cardiac structure and function. Similarly, the TRAMI registry demonstrated that baseline serum creatinine ≥ 1.5 mg/dl is an independent predictor of 1-year mortality (HR: 1.77; p = 0.002) [ 25 , 27 ]. In addition, NT-proBNP levels independently correlated with recurrent MR, reflecting the volumetric overload of the left ventricular and consequences of myocardial systolic dysfunction. Triantafyllis et al. demonstrated a similar correlation between NT-proBNP levels and cardiac mortality after TEER (n = 136; HR: 1.5; 95% CI: 1.1 to 2.1; p = 0.018) [ 28 ]. Thus, individuals with brief follow-up durations should regulate their fluid intake and use medications known to enhance myocardial systolic function [ 29 , 30 ]. Additionally, hemoglobin levels play a crucial role in postoperative prognosis. Decreased hemoglobin can lead to reduced blood oxygen-carrying capacity and ventricular dysfunction, emphasizing the importance of minimizing bleeding during interventions. Third, heart chamber volumes are significant predictors of outcomes. All cases without significant MR exhibited favorable remodeling of left atrial volume and left ventricular volume at follow-up. Left atrial volume and left ventricular endsystolic volume are predictors of outcomes after clip implantation, as reinforced by recent studies assessing TEER with secondary MR and the GRASP-IT registry [ 31 , 32 ]. Increased left atrial volume may lead to atrial fibrillation, while enlarged left ventricular endsystolic volume is associated with advanced cardiac remodeling and left ventricular dysfunction. Fourth, our model showed that higher pulmonary hypertension predicted MR recurrence (≥ 3+) after MitraClip intervention. Pulmonary hypertension, defined as pressures exceeding 50 mmHg, indicates advanced cardiomyopathy and is predictive of MR recurrence following TEER [ 33 , 34 ]. Fifth and importantly, we identified operation time and residual MR (post-clip) as independent predictors of MR recurrence after MitraClip intervention. Although MitraClip intervention is a safe procedure with a relatively low complication rate, longer operation times may lead to procedural complications such as acute heart failure, pulmonary embolism, stroke, pericardial effusion, pericardial tamponade, cardiogenic shock, and bleeding requiring blood transfusions. However, our model highlighted that only two mitral valve anatomical structures (medio-lateral mitral annulus diameter and mitral valve area), were significant for predicting outcomes. Increased annular distortion was associated with a high rate of clip failure due to progressive left ventricular remodeling [ 35 ]. This study has several constraints. Primarily, the applicability and efficacy of the machine learning methodologies were verified within a single-center cohort, and the examination was conducted post hoc, with most limitations stemming from the retrospective nature of the dataset. The relatively small number of patients limits the ability to detect statistically significant differences in clinical outcomes in the validation group. We acknowledge the limited sample size and the lack of multivariable analysis (e.g., due to the limited number of events), which limits the strength of the results. A multi-center study will be conducted in the future as more hospitals adopt MitraClip intervention. Furthermore, assessing MR grading following a double orifice repair presents challenges. Nonetheless, the severity of MR post-MitraClip intervention was evaluated using the integrative approach advised by the guidelines18, allowing for better reproducibility. Moreover, the sensitivity (~ 0.5) of the predictive model remains unsatisfactory, restricted by the limited sample size and inadequate follow-up period. Nearly half of the patients were followed up within 100 days, which is a short period to observe the recurrence of MR after MitraClip intervention. Some recurrences or events may not have yet occurred, preventing the classifiers from accurately capturing the typical characteristics of positive prognosis. 5 Conclusions Our research demonstrates significant advantages of our model in predicting the prognosis of mitral valve interventions: (1) high predictive accuracy in determining outcomes, validated by the cohort; and (2) reliance solely on non-invasive echocardiographic assessments and common clinical parameters. In conclusion, our innovative model presents a more effective and non-invasive approach to forecasting the prognosis of mitral valve interventions. This model could serve as a valuable tool for advancing clinical understanding and enhancing criteria for selecting appropriate candidates for mitral valve interventions. Abbreviations MR mitral regurgitation TEER transcatheter edge-to-edge mitral valve repair SVM support vector machine KNN instance-based K-nearest neighbors PAPS pulmonary arterial pressure MDA mean decrease accuracy Declarations Ethics approval and consent to participate This prospective clinical trial was conducted from January 2021 to February 2024 at Fuwai Hospital in Beijing, China. The study adhered to the Declaration of Helsinki, received approval from the Fuwai Hospital Ethics Committee (reference number: 2024BJYYEC-KY067-01), and all participants provided written informed consent. Competing interests The authors declare that they have no conflict of interest. Funding This study was financially supported by the National High Level Hospital Clinical Research Funding (2023-GSP-QN-40, 2022-GSP-PT-7, 2022-GSP-QN-18), the Clinical and Translational Medicine Research Program of the Chinese Academy of Medical Sciences (2023-I2M-C&T-B-056, 2023-I2M-C&T-B-117) and CAMS Innovation Fund for Medical Sciences (2021-I2M-1-065). Author Contribution Hui Li and Ying Guo contributed significantly to writing the manuscript. Hui Li and Junsong Gong analyzed the echocardiograms. Yiran Hu and Fengwen Zhang collected patient information. Fujian Duan and Xiangbin Pang revised the manuscript, conceived the study and supervised the project. All authors read and approved of the final manuscript. Acknowledgement We thank Yiliang Wei and Yongqiang Kong for their assistance in constructing and optimizing predictive models. Hui Li had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Data Availability Data is provided within the supplementary information files. References Nkomo, V. T. et al. Burden of valvular heart diseases: a population-based study. :368. (2006). Dziadzko, V., MikhailMedina-Inojosa, J. R. B. & GiovanniMichelena, H. I. C. Juan A.Maalouf, JosephThapa, PrabinEnriquez-Sarano, Maurice %J European Heart Journal: The Journal of the European Society of Cardiology. Causes Mech. isolated mitral regurgitation community: Clin. context outcome ; 40 . (2019). St Goar, F. G. et al. Endovascular edge-to-edge mitral valve repair: short-term results in a porcine model. Circulation . 108 , 1990–1993 (2003). Mauri, L. et al. 4-year results of a randomized controlled trial of percutaneous repair versus surgery for mitral regurgitation. J. Am. Coll. Cardiol. 62 , 317–328 (2013). De Bonis, M. et al. Edge-to-edge surgical mitral valve repair in the era of MitraClip: what if the annuloplasty ring is missed? Curr. Opin. Cardiol. 30 , 155–160 (2015). Tamborini, G. et al. Predictive Value of Pre-Operative 2D and 3D Transthoracic Echocardiography in Patients Undergoing Mitral Valve Repair: Long Term Follow Up of Mitral Valve Regurgitation Recurrence and Heart Chamber Remodeling. J. Cardiovasc. Dev. Dis. ; 7 . (2020). Verma, S., Latter, D. A., Bonow, R. O. & Failed Mitral, T. E. E. R. Are There Lessons for Decision Making? J. Am. Coll. Cardiol. 78 , 10–13 (2021). Hassan, A. & Eleid, M. F. Recurrent Mitral Regurgitation After MitraClip: Defining Success and Predicting Outcomes. Circ. Cardiovasc. Interv . 15 , e011837 (2022). Johnson, K. W. et al. Artificial Intelligence in Cardiology. J. Am. Coll. Cardiol. 71 , 2668–2679 (2018). Sahni, N., Simon, G. & Arora, R. Development and Validation of Machine Learning Models for Prediction of 1-Year Mortality Utilizing Electronic Medical Record Data Available at the End of Hospitalization in Multicondition Patients: a Proof-of-Concept Study. J. Gen. Intern. Med. 33 , 921–928 (2018). Ambale-Venkatesh, B. et al. Cardiovascular Event Prediction by Machine Learning: The Multi-Ethnic Study of Atherosclerosis. Circ. Res. 121 , 1092–1101 (2017). Baumgartner, H. et al. 2017 ESC/EACTS Guidelines for the management of valvular heart disease. Eur. Heart J. 38 , 2739–2791 (2017). Rick, A. et al. 2017 AHA/ACC Focused Update of the 2014 AHA/ACC Guideline for the Management of Patients With Valvular Heart Disease: A Report of the American College of Cardiology/American Heart Association Task Force on. Clin. Pract. Guidelines . 70 , 252–289 (2017). Katz, W. E. et al. Echocardiographic evaluation and guidance for MitraClip procedure. :616 – 32. (2017). Zoghbi, W. A. et al. Recommendations for Noninvasive Evaluation of Native Valvular Regurgitation, A Report from the American Society of Echocardiography Developed in Collaboration with the Society for. Cardiovasc. Magn. Reson. :4. (2020). Lang, R. M. et al. Recommendations for Cardiac Chamber Quantification by Echocardiography in Adults: An Update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. (2013). John, D. et al. Appropriate Use Criteria for Multimodality Imaging in Valvular Heart Disease: A Report of the American College of Cardiology Appropriate Use Criteria Task Force, American Association for Thoracic Surgery, American Heart Association, American Society of Echocardiography, American Society of Nuclear Cardiology, Heart Rhythm Society, Society for Cardiovascular Angiography and Interventions, Society of Cardiovascular Computed Tomography, Society for Cardiovascu. 2017. (2017). Nunes, M. C. P. et al. Role of LA Shape in Predicting Embolic Cerebrovascular Events in Mitral Stenosis: Mechanistic Insights From 3D Echocardiography. (2014). Adamo, M. et al. Five-year clinical outcomes after percutaneous edge-to-edge mitral valve repair: Insights from the multicenter GRASP-IT registry. Am. Heart J. 217 , 32–41 (2019). Mack, M. J. et al. 3-Year Outcomes of Transcatheter Mitral Valve Repair in Patients With Heart Failure. J. Am. Coll. Cardiol. 77 , 1029–1040 (2021). Glower, D. D. et al. Percutaneous mitral valve repair for mitral regurgitation in high-risk patients: results of the EVEREST II study. J. Am. Coll. Cardiol. 64 , 172–181 (2014). Boekstegers, P. et al. Percutaneous interventional mitral regurgitation treatment using the Mitra-Clip system. Clin. Res. Cardiol. 103 , 85–96 (2014). Dey, D. et al. Artificial Intelligence in Cardiovascular Imaging: JACC State-of-the-Art Review. J. Am. Coll. Cardiol. 73 , 1317–1335 (2019). Weng, S. F. et al. Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLoS One . 12 , e0174944 (2017). Puls, M. et al. One-year outcomes and predictors of mortality after MitraClip therapy in contemporary clinical practice: results from the German transcatheter mitral valve interventions registry. Eur. Heart J. 37 , 703–712 (2016). Boerlage-vanDijk, K. et al. Predictors of outcome in patients undergoing MitraClip implantation: An aid to improve patient selection. Int. J. Cardiol. 189 , 238–243 (2015). Zuern, C. S. et al. Influence of non-cardiac comorbidities on outcome after percutaneous mitral valve repair: results from the German transcatheter mitral valve interventions (TRAMI) registry. Clin. Res. Cardiol. 104 , 1044–1053 (2015). Triantafyllis, A. S. et al. Long-term survival and preprocedural predictors of mortality in high surgical risk patients undergoing percutaneous mitral valve repair. Catheter Cardiovasc. Interv . 87 , 467–475 (2016). Nishimura, R. A. et al. 2014 AHA/ACC guideline for the management of patients with valvular heart disease: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. J. Am. Coll. Cardiol. 63 , e57–185 (2014). Sarnak, M. J. et al. Anemia as a risk factor for cardiovascular disease in The Atherosclerosis Risk in Communities (ARIC) study. J. Am. Coll. Cardiol. 40 , 27–33 (2002). Baldi, C. et al. Predictors of outcome in heart failure patients with severe functional mitral regurgitation undergoing MitraClip treatment. Int. J. Cardiol. 284 , 50–58 (2019). Capodanno, D. et al. Predictors of clinical outcomes after edge-to-edge percutaneous mitral valve repair. Am. Heart J. 170 , 187–195 (2015). Taramasso, M. et al. Clinical and anatomical predictors of MitraClip therapy failure for functional mitral regurgitation: single central clip strategy in asymmetric tethering. Int. J. Cardiol. 186 , 286–288 (2015). Matsumoto, T. et al. Impact of pulmonary hypertension on outcomes in patients with functional mitral regurgitation undergoing percutaneous edge-to-edge repair. Am. J. Cardiol. 114 , 1735–1739 (2014). Asgar, A. W., Mack, M. J. & Stone, G. W. Secondary mitral regurgitation in heart failure: pathophysiology, prognosis, and therapeutic considerations. J. Am. Coll. Cardiol. 65 , 1231–1248 (2015). Additional Declarations No competing interests reported. Supplementary Files SupplementaryTableS1.xlsx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5370589","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":389133960,"identity":"4b63a564-8514-4de7-8b0d-42f6aabc8ebf","order_by":0,"name":"Hui Li","email":"","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hui","middleName":"","lastName":"Li","suffix":""},{"id":389133961,"identity":"ccd9be38-915c-40ca-b097-e67eb0851971","order_by":1,"name":"Ying Guo","email":"","orcid":"","institution":"Beijing Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ying","middleName":"","lastName":"Guo","suffix":""},{"id":389133962,"identity":"a5640fb6-b3fd-494c-b3b2-34b35030a16e","order_by":2,"name":"Junsong Gong","email":"","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Junsong","middleName":"","lastName":"Gong","suffix":""},{"id":389133963,"identity":"abae6fb4-89ae-4d54-8494-2fc2cbf97d47","order_by":3,"name":"Yiran Hu","email":"","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yiran","middleName":"","lastName":"Hu","suffix":""},{"id":389133964,"identity":"74245114-dfaf-48cf-8f2f-0f33ddecb734","order_by":4,"name":"Hongxia Qi","email":"","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hongxia","middleName":"","lastName":"Qi","suffix":""},{"id":389133965,"identity":"e2ba5305-2487-4d98-be49-4542b3132e3f","order_by":5,"name":"Fengwen Zhang","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences","correspondingAuthor":false,"prefix":"","firstName":"Fengwen","middleName":"","lastName":"Zhang","suffix":""},{"id":389133966,"identity":"450ded2a-eccd-4991-8171-bb04a7bd33e8","order_by":6,"name":"Xiangbin Pang","email":"","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiangbin","middleName":"","lastName":"Pang","suffix":""},{"id":389133967,"identity":"b7dd680b-785d-4955-b7bd-e2552ed424ed","order_by":7,"name":"Fujian Duan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4UlEQVRIiWNgGAWjYDCCA0BcwcDGw8DAfIDhgQGxWs6AtbAlMCSQoAUEeAwYEojRwXcj+dmDg218Mvyzez5+SCg4nLidgfnhoxt4tEjeSDM3ONjGxiNx5+xmiQSDw4k7G9iMjXPwaDG4kWAm/RGoheFG7gawlg0HeNik8WtJ/yYBskX+Rs7jH0RqyTEDawEy2IizRfLMmzKJA+fYeAxvpJlZJBikG284TMAvfMfTt0kcKDtmL3cj+fGND3+sZTccb374GJ8WKDgGYzQDUwFh5SBQA2PUEad+FIyCUTAKRhQAAOR/U8fVPcr2AAAAAElFTkSuQmCC","orcid":"","institution":"Fuwai Hospital","correspondingAuthor":true,"prefix":"","firstName":"Fujian","middleName":"","lastName":"Duan","suffix":""}],"badges":[],"createdAt":"2024-11-01 04:53:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5370589/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5370589/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71689010,"identity":"9c747458-aaee-412d-a134-01772905e9bb","added_by":"auto","created_at":"2024-12-17 17:48:42","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":824824,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eStudy flowchart for constructing the machine-learning predictive model.\u003c/strong\u003ePatients with MR of degree ≥3+ who underwent MitraClip intervention were monitored during the follow-up period for recurrence of MR (degree ≥2+) or other adverse cardiac events. A machine learning model were constructed to predict these unfavorable outcomes post-MitraClip intervention. The process involved testing and selecting training datasets, classifiers, and feature combinations that demonstrated stable performance.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/9e45cee90276542408c79c44.png"},{"id":71689012,"identity":"c8bded8b-3f1d-4ea0-a236-d573ab4a3d3c","added_by":"auto","created_at":"2024-12-17 17:48:42","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1579283,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical and echocardiographic characteristics of patients with significant differences between pre-operation and follow-up periods.\u003c/strong\u003eAfter MitraClip intervention, 13 characteristics decreased: systolic and diastolic blood pressure, hemoglobin, low-density lipoprotein cholesterol, NT-proBNP, left atrial anteroposterior diameter, left atrial volume, left ventricular end diastolic diameter, left ventricular end-diastolic volume, left ventricular ejection fraction, systolic pulmonary artery pressure, tricuspid regurgitation velocity, and tricuspid valve abnormality. Three characteristics increased: homocysteine levels, tricuspid E wave and mean pressure gradient.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/f67a2c10a30532c53452528c.png"},{"id":71689011,"identity":"a4d35046-bd4b-49ba-8e45-535ed82c16a5","added_by":"auto","created_at":"2024-12-17 17:48:42","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":643532,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMean decrease accuracy (MDA) and feature importance.\u003c/strong\u003e It presents the top 30 features based on MDA and importance evaluations, along with the covariance between these metrics. Many features show significant roles in both MDA and importance evaluations, such as operation time, uric acid levels, post-clip MR, mitral E wave, heart rate, systolic pulmonary artery pressure (SPAP), left atrial volume (LAV), hemoglobin, low-density lipoprotein cholesterol (LDL-C), fasting plasma glucose (FPG), mitral valve area (MVA), left ventricular endsystolic volume (LVESV), tricuspid regurgitation velocity (TRV), et al.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/e7cd556547705f917c3a8071.png"},{"id":71690078,"identity":"c056cea0-6eee-4ffe-9735-7a6404937b62","added_by":"auto","created_at":"2024-12-17 17:56:42","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1837239,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTesting and validating accuracy of 14 different machine-learning classifiers.\u003c/strong\u003eWe estimated precision, recall, f1 score, and accuracy for each classifier. The superior model was selected by considering both balanced mean levels and high maximum limits for classification parameters. LightGBM, logistic regression, ridge regression, softmax, support vector machine (SVM), and XGBoost classifiers exhibited great performance for the testing dataset (blue box). Among these six classifiers, logistic regression, softmax, and SVM demonstrated high maximum limits for the validation dataset parameters (red box).\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/dcd417c63025d1c944eaa88a.png"},{"id":71689014,"identity":"cfb3deaa-ae1b-4cc0-8474-050f2f553139","added_by":"auto","created_at":"2024-12-17 17:48:42","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1687835,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eProcess of building the best predictive model.\u003c/strong\u003e (A) High-performance feature combinations for three classifiers: 81 features for logistic regression, 71 features for softmax and 25 features for support vector machine (SVM). The training (80%) and testing (20%) dataset underwent 10-fold cross-validation. Blue violin plots represent testing datasets; red violin plots represent validation datasets. (B) Optimal arrangement of training (80%) and testing (20%) datasets. Among the 10 cross-validation runs, specific arrangements of training (80%) and testing (20%) datasets achieved the best performance for each of the three predictive models. (C) Area under the curve of the receiver operating characteristic (ROC AUC) for the three high-performance dataset arrangement models using all 164 patients. (D) Convergence of model performance as features are added. The best and most balanced performance was achieved using all 81 features.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/733a55eee1a66e36f8ef03ff.png"},{"id":75864084,"identity":"890c8107-2ddb-4357-8b20-d4405d03221a","added_by":"auto","created_at":"2025-02-10 05:41:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7932366,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/47db6ab4-e746-49d8-9581-fcef7d10b723.pdf"},{"id":71689015,"identity":"3f0a18db-7280-4a48-9e3d-d3d9ecd4cd24","added_by":"auto","created_at":"2024-12-17 17:48:42","extension":"xlsx","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":80070,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-5370589/v1/4c151bf88a630498662712ba.xlsx"}],"financialInterests":"No competing interests reported.","formattedTitle":"A machine learning model for predicting outcomes of MitraClip therapy","fulltext":[{"header":"1 Background","content":"\u003cp\u003eMitral valve regurgitation (MR) is the most common valvular heart condition, often caused by degenerative changes in the aging population and linked to increased mortality rates [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. While surgery is a treatment option, about half of the patients are ineligible due to factors like frailty, comorbidities, or left ventricular dysfunction [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. A viable alternative is transcatheter edge-to-edge mitral valve repair (TEER) using the MitraClip delivery system (Abbott Vascular, Abbott Park, IL, USA). The effectiveness of this approach has been evaluated in clinical trials [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe ability to predict outcomes after TEER can enhance candidate selection and provide patients with clearer expectations for their short- and long-term prognosis. Currently, the success of MitraClip intervention largely depends on the cardiologist's expertise. Several potential predictors for MR recurrence and major adverse events include clinical, echocardiographic, and procedural factors such as age, preoperative ventricular function, and procedural complexity [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, the discussion around echocardiographic and procedural indicators associated with MR recurrence continues, highlighting the need for a deeper understanding of the underlying causes of MR recurrence [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Recently, the use of machine learning algorithms, a subset of artificial intelligence, has gained prominence in handling the extensive diversity of clinical data and has shown promising results across various medical fields [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Machine learning techniques are powerful analytical tools capable of recognizing complex data patterns for categorization. Many machine learning models have demonstrated superior ability to accurately identify high-risk patients for mortality, outperforming traditional statistical methods in classification and prognosis in general healthcare settings [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, previous studies have not explored the potential of these methodologies to understand the factors contributing to unsuccessful MitraClip interventions.\u003c/p\u003e \u003cp\u003eWe hypothesized that machine learning approaches could identify new potential predictors and improve the comprehensive prediction of MR recurrence and major adverse events. The primary objective of this study was to construct a machine learning based prognostic model that enhances the interpretability of results and provides a deeper understanding of the underlying dataset.\u003c/p\u003e"},{"header":"2 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Patient Cohorts\u003c/h2\u003e \u003cp\u003eWe enrolled 164 patients (aged 18 years or older) with symptomatic moderate-to-severe or severe MR who met the echocardiographic inclusion criteria and were deemed inoperable or at high surgical risk by the heart team, with a life expectancy exceeding one year [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Exclusion criteria included patients with: (a) mitral stenosis requiring surgical intervention due to conditions such as rheumatic heart disease, endocarditis, or congenital causes; (b) simultaneous aortic valve replacement or repair, tricuspid valve replacement, or previous mitral, aortic, or tricuspid valve surgery; (c) concomitant congenital conditions (except for closure of patent foramen ovale) and bypass for obstructive coronary disease; (d) severe comorbid organ dysfunction. Patients requiring tricuspid valve repair for functional tricuspid regurgitation were not excluded from the study and met the same inclusion and exclusion criteria as the training cohort.\u003c/p\u003e \u003cp\u003eAfter evaluation by the Heart Team, all 164 patients underwent TEER using the MitraClip system, with procedural details previously described [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Echocardiography\u003c/h2\u003e \u003cp\u003eSkilled cardiologists conducted thorough transthoracic echocardiography evaluations on each patient before the procedure, immediately after the procedure, and during follow-up. These assessments were performed using either the EPIQ 7C or EPIQ CVX ultrasound platforms (Philips, Bothell, WA, USA). Echocardiographic imaging followed established guidelines for assessing valvular heart disease [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Left ventricular end-diastolic and end-systolic volumes, left ventricular ejection fraction, and left atrial volume were calculated using the biplane Simpson's method[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The severity of MR was graded into four categories (mild\u0026thinsp;=\u0026thinsp;1+, moderate\u0026thinsp;=\u0026thinsp;2+, moderate to severe\u0026thinsp;=\u0026thinsp;3+, severe\u0026thinsp;=\u0026thinsp;4+) and quantified by the effective regurgitant orifice area, regurgitant volume, and regurgitant fraction according to guidelines [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. To determine the etiology and mechanism of regurgitation (primary or secondary, using Carpentier nomenclature), we assessed the anatomical characteristics of the mitral valve to evaluate suitability for TEER. This included measurements of the mitral valve area, mitral annulus diameter, coaptation length and depth, as well as leaflet thickness and length. Additionally, the function of the aortic and tricuspid valves was evaluated, and systolic pulmonary arterial pressure (PAPS) was calculated.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Prognostic follow-up\u003c/h2\u003e \u003cp\u003eUp to February 2024, follow-up information was collected through medical assessments, echocardiographic examinations, and telephone inquiries. The mean follow-up duration was 281 days (ranging from 30 to 872 days), with a 100% follow-up completion rate. The primary endpoint was the recurrence of residual MR (graded as 2\u0026thinsp;+\u0026thinsp;or higher) during the follow-up period. Early-onset mitral regurgitation was defined as occurring within the first 3 days post-procedure, while late-onset mitral regurgitation was identified as occurring after the initial follow-up period. Grading of residual mitral regurgitation followed the guidelines set by the American Society of Echocardiography [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Secondary clinical endpoints included major adverse events such as mitral valve stenosis (defined as a mean pressure difference\u0026thinsp;\u0026gt;\u0026thinsp;5 mmHg), overall mortality, and failed MitraClip intervention necessitating surgical repair or replacement of the mitral valve. Additionally, causes of death were determined through a detailed review of medical records.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Data process and model construction\u003c/h2\u003e \u003cp\u003eThe study divided the 164 patients who underwent MitraClip intervention into two groups: 147 patients for model training and testing, and 17 patients for validation of the predictive model's efficiency. An endpoint prediction model was constructed using imaging and clinical features through machine learning methods. A total of 81 features from the 147 patients were used to establish the machine learning model. A 10-fold cross-validation test was applied during model training. The MR patient dataset was randomly divided into an 80% training dataset and a 20% testing dataset, and this process was repeated 10 times. Then, various machine learning classifiers and feature combinations were tested for each arrangement of the training and testing datasets. The performance consistency between the testing and validation sets was evaluated to select the best models. The workflow was shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5 Machine learning classifiers\u003c/h2\u003e \u003cp\u003eIn this study, we employed a comprehensive range of machine learning classifiers to identify the best-performing algorithm. These classifiers are based on distinct mathematical principles and logical frameworks and can be categorized into various types: linear regression algorithms, tree-based algorithms, kernel-based learning algorithms, et al. After rigorous testing and evaluation, we employed 14 diverse machine learning classifiers, including: tree-based algorithms (such as adaBoost, catBoost, extra trees, gradient boosting, lightGBM, random forest, and XGBoost), linear regression algorithms (such as logistic regression, ridge regression, and softmax regression), kernel-based support vector machine (SVM), instance-based K-nearest neighbors (KNN), and probabilistic models (naive Bayes and quadratic discriminant analysis).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003e2.5.1 Tree-based algorithms\u003c/h2\u003e \u003cp\u003eFirstly, we focused on a selection of sophisticated tree-based algorithms. AdaBoost enhances model performance by focusing on difficult-to-classify samples. CatBoost excels in handling categorical data without extensive preprocessing. Extra trees algorithm introduces more randomness in selecting split points, offering enhanced robustness. Gradient boosting sequentially corrects errors from previous models, reducing bias and variance. LightGBM optimizes for large-scale data with its efficient histogram-based algorithm. Random Forest aggregates multiple decision trees to mitigate overfitting. XGBoost stands out for its performance and speed across diverse datasets. These models collectively are powerful tools for addressing complex prediction tasks with high accuracy and efficiency.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e2.5.2 Linear regression models\u003c/h2\u003e \u003cp\u003eNext, we turned to linear models to predict outcomes by establishing linear relationships between features and target variables. Logistic regression excels at binary classification, providing probabilities for two distinct classes. To address model complexity and prevent overfitting, we employed ridge regression, which introduces \u003cem\u003eL2\u003c/em\u003e regularization into the linear regression framework, enhancing stability and performance. For multi-class classification, we employed softmax regression, an extension of logistic regression that efficiently assigns probabilities across multiple classes. This suite of linear models demonstrates their simplicity and adaptability, offering a solid foundation for both binary and multi-class prediction tasks with precision.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e2.5.3 Kernel-based methods\u003c/h2\u003e \u003cp\u003eWe also used a kernel-based method, the SVM, which excels at finding optimal boundaries between different categories. SVM is particularly notable for its ability to perform classification in higher-dimensional spaces by applying kernel tricks. This method effectively manages complex datasets where linear separability is not feasible, making it a powerful tool for nuanced classification tasks.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e \u003ch2\u003e2.5.4 Instance-based learning\u003c/h2\u003e \u003cp\u003eFor instance-based learning, we adopted the KNN algorithm due to its simplicity and effectiveness. KNN predicts the category of new samples based on the closest training examples in the feature space, bypassing the need for an explicit model training phase. This method excels in its straightforwardness and the intuitive principle that similar samples are likely to belong to the same class, making it a valuable component of our classification toolkit.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003e2.5.5 Probabilistic models\u003c/h2\u003e \u003cp\u003eLastly, we explored probabilistic models to leverage probability theories for classification and prediction. Naive Bayes, a simplistic yet efficient probabilistic classifier, operates on the principle of feature independence as grounded in Bayes' theorem. Additionally, quadratic discriminant analysis incorporates quadratic terms to fit data distributions, making it suitable for scenarios where linear separability assumptions are inadequate. These probabilistic models provide a robust framework for understanding and predicting outcomes based on underlying probabilities, enriching our analytical toolkit.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Probabilistic models\u003c/h2\u003e \u003cp\u003eThe dataset of 147 samples was divided into 10 folds for training and testing, with an 8:2 split for each fold. During model training, the folds were rotated 10 times between the training and testing sets to account for variations in model accuracy. This rotation led to fluctuations in model performance. Our goal was to identify the optimal combination of training and testing sets that consistently achieved high accuracy in both the testing and validation datasets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e2.7 Feature selection\u003c/h2\u003e \u003cp\u003eEighty-one clinical and echocardiographic features were used to train and optimize the prediction models. The importance of the 81 variables was assessed using the XGboost classifier, while the mean decrease accuracy (MDA) was evaluated using random forest. Both importance and MDA values were co-evaluated for each variable. The top-ranked features were incrementally added to the model based on their priority. The model was then trained with these evolving features, and classification accuracy was monitored to observe the accuracy curve and ensure model convergence. Finally, an optimal combination of features was identified to maximize classification efficiency.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e2.8 Statistical analysis\u003c/h2\u003e \u003cp\u003eNormally distributed variables were compared between groups using an independent-sample t test, while non-normally distributed variables were assessed with the Mann-Whitney \u003cem\u003eu\u003c/em\u003e test. Categorical data were analyzed using the chi-square test or Fischer\u0026rsquo;s exact test. Statistical significance was defined as \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05. All statistical analyses were performed using SPSS version 23.0 (IBM Corp, Armonk, NY).\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Clinical and echocardiographic information of patients\u003c/h2\u003e \u003cp\u003eA total of 164 patients (mean age 68.4\u0026thinsp;\u0026plusmn;\u0026thinsp;10.5 years; 58 women, 35.4%) with \u0026ge;\u0026thinsp;3\u0026thinsp;+\u0026thinsp;MR (3+: 28.7%, 4+: 71.3%) were enrolled in this study. Post-MitraClip intervention, all patients had their MR reduced to \u0026le;\u0026thinsp;2+. Follow-up was completed by 100% of the participants. Baseline and follow-up characteristics are detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Common comorbidities included hypertension (46.3%), hyperlipemia (22.0%), atrial fibrillation (28.7%), diabetes mellitus (17.7%), coronary artery disease (17.7%) and so on. The mean plasma NT-proBNP value was 2328.5\u0026thinsp;\u0026plusmn;\u0026thinsp;4227.9 pg/mL. During follow-up, significant reductions were observed in systolic blood pressure, diastolic blood pressure, plasma hemoglobin, low-density lipoprotein cholesterol, and NT-proBNP levels (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, P\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSummary of clinical and echocardiographic characteristics of the subjects.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePre-operation (n\u0026thinsp;=\u0026thinsp;164)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFollow-up (n\u0026thinsp;=\u0026thinsp;164)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSignificance\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGeneral information\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender-Female, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e58 (35.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge-year, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e68.4\u0026thinsp;\u0026plusmn;\u0026thinsp;10.5 (28\u0026ndash;92), 69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody surface area (BSA)-m\u003csup\u003e2\u003c/sup\u003e, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.73\u0026thinsp;\u0026plusmn;\u0026thinsp;0.19 (1.34\u0026ndash;2.15), 1.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBody mass index (BMI)-kg/m\u003csup\u003e2\u003c/sup\u003e, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24.0\u0026thinsp;\u0026plusmn;\u0026thinsp;3.6 (13.8\u0026ndash;33.2), 24.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical information\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHypertension, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e76 (46.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmoking, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39 (23.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHyperlipemia, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36 (22.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes mellitus, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29 (17.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoronary artery disease, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29 (17.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoronary Artery Bypass Grafting, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8 (4.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValve surgery, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (3.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAtrial fibrillation, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47 (28.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (27.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.273\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSystolic blood pressure-mmHg, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e121.4\u0026thinsp;\u0026plusmn;\u0026thinsp;18.0 (79\u0026ndash;172), 120.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e113.2\u0026thinsp;\u0026plusmn;\u0026thinsp;12.6 (70\u0026ndash;143), 112\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.54\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiastolic blood pressure-mmHg, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71.9\u0026thinsp;\u0026plusmn;\u0026thinsp;12.6 (46\u0026ndash;143), 71.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e62.5\u0026thinsp;\u0026plusmn;\u0026thinsp;9.1 (44\u0026ndash;111), 62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.72\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;13\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHaemoglobin-g/L, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max) Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e134.0\u0026thinsp;\u0026plusmn;\u0026thinsp;20.3 (79\u0026ndash;177), 135\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e119.5\u0026thinsp;\u0026plusmn;\u0026thinsp;21.8 (60\u0026ndash;173), 120\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.38\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;9\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCreatinine-umol/L, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e99.2\u0026thinsp;\u0026plusmn;\u0026thinsp;54.5 (37.3-419.26), 86.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e108.2\u0026thinsp;\u0026plusmn;\u0026thinsp;71.7 (36.4-589.89), 88.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUric acid-umol/L, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e436.1\u0026thinsp;\u0026plusmn;\u0026thinsp;159.7 (134.14-1372.3), 408.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e406.6\u0026thinsp;\u0026plusmn;\u0026thinsp;159.1 (84.58.05-1217.6), 390.37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHomocysteine-umol/L, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e18.5\u0026thinsp;\u0026plusmn;\u0026thinsp;7.7 (6.56\u0026ndash;51.21), 16.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21.4\u0026thinsp;\u0026plusmn;\u0026thinsp;8.3 (1.42\u0026ndash;50.34), 20.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLow-density lipoprotein cholesterol (LDL-C)-mmol/L, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.55\u0026thinsp;\u0026plusmn;\u0026thinsp;0.95 (0.87\u0026ndash;5.52), 2.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.28\u0026thinsp;\u0026plusmn;\u0026thinsp;0.92 (0.52\u0026ndash;5.96), 2.20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.011\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNT proBNP-pg/mL, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2328.5\u0026thinsp;\u0026plusmn;\u0026thinsp;4227.9 (12.6-36763), 836\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e685.6\u0026thinsp;\u0026plusmn;\u0026thinsp;926.9 (6.6\u0026ndash;5465), 334.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.55\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConventional Echocardiographic parameters\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft atrial anteroposterior diameter (LAD)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e48.7\u0026thinsp;\u0026plusmn;\u0026thinsp;9.2 (33\u0026ndash;89), 47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e44.3\u0026thinsp;\u0026plusmn;\u0026thinsp;8.7 (27\u0026ndash;76), 43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.72\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;5\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft atrial volume (LAV)-mL, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e119.1\u0026thinsp;\u0026plusmn;\u0026thinsp;60.5 (42\u0026ndash;413), 102.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e94.8\u0026thinsp;\u0026plusmn;\u0026thinsp;58.3 (28\u0026ndash;366), 76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.39\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;4\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft ventricular end diastolic diameter (LVDd)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e57.6\u0026thinsp;\u0026plusmn;\u0026thinsp;7.4 (37\u0026ndash;81), 57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e53.1\u0026thinsp;\u0026plusmn;\u0026thinsp;7.8 (35\u0026ndash;79), 52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.84\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;7\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft ventricular end systolic diameter (LVDs)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38.8\u0026thinsp;\u0026plusmn;\u0026thinsp;9.4 (23\u0026ndash;73), 37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37.2\u0026thinsp;\u0026plusmn;\u0026thinsp;9.4 (20\u0026ndash;72), 34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.127\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft ventricular ejection fraction (LVEF)-%, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e59.3\u0026thinsp;\u0026plusmn;\u0026thinsp;12.6 (28\u0026ndash;81), 63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e56.4\u0026thinsp;\u0026plusmn;\u0026thinsp;12.7 (9\u0026ndash;79), 60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight ventricular enddiastolic diameter (RVEDD)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e25.9\u0026thinsp;\u0026plusmn;\u0026thinsp;4.3 (17\u0026ndash;44), 26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e25.6\u0026thinsp;\u0026plusmn;\u0026thinsp;5.1 (2\u0026ndash;60), 25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.665\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMitral E wave, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.32\u0026thinsp;\u0026plusmn;\u0026thinsp;0.35 (0.5\u0026ndash;2.1), 1.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.40\u0026thinsp;\u0026plusmn;\u0026thinsp;0.44 (0.5\u0026ndash;3.2), 1.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.072\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTricuspid E wave, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.532\u0026thinsp;\u0026plusmn;\u0026thinsp;0.138 (0.3\u0026ndash;1.2), 0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.570\u0026thinsp;\u0026plusmn;\u0026thinsp;0.151 (0.3\u0026ndash;1.3), 0.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSystolic pulmonary artery pressure (SPAP)-mmHg, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e42.2\u0026thinsp;\u0026plusmn;\u0026thinsp;14.2 (18\u0026ndash;90), 41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e35.4\u0026thinsp;\u0026plusmn;\u0026thinsp;12.8 (6\u0026ndash;92), 32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6.81\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTricuspid regurgitation velocity-m/s, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.96\u0026thinsp;\u0026plusmn;\u0026thinsp;0.54 (1.8\u0026ndash;4.5) 2.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.67\u0026thinsp;\u0026plusmn;\u0026thinsp;0.54 (0.6\u0026ndash;4.5), 2.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.13\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMitral valve etiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMitral valve area (MVA)-cm\u003csup\u003e2\u003c/sup\u003e, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e5.98\u0026thinsp;\u0026plusmn;\u0026thinsp;1.47 (3.2\u0026ndash;14.3), 5.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnterior mitral valve leaflet (AML)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e27.2\u0026thinsp;\u0026plusmn;\u0026thinsp;4.3 (11\u0026ndash;38), 27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePosterior mitral valve leaflet (PML)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e15.9\u0026thinsp;\u0026plusmn;\u0026thinsp;4.2 (8\u0026ndash;36), 15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMitral valve leaflet thickening, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e111 (67.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMitral annulus calcification (MAC), n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13 (7.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAntero-posterior mitral annulus diameter (MAD)-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33.1\u0026thinsp;\u0026plusmn;\u0026thinsp;5.0 (21\u0026ndash;49), 33\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMedio-lateral mitral annulus diameter-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e33.2\u0026thinsp;\u0026plusmn;\u0026thinsp;4.6 (15\u0026ndash;46), 33.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVena contracta-mm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6.71\u0026thinsp;\u0026plusmn;\u0026thinsp;1.66 (2.3\u0026ndash;13), 6.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProximal isovelocity surface area (PISA)-cm, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.946\u0026thinsp;\u0026plusmn;\u0026thinsp;0.325 (0.4-3), 0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEffective regurgitant orifice area-cm\u003csup\u003e2\u003c/sup\u003e, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.484\u0026thinsp;\u0026plusmn;\u0026thinsp;0.340 (0.04\u0026ndash;2.49), 0.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegurgitant volume-mL/beat, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e66.7\u0026thinsp;\u0026plusmn;\u0026thinsp;40.3 (7-259), 58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDegree of regurgitant, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e38 (23.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.00\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;58\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e85 (51.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28 (17.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47 (28.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8 (4.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e117 (71.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5 (3.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEtiology, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDegenerative mitral regurgitant (DMR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e108 (65.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFunctional mitral regurgitant (FMR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e47 (28.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBoth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9 (5.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMean pressure gradient-mmHg, mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD (min-max), Median\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.16\u0026thinsp;\u0026plusmn;\u0026thinsp;1.09 (1\u0026ndash;6), 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4.02\u0026thinsp;\u0026plusmn;\u0026thinsp;2.72 (1\u0026ndash;23), 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.26\u0026times;10\u003csup\u003e\u0026minus;\u0026thinsp;14\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Building the MR recurrence risk prediction model\u003c/h2\u003e \u003cp\u003eWe developed and optimized a high-performance predictive model at three levels: (a) selecting an appropriate combination of features to avoid underfitting nor overfitting; (b) identifying the best-performing classifier by testing various machine learning classifiers; (c) establishing the most representative training set to achieve high predictive accuracy in both testing and validating datasets.\u003c/p\u003e \u003cp\u003eInitially, we tested the appropriate combination of features by incrementally adding the most important features to the model. The overall set of 81 features comprised clinical variables obtained before and during the MitraClip intervention, aimed at predicting prognosis. These features were ranked based on MDA and importance values (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The sequences of MDA and importance were simultaneously considered when adding features to the predictive model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAll 14 machine learning classifiers were tested by incrementally adding features from 1 to 81, using 10-fold cross-validation between training (80%) and testing (20%) datasets. Predictive parameters, including precision, recall, f1 score for both positive and negative patients, and model accuracy, were analyzed and presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. Both mean and maximum limits were considered for selecting superior classifiers. Positive recall, being the lowest, reflects the model\u0026rsquo;s performance. LightGBM, logistic regression, ridge regression, softmax, SVM, and XGBoost demonstrated higher positive recall values compared to other classifiers in the testing dataset. To build the high-performance model, logistic regression, softmax, and SVM were selected from these six classifiers for their superior performance in the validation dataset, as their maximum parameter limits were higher (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe best-performance combination of features for each of the three selected classifiers, logistic regression, softmax and SVM, was presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA. For example, using all 81 features in the logistic regression model (logistic regression-81) with 10-fold cross-validation on an 80% training and 20% testing dataset, specificity increased to approximately 0.797 (0.739, 0.854) and sensitivity to about 0.459 (0.370, 0.549), resulting in an overall accuracy of 0.688 (0.632, 0.745) for the validation dataset. In one of the 10 cross-validation runs, a particular arrangement of the training (80%) and testing (20%) dataset (the best model of logistic regression-81) achieved a specificity of 0.900 and a sensitivity of 0.571, resulting in an accuracy of 0.765 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB). The overall ROC AUC for the best models of logistic regression-81, softmax-71, and SVM-25 using all 164 patients is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC. Among the three classifiers, logistic regression demonstrated the highest diagnostic efficacy (overall ROC AUC: 0.909; test and validation dataset ROC AUC: 0.774). Ultimately, the logistic regression-81 model was selected for its superior performance. This model's parameters improved and converged as more features were added, achieving the best and most balanced performance for both the testing and validation datasets when using all 81 features (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eD).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance of three superior classifiers, each with a specific combination of features.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eClassifiers-No. of features, mean (95% CI); min-max\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStatistics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLogistic regression-81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSoftmax-71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSupport vector machine-25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTesting set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.683 (0.639, 0.727); 0.567\u0026ndash;0.767\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.713 (0.664, 0.763); 0.600\u0026ndash;0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.553 (0.418, 0.689); 0.267-0.800\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_precision (Specificity)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.772 (0.731, 0.813); 0.682\u0026ndash;0.857\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.780 (0.728, 0.833); 0.667\u0026ndash;0.905\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.684 (0.577, 0.791); 0.455-0.900\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.765 (0.692, 0.837); 0.565\u0026ndash;0.938\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.805 (0.735, 0.875); 0.609\u0026ndash;0.905\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.537 (0.333, 0.741); 0.095\u0026ndash;0.833\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_f1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.763 (0.725, 0.802); 0.667\u0026ndash;0.821\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.788 (0.742, 0.835); 0.686\u0026ndash;0.864\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.580 (0.405, 0.756); 0.160\u0026ndash;0.857\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_precision (Sensitivity)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.516 (0.395, 0.636); 0.286\u0026ndash;0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.571 (0.480, 0.662); 0.308\u0026ndash;0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.410 (0.301, 0.518); 0.158\u0026ndash;0.625\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.518 (0.441, 0.595); 0.333-0.700\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.527 (0.447, 0.606); 0.333\u0026ndash;0.714\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.568 (0.447, 0.689); 0.286\u0026ndash;0.778\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_f1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.504 (0.427, 0.582); 0.353\u0026ndash;0.667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.535 (0.473, 0.598); 0.400-0.667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.457 (0.364, 0.551); 0.214\u0026ndash;0.667\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eValidation set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.688 (0.632, 0.745); 0.588\u0026ndash;0.824\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.588 (0.522, 0.654); 0.471\u0026ndash;0.765\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.547 (0.430, 0.664); 0.353\u0026ndash;0.765\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_precision (Specificity)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.797 (0.739, 0.854); 0.692\u0026ndash;0.909\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.718 (0.658, 0.777); 0.615-0.900\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.718 (0.646, 0.790); 0.571\u0026ndash;0.833\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.758 (0.724, 0.792); 0.667\u0026ndash;0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.700 (0.636, 0.764); 0.583\u0026ndash;0.917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.558 (0.385, 0.732); 0.250\u0026ndash;0.833\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0_f1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.775 (0.738, 0.813); 0.696\u0026ndash;0.870\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.705 (0.658, 0.752); 0.636\u0026ndash;0.818\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.611 (0.476, 0.745); 0.353\u0026ndash;0.833\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_precision (Sensitivity)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.459 (0.370, 0.549); 0.250\u0026ndash;0.667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.262 (0.129, 0.396); 0.000-0.571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.368 (0.271, 0.466); 0.200\u0026ndash;0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.520 (0.366, 0.674); 0.200\u0026ndash;0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.320 (0.139, 0.501); 0.000-0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.520 (0.420, 0.620); 0.400\u0026ndash;0.800\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1_f1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.484 (0.368, 0.599); 0.222\u0026ndash;0.727\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.287 (0.134, 0.440); 0.000-0.667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.415 (0.338, 0.491); 0.267-0.600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Clinical outcomes\u003c/h2\u003e \u003cp\u003eOur model demonstrated high specificity and sensitivity in predicting events after MitraClip intervention in MR patients. Follow-up completeness was 100%. During the follow-up period, the composite endpoint occurred in 53 patients (32.3%). These endpoints included the recurrence of moderate MR (2\u0026thinsp;+\u0026thinsp;or higher), mitral valve stenosis (mean pressure difference\u0026thinsp;\u0026gt;\u0026thinsp;5 mmHg), all-cause mortality, and TEER failure requiring subsequent mitral valve surgical repair or replacement. In this study, there were 4 deaths (3 cardiac and 1 non-cardiac). Causes of death included sudden cardiac death (1 patient), malignant arrhythmia (2 patients), and other (1 patient). Recurrence rates of severe MR (3\u0026thinsp;+\u0026thinsp;or higher) were 3.0% at 30 days and 4.9% at three years.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eThe recurrence of MR following mitral valve intervention has been linked to a notable increase in long-term mortality rates and a higher risk of hospital readmission due to heart failure during the follow-up period [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. In our study, severe recurrent MR after MitraClip intervention occurred in 7.9% of patients. In comparison, other studies reported recurrence rates of 18% in Everest 2 and 21% in ACESS-EU at one year [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Recently, various factors with prognostic significance following mitral valve interventions have been increasingly recognized. These factors encompass demographic, clinical, anatomic, and procedural features. The complexity of assessing recurrent MR is heightened by the large number of variables involved, making it difficult for clinicians to assess the risks for individual patients comprehensively. Despite this, predictors of MR recurrence remain poorly defined [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Previous studies on predicting MR recurrence after MitraClip intervention mainly relied on classic statistical modeling techniques constrained by assumptions such as distribution normality, non-informative or random censoring, and hazard risk linearity [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. However, these traditional methods usually focus one or a few clinical features and overlook the potential effects of complex and hidden interactions among several weaker predictors. Machine learning algorithms, a subfield of artificial intelligence, can overcome these limitations by capturing high-dimensional nonlinear relationships among many clinical features [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Recently, several machine learning models have been introduced to handle the significant variability in clinical data and have demonstrated efficacy in various medical applications for cardiovascular diseases [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn our study, we employed logistic regression classifiers due to their state-of-the-art accuracy and interpretability. We used softmax regression because it is a popular method for handling high-dimensional predictors. Logistic regression was particularly useful for building risk models to address binary classification problems. Therefore, we utilized these methods for variable selection and risk model construction, ultimately choosing the best-performing model. Logistic regression is particularly suited for small datasets due to its computational efficiency. Compared to decision tree models, logistic regression offers several advantages, including faster runtime, better handling of outliers, greater flexibility in data processing, and superior performance. These benefits make logistic regression preferable, especially when dealing with high model complexity relative to dataset size. This efficiency and reduced generalization error are crucial in such scenarios.\u003c/p\u003e \u003cp\u003eThis study is likely the first to develop a predictive model for MitraClip intervention using echocardiography integrated with machine learning. The superior prediction model demonstrated an ROC AUC value of 0.909, with a sensitivity of 0.900 and specificity of 0.571 in the test group, and an ROC AUC value of 0.773 in the testing and validation dataset. Logistic regression exhibited excellent discriminatory performance in predicting the recurrence of MR and major adverse cardiovascular events. These findings highlight the potential of machine learning in assessing prognostic risk for MR patients undergoing MitraClip intervention. Accurately predicting the timeline for significant recurrent MR and major adverse cardiovascular events based on preoperative clinical and echocardiographic parameters is complex. Our results support expanding indications and refining patient selection criteria for those undergoing MitraClip intervention.\u003c/p\u003e \u003cp\u003eIn our machine learning model, we found various clinical, anatomic, and procedural factors linked to recurrent MR following MitraClip intervention. Conversely, traditional cardiovascular risk factors (such as hypertension, diabetes mellitus, and chronic obstructive pulmonary disease) and specific parameters used to evaluate MR severity (such as proximal isovelocity surface area, effective regurgitation area, and regurgitant volume) did not emerge as the most predictive factors for the endpoint in this specific patient population.\u003c/p\u003e \u003cp\u003eFirst, a history of surgery is a predictor of recurrent MR after MitraClip intervention. Similar findings from the TRAMI registry and Boerlage-van Dijk indicate that patients with previous valve surgery have a poor prognosis following transcatheter edge-to-edge mitral valve repair [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Second, certain serum biomarkers reflecting pathophysiological states are related to clinical outcomes. Our machine learning model identified serum creatinine, which reflects renal function, as associated with worse outcomes after MitraClip intervention. Chronic kidney insufficiency can cause myocardial damage due to changes in cardiac structure and function. Similarly, the TRAMI registry demonstrated that baseline serum creatinine\u0026thinsp;\u0026ge;\u0026thinsp;1.5 mg/dl is an independent predictor of 1-year mortality (HR: 1.77; p\u0026thinsp;=\u0026thinsp;0.002) [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. In addition, NT-proBNP levels independently correlated with recurrent MR, reflecting the volumetric overload of the left ventricular and consequences of myocardial systolic dysfunction. Triantafyllis et al. demonstrated a similar correlation between NT-proBNP levels and cardiac mortality after TEER (n\u0026thinsp;=\u0026thinsp;136; HR: 1.5; 95% CI: 1.1 to 2.1; p\u0026thinsp;=\u0026thinsp;0.018) [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Thus, individuals with brief follow-up durations should regulate their fluid intake and use medications known to enhance myocardial systolic function [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Additionally, hemoglobin levels play a crucial role in postoperative prognosis. Decreased hemoglobin can lead to reduced blood oxygen-carrying capacity and ventricular dysfunction, emphasizing the importance of minimizing bleeding during interventions. Third, heart chamber volumes are significant predictors of outcomes. All cases without significant MR exhibited favorable remodeling of left atrial volume and left ventricular volume at follow-up. Left atrial volume and left ventricular endsystolic volume are predictors of outcomes after clip implantation, as reinforced by recent studies assessing TEER with secondary MR and the GRASP-IT registry [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Increased left atrial volume may lead to atrial fibrillation, while enlarged left ventricular endsystolic volume is associated with advanced cardiac remodeling and left ventricular dysfunction. Fourth, our model showed that higher pulmonary hypertension predicted MR recurrence (\u0026ge;\u0026thinsp;3+) after MitraClip intervention. Pulmonary hypertension, defined as pressures exceeding 50 mmHg, indicates advanced cardiomyopathy and is predictive of MR recurrence following TEER [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Fifth and importantly, we identified operation time and residual MR (post-clip) as independent predictors of MR recurrence after MitraClip intervention. Although MitraClip intervention is a safe procedure with a relatively low complication rate, longer operation times may lead to procedural complications such as acute heart failure, pulmonary embolism, stroke, pericardial effusion, pericardial tamponade, cardiogenic shock, and bleeding requiring blood transfusions.\u003c/p\u003e \u003cp\u003eHowever, our model highlighted that only two mitral valve anatomical structures (medio-lateral mitral annulus diameter and mitral valve area), were significant for predicting outcomes. Increased annular distortion was associated with a high rate of clip failure due to progressive left ventricular remodeling [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis study has several constraints. Primarily, the applicability and efficacy of the machine learning methodologies were verified within a single-center cohort, and the examination was conducted post hoc, with most limitations stemming from the retrospective nature of the dataset. The relatively small number of patients limits the ability to detect statistically significant differences in clinical outcomes in the validation group. We acknowledge the limited sample size and the lack of multivariable analysis (e.g., due to the limited number of events), which limits the strength of the results. A multi-center study will be conducted in the future as more hospitals adopt MitraClip intervention. Furthermore, assessing MR grading following a double orifice repair presents challenges. Nonetheless, the severity of MR post-MitraClip intervention was evaluated using the integrative approach advised by the guidelines18, allowing for better reproducibility. Moreover, the sensitivity (~\u0026thinsp;0.5) of the predictive model remains unsatisfactory, restricted by the limited sample size and inadequate follow-up period. Nearly half of the patients were followed up within 100 days, which is a short period to observe the recurrence of MR after MitraClip intervention. Some recurrences or events may not have yet occurred, preventing the classifiers from accurately capturing the typical characteristics of positive prognosis.\u003c/p\u003e"},{"header":"5 Conclusions","content":"\u003cp\u003eOur research demonstrates significant advantages of our model in predicting the prognosis of mitral valve interventions: (1) high predictive accuracy in determining outcomes, validated by the cohort; and (2) reliance solely on non-invasive echocardiographic assessments and common clinical parameters. In conclusion, our innovative model presents a more effective and non-invasive approach to forecasting the prognosis of mitral valve interventions. This model could serve as a valuable tool for advancing clinical understanding and enhancing criteria for selecting appropriate candidates for mitral valve interventions.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emitral regurgitation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTEER\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003etranscatheter edge-to-edge mitral valve repair\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSVM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003esupport vector machine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eKNN\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003einstance-based K-nearest neighbors\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePAPS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003epulmonary arterial pressure\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eMDA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003emean decrease accuracy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eThis prospective clinical trial was conducted from January 2021 to February 2024 at Fuwai Hospital in Beijing, China. The study adhered to the Declaration of Helsinki, received approval from the Fuwai Hospital Ethics Committee (reference number: 2024BJYYEC-KY067-01), and all participants provided written informed consent.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare that they have no conflict of interest.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis study was financially supported by the National High Level Hospital Clinical Research Funding (2023-GSP-QN-40, 2022-GSP-PT-7, 2022-GSP-QN-18), the Clinical and Translational Medicine Research Program of the Chinese Academy of Medical Sciences (2023-I2M-C\u0026amp;T-B-056, 2023-I2M-C\u0026amp;T-B-117) and CAMS Innovation Fund for Medical Sciences (2021-I2M-1-065).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eHui Li and Ying Guo contributed significantly to writing the manuscript. Hui Li and Junsong Gong analyzed the echocardiograms. Yiran Hu and Fengwen Zhang collected patient information. Fujian Duan and Xiangbin Pang revised the manuscript, conceived the study and supervised the project. All authors read and approved of the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe thank Yiliang Wei and Yongqiang Kong for their assistance in constructing and optimizing predictive models. Hui Li had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eData is provided within the supplementary information files.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eNkomo, V. T. et al. Burden of valvular heart diseases: a population-based study. :368. (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDziadzko, V., MikhailMedina-Inojosa, J. R. B. \u0026amp; GiovanniMichelena, H. I. C. Juan A.Maalouf, JosephThapa, PrabinEnriquez-Sarano, Maurice %J European Heart Journal: The Journal of the European Society of Cardiology. \u003cem\u003eCauses Mech. isolated mitral regurgitation community: Clin. context outcome\u003c/em\u003e ;\u003cb\u003e40\u003c/b\u003e. (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSt Goar, F. G. et al. Endovascular edge-to-edge mitral valve repair: short-term results in a porcine model. \u003cem\u003eCirculation\u003c/em\u003e. \u003cb\u003e108\u003c/b\u003e, 1990\u0026ndash;1993 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMauri, L. et al. 4-year results of a randomized controlled trial of percutaneous repair versus surgery for mitral regurgitation. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e62\u003c/b\u003e, 317\u0026ndash;328 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Bonis, M. et al. Edge-to-edge surgical mitral valve repair in the era of MitraClip: what if the annuloplasty ring is missed? \u003cem\u003eCurr. Opin. Cardiol.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e, 155\u0026ndash;160 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTamborini, G. et al. Predictive Value of Pre-Operative 2D and 3D Transthoracic Echocardiography in Patients Undergoing Mitral Valve Repair: Long Term Follow Up of Mitral Valve Regurgitation Recurrence and Heart Chamber Remodeling. \u003cem\u003eJ. Cardiovasc. Dev. Dis.\u003c/em\u003e ;\u003cb\u003e7\u003c/b\u003e. (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVerma, S., Latter, D. A., Bonow, R. O. \u0026amp; Failed Mitral, T. E. E. R. Are There Lessons for Decision Making? \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e78\u003c/b\u003e, 10\u0026ndash;13 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHassan, A. \u0026amp; Eleid, M. F. Recurrent Mitral Regurgitation After MitraClip: Defining Success and Predicting Outcomes. \u003cem\u003eCirc. Cardiovasc. Interv\u003c/em\u003e. \u003cb\u003e15\u003c/b\u003e, e011837 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson, K. W. et al. Artificial Intelligence in Cardiology. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e71\u003c/b\u003e, 2668\u0026ndash;2679 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSahni, N., Simon, G. \u0026amp; Arora, R. Development and Validation of Machine Learning Models for Prediction of 1-Year Mortality Utilizing Electronic Medical Record Data Available at the End of Hospitalization in Multicondition Patients: a Proof-of-Concept Study. \u003cem\u003eJ. Gen. Intern. Med.\u003c/em\u003e \u003cb\u003e33\u003c/b\u003e, 921\u0026ndash;928 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmbale-Venkatesh, B. et al. Cardiovascular Event Prediction by Machine Learning: The Multi-Ethnic Study of Atherosclerosis. \u003cem\u003eCirc. Res.\u003c/em\u003e \u003cb\u003e121\u003c/b\u003e, 1092\u0026ndash;1101 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaumgartner, H. et al. 2017 ESC/EACTS Guidelines for the management of valvular heart disease. \u003cem\u003eEur. Heart J.\u003c/em\u003e \u003cb\u003e38\u003c/b\u003e, 2739\u0026ndash;2791 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRick, A. et al. 2017 AHA/ACC Focused Update of the 2014 AHA/ACC Guideline for the Management of Patients With Valvular Heart Disease: A Report of the American College of Cardiology/American Heart Association Task Force on. \u003cem\u003eClin. Pract. Guidelines\u003c/em\u003e. \u003cb\u003e70\u003c/b\u003e, 252\u0026ndash;289 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatz, W. E. et al. Echocardiographic evaluation and guidance for MitraClip procedure. :616\u0026thinsp;\u0026ndash;\u0026thinsp;32. (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZoghbi, W. A. et al. Recommendations for Noninvasive Evaluation of Native Valvular Regurgitation, A Report from the American Society of Echocardiography Developed in Collaboration with the Society for. \u003cem\u003eCardiovasc. Magn. Reson.\u003c/em\u003e :4. (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLang, R. M. et al. Recommendations for Cardiac Chamber Quantification by Echocardiography in Adults: An Update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohn, D. et al. Appropriate Use Criteria for Multimodality Imaging in Valvular Heart Disease: A Report of the American College of Cardiology Appropriate Use Criteria Task Force, American Association for Thoracic Surgery, American Heart Association, American Society of Echocardiography, American Society of Nuclear Cardiology, Heart Rhythm Society, Society for Cardiovascular Angiography and Interventions, Society of Cardiovascular Computed Tomography, Society for Cardiovascu. 2017. (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNunes, M. C. P. et al. Role of LA Shape in Predicting Embolic Cerebrovascular Events in Mitral Stenosis: Mechanistic Insights From 3D Echocardiography. (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAdamo, M. et al. Five-year clinical outcomes after percutaneous edge-to-edge mitral valve repair: Insights from the multicenter GRASP-IT registry. \u003cem\u003eAm. Heart J.\u003c/em\u003e \u003cb\u003e217\u003c/b\u003e, 32\u0026ndash;41 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMack, M. J. et al. 3-Year Outcomes of Transcatheter Mitral Valve Repair in Patients With Heart Failure. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e77\u003c/b\u003e, 1029\u0026ndash;1040 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGlower, D. D. et al. Percutaneous mitral valve repair for mitral regurgitation in high-risk patients: results of the EVEREST II study. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e64\u003c/b\u003e, 172\u0026ndash;181 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoekstegers, P. et al. Percutaneous interventional mitral regurgitation treatment using the Mitra-Clip system. \u003cem\u003eClin. Res. Cardiol.\u003c/em\u003e \u003cb\u003e103\u003c/b\u003e, 85\u0026ndash;96 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDey, D. et al. Artificial Intelligence in Cardiovascular Imaging: JACC State-of-the-Art Review. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e73\u003c/b\u003e, 1317\u0026ndash;1335 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeng, S. F. et al. Can machine-learning improve cardiovascular risk prediction using routine clinical data? \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e12\u003c/b\u003e, e0174944 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePuls, M. et al. One-year outcomes and predictors of mortality after MitraClip therapy in contemporary clinical practice: results from the German transcatheter mitral valve interventions registry. \u003cem\u003eEur. Heart J.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 703\u0026ndash;712 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoerlage-vanDijk, K. et al. Predictors of outcome in patients undergoing MitraClip implantation: An aid to improve patient selection. \u003cem\u003eInt. J. Cardiol.\u003c/em\u003e \u003cb\u003e189\u003c/b\u003e, 238\u0026ndash;243 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZuern, C. S. et al. Influence of non-cardiac comorbidities on outcome after percutaneous mitral valve repair: results from the German transcatheter mitral valve interventions (TRAMI) registry. \u003cem\u003eClin. Res. Cardiol.\u003c/em\u003e \u003cb\u003e104\u003c/b\u003e, 1044\u0026ndash;1053 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTriantafyllis, A. S. et al. Long-term survival and preprocedural predictors of mortality in high surgical risk patients undergoing percutaneous mitral valve repair. \u003cem\u003eCatheter Cardiovasc. Interv\u003c/em\u003e. \u003cb\u003e87\u003c/b\u003e, 467\u0026ndash;475 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNishimura, R. A. et al. 2014 AHA/ACC guideline for the management of patients with valvular heart disease: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e63\u003c/b\u003e, e57\u0026ndash;185 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarnak, M. J. et al. Anemia as a risk factor for cardiovascular disease in The Atherosclerosis Risk in Communities (ARIC) study. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e40\u003c/b\u003e, 27\u0026ndash;33 (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaldi, C. et al. Predictors of outcome in heart failure patients with severe functional mitral regurgitation undergoing MitraClip treatment. \u003cem\u003eInt. J. Cardiol.\u003c/em\u003e \u003cb\u003e284\u003c/b\u003e, 50\u0026ndash;58 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCapodanno, D. et al. Predictors of clinical outcomes after edge-to-edge percutaneous mitral valve repair. \u003cem\u003eAm. Heart J.\u003c/em\u003e \u003cb\u003e170\u003c/b\u003e, 187\u0026ndash;195 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaramasso, M. et al. Clinical and anatomical predictors of MitraClip therapy failure for functional mitral regurgitation: single central clip strategy in asymmetric tethering. \u003cem\u003eInt. J. Cardiol.\u003c/em\u003e \u003cb\u003e186\u003c/b\u003e, 286\u0026ndash;288 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatsumoto, T. et al. Impact of pulmonary hypertension on outcomes in patients with functional mitral regurgitation undergoing percutaneous edge-to-edge repair. \u003cem\u003eAm. J. Cardiol.\u003c/em\u003e \u003cb\u003e114\u003c/b\u003e, 1735\u0026ndash;1739 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAsgar, A. W., Mack, M. J. \u0026amp; Stone, G. W. Secondary mitral regurgitation in heart failure: pathophysiology, prognosis, and therapeutic considerations. \u003cem\u003eJ. Am. Coll. Cardiol.\u003c/em\u003e \u003cb\u003e65\u003c/b\u003e, 1231\u0026ndash;1248 (2015).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"mitral regurgitation, MitraClip intervention, machine learning model","lastPublishedDoi":"10.21203/rs.3.rs-5370589/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5370589/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e \u003cp\u003eSevere mitral regurgitation (MR) is a life-threatening mitral valve disease. The MitraClip intervention offers a percutaneous solution for patients who are unsuitable for surgery. However, limited information is available on outcomes post-MitraClip intervention. This study aims to develop an approach for predicting MR outcomes after MitraClip intervention using machine learning-enhanced echocardiography.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e \u003cp\u003eWe enrolled 164 patients with MR\u0026thinsp;\u0026ge;\u0026thinsp;3\u0026thinsp;+\u0026thinsp;degree who underwent MitraClip intervention at our institution between 2021 and 2024. Patients were monitored for approximately three years. The analysis included clinical data and echocardiographic parameters. Study endpoints were the recurrence of MR (2\u0026thinsp;+\u0026thinsp;or above) and major adverse events during follow-up. A total of 147 patients were randomly divided into training (80%) and testing (90%) sets. An additional 17 patients comprised the validation cohort.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e \u003cp\u003eThe best-performing model for predicting clinical outcomes utilized 81 features in a logistic regression classifier. Using all 81 features in the logistic regression model, specificity increased to approximately 0.797 (95% confidence interval: 0.739\u0026thinsp;~\u0026thinsp;0.854) and sensitivity to about 0.459 (0.370\u0026thinsp;~\u0026thinsp;0.549), resulting in an overall accuracy of 0.688 (0.632\u0026thinsp;~\u0026thinsp;0.745) for the validation dataset. The best-performing model achieved a receiver operating characteristic area under the curve value of 0.773 in both the test and validation groups.\u003c/p\u003e\u003ch2\u003eConclusions:\u003c/h2\u003e \u003cp\u003eOur machine learning model, leveraging echocardiographic characteristics, demonstrated superior predictive performance. This model effectively forecasts patient outcomes following MitraClip intervention, proving beneficial within a clinical setting.\u003c/p\u003e","manuscriptTitle":"A machine learning model for predicting outcomes of MitraClip therapy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-17 17:48:37","doi":"10.21203/rs.3.rs-5370589/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"09c4febf-222a-451a-9a45-db52a1de900a","owner":[],"postedDate":"December 17th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":41473493,"name":"Health sciences/Cardiology"},{"id":41473494,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2025-02-10T05:24:19+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-17 17:48:37","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5370589","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5370589","identity":"rs-5370589","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}