Overview of AI and Machine Learning Methods for Outcome Prediction in Pediatric Congenital Heart Surgery

Overview of AI and Machine Learning Methods for Outcome Prediction in Pediatric Congenital Heart Surgery | 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 Systematic Review Overview of AI and Machine Learning Methods for Outcome Prediction in Pediatric Congenital Heart Surgery Paschalina Lialiou, Ilias Maglogiannis This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9055427/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 Purpose Congenital heart disease (CHD) constitutes the most common major congenital malformation. Despite the improvement of diagnostic technologies and the advances in pediatric cardiovascular surgery, the utilization of artificial intelligence (AI) holds a critical role in predicting post-operative outcomes in pediatric population. The primary aim of the study was to define the utilization of AI machine learning models in the prediction of post-operative outcome in children undergoing a congenital heart surgery. Methods Following the PRISMA guidelines, a systematic literature review was conducted by a comprehensive retrieval of two large databases. The inclusion and exclusion criteria were predefined. Two independent reviewers screened the articles. Results 12 articles included in the review published the last seven years. The included research papers were retrospective cohort studies with a range of size population from 71 to 24.685 pediatric patients. The majority of them examined the prediction performance of AI machine learning algorithms in mortality and other post-operative complications. Various types of congenital heart surgeries were described. The area under the curve (AUC) was used for model performance, ranged from 0.642 to 0.970. LightGBM outperformed with AUC 0.970 for the prediction of deep venous thrombosis (DVT). Conclusion AI machine learning models show greater discriminative power for the prediction of post-operative outcomes in pediatric patients undergoing a CHS surpass the traditional prediction risk tools. Further studies must be conducted to strengthen the explainable role of AI applications in clinical decision making. Congenital Heart Surgery Cardiac Surgery Children Artificial Intelligence Machine Learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1. Introduction Congenital heart disease (CHD) accounts almost one-third of all major congenital anomalies. CHD birth prevalence worldwide varies and may also be observed differences that are due to genetic, environmental, socioeconomical or ethnic origins[ 1 ]. Nonetheless, there was an increasing birth prevalence over the total reported CHD births till 1995, over the last 15 years, a stabilization occurred, corresponding to 1.35 million newborns with CHD every year. Asia reported the highest CHD birth prevalence with 9.3 per 1,000 live births, followed by Europe and North America that reported 8.2 and 6.9 respectively[ 1 ], [ 2 ], [ 3 ]. Over 97% of them can be expected to reach adulthood through a life-time health care management[ 4 ], [ 5 ]. Despite the innovative advances in cardiovascular diagnostics, cardiothoracic surgery and perioperative care, children with CHD receiving anatomical radical treatment were still associated with high morbidity and mortality rates[ 6 ]. Previous studies have mainly focused on in-hospital mortality and importantly, on morbidity, as a leading cause of mortality that measure the performance and the direct quality of the improvement initiatives[ 7 ], [ 8 ]. Nonetheless, the complex prognosis and treatment after a congenital heart surgery remains ubiquitous due to the complex nature of surgery and influence the quality of treatment. Due to the extraordinary advances in cardiac surgery, the postoperative complications are many. Some of them have been reported, having important contributions in mortality, hospital stay, cost and quality of life. In recent years, the scientific community especially cardiac surgeons evolve with the assessment of the risk of congenital heart surgery using empirical tools, scales and predictors which can estimate the risk mortality of pediatric patients[ 9 ]. Machine learning (ML), a field of artificial intelligence (AI), enables the discovery of important patterns within complex multidimensional data and highlight the utility of its applications for clinical risk and scope to improve patient surveillance[ 10 ]. There is an increasing interest in the use of large claims databases in medical practice and research, the essential role of ML identify patients with any kind of diseases[ 11 ]. ML approaches such as tree-based models and deep learning are able to identify complex data patterns among variables exploring complex interactions and disease complications[ 11 ]. These ML methods have the potential to generate more accurate prediction and classification beyond traditional methods, with impressive beneficials in cardiovascular medicine[ 12 ], [ 13 ]. More precisely, in pediatric congenital heart surgery, ML models can accurately predict the risk of major adverse postoperative outcomes (APOs) affecting greatly the mortality, hospital stay, care management and planning, and quality of life[ 6 ]. These predictions provide reliable interpretations for high-risk contributor identification and informed clinical decisions. Also, previous studies have explored the potential of ML models as a tool for prediction mortality and other postoperative outcomes in cardiac surgery, showing great discriminative power to revolutionize cardiac surgery[ 14 ], [ 15 ], [ 16 ]. The aim of current study was to systematically review the relevant literature and identify all studies that have utilized AI to predict the post-operative outcomes and risk mortality in pediatric patients undergoing surgery for congenital heart defects. 2. Materials and Methods This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement[ 17 ], [ 18 ]. PRISMA statement is the most common guidance that has been used by authors and reviewers, and mentions the whole literature search procedure[ 19 ]. Moreover, PRISMA ensures the quality of reports and guide authors about the methodological strategy and study assessment[ 19 ]. Current SLR utilizes the 27 items of PRISMA checklist to verify that each search component is completely reported and reproducible. The proposed reviewing process is illustrated in Fig. 1 . Firstly, it is identified the purpose of present SLR which have been motivated this research. Three digital databases were recruited with predetermined search terms. Subsequently, an initial screening of identified studies was conducted, and the predefined inclusion and exclusion criteria were applied. The final set of eligible studies was coded, and relevant data aligned with the objectives of the present study were extracted. The extracted information was then systematically organized, compared, and synthesized to address the research questions. Zotero 7 for macOS (20,21), an open-access reference management software, was used to identify and remove duplicate records, support citation tracking, and manage source synthesis. 2.1 Search strategy A comprehensive search was conducted between July to September 2025 in three electronic databases, PubMed, ScienceDirect and Scopus by one author. Limitations about the English language and publication date within the past decade were obtained. Then, the search results were passed on to the second author for further evaluation. A series of keywords such as “transcatheter aortic valve replacement”, “congenital heart diseases”, “mortality”, “complications”, “machine learning”, “artificial intelligence algorithms”, “large language models”, “children”, pediatric patients” were identified and formed the search queries using the Boolean operators “AND” and “OR”. The search term were combined and formed separate search parameters as follows: (Transcatheter aortic valve replacement) OR (congenital heart diseases) AND (machine learning) AND (mortality), (Transcatheter aortic valve replacement) AND (AI algorithms) AND (mortality), (Transcatheter aortic valve replacement) AND (AI algorithms) AND (complications), Transcatheter aortic valve replacement) AND (Large language models) AND (complications), (congenital heart diseases) AND (machine learning) AND (complications) AND (children) OR (pediatric patients). 2.2 Inclusion and exclusion criteria The eligibility criteria were defined as follows: we included in this review those studies that investigates the adoption of machine learning or deep learning methods to predict the implications and mortality risk in pediatric patients undergoing a congenital heart surgery like heart transplantation, open heart surgery, transcatheter aortic valve implantation or narrow aortic valve (TAVI/TAVR) procedures or Norwood procedure. We used research studies, written in English language and published till September 2025, in which participants were infants and children up to the age of eighteen. For this review, we reclaimed studies that used clinical, laboratory data including images from MRIs and CTs, while only the research studies in which the accuracy of the predictive models was assessed. We excluded pilot studies, research articles that they did not specify the predictive analytics of congenital heart surgery outcomes or did not use machine learning methodologies. Also, we omitted studies involving both children and adult population. Additional screening of studies eliminated those that did not mentioned the performance measures of ML models. 2.3 Study selection After recruitment of the three selected databases, a total of 256 articles identified in the literature search. There were 32 duplicates that we extracted them manually. During the primary screening, we assessed the titles and the abstracts of the retrieved studies among the predetermined research objectives and the established inclusion/ exclusion criteria. Also, we exclude the articles that were not associated with the topic of present SLR. Additionally, we extracted all the randomized control trials, study protocols, reviews and meta-analysis. Finally, a total number of 121 articles were sought for the text retrieval. During the phase of screening 103 were excluded due to the target population (n = 75) not being pediatric patients, studied other pediatric syndrome (n = 9), were not associated with the aim of current study (n = 6) and due to the article type (n = 13). A total number of 15 studies included finally in the SLR. On a secondary stage, all the selected articles sent to the second author to carry out the data extraction independently. The structured data extraction approach was essential to ensure that the systematic review was precise and comprehensive. The following information was extracted: Publication information: First author’s name, source and publication year were listed. Study design: The study design such as prospective cohort, retrospective cohort was identified. Objectives: The research goals of each included study were outlined to ensure that coincide with the present review. Information about population: We collected information about sample size, demographics and country of research. Data source: We identified the data source of each study and the dataset that was derived. Model selection: We noted the AI prediction models used for outcome prediction. Top predictors of congenital heart surgery outcomes: The most powerful predictors of cardiac surgery outcomes were listed. Limitations: Any bias or constrains in the studies were noted, such as small sample size, selection bias or other. 3. Results 3.1 Study characteristics The characteristics of the included studied are summarized in Table 1 . In this table are outlined, the characteristics of included population in each study. The included studies were published from 2016 to 2024 and mostly from USA. The design of all included studies was retrospective cohort, which means that all the researchers examined the outcome of their interest of each study. A comparison of the reviewed studies according to the country distributed, source of data and data range is illustrated in Table 2 . Data of the majority of studies were derived from patients’ electronic health records, and ML models were trained and tested between 71 to 24,685 individuals in pediatric population. The majority of the reviewed population was males in a portion of almost 60%. Moreover, most of the studies utilized K-fold cross validation while in some studies is reported a train and test portion of 7:3 or 8:2. Among the studies, the specific type of heart surgery that investigated were 50% various congenital heart surgery, 25% Norwood procedure (stage 1 palliation), 9% first open-heart surgery in CHD patients, 8% heart transplantation and 8% transcatheter device closure of perimembranous ventricular septal defect (pmVSD). Table 3 depicts the AI prediction models constructed in each study. The most commonly utilized algorithm was Gradient Boosting Trees, in a portion of 26% (XGBoost, CatBoost, LightGBM), followed by Logistic Regression (LR) 15%, Random Forest (RF) 14%, Decision Trees (DT) and Support Vector Machine (SVM) in a portion of 9% for each of them, Radial Basis Function kernel, decision tree, Cox Proportional Hazards, other adaptive machine learning models, Self-Organizing Map, Fig. 2 . (SOM), Multilayeer Perceptton (MLP) and deep neural network (DNN). As we’ve seen, the most frequent postoperative outcome assessed by authors in the reviewed studies was mortality. Table 1 Characteristics of the population of the included studies in present systematic literature review. Abbreviations: RCS: Retrospective cohort study, CICU: Cardiac Intensive Care Unit, HLHS: hypoplastic left heart syndrome, CHD: Congenital Heart Diseases, CHS: Congenital Heart Surgery, pmVSD: perimembranous ventricular septal defects, VSD: ventricle heart disease, CPB: cardiopulmonary bypass. First author (Year) Population size Train/Test Proportion Gender (male) Population characteristics Du X. (2022)[ 20 ] 24,685 patients; number of in-hospital deaths: 595(2.4%) 7.5:2.5 57.59% Age: median 316 days (range 1–6,568 days) Boskovski (2020)[ 21 ] 2,517 CHD patients with WES; of these 1,268 underwent first open-heart surgery and had surgical follow-up data. NS 57.6% Various CHD patients; de novo damaging variant carriers: 294/2,517 (11.7%) Jalali (2018)[ 22 ] 71 neonates 1:6 47.89% Neonates undergoing cardiac surgery, including those with HLHS Jalali (2020)[ 23 ] 549 patients 7:3 61.74% Neonates undergoing the Norwood procedure for single ventricle CHD Ruiz-Fernández D. (2016)[ 24 ] NS 9:1 NS Pediatric patients undergoing CHS Smith A. (2024)[ 25 ] NS NS NS Infants with HLHS Sunthankar SD, (2023)[ 26 ] 3,267 infants 8:2 62% Infants with single VSD, including HLHS Chaoyang Tong, et al. (2024)[ 27 ] 23,000 pediatric patients Training cohort n = 13927 54% Pediatric patients undergoing CHS Wisotzkey BL. (2023)[ 28 ] NS 55% Pediatric heart transplant patients Yan L. (2024)[ 29 ] 384 children with pmVSD who underwent transcatheter closure 7:3 30.7% Pediatric patients with pmVSD treated with transcatheter device closure Xian Zeng (2021)[ 30 ] 1,964 patients included in the analysis (after exclusions) 7:3 52.8% Pediatric patients (median age ~ 11 months [IQR 4–26 months]) undergoing CHS; of these, 34.4% developed postoperative complications Zürn C. (2023)[ 31 ] 1765 patients 1:1.26 NS Children (ages 1 day to 18 years) undergoing CHS with CPB. Children with < 24 h PCICU stay or multiple ops only last stay counted Rogers et al, 2017[ 32 ] 21,838 4/1 NS Pediatric patients with various surgical procedures Normad et al 2022[ 33 ] 98.825 7:3 55% Pediatric patients (including premature neonates) with congenital or non-congenital abnormalities or chromosomal abnormality. Faerber 2021[ 34 ] 162 8:2 62.73% Children with TOF, some of them had a prior palliative procedure, 58.6% of them surgical repair Table 2 There is a comparison of the included studies, according to the country distributed, settings/ source of data and data range. Study Country Setting/Source of data Data range Du Y., et al. (2022) China Single-centre, LIS,HIS, ICU database, clinical data repository 2006–2017 Boskovski M., et al. (2020) USA Multi-centre, Pediatric Cardiac Genomics Consortium NS Jalali A., et al. (2018) USA Children's Hospital of Philadelphia, CICU-Local EMR NS Jalali A., et al. (2020) USA Pediatric Heart Network Single Ventricle Reconstruction Trial 2005–2009 Ruiz-Fernández D., et al. (2016) Colombia Children Heart Disease database from Cardiovascular Foundation of Colombia, local EMR NS Smith A., et al. (2024) USA The pediatric Heart Network Single Ventricle reconstruction trial 2005–2009 Sunthankar SD., et al (2023) USA, Canada National Pediatric Cardiology Quality Improvement Collaborative (NPC-QIC), Pediatric Electronic Data Capture (PEDCAP) database 2008–2019 Tong C., et al. (2024) China Congenital Heart Surgery Database Collection NS Wisotzkey BL., (2023) USA Pediatric Heart Transplant Society 2010–2019 Yan L., et al (2024) China Single-centre hospital in China, Platform for Epidiomological Investigation and Precision Treatment of Childhood Heart Disease 2002–2024 Zeng X., et al. (2021) China Single tertiary centre (Children's Hospital of Zhejiang University School of Medicine) NS Zurn C., et al. (2023) Germany Departments of pediatric cardiology and cardiac surgery in Freiburg and Heidelberg University Heart Center 2014–2019 Rogers L., et al. (2017) UK UK National Congenital Heart Audit 2009–2014 Normand SL., et al. (2022) USA The Society of Thoracic Surgeons (STS) Congenital Heart Surgery Database (CHSD) 2014–2018 Faerber J., et al. (2021) USA Children's Hospital of Philadelphia 2012–2018 Table 3 Overview of the AI Models that were utilized by authors in the reviewed papers according to the type of cardiac surgery and the evaluated postoperative outcome. Abbreviations: CHS: Congenital Heart Surgery, XGBoost:Extreme Gradient Boosting, SVN:Support Vector Network, RBFK: Radial Basis Function kernel, DNN: Deep neural network, RF:Random Forest, DT:Decision Tree, LR:Logistic Regression, MLP:Multilayer Perceptron, SOM:Self-Organizing Map, RBF:Radial Basis Function, GBT:Gradiant Boosting Trees, LightGBM: Light Gradiant Boosting Machine, CPH: Cox Proportional Hazards, pmVSD: perimembranous Ventricular Septal Defects, PVL: Periventricular leukomalacia, LCOS: Low cardiac output syndrome, DVT: deep venous thrombosis, MARS: Multivariate adaptive regression spline model. Authors Type of Procedure Outcome AI Algorithm Model Du et al., 2022[ 35 ] Various CHS Mortality XGBoost Boskovski et al, 2020[ 21 ] First open-heart surgery Mortality, extubation time MLP, RBFN, SOM, DT Jalali et al, 2018[ 22 ] Various CHS PVL SVN, RBF kernel Jalali et al, 2020[ 23 ] Norwood procedure Mortality, length of hospital stays DNN, XGBoost, RF, DT, ridge LR Ruiz-Fernández D, et al. 2016[ 24 ] Various CHS Surgical risk classification MLP, SOM, RBF networks, DT Smith A., et al.2024[ 25 ] Hypoplastic left heart syndrome Five-year transplant free survival Not specified Sunthankar SD, et al 2023[ 26 ] Single ventricle disease, Norwood procedure Mortality LR, RF, XGBoost, GBT, LightGBM Chaoyang Tong, et al. 2024[ 27 ] Various CHS LCOS, pneumonia, renal failure, DVT LightGBM, LR, SVM, RF, CatBoost [ 29 ]Wisotzkey BL, 2023[ 28 ] Heart transplantation 1-year allograft loss RF, CPH [ 30 ]Yan L., et al 2024[ 29 ] Transcatheter device closure of pmVSD Postoperative arrythmia SVM, LR, RF, XGBoost [ 31 ]Zeng X., et al. 2021[ 30 ] Various CHS Postoperative complications XGBoost, LR [ 32 ]Zürn C., et al. 2023[ 31 ] Various CHS Postoperative survival at 30 days LR Rogers., et al. 2017[ 32 ] Various CHS Postoperative survival at 30 days LR Normand et al 2022[ 33 ] Various CHS Postoperative mortality within 30-days Lasso, BART Faerber 2021[ 34 ] Tetralogy of Fallot Postoperative cardiac complications RF, GBM: two-stage with and without a MARS 3.2 Performance metrics and interpretation Overall AUCs above 0.65 were achieved by all the reviewed research papers. The most frequent investigated outcome was postoperative mortality followed by other postoperative complications associated with lung, cardiac rhythm and infections, (Table 4 ). Different ML models across the 15 different articles were used and tested. Among the best performance of prediction models was achieved by Tong et al. 2024 [ 6 ] with an AUC of 0.97 as reported by LightGBM for the prognosis of deep venous thrombosis, following 0.96 by renal failure by LightGBM. Also, an AUC portion of 94.86% was achieved by LR for the assessment of postoperative survival through a variety of clinical modalities by Zurn et al. 2023 [ 31 ]. On the other hand, the worst performance demonstrated by Sunthankar et al. 2023 using the LightGBM model with an AUC of 64% for the influential factors of mortality risk. Calibration was described for models n = 7 (47%), in which 4 of them [ 26 ], [ 30 ], [ 33 ], [ 34 ] used visual calibration plots to ensure risk mortality, risk complications across deciles, and compare predicted to observed probabilities. Two studies (n = 2) used brier score and Integrated calibration index(ICI) for grass loss risk[ 28 ] and risk of renal failure in children. Calibration curves was used by one model[ 29 ] and [ 31 ] used standard logistic calibration. Class probabilities, utilizing decision trees used by one study[ 24 ] to determine calibration via frequency, and time -dependent calibration, across 5 years of follow-up used by [ 25 ]. Importantly, the model of Zürn et al. [ 31 ] delivered excellent generalization due to the high AUC values (0.948) in test dataset with specificity 89.48% and sensitivity 85.00%, (slightly below of Freiburg train dataset), the model’s calibration assessments shows that for very high and very low probabilities correctly identifies the outcomes. But, for probabilities in between, the model trends to overestimate the mortality risk. In particular, the model of Rogers’ et al [ 32 ] had a median AUC of 0.83 (range 0.82 to 0.83), showing excellent discrimination, with a median calibration slope of 0.92 (range 0.64 to1.25) and median calibration intercept of -0.23 (range 1.08 to 0.85; perfect calibration slope = 1 and intercept = 0), indicating slight underprediction. Moreover, Normad et al., [ 33 ] indicates calibration for mortality predictions at the higher end of risk from the lasso and Bayesian additive regression trees models, which was better than the traditional STS CHSD model. On the other hand, the models of Faerber et al were generally well-calibrated, though RF models can sometimes be overconfident without proper tuning. Moreover, the calibration curves’ of the predictive model of the development of arrhythmia in pmVSD patients demonstrated that there was a little deviation between the ideal and real, which means that the predictive model had some predictive value following internal validation[ 29 ]. Tong et al [ 27 ] evaluated the interpretations of ML models by using SHAP ranking feature importance and representing them as patient-specific visualizations. The SHAP values showed important clinical factors associated with risk of major APOs, of which longer mechanical ventilation time was the most important factor leading to an increased risk of the complications. Predictive model of [ 28 ] performed 7.6% cumulative incidence of grass loss or mortality. Furthermore, random forests had favorable discrimination and calibration, while performed closer alignment between predicted and observed risk. Explainable AI (XAI) features and strategies used from (n = 5) of reviewed papers. SHAP (Shapley additive explanations) values used by (n = 3) of the models [ 26 ], [ 28 ], [ 30 ], feature importance analysis utilized by one reviewed model[ 31 ], while [ 24 ] adopted an intrinsic exlpainability design for model interpretation. The SHAP values of the predictive model of [ 28 ] estimated the most important predictors in the final model. In other words, the diagnosis of congenital heart disease increased one year predicted risk of grass loss by 1.7, the need for mechanical circulatory support increased predicted risk by 2 and single ventricle CHD increased predicted risk by 1.9. In other words, they concluded that risk prediction models used to facilitate patient selection for pediatric heart transplant can be improved without loss of interpretability using machine learning. 3.3 Model validation For the internal validation of machine learning models, most studies used cross validation (n = 3), hold-out split (n = 1), temporal split (n = 1), repeated split(n = 1), leave one out (LOOCV)(n = 1), a random split of the dataset (n = 2), or bootstrapping (n = 1). External validation was performed in 7 studies. Five studies [ 6 ], [ 26 ], [ 28 ], [ 31 ], [ 33 ] conducted bicentric or multicenter validation, and 3 studies [ 29 ], [ 30 ], [ 32 ]used temporal validation. The discriminative ability (AUCs) of these models ranged between 0.64 and 0.97, and the calibration was reported for 7 machine learning models. Temporal split of data validation from 2023 to 2015 was adopted, also, by Faerber et al[ 34 ]. Table 4 Above is cited the best performing ML algorithm due to the measured post-operative outcome. Abbreviations: XGBoost:Extreme Gradient Boosting, MLP:Multilayer Perceptron,, RBF kernel:Radial Basis Function kernel, SOM:Self-Organizing Map, DT:Decision Tree, SVN:Support Vector Network, DNN: Deep neural network, RBF:Radial Basis Function, GBT:Gradiant Boosting Trees, LightGBM: Light Gradiant Boosting Machine, RF:Random Forest, LR:Logistic Regression, pLOS:prolonged length of hospital stay, PVL:Periventricular leukomalacia, MMI: Modified Mutual Information, LCOS:Low cardiac output syndrome, DVT: deep venous thrombosis, PHN: Pediatric Heart Network., SVR: Single Ventricle Reconstruction, STS model: Society of Thoracic Surgeons model. Authors Outcome Best Performing Algorithm Best Performance metrics Comments (Small/Imbalanced data, feature selection strategies, lack of external validation) Du X., 2022[ 22 ] Mortality XGBoost AUC = 0.887 Outperformed traditional scores: STS-EACTS(0.748) and RACHS-1 (0.677). Large dataset(N = 24,684), mortality was only 2.4% (595 deaths). This represents a highly imbalanced scenario. The study primarily relies on internal validation (like a 70/30 split of the same institutional registry). No independent validation on a separate hospital cohort was reported. Boskovski et al, 2020[ 23 ] Mortality, time to final extubation MLP, RBFN, SOM, DT NS 11.7% of patients with de novo variants (imbalanced). Agnostic ML used to identify variants (CNVs). None validation (internal cohort only). Jalali A., et al. 2018[ 24 ] PVL SVN F1 score reported 0.81-1.00 Small dataset (n = 71), 33%PVL rate MMI ranking system. None validation (Small single-center scale). Jalali A., et al, 2020[ 25 ] Mortality, pLOS DNN For mortality prediction: DNN: AUROC = 0.95 For pLOS: DNN: AUROC = 0.94 High imbalance data (mortality ~ 7–8% in SVR trial). DNN used on preoperative clinical features. Internal validation (SVR trial dataset split). Ruiz-Fernández D, et al. 2016[ 26 ] Surgical risk classification (RACHS) MLP MLP: Accuracy = 0.999 3 categories of risk (low, medium, high). Manual feature selection (clinical factors). None validation (Single center/Colombia). Smith A., et al.2024[ 27 ] Mortality Not specified Td-AUC > 0.800 Trial cohorts PHN SVR I and II Incremental feature inclusion (pre-operative to post-operative). Internal (Wait-time cohort validation). Sunthankar SD, et al 2023[ 28 ] Mortality LightGBM AUC = 0.642 Imbalanced (6.4% mortality). Evaluated 180 clinical features. Internal validation (Multicenter NPC-QIC database). Chaoyang Tong, et al. 2024[ 20 ] LCOS, pneumonia, renal failure, DVT LightGBM, LR LCOS: LightGBM, AUC = 0.893 Pneumonia: LR, AUC = 0.929 Renal Failure: LightGBM, AUC = 0.963 DVT: LightGBM, AUC = 0.970 Large dataset (N = 23,000): Training (n = 13,927) and Testing (n = 9,073). While the cohort is large, specific complications like Renal Failure and DVT are inherently rare (imbalanced), though specific ratios were not highlighted as a failure point. The study uses a temporal split (data before 2019 for training, after 2019 for testing). While better than a random split, it remains Internal Validation as it uses data from the same institution (Shanghai Children's Medical Center). Wisotzkey BL, 2023[ 29 ] 1-year allograft loss RF NS 3,787 patients, 7.6% graft loss (imbalanced data). SHAP – top 15 features were selected. Internal validation (75:25 split). Yan L., et al 2024[ 30 ] Postoperative arrythmia LR LR: AUC = 0.863 1,384 patients (Retrospective) 5-variable nomogram. Internal validation (7:3 split). Zeng X., et al. 2021[ 31 ] Prediction/classification of postoperative complications XGBoost Prediction: AUC = 0.839. Classification: AUC = 0.850 Integrated HER and time series data. k-means, Dynamic Time Warping for series data. Internal validation only. Zürn C., et al. 2023[ 32 ] Mortality LR AUC = 0.9486 Retrospective (780 train/985 test). Knowledge-driven selection feature strategy: STAT score, age, clamp time, lactate. Bicentric validation (Heidelberg used to test Freiburg model). Rogers et al 2017[ 33 ] Mortality LR AUC = 0.83 Validation dataset 4,207 episodes with 97 deaths. Expert advisory panel to consider the relative importance of comorbidities and risk factors. External validation dataset from 2014 to 2015. Normad et al, 2022[ 34 ] Mortality BART, lasso AUC = 0.858 to 0.875 7 different models (3 BART, 3 lasso, current STS model). Clinical review group discuss clinical coherence. Highly robust multicenter internal validation. Faerber et al 2021[ 35 ] Postoperative cardiac complications GB AUC = 0.71 Imbalance handling: undersampling Feature selection: 48 candidate variables(data-driven). Temporal slip validation strategy. External validation on data from 2013–2015. 3.4 Risk of bias assessment Prediction models in health care use predictors to estimate the probability that a condition or a disease is already present or will occur in the future. Within this knowledge framework, the authors examined the risk of bias (ROB) of the prediction models in present SLR. The risk of bias of the selected articles was assessed using the PROBAST tool [ 36 ], [ 37 ] by two independent reviewers. PROBAST tool consists of 4 domains, participants, predictors, outcome, and analysis with a total of 20 questions to facilitate the overall ROB assessment. Each domain can be graded with a low-risk bias, a high-risk bias or an unclear risk of bias. Studies with a low risk of bias in all four domains were classified as low risk while having a high risk of bias in even one domain were judged as a high risk of bias. Moreover, if one or more domains characterized as unclear risk of bias and the other as low-risk then the overall judgement to the study would be declared as unclear. Overall, eight, three and one studies had high, low and unclear risk of bias respectively. The most common domain of bias was analysis (domain 4), while the domain appearing the least amount of bias was outcome (domain 3). Given that approximately 54% of the included studies were assessed as having a high risk of bias—primarily due to limitations in the analysis domain—concerns remain regarding the applicability of the reported predictive models. Consequently, further validation is required, underscoring the need for rigorously designed, high-quality studies in the field of congenital heart disease, particularly within pediatric populations, (Fig. 3 ). 3.5 Key findings and conclusions Overall, we summarize the main research findings of all the reviewed inquiries. In more detail, the XGBoost model outperformed traditional stratification scores, (STS-EACTS and RACHS-1) and offered better discrimination for in-hospital mortality in pediatric CHD surgery took place in China[ 22 ]. Moreover, the SVM model demonstrated potential in predicting PVL occurrence in neonates after cardiac surgery, with varying rates of PVl observed HLHS and HLHS groups [ 24 ]. Also, the DNN model outperformed traditional risk stratification tools in predicting one-year mortality or cardiac transplantation and prolonged length of hospital stay in neonates undergoing the Norwood procedure [ 25 ]. The same goes to the research of Ruiz-Fernadez et al [ 26 ], who concluded that AI-based algorithms, particularly MLP, are feasible for classifying surgical risk in pediatric congenital heart surgery, aiding in preoperative decision-making. The machine learning model of Smith A., et al [ 27 ] demonstrates that predicts five-year transplant-free survival in infants with HLHS undergoing the Norwood procedure. Furthermore, the model demonstrated satisfactory clinical performance with a C-index of 0.692. On the other hand, patients with congenital heart disease undergoing open-heart surgery, was found that the de novo variants were associated with worse transplant-free survival and longer times on the ventilator [ 23 ]. As we’ve seen the LGB machines provided the best predictive performance, highlighting the potential of advanced machine learning algorithms in clinical decision-making [ 28 ]. Additional, one research study concluded that the established ML models can accurately predict the risk of four major adverse postoperative outcomes in pediatric congenital heart surgery, providing reliable interpretations for high-risk contributor identification and informed clinical decision-making [ 6 ], notable, Random Forests, can improve 1-year risk assessment for pediatric heart transplant patients without loss of interpretability[ 29 ]. The logistic regression model accurately predicts postoperative arrhythmias after pmVSD transcatheter closure with the selected five variables performed best (AUC = 0.863). The nomogram derived can help risk stratify patients and guide clinical decision-making [ 30 ]. Logistic regression, also, achieved excellent discrimination (AUC ~ 94.9%) for 30-day survival, substantially better than using STAT score. Use of peri-/post-operative data (clamp time, lactate) improved prediction over preoperative risk alone, reducing prediction error by ~ 53.5%. STAT score and aortic cross-clamp time were highly significant predictors; age had a minimal effect; lactate dynamics (either persistently high or rising after 8 h) were associated with higher mortality[ 32 ]. However, although the majority of the predictive ML models achieved high performance metrics, there is a ubiquity across the reviewed studies related to the predictive value and clinical applicability of models. For instance, despite the excellent performance of the PRAiS2 model[ 32 ], it was underpredicted, the risk for the HLHS hybrid procedure, generally performed on the sickest patients. Similar, in the study of [ 33 ], all models overpredicted operative mortality in very high risk patients and the STS model had worse calibration in this group. Overall, overestimation of expected mortalities would lead to more favourable hospital performance classifications because the expected mortality rate would be inflated, making the observed-to-expected mortality ratios lower. In terms SHAP values provided reliable interpretations for high-risk contributor identification and informed clinical decisions-making[ 27 ], they was also used as a framework to provide explanations of the prediction[ 30 ]. To sum up, these novel applications are still a promising step for improving the care of children and neonates undergoing any kind of congenital heart surgery. 3.6 Open issues and challenges Several key challenges emerged across the studies reviewed. In children with congenital heart disease especially those who had surgery, the number of cases with specific cardiac defects, surgery procedure and postoperative outcomes is limited[ 22 ], [ 23 ], [ 28 ], [ 29 ]. These restrictions limit the ability to train large robust machine learning models. Also, because of the heterogeneity of clinical phenotypes and anatomical variability related to the age of this patient population, CHD covers a wide range of anatomic variations, surgical repairs customized to the specific patients’ age. Based on this, in Fig. 4 presents, all reviewed studies that employed different sets of variables for training the machine learning models, ranging from 4 to 538. This heterogeneity complicates the training of machine learning models and even more the generalization of results[ 31 ]. Congenital heart disease (CHD) is an excellent domain for AI to give the robust and diverse datasets extending from complex disease diagnosis and management to multimodality imaging[ 38 ]. At this baseline, the small datasets and the heterogeneity of pediatric datasets perform issues in internal testing limiting the clinical utility[ 24 ], (Fig. 5 ) . Data acquisition in the area of congenital heart surgery depends strongly on the quality and quantity of input data[ 39 ]. Moreover, children of different ages, centers and clinical domains, or patients from different ethical diversity, make machine learning models to adapt and develop solutions making them more explainable for clinicians in decision making. These challenges point out to the need for future research on congenital heart surgery and inform the scope of this review ensuring that AI prevention models for pediatric CHD surgery are applicable and introduce novel challenges to lifelong trajectories in pediatric clinical care[ 40 ]. 4 Discussion This systematic review provides insights into the application of artificial intelligence for predicting postoperative outcomes in children undergoing congenital cardiac surgery. In total, 15 studies published over the past seven years, were included, reflecting a growing interest in—and increasing demand for—automated and interpretable machine learning methods to support the prediction of postoperative outcomes in pediatric cardiac surgery. Present SLR, also, emphasizes in children population. Moreover, the population of all the selected studies was neonates, infants or children with a minimum age of 1 day to 18 years old, who were diagnosed with a congenital heart disease. All the included articles were retrospective cohort studies, the data collection was implemented between 2002 to 2024 from different pediatric cardiac or heart clinical domains, retrieved from local electronic medical records of each setting. 8 studies were implemented in USA and 4 studies in China. All the pediatric population were undertaken a congenital heart surgery including various procedures like open heart surgery (n = 1), Norwood procedure (n = 3), heart transplantation (n = 1), ventricular septal defects repair (n = 2), cardiopulmonary bypass (n = 2), various procedures (n = 6). The ML models incorporated in the selected studies were designed to forecast mortality and survival (n = 8) [ 21 ], [ 23 ], [ 25 ], [ 26 ], [ 32 ], [ 33 ], [ 35 ], [ 41 ], postoperative complications (n = 4)[ 6 ], [ 28 ], [ 29 ], [ 30 ], [ 34 ], periventricular leukomalacia (PVL) (n = 1) [ 22 ], surgical risk classification (n = 1) [ 24 ]. The postoperative complications included low cardiac output syndrome (LCOS), pneumonia, renal failure, deep venous thrombosis (DVT), one year allograft loss, postoperative arrythmia, lung complications, cardiac rhythm complications, infections, and other. One study [ 23 ] examined both mortality and the prolonged length of stay in neonates undergoing the Norwood procedure for single ventricle congenital heart defects. To evaluate the discrimination of the predictive models, we used the area under the curve (AUCs). The AUCs of the models ranged between 0.642 to 0.970, with most of the models achieving an AUC above 0.8, highlights the potential of artificial intelligence as a reliable tool in congenital heart surgery. More details about each assessed outcome are discussed below. The majority of models performed the area under the curve above 0.8 for mortality assessment, apart from the study of Sunthatnkar et al. [ 26 ] which mortality prediction between stages I and II of palliation in patients undergoing single ventricle surgery was ranged 0.642. Moreover, the accuracy of the mortality model of Jalali et al. [ 23 ] was indicated 89% for DNN, portion that linked with the results of Boskovski et al. [ 21 ] for SOM predictive model. The best performing predictive model was pointed out by Zürn et al. [ 31 ] with the area under the curve ranged 0.9486 for LR, results that are in agreement with Jalali et al. [ 23 ] for DNN. In the same study, the same model achieved an AUC 0.94 for the length of hospital stay. Tong et al. [ 6 ] used LR and LightGBM for pneumonia, renal failure and deep venous thrombosis prediction, indicating AUC 0.929, 0.963 and 0.970 respectively. Findings from this review leverage the significant predictors of each model to identify the examined post-operative outcomes. A core set of predictors used for each machine learning model, which varied greatly among the predictive value of mortality. In more details, preoperative variables, such as age, weight, oxygen saturation, preoperative mechanical ventilation, left atrial dimension, atrial shunt dimension, history of cardiac surgery and number of defects achieved high AUCs up to 0.80 compromises high descriptive ability[ 20 ], [ 21 ], [ 23 ], [ 25 ]. Also, intraoperative clinical and laboratory features, like weight, procedure time, defect diameter, pre-interventional arrhythmia, difference between occlude to defect diameter >2mm compromise a significant role in the ML predictive model in complications (LCOS, DVT, postoperative arrythmia), ranging accuracy between 80% and 90%[ 6 ], [ 29 ]. Moreover, post-operative indicators, such as serum lactate, prolonged mechanical ventilation, arrhythmias, further improves the model calibration and sensitivity[ 23 ], [ 27 ]. Genomic and clinical phenotype data, particularly de novo damaging variants, were highly predictive indicators of transplant free survival with the area under the curve up to 0.82[ 21 ]. The findings above confirm the growing body of literature that integrating clinical, operative, post-operative and genetic variables yields most accurate and generalizable predictive models in pediatric CHD surgery. Moreover, genomic analysis of 2517 patients with CHD and their parents revealed clinically significant de novo variants in 11.7% of patients[ 21 ]. These patients were more likely to extra-cardiac disfunctions which reinforcing prior existing evidence that de novo variants are more prevalent in syndromic than in isolated CHD. [ 42 ], [ 43 ]. More findings suggest the potential role of digenic interactions in CHD pathogenesis and provide insights into molecular diagnosis by enhancing a genetic discovery rates[ 44 ]. Additional biomarkers like tissue inhibitor of metalloproteinase-1 (TIMP‐1) constitute a critical role to identify the high risk of death in patients with aortic stenosis undergoing transcatheter aortic valve replacement[ 45 ]. This synthesizes the complexity of pediatric heart defects stratifying the need to develop machine learning applications based on a multimodal computational framework by combining both imaging and biologically relevant features[ 46 ]. Due to the fact that de novo genetic variants are associated with noncardiac phenotypes and negative outcomes after cardiac surgery[ 21 ], the establishment of a reliable machine-learning model which predict cardiac phenotype using genotype will facilitate effective management of surgical complications in CHS[ 47 ], [ 48 ]. Overall, the reviewed studies demonstrated strong predictive performance of artificial intelligence models for mortality prediction in pediatric patients undergoing congenital heart surgery. Previous research indicates that mortality among individuals with congenital heart disease is highest during the first year of life, with survival gradually declining beyond infancy and into childhood (5). However, more studies need to be conducted to perform quantitative measures of the performance of AI driven models in survival appraisal of CHD patients. Nonetheless, Du et al. [ 20 ] concluded that the XGBoost model outperformed the traditional stratification scores (STS-EACTS & RACHS-1) and offered better discrimination for in-hospital mortality in pediatric CHD surgery in China. Findings are confirmed by Jalali et al. [ 23 ], who found that the DNN model, also, outperformed the traditional risk stratification tools in predicting one-year mortality or cardiac transplantation and prolonged length of hospital stay in neonates undergoing the Norwood procedure. These results agree with Zaka et al., [ 49 ] who concluded that ML models outperformed traditional risk scores in the discrimination of all-cause mortality following transcatheter aortic valve implementation (TAVI). Furthermore, machine learning models are associated with greater post-operative outcomes and survival in neonates and infants undergoing a heart surgical procedure like open-heart surgery or Norwood procedure, such as longer transplant free survival and satisfactory clinical performance with C-index of 0.692[ 21 ], [ 22 ], [ 25 ]. More key findings from present study reveals that machine learning models, particularly Random Forests, can improve 1-year risk assessment for paediatric heart transplant patients without loss of interpretability[ 28 ], the prediction of post-operative arrhythmias after pmVSD transcatheter closure, are able to identify modifiable and non-modifiable risk factors for interstage mortality following Stage I palliation Further machine learning algorithms[ 26 ], [ 29 ]. In addition, AI based algorithms [ 24 ], particularly MLP are feasible to classify the surgical risk contributing in preoperative decision making. Also, the established ML models in the study of Tong et al. [ 27 ] can accurately predict the risk of four major adverse postoperative outcomes in pediatric congenital heart surgery, providing reliable interpretations for high-risk contributor identification and informed clinical decision-making. Further, LGB machines [ 26 ] provided the best predictive performance, confirming, additionally, the potential of advanced machine learning algorithms in clinical decision-making. Furthermore, Betsimas et al. [ 50 ] have shown that the machine learning methodology of optimal classification trees (OCTs) can accurately predict risk after congenital heart surgery. They, also, concluded that OCT benchmarking analysis can assess hospital-specific case-adjusted performance after CHS, both overall and patient cohort-specific, serving as a tool for hospital self-assessment and quality improvement. These findings underscores the importance of considering the integration of ML algorithms into electronic healthcare systems, though, they may improve periprocedural risk stratification, but immediate implementation in the clinical setting, still, remains uncertain[ 49 ]. From a clinical perspective, the distinct pattern of congenital heart dysfunctions raises several questions regarding the optimal management of patients and how these patients may impact from AI and machine learning in future risk of adverse events or response to surgical interventions[ 51 ], [ 52 ]. This review illustrates the capabilities of machine learning in prediction of several surgical outcomes. Promising discriminative abilities of ML models have been discovered. In surgical ML research, there is a need for standardized calibration assessment and reporting to facilitate the clinical adoption of ML models[ 53 ]. However, most studies included a retrospective study design without external validation or calibration. Large-scale data is warranted to bridge the gap between calibration and external validation[ 53 ], [ 54 ]. Clinical implementation is needed to demonstrate the contribution of ML within daily practice. We observed that there is a clear trend for older papers focus on performance (AUC), while papers from 2023–2025 prioritize calibration and XAI to make the models clinically useful. This approach reveals areas for potential improvement and transform ML as a powerful tool for quality improvement clinical initiatives[ 55 ]. To conclude, our systematic review has several limitations. Primarily, the diversity of congenital heart diseases, the nature of heart defects makes the models heterogeneous. Also, this is reinforced by certain characteristics of study population like age, weight, gestational age, although the nature of pediatric population meets some clinical particularities by its own. Unless, there was an effort to develop AI models specialized in a specific heart defect, very large sample size of pediatric population is needed to develop more reliable ML models. Another important issue that was reported by many of the included studies, was the high risk of bias resulted by the low quality or unclear data analysis. Further research is required to overcome methodological and validation limitations. 5 Conclusion In summary, with a novel interpretable machine learning algorithm, we can predict whether a pediatric patient perform complications after a heart congenital heart surgery or not. Also, what kind of complications will occur and explain the specific patient characteristics that led to this prediction. The majority of the developed AI prediction models achieved high accuracy and sensitivity. To the best of our knowledge, this systematic review is one of its kind and we believe that the combination of high model performance and interpretability could provide useful information for physicians being part of clinical decision making[ 30 ]. Continuing research in clinical settings provides evidence of the feasibility and advantages of offering genome sequencing to patients with cardiac phenotypes indicating a genetic cause. This enables the necessity to create a model of care that harnessed clinical and functional genomics to inform a future ML based clinical practice[ 56 ]. Physicians that care critically ill children with CHD might be influenced by genomic results by making recommendations about whether to forego or withdraw certain treatment option[ 57 ]. AI driven machine learning models motivate a personalized plan of care, giving strategies and recommendations. So, the abetment of a cardiac genetic testing and the approval for a further publicly genetic repository will improve the collaboration between stakeholders and clinical caregivers to improve the quality of care and outcomes for CHD pediatric patients and their families[ 58 ], [ 59 ]. Declarations Ethical approval Not Applicable. Consent to participate Not Applicable. Consent for publication Not Applicable. Clinical Trial number Not Applicable. Conflict of interest All authors have reported that they have no relationships relevant to the contents of this paper to disclose. Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Author contributions All authors contributed to the study conception and design. Paschalina Lialiou and Ilias Maglogiannis contributed to the data collection. Material preparation was performed by Paschalina Lialiou and Ilias Maglogiannis. The first draft of the manuscript was written by Paschalina Lialiou and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Data availability Not Applicable. Code availability Not Applicable. References van der Linde D et al (Nov. 2011) Birth Prevalence of Congenital Heart Disease Worldwide. JACC 58(21):2241–2247. 10.1016/j.jacc.2011.08.025 Liu Y et al (2019) Global birth prevalence of congenital heart defects 1970–2017: updated systematic review and meta-analysis of 260 studies, Int. J. Epidemiol. , vol. 48, no. 2, pp. 455–463, Apr. 10.1093/ije/dyz009 Hoffman JIE (May 2013) The global burden of congenital heart disease: review article. Cardiovasc J Afr 24(4):141–145. 10.10520/EJC137177 Mandalenakis Z et al (Nov. 2020) Survival in Children With Congenital Heart Disease: Have We Reached a Peak at 97%? J Am Heart Assoc 9:e017704. 10.1161/JAHA.120.017704 Best KE, Rankin J (Jun. 2016) Long-Term Survival of Individuals Born With Congenital Heart Disease: A Systematic Review and Meta‐Analysis. J Am Heart Assoc 5(6):e002846. 10.1161/JAHA.115.002846 Tong C et al (Apr. 2024) Machine learning prediction model of major adverse outcomes after pediatric congenital heart surgery: a retrospective cohort study. Int J Surg 110(4):2207. 10.1097/JS9.0000000000001112 Pasquali SK, He X, Jacobs JP, Jacobs ML, O’Brien SM, Gaynor JW (2012) Evaluation of failure to rescue as a quality metric in pediatric heart surgery: an analysis of the STS Congenital Heart Surgery Database. Ann Thorac Surg 94(2):573–579 discussion 579–580, Aug. 10.1016/j.athoracsur.2012.03.065 Pasquali SK et al (2012) Association of center volume with mortality and complications in pediatric heart surgery, Pediatrics , vol. 129, no. 2, pp. e370-376, Feb. 10.1542/peds.2011-1188 Ferry C, Fiery-Fraillon J, Togni M, Cook S (2025) Futility in TAVI: A scoping review of definitions, predictive criteria, and medical predictive models. PLoS ONE 20(1):e0313399. 10.1371/journal.pone.0313399 Meredith T et al (2025) Machine learning cluster analysis identifies increased 12-month mortality risk in transcatheter aortic valve replacement recipients. Front Cardiovasc Med 12:1444658. 10.3389/fcvm.2025.1444658 Marelli AJ et al (Feb. 2024) Machine Learning Informed Diagnosis for Congenital Heart Disease in Large Claims Data Source. JACC Adv 3(2):100801. 10.1016/j.jacadv.2023.100801 Razavian N, Marcus J, Sontag D Multi-task Prediction of Disease Onsets from Longitudinal Lab Tests, Sep. 20, 2016, arXiv : arXiv:1608.00647. 10.48550/arXiv.1608.00647 Mayourian J, Geggel R, La Cava WG, Ghelani SJ, Triedman JK (2025) Pediatric Electrocardiogram-Based Deep Learning to Predict Secundum Atrial Septal Defects, Pediatr. Cardiol. , vol. 46, no. 5, pp. 1235–1240, Jun. 10.1007/s00246-024-03540-7 Penny-Dimri JC, Bergmeir C, Perry L, Hayes L, Bellomo R, Smith JA (2022) Machine learning to predict adverse outcomes after cardiac surgery: A systematic review and meta-analysis, J. Card. Surg. , vol. 37, no. 11, pp. 3838–3845, Nov. 10.1111/jocs.16842 Leivaditis V et al (Jan. 2025) Artificial Intelligence in Cardiac Surgery: Transforming Outcomes and Shaping the Future. Clin Pract 15(1):17. 10.3390/clinpract15010017 Mestres CA, Quintana E, Pereda D (2022) Will artificial intelligence help us in predicting outcomes in cardiac surgery? J. Card. Surg. , vol. 37, no. 11, pp. 3846–3847, Nov. 10.1111/jocs.16844 Page MJ et al (2021) The PRISMA 2020 statement: an updated guideline for reporting systematic reviews, The BMJ , vol. 372, p. n71, Mar. 10.1136/bmj.n71 Rethlefsen ML et al (2021) PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews, Syst. Rev. , vol. 10, no. 1, Art. no. 1, Jan. 10.1186/s13643-020-01542-z Amir-Behghadami M, Janati A (2021) Reporting Systematic Review in Accordance With the PRISMA Statement Guidelines: An Emphasis on Methodological Quality, Disaster Med. Public Health Prep. , vol. 15, no. 5, Art. no. 5, Oct. 10.1017/dmp.2020.90 Du X et al (Nov. 2022) Machine Learning Model for Predicting Risk of In-Hospital Mortality after Surgery in Congenital Heart Disease Patients. Rev Cardiovasc Med 23 11, Art. 11. 10.31083/j.rcm2311376 Boskovski MT et al (Aug. 2020) De Novo Damaging Variants, Clinical Phenotypes, and Post-Operative Outcomes in Congenital Heart Disease. Circ Genomic Precis Med 13(4):e002836. 10.1161/CIRCGEN.119.002836 Jalali A, Simpao AF, Gálvez JA, Licht DJ, Nataraj C (Aug. 2018) Prediction of Periventricular Leukomalacia in Neonates after Cardiac Surgery Using Machine Learning Algorithms. J Med Syst 42(10):177. 10.1007/s10916-018-1029-z Jalali A et al (Jun. 2020) Deep Learning for Improved Risk Prediction in Surgical Outcomes. Sci Rep 10:9289. 10.1038/s41598-020-62971-3 Ruiz-Fernández D, Monsalve Torra A, Soriano-Payá A, Marín-Alonso O, Triana Palencia E (Apr. 2016) Aid decision algorithms to estimate the risk in congenital heart surgery. Comput Methods Programs Biomed 126:118–127. 10.1016/j.cmpb.2015.12.021 Smith AH, Gray GM, Ashfaq A, Asante-Korang A, Rehman MA, Ahumada LM (Feb. 2024) Using machine learning to predict five-year transplant-free survival among infants with hypoplastic left heart syndrome. Sci Rep 14(1):4512. 10.1038/s41598-024-55285-1 Sunthankar SD et al (Aug. 2023) Machine Learning to Predict Interstage Mortality Following Single Ventricle Palliation: A NPC-QIC Database Analysis. Pediatr Cardiol 44(6):1242–1250. 10.1007/s00246-023-03130-z Tong C et al (2024) Machine learning prediction model of major adverse outcomes after pediatric congenital heart surgery: a retrospective cohort study, Int. J. Surg. Lond. Engl. , vol. 110, no. 4, pp. 2207–2216, Jan. 10.1097/JS9.0000000000001112 Wisotzkey BL et al (Dec. 2023) Risk factors for 1-year allograft loss in pediatric heart transplant patients using machine learning: An analysis of the pediatric heart transplant society database. Pediatr Transpl 27(8):e14612. 10.1111/petr.14612 Yan L, Meng Y, Sun H, Liu X, Han B (2025) Application of machine learning in predicting postoperative arrhythmia following transcatheter closure of perimembranous ventricular septal defects, Pol. Heart J. Kardiologia Pol. , vol. 83, no. 3, Art. no. 3. 10.33963/v.phj.103535 Zeng X et al (Aug. 2021) Explainable machine-learning predictions for complications after pediatric congenital heart surgery. Sci Rep 11(1):17244. 10.1038/s41598-021-96721-w Zürn C et al (Sep. 2023) Model-driven survival prediction after congenital heart surgery. Interdiscip Cardiovasc Thorac Surg 37(3). 10.1093/icvts/ivad089 Rogers L et al (Jul. 2017) Improving Risk Adjustment for Mortality After Pediatric Cardiac Surgery: The UK PRAiS2 Model. Ann Thorac Surg 104(1):211–219. 10.1016/j.athoracsur.2016.12.014 Normand S-LT et al (Sep. 2022) Mortality Prediction After Cardiac Surgery in Children: An STS Congenital Heart Surgery Database Analysis. Ann Thorac Surg 114(3):785–798. 10.1016/j.athoracsur.2021.11.077 Faerber JA et al (Jul. 2021) Identifying Risk Factors for Complicated Post-operative Course in Tetralogy of Fallot Using a Machine Learning Approach. Front Cardiovasc Med 8. 10.3389/fcvm.2021.685855 Du Y, Antoniadi AM, McNestry C, McAuliffe FM, Mooney C (2022) The Role of XAI in Advice-Taking from a Clinical Decision Support System: A Comparative User Study of Feature Contribution-Based and Example-Based Explanations, Appl. Sci. , vol. 12, no. 20, Art. no. 20, Jan. 10.3390/app122010323 Moons KGM et al (Jan. 2019) PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration. Ann Intern Med 170(1):W1–W33. 10.7326/M18-1377 Wolff RF et al (Jan. 2019) PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med 170(1):51–58. 10.7326/M18-1376 Jone P-N et al (Dec. 2022) Artificial Intelligence in Congenital Heart Disease. JACC Adv 1(5):100153. 10.1016/j.jacadv.2022.100153 Auctores Advancements in AI Heart Model Technology: A Comprehensive Review, Auctores. Accessed: Nov. 11, 2025. [Online]. Available: https://auctoresonline.org/article/advancements-in-ai-heart-model-technology-a-comprehensive-review Kurath-Koller S (Jun. 2025) Artificial intelligence in pediatrics: promise, peril, and the path ahead. Front Pediatr 13. 10.3389/fped.2025.1631521 Salavati A et al (Dec. 2024) Artificial Intelligence Advancements in Cardiomyopathies: Implications for Diagnosis and Management of Arrhythmogenic Cardiomyopathy. Curr Heart Fail Rep 22(1). 10.1007/s11897-024-00688-4 Homsy J et al (2015) De novo mutations in congenital heart disease with neurodevelopmental and other congenital anomalies, Science , vol. 350, no. 6265, pp. 1262–1266, Dec. 10.1126/science.aac9396 Sifrim A et al (Sep. 2016) Distinct genetic architectures for syndromic and nonsyndromic congenital heart defects identified by exome sequencing. Nat Genet 48(9):1060–1065. 10.1038/ng.3627 Kars ME et al (Mar. 2025) Deciphering the digenic architecture of congenital heart disease using trio exome sequencing data. Am J Hum Genet 112(3):583–598. 10.1016/j.ajhg.2025.01.024 Boeckling F et al (Mar. 2025) Extracellular Matrix Proteins Improve Risk Prediction in Patients Undergoing Transcatheter Aortic Valve Replacement. J Am Heart Assoc 14(5):e037296. 10.1161/JAHA.124.037296 Ceschin R et al (Sep. 2018) A computational framework for the detection of subcortical brain dysmaturation in neonatal MRI using 3D Convolutional Neural Networks. NeuroImage 178:183–197. 10.1016/j.neuroimage.2018.05.049 Song L et al (2025) Establishment of a Stacking Machine Learning Model Predicting Cardiac Phenotype in Ectopia Lentis Patients Based on Genotype and Ocular Phenotype. Int J Med Sci 22(14):3501–3510. 10.7150/ijms.109657 Cornhill AK et al (2022) Machine Learning Patient-Specific Prediction of Heart Failure Hospitalization Using Cardiac MRI-Based Phenotype and Electronic Health Information. Front Cardiovasc Med 9:890904. 10.3389/fcvm.2022.890904 Zaka A et al (Jan. 2025) Machine-learning versus traditional methods for prediction of all-cause mortality after transcatheter aortic valve implantation: a systematic review and meta-analysis. Open Heart 12(1):e002779. 10.1136/openhrt-2024-002779 Bertsimas D et al (2022) Benchmarking in Congenital Heart Surgery Using Machine Learning-Derived Optimal Classification Trees, World J. Pediatr. Congenit. Heart Surg. , vol. 13, no. 1, pp. 23–35, Jan. 10.1177/21501351211051227 Simsic JM, Bradley SM, Stroud MR, Atz AM (2005) Risk Factors for Interstage Death After the Norwood Procedure, Pediatr. Cardiol. , vol. 26, no. 4, pp. 400–403, Aug. 10.1007/s00246-004-0776-4 Crabb BT et al (Summer 2025) Characteristics of left ventricular dysfunction in repaired tetralogy of Fallot: A multi-institutional deep learning analysis of regional strain and dyssynchrony. J Cardiovasc Magn Reson Off J Soc Cardiovasc Magn Reson 27(1):101886. 10.1016/j.jocmr.2025.101886 Bektaş M, Tuynman JB, Costa Pereira J, Burchell GL, van der Peet DL (2022) Machine Learning Algorithms for Predicting Surgical Outcomes after Colorectal Surgery: A Systematic Review. World J Surg 46(12). 10.1007/s00268-022-06728-1 Mayourian J, Geggel R, La Cava WG, Ghelani SJ, Triedman JK (2025) Pediatric Electrocardiogram-Based Deep Learning to Predict Secundum Atrial Septal Defects, Pediatr. Cardiol. , vol. 46, no. 5, pp. 1235–1240, Jun. 10.1007/s00246-024-03540-7 Sarris GE et al (Jul. 2024) Congenital Heart Surgery Machine Learning-Derived In-Depth Benchmarking Tool. Ann Thorac Surg 118(1):199–206. 10.1016/j.athoracsur.2023.10.034 Austin R et al (Jan. 2024) A multitiered analysis platform for genome sequencing: Design and initial findings of the Australian Genomics Cardiovascular Disorders Flagship. Genet Med Open 2:101842. 10.1016/j.gimo.2024.101842 Char DS, Deuitch NT, Berent MK, Chung WK, Krosnick JA, Magnus D (2025) Impact of Genomic Sequencing Information on Physicians’ Treatment Recommendations for Children with Congenital Heart Disease, Genet. Med. Open , p. 103470, Nov. 10.1016/j.gimo.2025.103470 Chen T et al (Dec. 2024) Genomic insights for personalised care in lung cancer and smoking cessation: motivating at-risk individuals toward evidence-based health practices. eBioMedicine 110:105441. 10.1016/j.ebiom.2024.105441 Singh M et al (Jul. 2024) Artificial intelligence for cardiovascular disease risk assessment in personalised framework: a scoping review. eClinicalMedicine 73:102660. 10.1016/j.eclinm.2024.102660 Additional Declarations The authors declare no competing interests. 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-9055427","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":602191738,"identity":"97d7f65a-3c60-46b9-8c71-b65be922ef94","order_by":0,"name":"Paschalina Lialiou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxklEQVRIiWNgGAWjYDACdsYGhgcGDAz8QPYBBgNitDADtSQAVUo2EK8FiBOA2OAAse4yZ2Zu/JBQYJO4+UbuwcMFBQx22xsIaLFsZmyWSDBIS9x2Iy/h8AwDhuQ5hKwzOMzYANRyGKglx+AwD1CLBCGHAbU0/wBp2TyDBC1tYFs2SEC02BHUAvRLmwXQL8YzzrwxAPpFIoGgFnP29sc3Pvyxke1vzzH+XPDHxp6ww5A5wDiSSGwgVQuDPSEdo2AUjIJRMPIAAOlmPnk8L0AaAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0009-5832-7915","institution":"","correspondingAuthor":true,"prefix":"","firstName":"Paschalina","middleName":"","lastName":"Lialiou","suffix":""},{"id":602191739,"identity":"8d396fbd-25ef-4c01-bcd9-1478e7bd25b6","order_by":1,"name":"Ilias Maglogiannis","email":"","orcid":"https://orcid.org/0000-0003-2860-399X","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Ilias","middleName":"","lastName":"Maglogiannis","suffix":""}],"badges":[],"createdAt":"2026-03-07 05:04:19","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9055427/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9055427/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104340242,"identity":"ab17e723-6275-4cfd-842e-5a87e0c6b96f","added_by":"auto","created_at":"2026-03-10 16:36:41","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":45468,"visible":true,"origin":"","legend":"\u003cp\u003eSystematic literature survey process and corresponding results in PRISMA 2020 diagram.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/ed1ec437845ff6134d050d10.jpg"},{"id":104340238,"identity":"1fc60239-d987-4ed7-9848-87a214998099","added_by":"auto","created_at":"2026-03-10 16:36:41","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":22134,"visible":true,"origin":"","legend":"\u003cp\u003eThe percentages of AI/ML methodologies used by the selected studies. Abbreviations: XGBoost:Extreme Gradient Boosting, SVN:Support Vector Network, RBF kernel:Radial Basis Function kernel, DNN: Deep neural network, RF:Random Forest, DT:Decision Tree, LR:Logistic Regression, MLP:Multilayer Perceptron, SOM:Self-Organizing Map, RBF:Radial Basis Function, GBT:Gradiant Boosting Trees, LightGBM: Light Gradiant Boosting Machine, CPH: Cox Proportional Hazards.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/4a126ed241405ae50af344d6.jpg"},{"id":104405496,"identity":"d71af03d-9172-48de-b068-49ff17e71880","added_by":"auto","created_at":"2026-03-11 12:23:07","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":36194,"visible":true,"origin":"","legend":"\u003cp\u003eSummary of Risk of Bias assessment, according to PROBAST tool.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/7d2b9c47518207130d38c0f8.jpg"},{"id":104340239,"identity":"3f184d84-b8d1-4321-8830-966b8841ddd4","added_by":"auto","created_at":"2026-03-10 16:36:41","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":71482,"visible":true,"origin":"","legend":"\u003cp\u003eAn overview of the heterogeneity of the included studies included according to the number of variables that was selected for ML models training. In the graph above, two studies were excluded as the number of variables was not clearly defined.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/1fb2301e2bad6c189b8a06c0.jpg"},{"id":104340240,"identity":"7d2f813c-239c-471c-88ba-d72371161dda","added_by":"auto","created_at":"2026-03-10 16:36:41","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":54290,"visible":true,"origin":"","legend":"\u003cp\u003eAn overview of the heterogeneity in the size of the datasets used in the studies included in the review. The sample size represents the number of patients who participated in the cohort studies included in this systematic literature review.\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/e73e94b61bc8c22651d7b7f2.jpg"},{"id":104779853,"identity":"e30af363-e9d6-480c-ac4c-c57455eaf3ce","added_by":"auto","created_at":"2026-03-17 07:46:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1561695,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9055427/v1/c3730861-ba79-4ba2-88df-032960ac1053.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eOverview of AI and Machine Learning Methods for Outcome Prediction in Pediatric Congenital Heart Surgery\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eCongenital heart disease (CHD) accounts almost one-third of all major congenital anomalies. CHD birth prevalence worldwide varies and may also be observed differences that are due to genetic, environmental, socioeconomical or ethnic origins[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Nonetheless, there was an increasing birth prevalence over the total reported CHD births till 1995, over the last 15 years, a stabilization occurred, corresponding to 1.35\u0026nbsp;million newborns with CHD every year. Asia reported the highest CHD birth prevalence with 9.3 per 1,000 live births, followed by Europe and North America that reported 8.2 and 6.9 respectively[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e], [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Over 97% of them can be expected to reach adulthood through a life-time health care management[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e], [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Despite the innovative advances in cardiovascular diagnostics, cardiothoracic surgery and perioperative care, children with CHD receiving anatomical radical treatment were still associated with high morbidity and mortality rates[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Previous studies have mainly focused on in-hospital mortality and importantly, on morbidity, as a leading cause of mortality that measure the performance and the direct quality of the improvement initiatives[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNonetheless, the complex prognosis and treatment after a congenital heart surgery remains ubiquitous due to the complex nature of surgery and influence the quality of treatment. Due to the extraordinary advances in cardiac surgery, the postoperative complications are many. Some of them have been reported, having important contributions in mortality, hospital stay, cost and quality of life. In recent years, the scientific community especially cardiac surgeons evolve with the assessment of the risk of congenital heart surgery using empirical tools, scales and predictors which can estimate the risk mortality of pediatric patients[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMachine learning (ML), a field of artificial intelligence (AI), enables the discovery of important patterns within complex multidimensional data and highlight the utility of its applications for clinical risk and scope to improve patient surveillance[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. There is an increasing interest in the use of large claims databases in medical practice and research, the essential role of ML identify patients with any kind of diseases[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. ML approaches such as tree-based models and deep learning are able to identify complex data patterns among variables exploring complex interactions and disease complications[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. These ML methods have the potential to generate more accurate prediction and classification beyond traditional methods, with impressive beneficials in cardiovascular medicine[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. More precisely, in pediatric congenital heart surgery, ML models can accurately predict the risk of major adverse postoperative outcomes (APOs) affecting greatly the mortality, hospital stay, care management and planning, and quality of life[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. These predictions provide reliable interpretations for high-risk contributor identification and informed clinical decisions. Also, previous studies have explored the potential of ML models as a tool for prediction mortality and other postoperative outcomes in cardiac surgery, showing great discriminative power to revolutionize cardiac surgery[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e The aim of current study was to systematically review the relevant literature and identify all studies that have utilized AI to predict the post-operative outcomes and risk mortality in pediatric patients undergoing surgery for congenital heart defects.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003eThis systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e], [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. PRISMA statement is the most common guidance that has been used by authors and reviewers, and mentions the whole literature search procedure[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Moreover, PRISMA ensures the quality of reports and guide authors about the methodological strategy and study assessment[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Current SLR utilizes the 27 items of PRISMA checklist to verify that each search component is completely reported and reproducible.\u003c/p\u003e \u003cp\u003eThe proposed reviewing process is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Firstly, it is identified the purpose of present SLR which have been motivated this research. Three digital databases were recruited with predetermined search terms. Subsequently, an initial screening of identified studies was conducted, and the predefined inclusion and exclusion criteria were applied. The final set of eligible studies was coded, and relevant data aligned with the objectives of the present study were extracted. The extracted information was then systematically organized, compared, and synthesized to address the research questions. Zotero 7 for macOS (20,21), an open-access reference management software, was used to identify and remove duplicate records, support citation tracking, and manage source synthesis.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Search strategy\u003c/h2\u003e \u003cp\u003eA comprehensive search was conducted between July to September 2025 in three electronic databases, PubMed, ScienceDirect and Scopus by one author. Limitations about the English language and publication date within the past decade were obtained. Then, the search results were passed on to the second author for further evaluation. A series of keywords such as \u0026ldquo;transcatheter aortic valve replacement\u0026rdquo;, \u0026ldquo;congenital heart diseases\u0026rdquo;, \u0026ldquo;mortality\u0026rdquo;, \u0026ldquo;complications\u0026rdquo;, \u0026ldquo;machine learning\u0026rdquo;, \u0026ldquo;artificial intelligence algorithms\u0026rdquo;, \u0026ldquo;large language models\u0026rdquo;, \u0026ldquo;children\u0026rdquo;, pediatric patients\u0026rdquo; were identified and formed the search queries using the Boolean operators \u0026ldquo;AND\u0026rdquo; and \u0026ldquo;OR\u0026rdquo;. The search term were combined and formed separate search parameters as follows: (Transcatheter aortic valve replacement) OR (congenital heart diseases) AND (machine learning) AND (mortality), (Transcatheter aortic valve replacement) AND (AI algorithms) AND (mortality), (Transcatheter aortic valve replacement) AND (AI algorithms) AND (complications), Transcatheter aortic valve replacement) AND (Large language models) AND (complications), (congenital heart diseases) AND (machine learning) AND (complications) AND (children) OR (pediatric patients).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Inclusion and exclusion criteria\u003c/h2\u003e \u003cp\u003eThe eligibility criteria were defined as follows: we included in this review those studies that investigates the adoption of machine learning or deep learning methods to predict the implications and mortality risk in pediatric patients undergoing a congenital heart surgery like heart transplantation, open heart surgery, transcatheter aortic valve implantation or narrow aortic valve (TAVI/TAVR) procedures or Norwood procedure. We used research studies, written in English language and published till September 2025, in which participants were infants and children up to the age of eighteen. For this review, we reclaimed studies that used clinical, laboratory data including images from MRIs and CTs, while only the research studies in which the accuracy of the predictive models was assessed.\u003c/p\u003e \u003cp\u003eWe excluded pilot studies, research articles that they did not specify the predictive analytics of congenital heart surgery outcomes or did not use machine learning methodologies. Also, we omitted studies involving both children and adult population. Additional screening of studies eliminated those that did not mentioned the performance measures of ML models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Study selection\u003c/h2\u003e \u003cp\u003eAfter recruitment of the three selected databases, a total of 256 articles identified in the literature search. There were 32 duplicates that we extracted them manually. During the primary screening, we assessed the titles and the abstracts of the retrieved studies among the predetermined research objectives and the established inclusion/ exclusion criteria. Also, we exclude the articles that were not associated with the topic of present SLR. Additionally, we extracted all the randomized control trials, study protocols, reviews and meta-analysis. Finally, a total number of 121 articles were sought for the text retrieval. During the phase of screening 103 were excluded due to the target population (n\u0026thinsp;=\u0026thinsp;75) not being pediatric patients, studied other pediatric syndrome (n\u0026thinsp;=\u0026thinsp;9), were not associated with the aim of current study (n\u0026thinsp;=\u0026thinsp;6) and due to the article type (n\u0026thinsp;=\u0026thinsp;13). A total number of 15 studies included finally in the SLR.\u003c/p\u003e \u003cp\u003eOn a secondary stage, all the selected articles sent to the second author to carry out the data extraction independently. The structured data extraction approach was essential to ensure that the systematic review was precise and comprehensive. The following information was extracted:\u003c/p\u003e \u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003ePublication information: First author\u0026rsquo;s name, source and publication year were listed.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eStudy design: The study design such as prospective cohort, retrospective cohort was identified.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e Objectives: The research goals of each included study were outlined to ensure that coincide with the present review.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eInformation about population: We collected information about sample size, demographics and country of research.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eData source: We identified the data source of each study and the dataset that was derived.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eModel selection: We noted the AI prediction models used for outcome prediction.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eTop predictors of congenital heart surgery outcomes: The most powerful predictors of cardiac surgery outcomes were listed.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eLimitations: Any bias or constrains in the studies were noted, such as small sample size, selection bias or other.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study characteristics\u003c/h2\u003e \u003cp\u003eThe characteristics of the included studied are summarized in Table \u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. In this table are outlined, the characteristics of included population in each study. The included studies were published from 2016 to 2024 and mostly from USA. The design of all included studies was retrospective cohort, which means that all the researchers examined the outcome of their interest of each study. A comparison of the reviewed studies according to the country distributed, source of data and data range is illustrated in Table \u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Data of the majority of studies were derived from patients\u0026rsquo; electronic health records, and ML models were trained and tested between 71 to 24,685 individuals in pediatric population. The majority of the reviewed population was males in a portion of almost 60%. Moreover, most of the studies utilized K-fold cross validation while in some studies is reported a train and test portion of 7:3 or 8:2. Among the studies, the specific type of heart surgery that investigated were 50% various congenital heart surgery, 25% Norwood procedure (stage 1 palliation), 9% first open-heart surgery in CHD patients, 8% heart transplantation and 8% transcatheter device closure of perimembranous ventricular septal defect (pmVSD).\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e depicts the AI prediction models constructed in each study. The most commonly utilized algorithm was Gradient Boosting Trees, in a portion of 26% (XGBoost, CatBoost, LightGBM), followed by Logistic Regression (LR) 15%, Random Forest (RF) 14%, Decision Trees (DT) and Support Vector Machine (SVM) in a portion of 9% for each of them, Radial Basis Function kernel, decision tree, Cox Proportional Hazards, other adaptive machine learning models, Self-Organizing Map, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. (SOM), Multilayeer Perceptton (MLP) and deep neural network (DNN). As we\u0026rsquo;ve seen, the most frequent postoperative outcome assessed by authors in the reviewed studies was mortality.\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\u003eCharacteristics of the population of the included studies in present systematic literature review. Abbreviations: RCS: Retrospective cohort study, CICU: Cardiac Intensive Care Unit, HLHS: hypoplastic left heart syndrome, CHD: Congenital Heart Diseases, CHS: Congenital Heart Surgery, pmVSD: perimembranous ventricular septal defects, VSD: ventricle heart disease, CPB: cardiopulmonary bypass.\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=\"left\" 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\u003eFirst author (Year)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePopulation size\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTrain/Test Proportion\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGender (male)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePopulation characteristics\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDu X. (2022)[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e24,685 patients; number of in-hospital deaths: 595(2.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7.5:2.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.59%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAge: median 316 days (range 1\u0026ndash;6,568 days)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBoskovski (2020)[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2,517 CHD patients with WES; of these 1,268 underwent first open-heart surgery and had surgical follow-up data.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e57.6%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eVarious CHD patients; de novo damaging variant carriers: 294/2,517 (11.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali (2018)[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71 neonates\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1:6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47.89%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeonates undergoing cardiac surgery, including those with HLHS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali (2020)[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e549 patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7:3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61.74%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNeonates undergoing the Norwood procedure for single ventricle CHD\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRuiz-Fern\u0026aacute;ndez D. (2016)[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9:1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients undergoing CHS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmith A. (2024)[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInfants with HLHS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSunthankar SD, (2023)[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3,267 infants\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8:2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e62%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInfants with single VSD, including HLHS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChaoyang Tong, et al. (2024)[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23,000 pediatric patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTraining cohort n\u0026thinsp;=\u0026thinsp;13927\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e54%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients undergoing CHS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWisotzkey BL. (2023)[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric heart transplant patients\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYan L. (2024)[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e384 children with pmVSD who underwent transcatheter closure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7:3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30.7%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients with pmVSD treated with transcatheter device closure\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXian Zeng (2021)[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1,964 patients included in the analysis (after exclusions)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7:3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e52.8%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients (median age\u0026thinsp;~\u0026thinsp;11 months [IQR 4\u0026ndash;26 months]) undergoing CHS; of these, 34.4% developed postoperative complications\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZ\u0026uuml;rn C. (2023)[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1765 patients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1:1.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eChildren (ages 1 day to 18 years) undergoing CHS with CPB.\u003c/p\u003e \u003cp\u003eChildren with \u0026lt;\u0026thinsp;24 h PCICU stay or multiple ops only last stay counted\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRogers et al, 2017[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e21,838\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4/1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients with various surgical procedures\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNormad et al 2022[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e98.825\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7:3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e55%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePediatric patients (including premature neonates) with congenital or non-congenital abnormalities or chromosomal abnormality.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFaerber 2021[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e162\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8:2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e62.73%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eChildren with TOF, some of them had a prior palliative procedure, 58.6% of them surgical repair\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 \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\u003eThere is a comparison of the included studies, according to the country distributed, settings/ source of data and data range.\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 \u003cp\u003eStudy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCountry\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSetting/Source of data\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData range\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDu Y., et al. (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSingle-centre, LIS,HIS, ICU database, clinical data repository\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2006\u0026ndash;2017\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBoskovski M., et al. (2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMulti-centre, Pediatric Cardiac Genomics Consortium\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali A., et al. (2018)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChildren's Hospital of Philadelphia, CICU-Local EMR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali A., et al. (2020)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePediatric Heart Network Single Ventricle Reconstruction Trial\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2005\u0026ndash;2009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRuiz-Fern\u0026aacute;ndez D., et al. (2016)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eColombia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChildren Heart Disease database from Cardiovascular Foundation of Colombia, local EMR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmith A., et al. (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe pediatric Heart Network Single Ventricle reconstruction trial\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2005\u0026ndash;2009\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSunthankar SD., et al (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA, Canada\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNational Pediatric Cardiology Quality Improvement Collaborative (NPC-QIC), Pediatric Electronic Data Capture (PEDCAP) database\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2008\u0026ndash;2019\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTong C., et al. (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCongenital Heart Surgery Database Collection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWisotzkey BL., (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePediatric Heart Transplant Society\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2010\u0026ndash;2019\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYan L., et al (2024)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSingle-centre hospital in China, Platform for Epidiomological Investigation and Precision Treatment of Childhood Heart Disease\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2002\u0026ndash;2024\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZeng X., et al. (2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eChina\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSingle tertiary centre (Children's Hospital of Zhejiang University School of Medicine)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZurn C., et al. (2023)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGermany\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDepartments of pediatric cardiology and cardiac surgery in Freiburg and Heidelberg University Heart Center\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2014\u0026ndash;2019\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRogers L., et al. (2017)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUK\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUK National Congenital Heart Audit\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2009\u0026ndash;2014\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNormand SL., et al. (2022)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eThe Society of Thoracic Surgeons (STS) Congenital Heart Surgery Database (CHSD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2014\u0026ndash;2018\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFaerber J., et al. (2021)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUSA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eChildren's Hospital of Philadelphia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2012\u0026ndash;2018\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 \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eOverview of the AI Models that were utilized by authors in the reviewed papers according to the type of cardiac surgery and the evaluated postoperative outcome. Abbreviations: CHS: Congenital Heart Surgery, XGBoost:Extreme Gradient Boosting, SVN:Support Vector Network, RBFK: Radial Basis Function kernel, DNN: Deep neural network, RF:Random Forest, DT:Decision Tree, LR:Logistic Regression, MLP:Multilayer Perceptron, SOM:Self-Organizing Map, RBF:Radial Basis Function, GBT:Gradiant Boosting Trees, LightGBM: Light Gradiant Boosting Machine, CPH: Cox Proportional Hazards, pmVSD: perimembranous Ventricular Septal Defects, PVL: Periventricular leukomalacia, LCOS: Low cardiac output syndrome, DVT: deep venous thrombosis, MARS: Multivariate adaptive regression spline model.\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 \u003cp\u003eAuthors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eType of Procedure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAI Algorithm\u003c/p\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDu et al., 2022[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBoskovski et al, 2020[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFirst open-heart surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMortality, extubation time\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMLP, RBFN, SOM, DT\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali et al, 2018[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePVL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVN, RBF kernel\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJalali et al, 2020[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNorwood procedure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMortality, length of hospital stays\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDNN, XGBoost, RF, DT, ridge LR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRuiz-Fern\u0026aacute;ndez D, et al. 2016[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSurgical risk classification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMLP, SOM, RBF networks, DT\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmith A., et al.2024[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHypoplastic left heart syndrome\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFive-year transplant free survival\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNot specified\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSunthankar SD, et al 2023[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSingle ventricle disease, Norwood procedure\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR, RF, XGBoost, GBT, LightGBM\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChaoyang Tong, et al. 2024[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLCOS, pneumonia, renal failure, DVT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLightGBM, LR, SVM, RF, CatBoost\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]Wisotzkey BL, 2023[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHeart transplantation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1-year allograft loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, CPH\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]Yan L., et al 2024[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTranscatheter device closure of pmVSD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative arrythmia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM, LR, RF, XGBoost\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]Zeng X., et al. 2021[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost, LR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]Z\u0026uuml;rn C., et al. 2023[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative survival at 30 days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRogers., et al. 2017[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative survival at 30 days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNormand et al 2022[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVarious CHS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative mortality within 30-days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLasso, BART\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFaerber 2021[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTetralogy of Fallot\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePostoperative cardiac complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF, GBM: two-stage with and without a MARS\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=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Performance metrics and interpretation\u003c/h2\u003e \u003cp\u003eOverall AUCs above 0.65 were achieved by all the reviewed research papers. The most frequent investigated outcome was postoperative mortality followed by other postoperative complications associated with lung, cardiac rhythm and infections, (Table \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Different ML models across the 15 different articles were used and tested. Among the best performance of prediction models was achieved by Tong et al. 2024 [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] with an AUC of 0.97 as reported by LightGBM for the prognosis of deep venous thrombosis, following 0.96 by renal failure by LightGBM. Also, an AUC portion of 94.86% was achieved by LR for the assessment of postoperative survival through a variety of clinical modalities by Zurn et al. 2023 [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. On the other hand, the worst performance demonstrated by Sunthankar et al. 2023 using the LightGBM model with an AUC of 64% for the influential factors of mortality risk.\u003c/p\u003e \u003cp\u003eCalibration was described for models n\u0026thinsp;=\u0026thinsp;7 (47%), in which 4 of them [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] used visual calibration plots to ensure risk mortality, risk complications across deciles, and compare predicted to observed probabilities. Two studies (n\u0026thinsp;=\u0026thinsp;2) used brier score and Integrated calibration index(ICI) for grass loss risk[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] and risk of renal failure in children. Calibration curves was used by one model[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] and [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] used standard logistic calibration. Class probabilities, utilizing decision trees used by one study[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] to determine calibration via frequency, and time -dependent calibration, across 5 years of follow-up used by [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eImportantly, the model of Z\u0026uuml;rn et al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] delivered excellent generalization due to the high AUC values (0.948) in test dataset with specificity 89.48% and sensitivity 85.00%, (slightly below of Freiburg train dataset), the model\u0026rsquo;s calibration assessments shows that for very high and very low probabilities correctly identifies the outcomes. But, for probabilities in between, the model trends to overestimate the mortality risk. In particular, the model of Rogers\u0026rsquo; et al [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e] had a median AUC of 0.83 (range 0.82 to 0.83), showing excellent discrimination, with a median calibration slope of 0.92 (range 0.64 to1.25) and median calibration intercept of -0.23 (range 1.08 to 0.85; perfect calibration slope\u0026thinsp;=\u0026thinsp;1 and intercept\u0026thinsp;=\u0026thinsp;0), indicating slight underprediction. Moreover, Normad et al., [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] indicates calibration for mortality predictions at the higher end of risk from the lasso and Bayesian additive regression trees models, which was better than the traditional STS CHSD model. On the other hand, the models of Faerber et al were generally well-calibrated, though RF models can sometimes be overconfident without proper tuning. Moreover, the calibration curves\u0026rsquo; of the predictive model of the development of arrhythmia in pmVSD patients demonstrated that there was a little deviation between the ideal and real, which means that the predictive model had some predictive value following internal validation[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTong et al [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] evaluated the interpretations of ML models by using SHAP ranking feature importance and representing them as patient-specific visualizations. The SHAP values showed important clinical factors associated with risk of major APOs, of which longer mechanical ventilation time was the most important factor leading to an increased risk of the complications. Predictive model of [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] performed 7.6% cumulative incidence of grass loss or mortality. Furthermore, random forests had favorable discrimination and calibration, while performed closer alignment between predicted and observed risk.\u003c/p\u003e \u003cp\u003eExplainable AI (XAI) features and strategies used from (n\u0026thinsp;=\u0026thinsp;5) of reviewed papers. SHAP (Shapley additive explanations) values used by (n\u0026thinsp;=\u0026thinsp;3) of the models [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], feature importance analysis utilized by one reviewed model[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], while [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] adopted an intrinsic exlpainability design for model interpretation. The SHAP values of the predictive model of [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e] estimated the most important predictors in the final model. In other words, the diagnosis of congenital heart disease increased one year predicted risk of grass loss by 1.7, the need for mechanical circulatory support increased predicted risk by 2 and single ventricle CHD increased predicted risk by 1.9. In other words, they concluded that risk prediction models used to facilitate patient selection for pediatric heart transplant can be improved without loss of interpretability using machine learning.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Model validation\u003c/h2\u003e \u003cp\u003eFor the internal validation of machine learning models, most studies used cross validation (n\u0026thinsp;=\u0026thinsp;3), hold-out split (n\u0026thinsp;=\u0026thinsp;1), temporal split (n\u0026thinsp;=\u0026thinsp;1), repeated split(n\u0026thinsp;=\u0026thinsp;1), leave one out (LOOCV)(n\u0026thinsp;=\u0026thinsp;1), a random split of the dataset (n\u0026thinsp;=\u0026thinsp;2), or bootstrapping (n\u0026thinsp;=\u0026thinsp;1). External validation was performed in 7 studies. Five studies [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] conducted bicentric or multicenter validation, and 3 studies [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]used temporal validation. The discriminative ability (AUCs) of these models ranged between 0.64 and 0.97, and the calibration was reported for 7 machine learning models. Temporal split of data validation from 2023 to 2015 was adopted, also, by Faerber et al[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAbove is cited the best performing ML algorithm due to the measured post-operative outcome. Abbreviations: XGBoost:Extreme Gradient Boosting, MLP:Multilayer Perceptron,, RBF kernel:Radial Basis Function kernel, SOM:Self-Organizing Map, DT:Decision Tree, SVN:Support Vector Network, DNN: Deep neural network, RBF:Radial Basis Function, GBT:Gradiant Boosting Trees, LightGBM: Light Gradiant Boosting Machine, RF:Random Forest, LR:Logistic Regression, pLOS:prolonged length of hospital stay, PVL:Periventricular leukomalacia, MMI: Modified Mutual Information, LCOS:Low cardiac output syndrome, DVT: deep venous thrombosis, PHN: Pediatric Heart Network., SVR: Single Ventricle Reconstruction, STS model: Society of Thoracic Surgeons model.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\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 \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eAuthors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eBest Performing Algorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eBest Performance metrics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eComments (Small/Imbalanced data, feature selection strategies, lack of external validation)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eDu X., 2022[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eOutperformed traditional scores: STS-EACTS(0.748) and RACHS-1 (0.677).\u003c/p\u003e \u003cp\u003eLarge dataset(N\u0026thinsp;=\u0026thinsp;24,684), mortality was only 2.4% (595 deaths). This represents a highly imbalanced scenario.\u003c/p\u003e \u003cp\u003eThe study primarily relies on internal validation (like a 70/30 split of the same institutional registry).\u003c/p\u003e \u003cp\u003eNo independent validation on a separate hospital cohort was reported.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eBoskovski et al, 2020[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality, time to final extubation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eMLP, RBFN, SOM, DT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e11.7% of patients with de novo variants (imbalanced).\u003c/p\u003e \u003cp\u003eAgnostic ML used to identify variants (CNVs).\u003c/p\u003e \u003cp\u003eNone validation (internal cohort only).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eJalali A., et al. 2018[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003ePVL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eSVN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eF1 score reported 0.81-1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eSmall dataset (n\u0026thinsp;=\u0026thinsp;71), 33%PVL rate\u003c/p\u003e \u003cp\u003eMMI ranking system.\u003c/p\u003e \u003cp\u003eNone validation (Small single-center scale).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eJalali A., et al, 2020[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality, pLOS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eDNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eFor mortality prediction: DNN: AUROC\u0026thinsp;=\u0026thinsp;0.95\u003c/p\u003e \u003cp\u003eFor pLOS: DNN: AUROC\u0026thinsp;=\u0026thinsp;0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eHigh imbalance data (mortality\u0026thinsp;~\u0026thinsp;7\u0026ndash;8% in SVR trial).\u003c/p\u003e \u003cp\u003eDNN used on preoperative clinical features.\u003c/p\u003e \u003cp\u003eInternal validation (SVR trial dataset split).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eRuiz-Fern\u0026aacute;ndez D, et al. 2016[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eSurgical risk classification (RACHS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eMLP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eMLP: Accuracy\u0026thinsp;=\u0026thinsp;0.999\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e3 categories of risk (low, medium, high).\u003c/p\u003e \u003cp\u003eManual feature selection (clinical factors).\u003c/p\u003e \u003cp\u003eNone validation (Single center/Colombia).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSmith A., et al.2024[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eNot specified\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eTd-AUC\u0026thinsp;\u0026gt;\u0026thinsp;0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eTrial cohorts PHN SVR I and II\u003c/p\u003e \u003cp\u003eIncremental feature inclusion (pre-operative to post-operative).\u003c/p\u003e \u003cp\u003eInternal (Wait-time cohort validation).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSunthankar SD, et al 2023[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.642\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eImbalanced (6.4% mortality).\u003c/p\u003e \u003cp\u003eEvaluated 180 clinical features.\u003c/p\u003e \u003cp\u003eInternal validation (Multicenter NPC-QIC database).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eChaoyang Tong, et al. 2024[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eLCOS, pneumonia, renal failure, DVT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLightGBM, LR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eLCOS: LightGBM, AUC\u0026thinsp;=\u0026thinsp;0.893\u003c/p\u003e \u003cp\u003ePneumonia: LR, AUC\u0026thinsp;=\u0026thinsp;0.929\u003c/p\u003e \u003cp\u003eRenal Failure: LightGBM, AUC\u0026thinsp;=\u0026thinsp;0.963\u003c/p\u003e \u003cp\u003eDVT: LightGBM, AUC\u0026thinsp;=\u0026thinsp;0.970\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eLarge dataset (N\u0026thinsp;=\u0026thinsp;23,000): Training (n\u0026thinsp;=\u0026thinsp;13,927) and Testing (n\u0026thinsp;=\u0026thinsp;9,073). While the cohort is large, specific complications like Renal Failure and DVT are inherently rare (imbalanced), though specific ratios were not highlighted as a failure point.\u003c/p\u003e \u003cp\u003eThe study uses a temporal split (data before 2019 for training, after 2019 for testing). While better than a random split, it remains Internal Validation as it uses data from the same institution (Shanghai Children's Medical Center).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eWisotzkey BL, 2023[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003e1-year allograft loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eNS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e3,787 patients, 7.6% graft loss (imbalanced data).\u003c/p\u003e \u003cp\u003eSHAP \u0026ndash; top 15 features were selected.\u003c/p\u003e \u003cp\u003eInternal validation (75:25 split).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eYan L., et al 2024[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003ePostoperative arrythmia\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eLR: AUC\u0026thinsp;=\u0026thinsp;0.863\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e1,384 patients (Retrospective)\u003c/p\u003e \u003cp\u003e5-variable nomogram.\u003c/p\u003e \u003cp\u003eInternal validation (7:3 split).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eZeng X., et al. 2021[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003ePrediction/classification of postoperative complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003ePrediction: AUC\u0026thinsp;=\u0026thinsp;0.839.\u003c/p\u003e \u003cp\u003eClassification: AUC\u0026thinsp;=\u0026thinsp;0.850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eIntegrated HER and time series data.\u003c/p\u003e \u003cp\u003ek-means, Dynamic Time Warping for series data.\u003c/p\u003e \u003cp\u003eInternal validation only.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eZ\u0026uuml;rn C., et al. 2023[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.9486\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eRetrospective (780 train/985 test).\u003c/p\u003e \u003cp\u003eKnowledge-driven selection feature strategy: STAT score, age, clamp time, lactate.\u003c/p\u003e \u003cp\u003eBicentric validation (Heidelberg used to test Freiburg model).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eRogers et al 2017[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eValidation dataset 4,207 episodes with 97 deaths.\u003c/p\u003e \u003cp\u003eExpert advisory panel to consider the relative importance of comorbidities and risk factors.\u003c/p\u003e \u003cp\u003eExternal validation dataset from 2014 to 2015.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eNormad et al, 2022[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eMortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eBART, lasso\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.858 to 0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003e7 different models (3 BART, 3 lasso, current STS model).\u003c/p\u003e \u003cp\u003eClinical review group discuss clinical coherence.\u003c/p\u003e \u003cp\u003eHighly robust multicenter internal validation.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eFaerber et al 2021[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003ePostoperative cardiac complications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eGB\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c8\" namest=\"c7\"\u003e \u003cp\u003eAUC\u0026thinsp;=\u0026thinsp;0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eImbalance handling: undersampling\u003c/p\u003e \u003cp\u003eFeature selection: 48 candidate variables(data-driven).\u003c/p\u003e \u003cp\u003eTemporal slip validation strategy. External validation on data from 2013\u0026ndash;2015.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"1\" nameend=\"c11\" namest=\"c11\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Risk of bias assessment\u003c/h2\u003e \u003cp\u003ePrediction models in health care use predictors to estimate the probability that a condition or a disease is already present or will occur in the future. Within this knowledge framework, the authors examined the risk of bias (ROB) of the prediction models in present SLR. The risk of bias of the selected articles was assessed using the PROBAST tool [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e], [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e] by two independent reviewers. PROBAST tool consists of 4 domains, participants, predictors, outcome, and analysis with a total of 20 questions to facilitate the overall ROB assessment. Each domain can be graded with a low-risk bias, a high-risk bias or an unclear risk of bias. Studies with a low risk of bias in all four domains were classified as low risk while having a high risk of bias in even one domain were judged as a high risk of bias. Moreover, if one or more domains characterized as unclear risk of bias and the other as low-risk then the overall judgement to the study would be declared as unclear.\u003c/p\u003e \u003cp\u003eOverall, eight, three and one studies had high, low and unclear risk of bias respectively. The most common domain of bias was analysis (domain 4), while the domain appearing the least amount of bias was outcome (domain 3). Given that approximately 54% of the included studies were assessed as having a high risk of bias\u0026mdash;primarily due to limitations in the analysis domain\u0026mdash;concerns remain regarding the applicability of the reported predictive models. Consequently, further validation is required, underscoring the need for rigorously designed, high-quality studies in the field of congenital heart disease, particularly within pediatric populations, (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Key findings and conclusions\u003c/h2\u003e \u003cp\u003eOverall, we summarize the main research findings of all the reviewed inquiries. In more detail, the XGBoost model outperformed traditional stratification scores, (STS-EACTS and RACHS-1) and offered better discrimination for in-hospital mortality in pediatric CHD surgery took place in China[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Moreover, the SVM model demonstrated potential in predicting PVL occurrence in neonates after cardiac surgery, with varying rates of PVl observed HLHS and HLHS groups [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Also, the DNN model outperformed traditional risk stratification tools in predicting one-year mortality or cardiac transplantation and prolonged length of hospital stay in neonates undergoing the Norwood procedure [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. The same goes to the research of Ruiz-Fernadez et al [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], who concluded that AI-based algorithms, particularly MLP, are feasible for classifying surgical risk in pediatric congenital heart surgery, aiding in preoperative decision-making. The machine learning model of Smith A., et al [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] demonstrates that predicts five-year transplant-free survival in infants with HLHS undergoing the Norwood procedure. Furthermore, the model demonstrated satisfactory clinical performance with a C-index of 0.692. On the other hand, patients with congenital heart disease undergoing open-heart surgery, was found that the de novo variants were associated with worse transplant-free survival and longer times on the ventilator [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAs we\u0026rsquo;ve seen the LGB machines provided the best predictive performance, highlighting the potential of advanced machine learning algorithms in clinical decision-making [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Additional, one research study concluded that the established ML models can accurately predict the risk of four major adverse postoperative outcomes in pediatric congenital heart surgery, providing reliable interpretations for high-risk contributor identification and informed clinical decision-making [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], notable, Random Forests, can improve 1-year risk assessment for pediatric heart transplant patients without loss of interpretability[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. The logistic regression model accurately predicts postoperative arrhythmias after pmVSD transcatheter closure with the selected five variables performed best (AUC\u0026thinsp;=\u0026thinsp;0.863). The nomogram derived can help risk stratify patients and guide clinical decision-making [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Logistic regression, also, achieved excellent discrimination (AUC\u0026thinsp;~\u0026thinsp;94.9%) for 30-day survival, substantially better than using STAT score. Use of peri-/post-operative data (clamp time, lactate) improved prediction over preoperative risk alone, reducing prediction error by ~\u0026thinsp;53.5%. STAT score and aortic cross-clamp time were highly significant predictors; age had a minimal effect; lactate dynamics (either persistently high or rising after 8 h) were associated with higher mortality[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eHowever, although the majority of the predictive ML models achieved high performance metrics, there is a ubiquity across the reviewed studies related to the predictive value and clinical applicability of models. For instance, despite the excellent performance of the PRAiS2 model[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], it was underpredicted, the risk for the HLHS hybrid procedure, generally performed on the sickest patients. Similar, in the study of [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], all models overpredicted operative mortality in very high risk patients and the STS model had worse calibration in this group. Overall, overestimation of expected mortalities would lead to more favourable hospital performance classifications because the expected mortality rate would be inflated, making the observed-to-expected mortality ratios lower. In terms SHAP values provided reliable interpretations for high-risk contributor identification and informed clinical decisions-making[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], they was also used as a framework to provide explanations of the prediction[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. To sum up, these novel applications are still a promising step for improving the care of children and neonates undergoing any kind of congenital heart surgery.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.6 Open issues and challenges\u003c/h2\u003e \u003cp\u003eSeveral key challenges emerged across the studies reviewed. In children with congenital heart disease especially those who had surgery, the number of cases with specific cardiac defects, surgery procedure and postoperative outcomes is limited[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. These restrictions limit the ability to train large robust machine learning models. Also, because of the heterogeneity of clinical phenotypes and anatomical variability related to the age of this patient population, CHD covers a wide range of anatomic variations, surgical repairs customized to the specific patients\u0026rsquo; age. Based on this, in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents, all reviewed studies that employed different sets of variables for training the machine learning models, ranging from 4 to 538. This heterogeneity complicates the training of machine learning models and even more the generalization of results[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eCongenital heart disease (CHD) is an excellent domain for AI to give the robust and diverse datasets extending from complex disease diagnosis and management to multimodality imaging[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. At this baseline, the small datasets and the heterogeneity of pediatric datasets perform issues in internal testing limiting the clinical utility[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e. Data acquisition in the area of congenital heart surgery depends strongly on the quality and quantity of input data[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. Moreover, children of different ages, centers and clinical domains, or patients from different ethical diversity, make machine learning models to adapt and develop solutions making them more explainable for clinicians in decision making.\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThese challenges point out to the need for future research on congenital heart surgery and inform the scope of this review ensuring that AI prevention models for pediatric CHD surgery are applicable and introduce novel challenges to lifelong trajectories in pediatric clinical care[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e].\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003e This systematic review provides insights into the application of artificial intelligence for predicting postoperative outcomes in children undergoing congenital cardiac surgery. In total, 15 studies published over the past seven years, were included, reflecting a growing interest in\u0026mdash;and increasing demand for\u0026mdash;automated and interpretable machine learning methods to support the prediction of postoperative outcomes in pediatric cardiac surgery. Present SLR, also, emphasizes in children population. Moreover, the population of all the selected studies was neonates, infants or children with a minimum age of 1 day to 18 years old, who were diagnosed with a congenital heart disease. All the included articles were retrospective cohort studies, the data collection was implemented between 2002 to 2024 from different pediatric cardiac or heart clinical domains, retrieved from local electronic medical records of each setting. 8 studies were implemented in USA and 4 studies in China. All the pediatric population were undertaken a congenital heart surgery including various procedures like open heart surgery (n\u0026thinsp;=\u0026thinsp;1), Norwood procedure (n\u0026thinsp;=\u0026thinsp;3), heart transplantation (n\u0026thinsp;=\u0026thinsp;1), ventricular septal defects repair (n\u0026thinsp;=\u0026thinsp;2), cardiopulmonary bypass (n\u0026thinsp;=\u0026thinsp;2), various procedures (n\u0026thinsp;=\u0026thinsp;6).\u003c/p\u003e \u003cp\u003eThe ML models incorporated in the selected studies were designed to forecast mortality and survival (n\u0026thinsp;=\u0026thinsp;8) [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e], [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e], [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e], postoperative complications (n\u0026thinsp;=\u0026thinsp;4)[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e], [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e], periventricular leukomalacia (PVL) (n\u0026thinsp;=\u0026thinsp;1) [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], surgical risk classification (n\u0026thinsp;=\u0026thinsp;1) [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The postoperative complications included low cardiac output syndrome (LCOS), pneumonia, renal failure, deep venous thrombosis (DVT), one year allograft loss, postoperative arrythmia, lung complications, cardiac rhythm complications, infections, and other. One study [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] examined both mortality and the prolonged length of stay in neonates undergoing the Norwood procedure for single ventricle congenital heart defects.\u003c/p\u003e \u003cp\u003eTo evaluate the discrimination of the predictive models, we used the area under the curve (AUCs). The AUCs of the models ranged between 0.642 to 0.970, with most of the models achieving an AUC above 0.8, highlights the potential of artificial intelligence as a reliable tool in congenital heart surgery. More details about each assessed outcome are discussed below. The majority of models performed the area under the curve above 0.8 for mortality assessment, apart from the study of Sunthatnkar et al. [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] which mortality prediction between stages I and II of palliation in patients undergoing single ventricle surgery was ranged 0.642. Moreover, the accuracy of the mortality model of Jalali et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] was indicated 89% for DNN, portion that linked with the results of Boskovski et al. [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] for SOM predictive model. The best performing predictive model was pointed out by Z\u0026uuml;rn et al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] with the area under the curve ranged 0.9486 for LR, results that are in agreement with Jalali et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] for DNN. In the same study, the same model achieved an AUC 0.94 for the length of hospital stay. Tong et al. [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] used LR and LightGBM for pneumonia, renal failure and deep venous thrombosis prediction, indicating AUC 0.929, 0.963 and 0.970 respectively.\u003c/p\u003e \u003cp\u003eFindings from this review leverage the significant predictors of each model to identify the examined post-operative outcomes. A core set of predictors used for each machine learning model, which varied greatly among the predictive value of mortality. In more details, preoperative variables, such as age, weight, oxygen saturation, preoperative mechanical ventilation, left atrial dimension, atrial shunt dimension, history of cardiac surgery and number of defects achieved high AUCs up to 0.80 compromises high descriptive ability[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Also, intraoperative clinical and laboratory features, like weight, procedure time, defect diameter, pre-interventional arrhythmia, difference between occlude to defect diameter \u0026gt;2mm compromise a significant role in the ML predictive model in complications (LCOS, DVT, postoperative arrythmia), ranging accuracy between 80% and 90%[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Moreover, post-operative indicators, such as serum lactate, prolonged mechanical ventilation, arrhythmias, further improves the model calibration and sensitivity[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Genomic and clinical phenotype data, particularly de novo damaging variants, were highly predictive indicators of transplant free survival with the area under the curve up to 0.82[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. The findings above confirm the growing body of literature that integrating clinical, operative, post-operative and genetic variables yields most accurate and generalizable predictive models in pediatric CHD surgery. Moreover, genomic analysis of 2517 patients with CHD and their parents revealed clinically significant de novo variants in 11.7% of patients[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. These patients were more likely to extra-cardiac disfunctions which reinforcing prior existing evidence that de novo variants are more prevalent in syndromic than in isolated CHD. [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e], [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. More findings suggest the potential role of digenic interactions in CHD pathogenesis and provide insights into molecular diagnosis by enhancing a genetic discovery rates[\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Additional biomarkers like tissue inhibitor of metalloproteinase-1 (TIMP‐1) constitute a critical role to identify the high risk of death in patients with aortic stenosis undergoing transcatheter aortic valve replacement[\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. This synthesizes the complexity of pediatric heart defects stratifying the need to develop machine learning applications based on a multimodal computational framework by combining both imaging and biologically relevant features[\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. Due to the fact that de novo genetic variants are associated with noncardiac phenotypes and negative outcomes after cardiac surgery[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], the establishment of a reliable machine-learning model which predict cardiac phenotype using genotype will facilitate effective management of surgical complications in CHS[\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e], [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOverall, the reviewed studies demonstrated strong predictive performance of artificial intelligence models for mortality prediction in pediatric patients undergoing congenital heart surgery. Previous research indicates that mortality among individuals with congenital heart disease is highest during the first year of life, with survival gradually declining beyond infancy and into childhood (5). However, more studies need to be conducted to perform quantitative measures of the performance of AI driven models in survival appraisal of CHD patients. Nonetheless, Du et al. [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] concluded that the XGBoost model outperformed the traditional stratification scores (STS-EACTS \u0026amp; RACHS-1) and offered better discrimination for in-hospital mortality in pediatric CHD surgery in China. Findings are confirmed by Jalali et al. [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], who found that the DNN model, also, outperformed the traditional risk stratification tools in predicting one-year mortality or cardiac transplantation and prolonged length of hospital stay in neonates undergoing the Norwood procedure. These results agree with Zaka et al., [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e] who concluded that ML models outperformed traditional risk scores in the discrimination of all-cause mortality following transcatheter aortic valve implementation (TAVI).\u003c/p\u003e \u003cp\u003eFurthermore, machine learning models are associated with greater post-operative outcomes and survival in neonates and infants undergoing a heart surgical procedure like open-heart surgery or Norwood procedure, such as longer transplant free survival and satisfactory clinical performance with C-index of 0.692[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. More key findings from present study reveals that machine learning models, particularly Random Forests, can improve 1-year risk assessment for paediatric heart transplant patients without loss of interpretability[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], the prediction of post-operative arrhythmias after pmVSD transcatheter closure, are able to identify modifiable and non-modifiable risk factors for interstage mortality following Stage I palliation Further machine learning algorithms[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn addition, AI based algorithms [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], particularly MLP are feasible to classify the surgical risk contributing in preoperative decision making. Also, the established ML models in the study of Tong et al. [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e] can accurately predict the risk of four major adverse postoperative outcomes in pediatric congenital heart surgery, providing reliable interpretations for high-risk contributor identification and informed clinical decision-making. Further, LGB machines [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e] provided the best predictive performance, confirming, additionally, the potential of advanced machine learning algorithms in clinical decision-making. Furthermore, Betsimas et al. [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e] have shown that the machine learning methodology of optimal classification trees (OCTs) can accurately predict risk after congenital heart surgery. They, also, concluded that OCT benchmarking analysis can assess hospital-specific case-adjusted performance after CHS, both overall and patient cohort-specific, serving as a tool for hospital self-assessment and quality improvement. These findings underscores the importance of considering the integration of ML algorithms into electronic healthcare systems, though, they may improve periprocedural risk stratification, but immediate implementation in the clinical setting, still, remains uncertain[\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. From a clinical perspective, the distinct pattern of congenital heart dysfunctions raises several questions regarding the optimal management of patients and how these patients may impact from AI and machine learning in future risk of adverse events or response to surgical interventions[\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e], [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. This review illustrates the capabilities of machine learning in prediction of several surgical outcomes. Promising discriminative abilities of ML models have been discovered. In surgical ML research, there is a need for standardized calibration assessment and reporting to facilitate the clinical adoption of ML models[\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. However, most studies included a retrospective study design without external validation or calibration. Large-scale data is warranted to bridge the gap between calibration and external validation[\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e], [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. Clinical implementation is needed to demonstrate the contribution of ML within daily practice.\u003c/p\u003e \u003cp\u003eWe observed that there is a clear trend for older papers focus on performance (AUC), while papers from 2023\u0026ndash;2025 prioritize calibration and XAI to make the models clinically useful. This approach reveals areas for potential improvement and transform ML as a powerful tool for quality improvement clinical initiatives[\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo conclude, our systematic review has several limitations. Primarily, the diversity of congenital heart diseases, the nature of heart defects makes the models heterogeneous. Also, this is reinforced by certain characteristics of study population like age, weight, gestational age, although the nature of pediatric population meets some clinical particularities by its own. Unless, there was an effort to develop AI models specialized in a specific heart defect, very large sample size of pediatric population is needed to develop more reliable ML models. Another important issue that was reported by many of the included studies, was the high risk of bias resulted by the low quality or unclear data analysis. Further research is required to overcome methodological and validation limitations.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eIn summary, with a novel interpretable machine learning algorithm, we can predict whether a pediatric patient perform complications after a heart congenital heart surgery or not. Also, what kind of complications will occur and explain the specific patient characteristics that led to this prediction. The majority of the developed AI prediction models achieved high accuracy and sensitivity. To the best of our knowledge, this systematic review is one of its kind and we believe that the combination of high model performance and interpretability could provide useful information for physicians being part of clinical decision making[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Continuing research in clinical settings provides evidence of the feasibility and advantages of offering genome sequencing to patients with cardiac phenotypes indicating a genetic cause. This enables the necessity to create a model of care that harnessed clinical and functional genomics to inform a future ML based clinical practice[\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e]. Physicians that care critically ill children with CHD might be influenced by genomic results by making recommendations about whether to forego or withdraw certain treatment option[\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. AI driven machine learning models motivate a personalized plan of care, giving strategies and recommendations. So, the abetment of a cardiac genetic testing and the approval for a further publicly genetic repository will improve the collaboration between stakeholders and clinical caregivers to improve the quality of care and outcomes for CHD pediatric patients and their families[\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e], [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e].\u003c/p\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cstrong\u003eEthical approval\u003c/strong\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent to participate\u003c/strong\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eClinical Trial number\u003c/h2\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eConflict of interest\u003c/h2\u003e \u003cp\u003eAll authors have reported that they have no relationships relevant to the contents of this paper to disclose.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThe authors declare that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e \u003cp\u003eAll authors contributed to the study conception and design. Paschalina Lialiou and Ilias Maglogiannis contributed to the data collection. Material preparation was performed by Paschalina Lialiou and Ilias Maglogiannis. The first draft of the manuscript was written by Paschalina Lialiou and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e\u003ch2\u003eCode availability\u003c/h2\u003e \u003cp\u003eNot Applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003evan der Linde D et al (Nov. 2011) Birth Prevalence of Congenital Heart Disease Worldwide. JACC 58(21):2241\u0026ndash;2247. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacc.2011.08.025\u003c/span\u003e\u003cspan address=\"10.1016/j.jacc.2011.08.025\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Y et al (2019) Global birth prevalence of congenital heart defects 1970\u0026ndash;2017: updated systematic review and meta-analysis of 260 studies, \u003cem\u003eInt. J. Epidemiol.\u003c/em\u003e, vol. 48, no. 2, pp. 455\u0026ndash;463, Apr. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/ije/dyz009\u003c/span\u003e\u003cspan address=\"10.1093/ije/dyz009\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoffman JIE (May 2013) The global burden of congenital heart disease: review article. Cardiovasc J Afr 24(4):141\u0026ndash;145. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.10520/EJC137177\u003c/span\u003e\u003cspan address=\"10.10520/EJC137177\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMandalenakis Z et al (Nov. 2020) Survival in Children With Congenital Heart Disease: Have We Reached a Peak at 97%? J Am Heart Assoc 9:e017704. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/JAHA.120.017704\u003c/span\u003e\u003cspan address=\"10.1161/JAHA.120.017704\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBest KE, Rankin J (Jun. 2016) Long-Term Survival of Individuals Born With Congenital Heart Disease: A Systematic Review and Meta‐Analysis. J Am Heart Assoc 5(6):e002846. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/JAHA.115.002846\u003c/span\u003e\u003cspan address=\"10.1161/JAHA.115.002846\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTong C et al (Apr. 2024) Machine learning prediction model of major adverse outcomes after pediatric congenital heart surgery: a retrospective cohort study. Int J Surg 110(4):2207. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/JS9.0000000000001112\u003c/span\u003e\u003cspan address=\"10.1097/JS9.0000000000001112\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePasquali SK, He X, Jacobs JP, Jacobs ML, O\u0026rsquo;Brien SM, Gaynor JW (2012) Evaluation of failure to rescue as a quality metric in pediatric heart surgery: an analysis of the STS Congenital Heart Surgery Database. Ann Thorac Surg 94(2):573\u0026ndash;579 discussion 579\u0026ndash;580, Aug. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.athoracsur.2012.03.065\u003c/span\u003e\u003cspan address=\"10.1016/j.athoracsur.2012.03.065\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePasquali SK et al (2012) Association of center volume with mortality and complications in pediatric heart surgery, \u003cem\u003ePediatrics\u003c/em\u003e, vol. 129, no. 2, pp. e370-376, Feb. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1542/peds.2011-1188\u003c/span\u003e\u003cspan address=\"10.1542/peds.2011-1188\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerry C, Fiery-Fraillon J, Togni M, Cook S (2025) Futility in TAVI: A scoping review of definitions, predictive criteria, and medical predictive models. PLoS ONE 20(1):e0313399. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0313399\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0313399\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeredith T et al (2025) Machine learning cluster analysis identifies increased 12-month mortality risk in transcatheter aortic valve replacement recipients. Front Cardiovasc Med 12:1444658. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2025.1444658\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2025.1444658\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarelli AJ et al (Feb. 2024) Machine Learning Informed Diagnosis for Congenital Heart Disease in Large Claims Data Source. JACC Adv 3(2):100801. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacadv.2023.100801\u003c/span\u003e\u003cspan address=\"10.1016/j.jacadv.2023.100801\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRazavian N, Marcus J, Sontag D Multi-task Prediction of Disease Onsets from Longitudinal Lab Tests, Sep. 20, 2016, \u003cem\u003earXiv\u003c/em\u003e: arXiv:1608.00647. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/arXiv.1608.00647\u003c/span\u003e\u003cspan address=\"10.48550/arXiv.1608.00647\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayourian J, Geggel R, La Cava WG, Ghelani SJ, Triedman JK (2025) Pediatric Electrocardiogram-Based Deep Learning to Predict Secundum Atrial Septal Defects, \u003cem\u003ePediatr. Cardiol.\u003c/em\u003e, vol. 46, no. 5, pp. 1235\u0026ndash;1240, Jun. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00246-024-03540-7\u003c/span\u003e\u003cspan address=\"10.1007/s00246-024-03540-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePenny-Dimri JC, Bergmeir C, Perry L, Hayes L, Bellomo R, Smith JA (2022) Machine learning to predict adverse outcomes after cardiac surgery: A systematic review and meta-analysis, \u003cem\u003eJ. Card. Surg.\u003c/em\u003e, vol. 37, no. 11, pp. 3838\u0026ndash;3845, Nov. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/jocs.16842\u003c/span\u003e\u003cspan address=\"10.1111/jocs.16842\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeivaditis V et al (Jan. 2025) Artificial Intelligence in Cardiac Surgery: Transforming Outcomes and Shaping the Future. Clin Pract 15(1):17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/clinpract15010017\u003c/span\u003e\u003cspan address=\"10.3390/clinpract15010017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMestres CA, Quintana E, Pereda D (2022) Will artificial intelligence help us in predicting outcomes in cardiac surgery? \u003cem\u003eJ. Card. Surg.\u003c/em\u003e, vol. 37, no. 11, pp. 3846\u0026ndash;3847, Nov. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/jocs.16844\u003c/span\u003e\u003cspan address=\"10.1111/jocs.16844\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePage MJ et al (2021) The PRISMA 2020 statement: an updated guideline for reporting systematic reviews, \u003cem\u003eThe BMJ\u003c/em\u003e, vol. 372, p. n71, Mar. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmj.n71\u003c/span\u003e\u003cspan address=\"10.1136/bmj.n71\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRethlefsen ML et al (2021) PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews, \u003cem\u003eSyst. Rev.\u003c/em\u003e, vol. 10, no. 1, Art. no. 1, Jan. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13643-020-01542-z\u003c/span\u003e\u003cspan address=\"10.1186/s13643-020-01542-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmir-Behghadami M, Janati A (2021) Reporting Systematic Review in Accordance With the PRISMA Statement Guidelines: An Emphasis on Methodological Quality, \u003cem\u003eDisaster Med. Public Health Prep.\u003c/em\u003e, vol. 15, no. 5, Art. no. 5, Oct. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1017/dmp.2020.90\u003c/span\u003e\u003cspan address=\"10.1017/dmp.2020.90\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu X et al (Nov. 2022) Machine Learning Model for Predicting Risk of In-Hospital Mortality after Surgery in Congenital Heart Disease Patients. Rev Cardiovasc Med 23 11, Art. 11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.31083/j.rcm2311376\u003c/span\u003e\u003cspan address=\"10.31083/j.rcm2311376\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoskovski MT et al (Aug. 2020) De Novo Damaging Variants, Clinical Phenotypes, and Post-Operative Outcomes in Congenital Heart Disease. Circ Genomic Precis Med 13(4):e002836. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCGEN.119.002836\u003c/span\u003e\u003cspan address=\"10.1161/CIRCGEN.119.002836\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJalali A, Simpao AF, G\u0026aacute;lvez JA, Licht DJ, Nataraj C (Aug. 2018) Prediction of Periventricular Leukomalacia in Neonates after Cardiac Surgery Using Machine Learning Algorithms. J Med Syst 42(10):177. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s10916-018-1029-z\u003c/span\u003e\u003cspan address=\"10.1007/s10916-018-1029-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJalali A et al (Jun. 2020) Deep Learning for Improved Risk Prediction in Surgical Outcomes. Sci Rep 10:9289. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-020-62971-3\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-62971-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuiz-Fern\u0026aacute;ndez D, Monsalve Torra A, Soriano-Pay\u0026aacute; A, Mar\u0026iacute;n-Alonso O, Triana Palencia E (Apr. 2016) Aid decision algorithms to estimate the risk in congenital heart surgery. Comput Methods Programs Biomed 126:118\u0026ndash;127. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.cmpb.2015.12.021\u003c/span\u003e\u003cspan address=\"10.1016/j.cmpb.2015.12.021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmith AH, Gray GM, Ashfaq A, Asante-Korang A, Rehman MA, Ahumada LM (Feb. 2024) Using machine learning to predict five-year transplant-free survival among infants with hypoplastic left heart syndrome. Sci Rep 14(1):4512. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-024-55285-1\u003c/span\u003e\u003cspan address=\"10.1038/s41598-024-55285-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSunthankar SD et al (Aug. 2023) Machine Learning to Predict Interstage Mortality Following Single Ventricle Palliation: A NPC-QIC Database Analysis. Pediatr Cardiol 44(6):1242\u0026ndash;1250. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00246-023-03130-z\u003c/span\u003e\u003cspan address=\"10.1007/s00246-023-03130-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTong C et al (2024) Machine learning prediction model of major adverse outcomes after pediatric congenital heart surgery: a retrospective cohort study, \u003cem\u003eInt. J. Surg. Lond. Engl.\u003c/em\u003e, vol. 110, no. 4, pp. 2207\u0026ndash;2216, Jan. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/JS9.0000000000001112\u003c/span\u003e\u003cspan address=\"10.1097/JS9.0000000000001112\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWisotzkey BL et al (Dec. 2023) Risk factors for 1-year allograft loss in pediatric heart transplant patients using machine learning: An analysis of the pediatric heart transplant society database. Pediatr Transpl 27(8):e14612. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/petr.14612\u003c/span\u003e\u003cspan address=\"10.1111/petr.14612\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan L, Meng Y, Sun H, Liu X, Han B (2025) Application of machine learning in predicting postoperative arrhythmia following transcatheter closure of perimembranous ventricular septal defects, \u003cem\u003ePol. Heart J. Kardiologia Pol.\u003c/em\u003e, vol. 83, no. 3, Art. no. 3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.33963/v.phj.103535\u003c/span\u003e\u003cspan address=\"10.33963/v.phj.103535\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng X et al (Aug. 2021) Explainable machine-learning predictions for complications after pediatric congenital heart surgery. Sci Rep 11(1):17244. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-021-96721-w\u003c/span\u003e\u003cspan address=\"10.1038/s41598-021-96721-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZ\u0026uuml;rn C et al (Sep. 2023) Model-driven survival prediction after congenital heart surgery. Interdiscip Cardiovasc Thorac Surg 37(3). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/icvts/ivad089\u003c/span\u003e\u003cspan address=\"10.1093/icvts/ivad089\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRogers L et al (Jul. 2017) Improving Risk Adjustment for Mortality After Pediatric Cardiac Surgery: The UK PRAiS2 Model. Ann Thorac Surg 104(1):211\u0026ndash;219. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.athoracsur.2016.12.014\u003c/span\u003e\u003cspan address=\"10.1016/j.athoracsur.2016.12.014\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNormand S-LT et al (Sep. 2022) Mortality Prediction After Cardiac Surgery in Children: An STS Congenital Heart Surgery Database Analysis. Ann Thorac Surg 114(3):785\u0026ndash;798. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.athoracsur.2021.11.077\u003c/span\u003e\u003cspan address=\"10.1016/j.athoracsur.2021.11.077\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaerber JA et al (Jul. 2021) Identifying Risk Factors for Complicated Post-operative Course in Tetralogy of Fallot Using a Machine Learning Approach. Front Cardiovasc Med 8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2021.685855\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2021.685855\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu Y, Antoniadi AM, McNestry C, McAuliffe FM, Mooney C (2022) The Role of XAI in Advice-Taking from a Clinical Decision Support System: A Comparative User Study of Feature Contribution-Based and Example-Based Explanations, \u003cem\u003eAppl. Sci.\u003c/em\u003e, vol. 12, no. 20, Art. no. 20, Jan. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/app122010323\u003c/span\u003e\u003cspan address=\"10.3390/app122010323\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoons KGM et al (Jan. 2019) PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration. Ann Intern Med 170(1):W1\u0026ndash;W33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7326/M18-1377\u003c/span\u003e\u003cspan address=\"10.7326/M18-1377\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWolff RF et al (Jan. 2019) PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med 170(1):51\u0026ndash;58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7326/M18-1376\u003c/span\u003e\u003cspan address=\"10.7326/M18-1376\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJone P-N et al (Dec. 2022) Artificial Intelligence in Congenital Heart Disease. JACC Adv 1(5):100153. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jacadv.2022.100153\u003c/span\u003e\u003cspan address=\"10.1016/j.jacadv.2022.100153\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAuctores Advancements in AI Heart Model Technology: A Comprehensive Review, Auctores. Accessed: Nov. 11, 2025. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://auctoresonline.org/article/advancements-in-ai-heart-model-technology-a-comprehensive-review\u003c/span\u003e\u003cspan address=\"https://auctoresonline.org/article/advancements-in-ai-heart-model-technology-a-comprehensive-review\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurath-Koller S (Jun. 2025) Artificial intelligence in pediatrics: promise, peril, and the path ahead. Front Pediatr 13. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fped.2025.1631521\u003c/span\u003e\u003cspan address=\"10.3389/fped.2025.1631521\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSalavati A et al (Dec. 2024) Artificial Intelligence Advancements in Cardiomyopathies: Implications for Diagnosis and Management of Arrhythmogenic Cardiomyopathy. Curr Heart Fail Rep 22(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11897-024-00688-4\u003c/span\u003e\u003cspan address=\"10.1007/s11897-024-00688-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHomsy J et al (2015) De novo mutations in congenital heart disease with neurodevelopmental and other congenital anomalies, \u003cem\u003eScience\u003c/em\u003e, vol. 350, no. 6265, pp. 1262\u0026ndash;1266, Dec. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1126/science.aac9396\u003c/span\u003e\u003cspan address=\"10.1126/science.aac9396\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSifrim A et al (Sep. 2016) Distinct genetic architectures for syndromic and nonsyndromic congenital heart defects identified by exome sequencing. Nat Genet 48(9):1060\u0026ndash;1065. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/ng.3627\u003c/span\u003e\u003cspan address=\"10.1038/ng.3627\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKars ME et al (Mar. 2025) Deciphering the digenic architecture of congenital heart disease using trio exome sequencing data. Am J Hum Genet 112(3):583\u0026ndash;598. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ajhg.2025.01.024\u003c/span\u003e\u003cspan address=\"10.1016/j.ajhg.2025.01.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoeckling F et al (Mar. 2025) Extracellular Matrix Proteins Improve Risk Prediction in Patients Undergoing Transcatheter Aortic Valve Replacement. J Am Heart Assoc 14(5):e037296. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/JAHA.124.037296\u003c/span\u003e\u003cspan address=\"10.1161/JAHA.124.037296\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCeschin R et al (Sep. 2018) A computational framework for the detection of subcortical brain dysmaturation in neonatal MRI using 3D Convolutional Neural Networks. NeuroImage 178:183\u0026ndash;197. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.neuroimage.2018.05.049\u003c/span\u003e\u003cspan address=\"10.1016/j.neuroimage.2018.05.049\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong L et al (2025) Establishment of a Stacking Machine Learning Model Predicting Cardiac Phenotype in Ectopia Lentis Patients Based on Genotype and Ocular Phenotype. Int J Med Sci 22(14):3501\u0026ndash;3510. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7150/ijms.109657\u003c/span\u003e\u003cspan address=\"10.7150/ijms.109657\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCornhill AK et al (2022) Machine Learning Patient-Specific Prediction of Heart Failure Hospitalization Using Cardiac MRI-Based Phenotype and Electronic Health Information. Front Cardiovasc Med 9:890904. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fcvm.2022.890904\u003c/span\u003e\u003cspan address=\"10.3389/fcvm.2022.890904\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZaka A et al (Jan. 2025) Machine-learning versus traditional methods for prediction of all-cause mortality after transcatheter aortic valve implantation: a systematic review and meta-analysis. Open Heart 12(1):e002779. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/openhrt-2024-002779\u003c/span\u003e\u003cspan address=\"10.1136/openhrt-2024-002779\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBertsimas D et al (2022) Benchmarking in Congenital Heart Surgery Using Machine Learning-Derived Optimal Classification Trees, \u003cem\u003eWorld J. Pediatr. Congenit. Heart Surg.\u003c/em\u003e, vol. 13, no. 1, pp. 23\u0026ndash;35, Jan. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/21501351211051227\u003c/span\u003e\u003cspan address=\"10.1177/21501351211051227\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSimsic JM, Bradley SM, Stroud MR, Atz AM (2005) Risk Factors for Interstage Death After the Norwood Procedure, \u003cem\u003ePediatr. Cardiol.\u003c/em\u003e, vol. 26, no. 4, pp. 400\u0026ndash;403, Aug. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00246-004-0776-4\u003c/span\u003e\u003cspan address=\"10.1007/s00246-004-0776-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrabb BT et al (Summer 2025) Characteristics of left ventricular dysfunction in repaired tetralogy of Fallot: A multi-institutional deep learning analysis of regional strain and dyssynchrony. J Cardiovasc Magn Reson Off J Soc Cardiovasc Magn Reson 27(1):101886. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jocmr.2025.101886\u003c/span\u003e\u003cspan address=\"10.1016/j.jocmr.2025.101886\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBektaş M, Tuynman JB, Costa Pereira J, Burchell GL, van der Peet DL (2022) Machine Learning Algorithms for Predicting Surgical Outcomes after Colorectal Surgery: A Systematic Review. World J Surg 46(12). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00268-022-06728-1\u003c/span\u003e\u003cspan address=\"10.1007/s00268-022-06728-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayourian J, Geggel R, La Cava WG, Ghelani SJ, Triedman JK (2025) Pediatric Electrocardiogram-Based Deep Learning to Predict Secundum Atrial Septal Defects, \u003cem\u003ePediatr. Cardiol.\u003c/em\u003e, vol. 46, no. 5, pp. 1235\u0026ndash;1240, Jun. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00246-024-03540-7\u003c/span\u003e\u003cspan address=\"10.1007/s00246-024-03540-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarris GE et al (Jul. 2024) Congenital Heart Surgery Machine Learning-Derived In-Depth Benchmarking Tool. Ann Thorac Surg 118(1):199\u0026ndash;206. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.athoracsur.2023.10.034\u003c/span\u003e\u003cspan address=\"10.1016/j.athoracsur.2023.10.034\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAustin R et al (Jan. 2024) A multitiered analysis platform for genome sequencing: Design and initial findings of the Australian Genomics Cardiovascular Disorders Flagship. Genet Med Open 2:101842. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.gimo.2024.101842\u003c/span\u003e\u003cspan address=\"10.1016/j.gimo.2024.101842\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChar DS, Deuitch NT, Berent MK, Chung WK, Krosnick JA, Magnus D (2025) Impact of Genomic Sequencing Information on Physicians\u0026rsquo; Treatment Recommendations for Children with Congenital Heart Disease, \u003cem\u003eGenet. Med. Open\u003c/em\u003e, p. 103470, Nov. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.gimo.2025.103470\u003c/span\u003e\u003cspan address=\"10.1016/j.gimo.2025.103470\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen T et al (Dec. 2024) Genomic insights for personalised care in lung cancer and smoking cessation: motivating at-risk individuals toward evidence-based health practices. eBioMedicine 110:105441. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ebiom.2024.105441\u003c/span\u003e\u003cspan address=\"10.1016/j.ebiom.2024.105441\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSingh M et al (Jul. 2024) Artificial intelligence for cardiovascular disease risk assessment in personalised framework: a scoping review. eClinicalMedicine 73:102660. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.eclinm.2024.102660\u003c/span\u003e\u003cspan address=\"10.1016/j.eclinm.2024.102660\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University of Piraeus","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":"Congenital Heart Surgery, Cardiac Surgery, Children, Artificial Intelligence, Machine Learning","lastPublishedDoi":"10.21203/rs.3.rs-9055427/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9055427/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eCongenital heart disease (CHD) constitutes the most common major congenital malformation. Despite the improvement of diagnostic technologies and the advances in pediatric cardiovascular surgery, the utilization of artificial intelligence (AI) holds a critical role in predicting post-operative outcomes in pediatric population. The primary aim of the study was to define the utilization of AI machine learning models in the prediction of post-operative outcome in children undergoing a congenital heart surgery.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003e Following the PRISMA guidelines, a systematic literature review was conducted by a comprehensive retrieval of two large databases. The inclusion and exclusion criteria were predefined. Two independent reviewers screened the articles.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003e 12 articles included in the review published the last seven years. The included research papers were retrospective cohort studies with a range of size population from 71 to 24.685 pediatric patients. The majority of them examined the prediction performance of AI machine learning algorithms in mortality and other post-operative complications. Various types of congenital heart surgeries were described. The area under the curve (AUC) was used for model performance, ranged from 0.642 to 0.970. LightGBM outperformed with AUC 0.970 for the prediction of deep venous thrombosis (DVT).\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eAI machine learning models show greater discriminative power for the prediction of post-operative outcomes in pediatric patients undergoing a CHS surpass the traditional prediction risk tools. Further studies must be conducted to strengthen the explainable role of AI applications in clinical decision making.\u003c/p\u003e","manuscriptTitle":"Overview of AI and Machine Learning Methods for Outcome Prediction in Pediatric Congenital Heart Surgery","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-10 16:36:30","doi":"10.21203/rs.3.rs-9055427/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":"60f6b22a-2e4a-4c1a-9701-38fdcb0e8e8a","owner":[],"postedDate":"March 10th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-10T16:36:31+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-10 16:36:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9055427","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9055427","identity":"rs-9055427","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}