Next Article in Journal
The Association Between Childhood Trauma, Emotional Dysregulation, and Depressive Symptoms’ Severity in Patients with Obesity Seeking Bariatric Surgery
Previous Article in Journal
Effect of Prior Laser-Assisted In Situ Keratomileusis on the Calibration Accuracy of Extended Depth of Focus Intraocular Lenses: A Direct Comparative Study
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
JPM
Volume 15
Issue 7
10.3390/jpm15070302
jpm-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Systematic Review

Artificial Intelligence in Risk Stratification and Outcome Prediction for Transcatheter Aortic Valve Replacement: A Systematic Review and Meta-Analysis

by
Shayan Shojaei
1,2,3,
Asma Mousavi
1,2,3,
Sina Kazemian
1,
Shiva Armani
4,
Saba Maleki
1,
Parisa Fallahtafti
1,
Farzin Tahmasbi Arashlow
5,
Yasaman Daryabari
6,
Mohammadreza Naderian
7,
Mohamad Alkhouli
7,8,
Jamal S. Rana
9,
Mehdi Mehrani
1,
Yaser Jenab
1 and
Kaveh Hosseini
1,*
1
Tehran Heart Center, Cardiovascular Diseases Research Institute, Tehran University of Medical Sciences, Tehran 1411713138, Iran
2
School of Medicine, Tehran University of Medical Sciences, Tehran 1936893813, Iran
3
Students’ Scientific Research Center, Tehran University of Medical Sciences, Tehran 1417755331, Iran
4
Heart Failure Research Center, Cardiovascular Research Institute, Isfahan University of Medical Sciences, Isfahan 7346181746, Iran
5
Medical Students Research Centre, Iran University of Medical Sciences, Tehran 1449614535, Iran
6
Pediatric Urology and Regenerative Medicine Research Center, Children’s Medical Center, Gene, Cell & Tissue Research Institute, Tehran University of Medical Sciences, Tehran 1419733151, Iran
7
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN 55905, USA
8
Division of Cardiovascular Diseases, Department of Medicine, West Virginia University, Morgantown, WV 26506, USA
9
Department of Cardiology, Oakland Medical Center, Kaiser Permanente Northern California, Oakland, CA 94611, USA
*
Author to whom correspondence should be addressed.
J. Pers. Med. 2025, 15(7), 302; https://doi.org/10.3390/jpm15070302
Submission received: 10 May 2025 / Revised: 29 June 2025 / Accepted: 2 July 2025 / Published: 11 July 2025
(This article belongs to the Special Issue Artificial Intelligence in Diagnosis, Treatment and Prognosis of Cardiovascular Diseases)
Browse Figures
Review Reports Versions Notes

Abstract

Background/Objectives: Transcatheter aortic valve replacement (TAVR) has been introduced as an optimal treatment for patients with severe aortic stenosis, offering a minimally invasive alternative to surgical aortic valve replacement. Predicting these outcomes following TAVR is crucial. Artificial intelligence (AI) has emerged as a promising tool for improving post-TAVR outcome prediction. In this systematic review and meta-analysis, we aim to summarize the current evidence on utilizing AI in predicting post-TAVR outcomes. Methods: A comprehensive search was conducted to evaluate the studies focused on TAVR that applied AI methods for risk stratification. We assessed various ML algorithms, including random forests, neural networks, extreme gradient boosting, and support vector machines. Model performance metrics—recall, area under the curve (AUC), and accuracy—were collected with 95% confidence intervals (CIs). A random-effects meta-analysis was conducted to pool effect estimates. Results: We included 43 studies evaluating 366,269 patients (mean age 80 ± 8.25; 52.9% men) following TAVR. Meta-analyses for AI model performances demonstrated the following results: all-cause mortality (AUC = 0.78 (0.74–0.82), accuracy = 0.81 (0.69–0.89), and recall = 0.90 (0.70–0.97); permanent pacemaker implantation or new left bundle branch block (AUC = 0.75 (0.68–0.82), accuracy = 0.73 (0.59–0.84), and recall = 0.87 (0.50–0.98)); valve-related dysfunction (AUC = 0.73 (0.62–0.84), accuracy = 0.79 (0.57–0.91), and recall = 0.54 (0.26–0.80)); and major adverse cardiovascular events (AUC = 0.79 (0.67–0.92)). Subgroup analyses based on the model development approaches indicated that models incorporating baseline clinical data, imaging, and biomarker information enhanced predictive performance. Conclusions: AI-based risk prediction for TAVR complications has demonstrated promising performance. However, it is necessary to evaluate the efficiency of the aforementioned models in external validation datasets.

Graphical Abstract

1. Introduction

Transcatheter aortic valve replacement (TAVR) has become a cornerstone therapy for patients with severe aortic stenosis, especially those at elevated surgical risk [1]. By employing small surgical incisions cuts and delivering the valve via catheter, this procedure is significantly less invasive and carries a lower risk for patients compared with traditional approaches, such as surgical aortic valve replacement (SAVR) [2]. Despite its widespread adoption, TAVR is associated with various complications and adverse events, including all-cause mortality, stroke, heart failure-related rehospitalization, and conduction disturbances [3]. As procedural techniques and device technologies evolve, TAVR is increasingly offered to broader patient populations, including those at low risk [4,5]. However, optimizing outcomes remains challenging due to the multifactorial nature of post-TAVR complications, which can significantly affect patients’ survival and quality of life [6].
The recent literature demonstrated that the occurrence of specific complications following TAVR significantly impacts long-term mortality and quality of life. These adverse events often exert a more profound influence on patient outcomes than the cumulative burden of pre-existing comorbidities as captured by traditional surgical risk scores [7,8]. Consequently, achieving an “event-free” TAVR procedure should be considered a primary objective to enhance and optimize clinical outcomes. Traditional surgical risk scores, such as the EuroSCORE and Society of Thoracic Surgeons (STS) score, have demonstrated limited discriminative ability in predicting mortality after TAVR [9].
In contrast, artificial intelligence (AI) and machine learning (ML) algorithms have demonstrated superior predictive capabilities in the TAVR setting [10]. These models can process complex, high-dimensional data and uncover intricate, non-linear relationships among variables. For example, AI models have achieved a pooled mean AUC of 0.79 in predicting post-TAVR mortality, significantly outperforming traditional scores [11]. The enhanced performance of AI models can be attributed to their ability to integrate diverse data types, including clinical, imaging, biomarker, and procedural variables [12]. By leveraging these multidimensional predictors, AI models can provide more individualized risk assessments, facilitating better patient selection and personalized treatment strategies [13]. However, AI usage should be guided under specific frameworks to ensure the effective and responsible integration of these innovative methods into healthcare [14].
Given the growing interest in AI-driven prediction models in the context of TAVR, a systematic review and meta-analysis is warranted to synthesize the current evidence and evaluate the performance of AI-based models across various clinical endpoints. The objective of this study is to systematically review and quantitatively assess the predictive performance of AI and ML algorithms used to forecast outcomes following TAVR.

2. Methods

The current study is reported in accordance with the preferred reporting items for systematic reviews and meta-analyses protocols (PRISMA) (Table S1) [15]. The review protocol was registered on the Prospective Register of Systematic Reviews (PROSPERO; CRD42025636245). Due to analyzing data from previously published studies, we did not require formal ethical approval.

2.1. Search Strategy

A comprehensive literature search was conducted through the PubMed and Embase databases from inception to 24 September 2024 to identify relevant studies. The search strategy included the following keywords: [“TAVI” OR “TAVR”] AND [“AI”, “ML”]. Synonyms and equivalent terms for these keywords were also included to ensure a broad and inclusive search. Detailed search strategy for each database is provided in Table S2. Reference lists of review articles and included studies were manually screened for additional relevant citations.

2.2. Eligibility Criteria and Screening

Studies were eligible for inclusion if they focused on TAVR and the use of ML algorithms for predicting post-TAVR outcomes. Exclusion criteria included abstracts, case reports, review articles, editorials, and animal studies. After removing duplications, two independent reviewers (SS and AM) screened titles and abstracts for relevance. Full texts of potentially eligible studies were assessed by the same reviewers. Any discrepancies between reviewers were resolved through mutual consensus and, if necessary, consultation with a third expert reviewer (KH).

2.3. Data Extraction

Key data were extracted by two independent authors (SS and AM), focusing on study characteristics (e.g., first author, publication year, country, design, inclusion and exclusion criteria, and sample size), population demographics (e.g., mean age and percentage of males), AI methodologies (e.g., algorithms used, feature selection processes, evaluation and validation method, and model development techniques), imaging data (e.g., cardiac magnetic resonance imaging (CMRI), echocardiography, computed tomography (CT), and electrocardiography (ECG)), study outcomes, as well as recall, specificity, negative predictor value, and positive predictive value. Performance metrics, such as AUC, accuracy, and recall, were systematically collected. Additional data on clinical endpoints and external validation were also recorded. To avoid including the same study population multiple times in our analysis, we selected the best-performing model (highest AUC) from each study among the various AUCs reported and machine learning algorithms employed. Furthermore, for studies that provided both internal and external validation data, priority was given to external validation. Discrepancies in data extraction were resolved through mutual consensus or consultation with a third reviewer (KH) when necessary.

2.4. Quality Assessment

Three reviewers (SM, FTA, and YD) evaluated the quality and risk of bias of the included studies using the prediction model risk of bias assessment tool (PROBAST), which assesses studies in four domains: participants, predictors, outcome, and analysis. Studies were rated as having a low, high, or unclear risk of bias based on these criteria [16]. Adherence to the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines was also evaluated to ensure transparent reporting of model development, validation, and performance [17].

2.5. Outcomes

Our primary outcome was all-cause mortality, while secondary outcomes included study-defined major adverse cardiovascular events (MACE), new permanent pacemaker implantation or new left bundle branch block (LBBB), valve-related dysfunction, stroke, and heart failure-related rehospitalization.

2.6. Statistical Analysis

A random-effects meta-analysis was conducted to pool effect model performance metrics such as AUC, accuracy, and recall across studies, accounting for potential heterogeneity. Heterogeneity was quantified using the I2 statistic, with values above 75% indicating substantial heterogeneity. Sensitivity analysis using a leave-one-out method was performed to assess the robustness of pooled estimates by systematically excluding individual studies and evaluating their influence on the overall effect size.
Subgroup analyses were conducted to explore potential sources of heterogeneity and differences in model performance. Studies were stratified based on the type of features (baseline and clinical data, imaging data, biomarkers, and procedural data) utilized in their predictive models. We also evaluated our results based on the best performance reported for each ML model to better identify sources of heterogeneity. Publication bias was assessed using visual inspection of funnel plots for asymmetry and further evaluated statistically using the Egger’s test and Begg’s test, where applicable. All statistical analyses were performed using R software version 4.1.2 (The R Foundation, Vienna, Austria), utilizing the meta, metafor, and ggplot2 packages for meta-analysis. The summary effect sizes were reported with 95% confidence intervals (CI). We generated forest plots to visualize individual study estimates and the pooled effect.

3. Results

3.1. Study Selection and Characteristics

Our database search initially yielded 7287 studies and, after removing duplicates, 5131 studies were considered for title and abstract screening. Eventually, 158 full texts were evaluated by two independent reviewers, and 43 articles (366,269 patients) were included in the final analysis. The corresponding PRISMA flow chart and study selection process is outlined in Figure 1.
The included studies were published between 2017 and 2024 and provided original data on AI model-based prediction outcomes following TAVR, with 40 studies using internal validation models [10,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56] and only 3 studies utilizing external validation models [57,58,59]. These studies included diverse cohorts with sample sizes ranging from 129 [48] to 117,389 [25] patients, with a mean age of 80 ± 8.25 years. The proportion of male participants varied between 36.3% [20] and 62% [44]. Most studies were retrospective studies, with four prospective cohorts [18,31,44,48] and only one post hoc analysis [39].
A wide range of machine learning algorithms were used in the included studies, whereas the best algorithm based on AUC was selected for each study to be included in the analyses. These included models utilized various algorithms, such as random forest (n = 4), extreme gradient boosting (n = 5), multilayer perceptron (n = 2), support vector machine [60], neural network (n = 3), logistic regression (n = 7), etc. These models were applied to different data sources, including baseline and clinical data, imaging data, biomarkers, procedural data, and a combination of them. The studies were conducted in various countries, including the United States (n = 16), Germany (n = 7), the Netherlands (n = 6), France (n = 2), etc. Details on the study characteristics are reported in Table 1.

3.2. Risk of Bias Assessment

The risk of bias of assessment based on TRIPOD guidelines showed that most studies met key reporting standards, though some (n = 4) lacked details on model calibration and validation (details are presented in Table S3). The PROBAST risk assessment indicated that 4 studies had high concerns regarding overall applicability due to outcome applicability, while 40 studies had a low risk of bias. The detailed results are presented in Table S4.

3.3. All-Cause Mortality

The pooled AUC of 26 models predicting all-cause mortality was 0.78 (95% CI: 0.74–0.82, I2=98.1%; Figure 2a), with individual model AUCs ranging from 0.55 to 0.97 [10,20,21,22,26,31,35,37,38,39,41,42,44,46,47,48,49,50,51,52,53,54,56,57,58,59]. The pooled accuracy of six models was 0.81 (95% CI: 0.69–0.89, I2 = 98.2%; Figure 3a) [20,35,37,46,50]. The pooled recall was 0.90 (95% CI: 0.70–0.97, I2 = 99.0%; Figure 4a), with recall values ranging from 0.33 to 1.00 (Table 2) [20,35,37,42,46,47,49,56].
The subgroup analysis demonstrated that the models trained exclusively on baseline and clinical data (n = 6) had a pooled AUC of 0.77 (95% CI: 0.69–0.85, I2 = 94.1%) [10,22,37,38,40,54]. The three models utilizing only imaging data showed a pooled AUC of 0.73 (95% CI: 0.71–0.74; I2 = 0%) [26,41,44]. The pooled AUC for the models incorporating baseline, clinical, and imaging data (n = 2) was 0.77 (95% CI: 0.69–0.84; I2 = 48.7%; Figure S1) [20,50], while the models integrating baseline and clinical, and biomarker data (n = 2) achieved a higher pooled AUC of 0.91 (95% CI: 0.88–0.95; I2 = 0.0%; Table 3) [39,48].

3.4. New Permanent Pacemaker Implantation or New Left Bundle Branch Block

For the prediction of pacemaker implantation need or left bundle branch block, the pooled AUC of nine models was 0.75 (95% CI: 0.68–0.82; I2 = 93.2%, Figure 2b), with individual AUC values ranging from 0.61 to 0.92 [19,27,28,33,34,37,40,45,46]. The pooled model accuracy was 0.73 (95% CI: 0.59–0.84; I2 = 99.1%, Figure 3b) [19,27,29,37,45,46] and the pooled recall was 0.87 (95% CI: 0.50–0.98; I2 = 99.1%, Figure 4b) (Table 2) [19,33,37,40,45,46].
The subgroup analysis indicated that the models trained on baseline, clinical, and imaging data (n = 2) had a pooled AUC of 0.77 (95% CI: 0.68–0.85; I2 = 90.1%, Figure S2) [27,34], while the addition of procedural data instead of imaging data (n = 3) further improved the pooled AUC to 0.75 (95% CI: 0.62–0.88; I2 = 73%, Figure S2) [28,29,40]. The highest AUC of 0.92 (95% CI: 0.84–1.00, Figure S2) was achieved in the model developed by Ouahidi et al., which integrated baseline, clinical, imaging, and procedural data (Table 3) [19].

3.5. Valve-Related Dysfunction

The combined AUC of the four models predicting valve-related dysfunction was 0.73 (95% CI: 0.62–0.84; I2 =96%, Figure 2c), with individual model AUCs ranging from 0.57 to 0.80 [23,33,37,46]. The overall accuracy of the three models was 0.79 (95% CI: 0.57–0.91; I2 =98.7%, Figure 3c) [23,37,46]. The pooled recall (n = 4) was 0.54 (95% CI: 0.26–0.80; I2 =99.0%), with individual recall values varying from 0.12 to 0.73 (Table 2) (Figure 4c) [23,33,37,46].
Abdelkhalek et al. used a model trained on baseline and clinical data, achieving an AUC of 0.74 (95% CI: 0.67–0.80) [33]. Another model utilized in the study by Shi et al., which relied solely on imaging data, had a higher AUC of 0.80 (95% CI: 0.73–0.80) [23]. However, the study by Gomes et al. used a model incorporating baseline, clinical, and imaging data, which had a lower AUC of 0.57 (95% CI: 0.52–0.62) (Table 3) (Figure S3) [46].

3.6. MACE

For predicting MACE, the combined AUC of five models was 0.79 (95% CI: 0.67–0.92; I2 = 89.9%, Figure 2d), with individual model AUCs ranging from 0.63 to 0.95 (Table 2) [24,32,36,43,55].
In a subgroup analysis, Stan et al. used a model trained on baseline and clinical data, achieving a high AUC of 0.92 (95% CI: 0.85–0.99) [24]. When models incorporated baseline, clinical, imaging, and biomarker data, the pooled AUC of two models improved to 0.84 (95% CI: 0.60–1; I2 = 91%) [32,43]. However, for models relying solely on imaging data (n = 2), the pooled AUC was lower at 0.67 (95% CI: 0.58–0.76; I2 = 91%) (Table 3) (Figure S4) [36,55].

3.7. Stroke

The pooled AUC of the three models predicting stroke was 0.73 (95%CI: 0.59–0.88; I2 = 97.1%), with individual model AUCs ranging from 0.60 to 0.82 (Figure 2e) (Table 2) [18,37,46].

3.8. Heart Failure-Related Re-Hospitalization

Pooling the three AI models for predicting heart failure [61]-related rehospitalization yielded an overall AUC of 0.70 (95% CI: 0.60–0.81; I2 = 83.3%, Figure 2f), with individual model AUCs ranging from 0.57 to 0.76 (Table 2) [22,25,30].
For the subgroup analysis based on baseline and clinical data, pooling two models resulted in an AUC of 0.67 (95% CI: 0.49–0.86; I2 = 91.2%) [22,30]. Sulaiman et al. reported a higher AUC of 0.74 (95% CI: 0.70–0.78) when integrating baseline, clinical, and procedural data (Table 3) (Figure S5) [25].

3.9. Sensitivity Analysis and Publication Bias

The subgroup analysis based on the type of ML algorithm regarding all-cause mortality and new PPI—the outcomes with the most included studies—also demonstrated a high I2. The details are demonstrated in Figures S6 and S7. The leave-one-out method also showed that removing any of the studies did not reduce the overall heterogeneity. We found no evidence of publication bias for all-cause mortality (Egger’s test p-value: 0.19) and new PPI or LBBB (Egger’s test p-value: 0.38). The funnel plots are presented in Figures S6 and S7.

4. Discussion

This systematic review of 43 studies, including 366,269 patients with severe AS undergoing TAVR, highlights the potential of AI models in predicting various outcomes following TAVR. Our findings showed that these models exhibit vigorous performance in predicting all-cause mortality, the need for new permanent pacemaker implantation, valve-related dysfunction, and major adverse cardiac events. Despite some concerns regarding bias and applicability in certain studies, the overall results demonstrate that integrating diverse clinical, imaging, and biomarker data can enhance predictive accuracy. Additionally, the absence of publication bias reinforces the reliability of the findings, which emphasize the promising role of AI in improving patient management and decision-making in TAVR procedures. However, the high heterogeneity observed in most of our analyses might affect the overall interpretation of our findings.
The usage of AI in predicting post-TAVR outcomes is a topic of concern which was evaluated in notable studies. Hu et al. conducted a study utilizing LR to predict post-TAVR outcomes. In this model, they demonstrated that some preoperative parameters, such as the duration of QRS in the ECG or the calcification score of the aortic valve can be predictive factors for high degree AV block after the procedure [62]. Moreover, in a study conducted by Kurmanaliyev et al., employing fine-tuned machine learning models suggested that the diameter of the left femoral artery, besides the aortic valve calcification score, was a predicting factor of early safety outcomes after TAVR. They observed that patients with lower diameter and higher calcification scores are more prone to early post-procedure adverse outcomes [63]. Whereas, in a systematic review by Sulaiman et al., it was demonstrated that various machine learning algorithms could potentially predict post-TAVR outcomes which could have been utilized in clinical settings and elevating patient-centered care [64].
The application of ML to predict patient outcomes extends beyond the realm of TAVR to a wide range of procedures. For example, in PCI, ML models have proven remarkably effective at forecasting risks like long-term all-cause mortality [65] and MACE in STEMI patients [66]. One study involving over 4500 participants demonstrated that various ML models, such as distributed random forest (DRF) and GBM, could identify high-risk STEMI patients with a high accuracy (AUCs of 0.92 and 0.91, respectively) [67]. Similarly, in patients with STEMI and diabetes, the CatBoost model outperformed traditional risk scores like GRACE, achieving an AUC of 0.92 for predicting in-hospital mortality [68]. These findings emphasize ML’s ability to refine risk assessment in time-sensitive cardiac emergencies. Regarding CABG and SAVR, algorithms like decision trees and random forests have consistently outperformed conventional methods. Decision tree models, for instance, have shown impressive accuracy in predicting short-term mortality after on-pump CABG, achieving AUCs of 0.90 and 0.86 [69]. Moreover, in patients with rheumatic heart disease undergoing valve surgery, ML models such as random forest and artificial neural networks (ANNs) have achieved perfect accuracy (AUCs of 0.98 and 0.952, respectively) in predicting in-hospital mortality [70]. Furthermore, ML’s applicability extends beyond predicting clinical outcomes. A study by Zea-Vera et al. demonstrated that extreme gradient boosting (XGBoost) algorithms can accurately predict not only operative mortality (accuracy 95%) and major morbidity/mortality (accuracy 71%), but also high hospitalization costs (accuracy 84%) across a diverse range of cardiac surgeries, including CABG, valve, and aortic procedures [71]. This reinforces the generalizability of ML across various cardiac procedures, supporting its potential as a versatile tool for risk stratification and outcome prediction in interventional cardiology. This capability to predict resource utilization offers a significant advantage for healthcare systems, enabling better planning and resource allocation.
Our systematic review and meta-analysis primarily focused on key outcomes such as mortality, MACE, PPI, hospitalization for heart failure, stroke, bundle branch block, and valve-related dysfunctions. However, the broader literature indicates that ML models are increasingly used to predict a wider range of outcomes in cardiovascular interventions, including TAVR and other procedures. This expansion reflects a more holistic approach to patient care and risk assessment, moving beyond traditional endpoints to include complications like acute kidney injury (AKI) and prolonged ventilation [72,73,74]. For instance, a study by Chong et al. used ANNs to predict reintubation and prolonged mechanical ventilation after CABG, achieving AUCs of 0.65 and 0.72, respectively [75]. Furthermore, AI models are being utilized in congenital heart surgery to predict not only mortality but also prolonged hospital or ICU stays and postoperative complications. This expanded scope is particularly relevant in complex surgical populations where traditional risk assessments may be less effective [71]. In the context of transcatheter mitral valve replacement (TMVR), ML is being explored to predict early safety outcomes, including all-cause mortality, stroke, life-threatening bleeding, AKI, coronary artery obstruction, major vascular complications, and valve-related dysfunction requiring repeat procedures. A retrospective study involving 224 participants with severe aortic stenosis found that ML models outperformed established risk scores in predicting TMVR success [63]. This capability to predict procedural outcomes represents a significant advancement beyond traditional risk stratification, which primarily focuses on adverse events. By guiding patient selection and procedural planning, ML can play a crucial role in optimizing TMVR outcomes and tailoring patient-based therapeutic plans. It was suggested that these AI-based quantification tools demonstrate superior performance to traditional previous risk scores, such as EuroSCORE and STS score, in predicting comprehensive, varied, and long-term outcomes [76]. Moreover, a fully automated prediction approach significantly reduced the time consumed per patient, which is crucial in the holistic view of clinical workflows [44]. Furthermore, physicians should be properly educated on how to use these novel methods in the most efficient and productive way, which requires dedicated training programs [77]. Notably, strict guidelines should be employed in order to prohibit the unregulated and potentially harmful use of AI technologies in ethical, legal, and professional manners [14].
In this study, we observed that ML models were employed to integrate a wide range of features including baseline clinical characteristics, imaging data, laboratory biomarkers, and procedural variables. This capacity to integrate different feature classes into a single model represents a key strength of ML over traditional risk scores, which are often limited by static variables and linear assumptions. ML models can capture complex, non-linear relationships and interactions among features, offering more nuanced risk stratification. Algorithms such as LR, SVM, and gradient boosting are particularly effective for structured data and have been widely used in predictive modeling for clinical outcomes [78]. However, these ML approaches may be insufficient for unstructured data, such as medical imaging or physiological time series. In such contexts, deep learning architectures offer a distinct advantage. For instance, convolutional neural networks (CNNs) have demonstrated efficacy in extracting morphological features from modalities like CMRI, CT, and echocardiography such as aortic valve calcification or leaflet motion directly from raw pixel data [79,80,81]. Likewise, recurrent neural networks (RNNs) are well suited to model sequential data, including intraoperative hemodynamics or ECG waveforms [82,83]. Furthermore, more advanced multimodal approaches, including ensemble learning, late fusion, and transformer-based architectures, are now being applied to combine structured and unstructured data streams, further enhancing predictive accuracy and supporting personalized decision-making [84,85,86,87,88].
The development of clinically applicable ML models for TAVR requires a comprehensive and methodologically rigorous approach. Robust performance depends not only on algorithmic design but also on access to high-quality, diverse datasets that reflect the heterogeneity of real-world TAVR populations [22]. In contrast to conventional risk scores that are derived from limited cohorts, ML models can adapt to a broader array of input features and dynamically recalibrate based on new data. Nonetheless, the clinical utility of these models extends beyond predictive accuracy. Interpretability is essential for clinician trust and uptake. Explainable AI (XAI) techniques, such as SHapley Additive exPlanations [89] and Local Interpretable Model-Agnostic Explanations (LIME), help elucidate the contribution of individual variables to model predictions, fostering transparency and clinical confidence [90]. To ensure generalizability and clinical relevance, future studies must focus on external validation in diverse populations and integration within real-world clinical workflows [91].
Future studies are needed to assess the performance of ML models in predicting the similar key outcomes we evaluated across various cardiac procedures. Additionally, further studies are required to compare the efficacy of the different ML algorithms using various metrics (e.g., AUC, recall, and calibration). Moreover, a comparison will clarify which model in specific clinical contexts was the best and the most potent. Beyond the key outcomes, studies should evaluate the utilization of ML tools for other critical outcomes, such as hospital readmission rate, cost, quality of life, and pre-procedural complications. Integrating multimodal data (e.g., imaging, biomarkers, and baseline characteristics) could further enhance predictions of outcomes. Additionally, trials with larger populations are needed to validate these models in real-world settings and assess their impact on clinical decision-making. Ultimately, detecting the most effective model of ML for each outcome and procedure might be critical for utilizing ML as a valuable predictive tool.
While AI offers promising advantages in medicine, particularly in the cardiovascular domain, there are some notable challenges that should be addressed. Incomplete medical datasets are major setbacks for the generalizability and scalability of AI. As healthcare datasets become larger and more complex, there is a need to develop effective and more efficient AI models to better perform in medical applications [92]. Additionally, the lack of a standardized framework to validate the aforementioned AI tools in real-world settings is another important issue to deal with [93]. Finally, the high computational cost and the need to improve infrastructure limits the employment of these novel methods in resource-poor settings.
Our study represents the most comprehensive systematic review and meta-analysis of data assessing the efficacy of ML models in predicting outcomes following the TAVR procedure. We conducted subgroup analyses based on multimodal data, including imaging data, clinical data, and baseline variables, which helped mitigate biases and provide more specific results. This approach aimed to identify the best-validated model for outcome prediction. However, the limited number of included studies restricted our ability to fully address the heterogeneity across all results. We also employed various approaches to address the observed heterogeneity, such as subgroup analyses stratified by ML algorithm and the leave-one-out sensitivity analyses. Despite these efforts, the overall heterogeneity remains substantial. The inability to perform additional subgroup analyses based on the type of model validation or study design, due to the limited number of studies with external validation or prospective design, likely contributed to the persistent heterogeneity. Moreover, the differences between the cut-off values among the included studies for reporting performance metrics might also be an important limitation in our analysis. Furthermore, one of the limitations of our study design is publication bias, as systematic reviews often rely on non-population-based data utilization. Finally, due to the limited data in our studies, post-treatment TAVR strategies were not adjusted in the analysis.

5. Conclusions

Overall, our findings emphasize the potential role of AI in patient management after TAVR. Healthcare providers can utilize these advanced predictive models to better identify high-risk patients and tailor person-centered interventions. These differences could ultimately lead to better clinical outcomes. These outcomes might potentially revolutionize routine clinical practice and cardiovascular care.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/jpm15070302/s1, Table S1. PRISMA checklist; Table S2. Search terms of databases; Table S3. Quality assessment of included studies with TRIPOD scoring [10,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59]; Table S4. Quality assessment of included studies with PROBAST scoring [10,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59]; Figure S1: Forest plot for all-cause mortality subgroup analysis [20,21,22,26,34,35,37,38,39,41,42,44,46,47,48,49,50,51,52,53,54,57,58,59]; Figure S2: Forest plot for new pacemaker implantation/left bundle branch block outcome subgroup analysis [19,27,28,33,34,37,40,45,46]; Figure S3. Forest plot for valve-related dysfunction subgroup analysis [23,33,46]; Figure S4. Forest plot for MACE subgroup analysis [24,32,36,43,55]; Figure S5. Forest plot for heart failure hospitalization subgroup analysis [22,25,30]; Figure S6. Funnel plot for all-cause mortality outcome; Figure S7. Funnel plot for new pacemaker implantation/left bundle branch block outcome.

Author Contributions

Conceptualization, S.S. and A.M.; Methodology, S.K.; Validation, S.S., A.M. and K.H.; Formal analysis, S.K.; Investigation, S.S. and A.M.; Writing—original draft, S.S., A.M., S.K., S.A., S.M., P.F., F.T.A. and Y.D.; Writing—review and editing, M.A., J.S.R., M.M., Y.J. and K.H.; Visualization, S.A., S.M. and P.F.; Supervision, M.N. and K.H.; Project administration, S.S., A.M. and K.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The data supporting the results of this study are available in each included study.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

Transcatheter aortic valve replacement (TAVR), artificial intelligence (AI), machine learning (ML), preferred reporting items for systematic reviews and meta-analyses protocols (PRISMA), cardiac magnetic resonance imaging (CMRI), computed tomography (CT), electrocardiography (ECG), prediction model risk of bias assessment tool (PROBAST), transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD), major adverse cardiovascular events (MACE), left bundle branch block (LBBB), confidence intervals (CIs), distributed random forest (DRF), artificial neural networks (ANNs), extreme gradient boosting (XGBoost), acute kidney injury (AKI), transcatheter mitral valve replacement (TMVR), convolutional neural networks (CNNs), recurrent neural networks (RNNs), explainable AI (XAI), local interpretable model-agnostic explanations (LIME).

References

  1. Baumgartner, H.; Falk, V.; Bax, J.J.; De Bonis, M.; Hamm, C.; Holm, P.J.; Iung, B.; Lancellotti, P.; Lansac, E.; Munoz, D.R. 2017 ESC/EACTS Guidelines for the management of valvular heart disease. Pol. Heart J. 2018, 76, 1–62. [Google Scholar] [CrossRef] [PubMed]
  2. Ajmal, Z.; Rehman, Z.U.; Ishtiaq, A.; Iftikhar, H.; Khokhar, M.M.; Khan, B.; Asad, A.; Nasir, H.; Athar, S.M.; Hassan, A.; et al. Comparative Outcomes of Transcatheter Versus Surgical Aortic Valve Replacement in Moderate-Risk Patients with Aortic Stenosis: A Systematic Review of Clinical Trials. Cureus 2024, 16, e70268. [Google Scholar] [CrossRef]
  3. Zou, Q.; Wei, Z.; Sun, S. Complications in transcatheter aortic valve replacement: A comprehensive analysis and management strategies. Curr. Probl. Cardiol. 2024, 49, 102478. [Google Scholar] [CrossRef] [PubMed]
  4. Tang, G.H.L.; Verma, S.; Bhatt, D.L. Transcatheter Aortic Valve Replacement in Low-Risk Patients. Circulation 2019, 140, 801–803. [Google Scholar] [CrossRef] [PubMed]
  5. Kazemian, S.; Fallahtafti, P.; Sharifi, M.; Mohammadi, N.S.H.; Soleimani, H.; Moghadam, A.S.; Karimi, E.; Sattar, Y.; Jenab, Y.; Mehrani, M.; et al. Trends in Transcatheter Versus Surgical Aortic Valve Replacement Outcomes in Patients With Low-Surgical Risk: A Systematic Review and Meta-Analysis of Randomized Trials. J. Am. Heart Assoc. 2024, 13, e036179. [Google Scholar] [CrossRef]
  6. Tang, M.; Wang, D.; Xue, F.; Chen, Y.; Zhang, N.; Wang, J.; Zhao, P.; Zhou, T. Analysis of potential categories of quality of life of transcatheter aortic valve replacement patients and discussion of their influencing factors. BMC Cardiovasc. Disord. 2025, 25, 259. [Google Scholar] [CrossRef]
  7. Arnold Suzanne, V.; Zhang, Y.; Baron Suzanne, J.; McAndrew Thomas, C.; Alu Maria, C.; Kodali Susheel, K.; Kapadia, S.; Thourani Vinod, H.; Miller, D.C.; Mack Michael, J.; et al. Impact of Short-Term Complications on Mortality and Quality of Life After Transcatheter Aortic Valve Replacement. JACC Cardiovasc. Interv. 2019, 12, 362–369. [Google Scholar] [CrossRef]
  8. AziziKia, H.; Mousavi, A.; Shojaei, S.; Shaker, F.; Salabat, D.; Bahri, R.A.; Dolama, R.H.; Radkhah, H. Predictive potential of pre-procedural cardiac and inflammatory biomarkers regarding mortality following transcatheter aortic valve implantation: A systematic review and meta-analysis. Heart Lung 2025, 69, 229–240. [Google Scholar] [CrossRef]
  9. Siddiqi, T.J.; Usman, M.S.; Khan, M.S.; Khan, M.A.A.; Riaz, H.; Khan, S.U.; Murad, M.H.; Kavinsky, C.J.; Doukky, R.; Kalra, A.; et al. Systematic review and meta-analysis of current risk models in predicting short-term mortality after transcatheter aortic valve replacement. EuroIntervention 2020, 15, 1497–1505. [Google Scholar] [CrossRef]
  10. Agasthi, P.; Ashraf, H.; Pujari, S.H.; Girardo, M.E.; Tseng, A.; Mookadam, F.; Venepally, N.R.; Buras, M.; Khetarpal, B.K.; Allam, M.; et al. Artificial Intelligence Trumps TAVI(2)-SCORE and CoreValve Score in Predicting 1-Year Mortality Post-Transcatheter Aortic Valve Replacement. Cardiovasc. Revasc. Med. 2021, 24, 33–41. [Google Scholar] [CrossRef]
  11. Sazzad, F.; Ler, A.A.L.; Furqan, M.S.; Tan, L.K.Z.; Leo, H.L.; Kuntjoro, I.; Tay, E.; Kofidis, T. Harnessing the power of artificial intelligence in predicting all-cause mortality in transcatheter aortic valve replacement: A systematic review and meta-analysis. Front. Cardiovasc. Med. 2024, 11, 1343210. [Google Scholar] [CrossRef] [PubMed]
  12. Parvin, N.; Joo, S.W.; Jung, J.H.; Mandal, T.K. Multimodal AI in Biomedicine: Pioneering the Future of Biomaterials, Diagnostics, and Personalized Healthcare. Nanomaterials 2025, 15, 895. [Google Scholar] [CrossRef]
  13. Xu, Y.; Liu, X.; Cao, X.; Huang, C.; Liu, E.; Qian, S.; Liu, X.; Wu, Y.; Dong, F.; Qiu, C.-W.; et al. Artificial intelligence: A powerful paradigm for scientific research. Innovation 2021, 2, 100179. [Google Scholar] [CrossRef] [PubMed]
  14. Mennella, C.; Maniscalco, U.; De Pietro, G.; Esposito, M. Ethical and regulatory challenges of AI technologies in healthcare: A narrative review. Heliyon 2024, 10, e26297. [Google Scholar] [CrossRef]
  15. Moher, D.; Liberati, A.; Tetzlaff, J.; Altman, D.G. Preferred reporting items for systematic reviews and meta-analyses: The PRISMA statement. PLoS Med. 2009, 6, e1000097. [Google Scholar] [CrossRef]
  16. Wolff, R.F.; Moons, K.G.M.; Riley, R.D.; Whiting, P.F.; Westwood, M.; Collins, G.S.; Reitsma, J.B.; Kleijnen, J.; Mallett, S. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann. Intern. Med. 2019, 170, 51–58. [Google Scholar] [CrossRef]
  17. Collins, G.S.; Reitsma, J.B.; Altman, D.G.; Moons, K.G.M. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD Statement. BMC Med. 2015, 13, 1. [Google Scholar] [CrossRef] [PubMed]
  18. Okuno, T.; Overtchouk, P.; Asami, M.; Tomii, D.; Stortecky, S.; Praz, F.; Lanz, J.; Siontis, G.C.M.; Gräni, C.; Windecker, S.; et al. Deep learning-based prediction of early cerebrovascular events after transcatheter aortic valve replacement. Sci. Rep. 2021, 11, 18754. [Google Scholar] [CrossRef]
  19. El Ouahidi, A.; El Ouahidi, Y.; Nicol, P.-P.; Hannachi, S.; Benic, C.; Mansourati, J.; Pasdeloup, B.; Didier, R. Machine learning for pacemaker implantation prediction after TAVI using multimodal imaging data. Sci. Rep. 2024, 14, 25008. [Google Scholar] [CrossRef]
  20. Penso, M.; Pepi, M.; Fusini, L.; Muratori, M.; Cefalù, C.; Mantegazza, V.; Gripari, P.; Ali, S.G.; Fabbiocchi, F.; Bartorelli, A.L.; et al. Predicting Long-Term Mortality in TAVI Patients Using Machine Learning Techniques. J. Cardiovasc. Dev. Dis. 2021, 8, 44. [Google Scholar] [CrossRef]
  21. Pollari, F.; Hitzl, W.; Rottmann, M.; Vogt, F.; Ledwon, M.; Langhammer, C.; Eckner, D.; Jessl, J.; Bertsch, T.; Pauschinger, M.; et al. A Machine Learning Model for the Accurate Prediction of 1-Year Survival in TAVI Patients: A Retrospective Observational Cohort Study. J. Clin. Med. 2023, 12, 5481. [Google Scholar] [CrossRef]
  22. Savitz, S.T.; Leong, T.; Sung, S.H.; Kitzman, D.W.; McNulty, E.; Mishell, J.; Rassi, A.; Ambrosy, A.P.; Go, A.S. Predicting short-term outcomes after transcatheter aortic valve replacement for aortic stenosis. Am. Heart J. 2023, 256, 60–72. [Google Scholar] [CrossRef] [PubMed]
  23. Shi, W.; Feng, D.; Hu, X.; Wang, C.; Niu, G.; Zhao, Z.; Zhang, H.; Wang, M.; Wu, Y. Prediction of hypoattenuating leaflet thickening in patients undergoing transcatheter aortic valves replacement based on clinical factors and 4D-computed tomography morphological characteristics: A retrospective cross-sectional study. Int. J. Cardiol. 2024, 410, 132219. [Google Scholar] [CrossRef]
  24. Stan, A.; Călburean, P.A.; Drinkal, R.K.; Harpa, M.; Elkahlout, A.; Nicolae, V.C.; Tomșa, F.; Hadadi, L.; Brînzaniuc, K.; Suciu, H.; et al. Inflammatory Status Assessment by Machine Learning Techniques to Predict Outcomes in Patients with Symptomatic Aortic Stenosis Treated by Transcatheter Aortic Valve Replacement. Diagnostics 2023, 13, 2907. [Google Scholar] [CrossRef] [PubMed]
  25. Sulaiman, S.; Kawsara, A.; Mahayni, A.A.; El Sabbagh, A.; Singh, M.; Crestanello, J.; Gulati, R.; Alkhouli, M. Development and Validation of a Machine Learning Score for Readmissions After Transcatheter Aortic Valve Implantation. JACC Adv. 2022, 1, 100060. [Google Scholar] [CrossRef]
  26. Theis, M.; Block, W.; Luetkens, J.A.; Attenberger, U.I.; Nowak, S.; Sprinkart, A.M. Direct deep learning-based survival prediction from pre-interventional CT prior to transcatheter aortic valve replacement. Eur. J. Radiol. 2023, 168, 111150. [Google Scholar] [CrossRef] [PubMed]
  27. Truong, V.T.; Beyerbach, D.; Mazur, W.; Wigle, M.; Bateman, E.; Pallerla, A.; Ngo, T.N.M.; Shreenivas, S.; Tretter, J.T.; Palmer, C.; et al. Machine learning method for predicting pacemaker implantation following transcatheter aortic valve replacement. Pacing Clin. Electrophysiol. 2021, 44, 334–340. [Google Scholar] [CrossRef]
  28. Tsushima, T.; Nadeem, F.; Al-Kindi, S.; Clevenger, J.R.; Bansal, E.J.; Wheat, H.L.; Kalra, A.; Attizzani, G.F.; Elgudin, Y.; Markowitz, A.; et al. Risk Prediction Model for Cardiac Implantable Electronic Device Implantation After Transcatheter Aortic Valve Replacement. JACC Clin. Electrophysiol. 2020, 6, 295–303. [Google Scholar] [CrossRef]
  29. Vejpongsa, P.; Zhang, X.; Bhise, V.; Kitkungvan, D.; Shivamurthy, P.; Anderson, H.V.; Balan, P.; Nguyen, T.C.; Estrera, A.L.; Dougherty, A.H.; et al. Risk Prediction Model for Permanent Pacemaker Implantation after Transcatheter Aortic Valve Replacement. Struct. Heart 2018, 2, 328–335. [Google Scholar] [CrossRef]
  30. Zahid, S.; Agrawal, A.; Salman, F.; Khan, M.Z.; Ullah, W.; Teebi, A.; Khan, S.U.; Sulaiman, S.; Balla, S. Development and Validation of a Machine Learning Risk-Prediction Model for 30-Day Readmission for Heart Failure Following Transcatheter Aortic Valve Replacement (TAVR-HF Score). Curr. Probl. Cardiol. 2024, 49, 102143. [Google Scholar] [CrossRef]
  31. Zisiopoulou, M.; Berkowitsch, A.; Redlich, L.; Walther, T.; Fichtlscherer, S.; Leistner, D.M. Personalised preinterventional risk stratification of mortality, length of stay and hospitalisation costs in transcatheter aortic valve implantation using a machine learning algorithm: A pilot trial. Open Heart 2024, 11, e002540. [Google Scholar] [CrossRef]
  32. Zusman, O.; Kornowski, R.; Witberg, G.; Lador, A.; Orvin, K.; Levi, A.; Assali, A.; Vaknin-Assa, H.; Sharony, R.; Shapira, Y.; et al. Transcatheter Aortic Valve Implantation Futility Risk Model Development and Validation Among Treated Patients With Aortic Stenosis. Am. J. Cardiol. 2017, 120, 2241–2246. [Google Scholar] [CrossRef] [PubMed]
  33. Abdelkhalek, M.; Bahadormanesh, N.; Ganame, J.; Keshavarz-Motamed, Z. Incremental prognostic value of intensity-weighted regional calcification scoring using contrast CT imaging in TAVR. Eur. Heart J. Imaging Methods Pract. 2023, 1, qyad027. [Google Scholar] [CrossRef]
  34. Agasthi, P.; Ashraf, H.; Pujari, S.H.; Girardo, M.; Tseng, A.; Mookadam, F.; Venepally, N.; Buras, M.R.; Abraham, B.; Khetarpal, B.K.; et al. Prediction of permanent pacemaker implantation after transcatheter aortic valve replacement: The role of machine learning. World J. Cardiol. 2023, 15, 95–105. [Google Scholar] [CrossRef] [PubMed]
  35. Alhwiti, T.; Aldrugh, S.; Megahed, F.M. Predicting in-hospital mortality after transcatheter aortic valve replacement using administrative data and machine learning. Sci. Rep. 2023, 13, 10252. [Google Scholar] [CrossRef]
  36. Aquino, G.J.; Abadia, A.F.; Schoepf, U.J.; Emrich, T.; Yacoub, B.; Kabakus, I.; Violette, A.; Wiley, C.; Moreno, A.; Sahbaee, P.; et al. Coronary CT Fractional Flow Reserve before Transcatheter Aortic Valve Replacement: Clinical Outcomes. Radiology 2022, 302, 50–58. [Google Scholar] [CrossRef] [PubMed]
  37. Asif, N.; Ayoade, P.; Razzouk, J.; Bohen, D.; Tooker, M.; Gladstone, L.; Hoff, J.; Mohsen, A.; Arnold, S.; Rabkin, D.G. Multilayer Perceptron Neural Network Analysis of Fluoroscopic Working Angle on Transcatheter Aortic Valve Implantation Complications. Cureus 2024, 16, e59144. [Google Scholar] [CrossRef]
  38. Bansal, A.; Mentias, A.; Jaber, W.; Zmaili, M.A.; Svensson, L.G.; Krishnaswamy, A.; Reed, G.W.; Puri, R.; Kapadia, S.R.; Xu, B. Machine-Learning Risk Model for Predicting In-Hospital Mortality for Patients With Infective Endocarditis After Transcatheter Aortic Valve Replacement. Cardiovasc. Revasc. Med. 2022, 39, 121–122. [Google Scholar] [CrossRef]
  39. Barbieri, F.; Pfeifer, B.E.; Senoner, T.; Dobner, S.; Spitaler, P.; Semsroth, S.; Lambert, T.; Zweiker, D.; Neururer, S.B.; Scherr, D.; et al. A Neuronal Network-Based Score Predicting Survival in Patients Undergoing Aortic Valve Intervention: The ABC-AS Score. J. Clin. Med. 2024, 13, 3691. [Google Scholar] [CrossRef]
  40. Barrett, C.D.; Nickel, A.; Rosenberg, M.A.; Ream, K.; Tzou, W.S.; Aleong, R.; Tumolo, A.; Garg, L.; Zipse, M.; West, J.J.; et al. PRIME score for prediction of permanent pacemaker implantation after transcatheter aortic valve replacement. Catheter. Cardiovasc. Interv. 2023, 102, 1357–1363. [Google Scholar] [CrossRef]
  41. Brüggemann, D.; Kuzo, N.; Anwer, S.; Kebernik, J.; Eberhard, M.; Alkadhi, H.; Tanner, F.C.; Konukoglu, E. Predicting mortality after transcatheter aortic valve replacement using preprocedural CT. Sci. Rep. 2024, 14, 12526. [Google Scholar] [CrossRef] [PubMed]
  42. van Erck, D.; Moeskops, P.; Schoufour, J.D.; Weijs, P.J.M.; Scholte op Reimer, W.J.M.; van Mourik, M.S.; Planken, R.N.; Vis, M.M.; Baan, J.; Išgum, I.; et al. Low muscle quality on a procedural computed tomography scan assessed with deep learning as a practical useful predictor of mortality in patients with severe aortic valve stenosis. Clin. Nutr. ESPEN 2024, 63, 142–147. [Google Scholar] [CrossRef]
  43. Erdogan, A.; Genc, O.; Inan, D.; Yildirim, A.; Ibisoglu, E.; Guler, Y.; Genc, D.; Guler, A.; Karagoz, A.; Kurt, I.H.; et al. Prediction of Major Adverse Cardiac Events After Transcatheter Aortic Valve Implantation: A Machine Learning Approach with GRACE Score. Sisli Etfal Hastan. Tip. Bul. 2024, 58, 216–225. [Google Scholar] [CrossRef] [PubMed]
  44. Evertz, R.; Lange, T.; Backhaus, S.J.; Schulz, A.; Beuthner, B.E.; Topci, R.; Toischer, K.; Puls, M.; Kowallick, J.T.; Hasenfuß, G.; et al. Artificial Intelligence Enabled Fully Automated CMR Function Quantification for Optimized Risk Stratification in Patients Undergoing Transcatheter Aortic Valve Replacement. J. Interv. Cardiol. 2022, 2022, 1368878. [Google Scholar] [CrossRef]
  45. Galli, V.; Loncaric, F.; Rocatello, G.; Astudillo, P.; Sanchis, L.; Regueiro, A.; De Backer, O.; Swaans, M.; Bosmans, J.; Ribeiro, J.M.; et al. Towards patient-specific prediction of conduction abnormalities induced by transcatheter aortic valve implantation: A combined mechanistic modelling and machine learning approach. Eur. Heart J. Digit. Health 2021, 2, 606–615. [Google Scholar] [CrossRef] [PubMed]
  46. Gomes, B.; Pilz, M.; Reich, C.; Leuschner, F.; Konstandin, M.; Katus, H.A.; Meder, B. Machine learning-based risk prediction of intrahospital clinical outcomes in patients undergoing TAVI. Clin. Res. Cardiol. 2021, 110, 343–356. [Google Scholar] [CrossRef]
  47. Hernandez-Suarez, D.F.; Kim, Y.; Villablanca, P.; Gupta, T.; Wiley, J.; Nieves-Rodriguez, B.G.; Rodriguez-Maldonado, J.; Feliu Maldonado, R.; da Luz Sant’Ana, I.; Sanina, C.; et al. Machine Learning Prediction Models for In-Hospital Mortality After Transcatheter Aortic Valve Replacement. JACC Cardiovasc. Interv. 2019, 12, 1328–1338. [Google Scholar] [CrossRef]
  48. Hoffmann, J.; Mas-Peiro, S.; Berkowitsch, A.; Boeckling, F.; Rasper, T.; Pieszko, K.; De Rosa, R.; Hiczkiewicz, J.; Burchardt, P.; Fichtlscherer, S.; et al. Inflammatory signatures are associated with increased mortality after transfemoral transcatheter aortic valve implantation. ESC Heart Fail. 2020, 7, 2597–2610. [Google Scholar] [CrossRef]
  49. Kwiecinski, J.; Dabrowski, M.; Nombela-Franco, L.; Grodecki, K.; Pieszko, K.; Chmielak, Z.; Pylko, A.; Hennessey, B.; Kalinczuk, L.; Tirado-Conte, G.; et al. Machine learning for prediction of all-cause mortality after transcatheter aortic valve implantation. Eur. Heart J. Qual. Care Clin. Outcomes 2023, 9, 768–777. [Google Scholar] [CrossRef]
  50. Lertsanguansinchai, P.; Chokesuwattanaskul, R.; Petchlorlian, A.; Suttirut, P.; Buddhari, W. Machine learning-based predictive risk models for 30-day and 1-year mortality in severe aortic stenosis patients undergoing transcatheter aortic valve implantation. Int. J. Cardiol. 2023, 374, 20–26. [Google Scholar] [CrossRef]
  51. Lopes, R.R.; Mamprin, M.; Zelis, J.M.; Tonino, P.A.L.; van Mourik, M.S.; Vis, M.M.; Zinger, S.; de Mol, B.A.J.M.; de With, P.H.N.; Marquering, H.A. Local and Distributed Machine Learning for Inter-hospital Data Utilization: An Application for TAVI Outcome Prediction. Front. Cardiovasc. Med. 2021, 8, 787246. [Google Scholar] [CrossRef] [PubMed]
  52. Lopes, R.R.; van Mourik, M.S.; Schaft, E.V.; Ramos, L.A.; Baan, J., Jr.; Vendrik, J.; de Mol, B.; Vis, M.M.; Marquering, H.A. Value of machine learning in predicting TAVI outcomes. Neth. Heart J. 2019, 27, 443–450. [Google Scholar] [CrossRef] [PubMed]
  53. Mamprin, M.; Zelis, J.M.; Tonino, P.A.L.; Zinger, S.; de With, P.H.N. Decision Trees for Predicting Mortality in Transcatheter Aortic Valve Implantation. Bioengineering 2021, 8, 22. [Google Scholar] [CrossRef]
  54. Abdul Ghffar, Y.; Osman, M.; Shrestha, S.; Shaukat, F.; Kagiyama, N.; Alkhouli, M.; Raybuck, B.; Badhwar, V.; Sengupta, P.P. Usefulness of Semisupervised Machine-Learning-Based Phenogrouping to Improve Risk Assessment for Patients Undergoing Transcatheter Aortic Valve Implantation. Am. J. Cardiol. 2020, 136, 122–130. [Google Scholar] [CrossRef]
  55. Tremamunno, G.; Vecsey-Nagy, M.; Schoepf, U.J.; Zsarnoczay, E.; Aquino, G.J.; Kravchenko, D.; Laghi, A.; Jacob, A.; Sharma, P.; Rapaka, S.; et al. Artificial Intelligence Improves Prediction of Major Adverse Cardiovascular Events in Patients Undergoing Transcatheter Aortic Valve Replacement Planning CT. Acad. Radiol. 2024, 32, 702–711. [Google Scholar] [CrossRef]
  56. Chen, Y.; Zhou, J.; Kai Chan, J.S.; Liu, T.; Hothi, S.S.; Roever, L.; Rajan, R.; Kei Wong, I.C.; Zhang, Q.; Tse, G.; et al. Development of an Electronic Frailty Index for Predicting Mortality in Patients Undergoing Transcatheter Aortic Valve Replacement using Machine Learning. Ann. Clin. Cardiol. 2023, 5, 17–26. [Google Scholar] [CrossRef]
  57. Mamprin, M.; Lopes, R.R.; Zelis, J.M.; Tonino, P.A.L.; van Mourik, M.S.; Vis, M.M.; Zinger, S.; de Mol, B.; de With, P.H.N. Machine Learning for Predicting Mortality in Transcatheter Aortic Valve Implantation: An Inter-Center Cross Validation Study. J. Cardiovasc. Dev. Dis. 2021, 8, 65. [Google Scholar] [CrossRef]
  58. Leha, A.; Huber, C.; Friede, T.; Bauer, T.; Beckmann, A.; Bekeredjian, R.; Bleiziffer, S.; Herrmann, E.; Möllmann, H.; Walther, T.; et al. Development and validation of explainable machine learning models for risk of mortality in transcatheter aortic valve implantation: TAVI risk machine scores. Eur. Heart J. Digit. Health 2023, 4, 225–235. [Google Scholar] [CrossRef]
  59. Yordanov, T.R.; Ravelli, A.C.J.; Amiri, S.; Vis, M.; Houterman, S.; Van der Voort, S.R.; Abu-Hanna, A. Performance of federated learning-based models in the Dutch TAVI population was comparable to central strategies and outperformed local strategies. Front. Cardiovasc. Med. 2024, 11, 1399138. [Google Scholar] [CrossRef]
  60. McBane, R.D., 2nd; Wysokinski, W.E.; Le-Rademacher, J.G.; Zemla, T.; Ashrani, A.; Tafur, A.; Perepu, U.; Anderson, D.; Gundabolu, K.; Kuzma, C.; et al. Apixaban and dalteparin in active malignancy-associated venous thromboembolism: The ADAM VTE trial. J. Thromb. Haemost. 2020, 18, 411–421. [Google Scholar] [CrossRef]
  61. Mancia, G.; Kreutz, R.; Brunström, M.; Burnier, M.; Grassi, G.; Januszewicz, A.; Muiesan, M.L.; Tsioufis, K.; Agabiti-Rosei, E.; Algharably, E.A.E.; et al. 2023 ESH Guidelines for the management of arterial hypertension The Task Force for the management of arterial hypertension of the European Society of Hypertension: Endorsed by the International Society of Hypertension (ISH) and the European Renal Association (ERA). J. Hypertens. 2023, 41, 1874–2071. [Google Scholar] [CrossRef]
  62. Hu, P.; Lin, N.; Wu, Z. Research and Prediction of Factors Related to High Degree Atrioventricular Block after TAVI Surgery Based on Logistic Regression Model. Heart Surg. Forum 2023, 26, E531–E536. [Google Scholar] [CrossRef] [PubMed]
  63. Kurmanaliyev, A.; Sutiene, K.; Braukylienė, R.; Aldujeli, A.; Jurenas, M.; Kregzdyte, R.; Braukyla, L.; Zhumagaliyev, R.; Aitaliyev, S.; Zhanabayev, N. An Integrative Machine Learning Model for Predicting Early Safety Outcomes in Patients Undergoing Transcatheter Aortic Valve Implantation. Medicina 2025, 61, 374. [Google Scholar] [CrossRef]
  64. Sulaiman, R.; Faisal, M.A.A.; Hasan, M.; Chowdhury, M.E.; Bensaali, F.; Alnabti, A.; Yalcin, H.C. Machine learning for predicting outcomes of transcatheter aortic valve implantation: A systematic review. Int. J. Med. Inform. 2025, 197, 105840. [Google Scholar] [CrossRef] [PubMed]
  65. Liu, S.; Yang, S.; Xing, A.; Zheng, L.; Shen, L.; Tu, B.; Yao, Y. Machine learning-based long-term outcome prediction in patients undergoing percutaneous coronary intervention. Cardiovasc. Diagn. Ther. 2021, 11, 736–743. [Google Scholar] [CrossRef]
  66. Huang, L.; Zhang, J.; Huang, Q.; Cui, R.; Chen, J. In-hospital major adverse cardiovascular events after primary percutaneous coronary intervention in patients with acute ST-segment elevation myocardial infarction: A retrospective study under the China chest pain center (standard center) treatment system. BMC Cardiovasc. Disord. 2023, 23, 198. [Google Scholar] [CrossRef] [PubMed]
  67. Tofighi, S.; Poorhosseini, H.; Jenab, Y.; Alidoosti, M.; Sadeghian, M.; Mehrani, M.; Tabrizi, Z.; Hashemi, P. Comparison of machine-learning models for the prediction of 1-year adverse outcomes of patients undergoing primary percutaneous coronary intervention for acute ST-elevation myocardial infarction. Clin. Cardiol. 2024, 47, e24157. [Google Scholar] [CrossRef]
  68. Chen, P.; Wang, B.; Zhao, L.; Ma, S.; Wang, Y.; Zhu, Y.; Zeng, X.; Bai, Z.; Shi, B. Machine learning for predicting intrahospital mortality in ST-elevation myocardial infarction patients with type 2 diabetes mellitus. BMC Cardiovasc. Disord. 2023, 23, 585. [Google Scholar] [CrossRef]
  69. Ghavidel, A.A.; Javadikasgari, H.; Maleki, M.; Karbassi, A.; Omrani, G.; Noohi, F. Two new mathematical models for prediction of early mortality risk in coronary artery bypass graft surgery. J. Thorac. Cardiovasc. Surg. 2014, 148, 1291–1298.e1291. [Google Scholar] [CrossRef]
  70. Mejia, O.A.V.; Antunes, M.J.; Goncharov, M.; Dallan, L.R.P.; Veronese, E.; Lapenna, G.A.; Lisboa, L.A.F.; Dallan, L.A.O.; Brandão, C.M.A.; Zubelli, J.; et al. Predictive performance of six mortality risk scores and the development of a novel model in a prospective cohort of patients undergoing valve surgery secondary to rheumatic fever. PLoS ONE 2018, 13, e0199277. [Google Scholar] [CrossRef]
  71. Zea-Vera, R.; Ryan, C.T.; Navarro, S.M.; Havelka, J.; Wall, M.J.; Coselli, J.S.; Rosengart, T.K.; Chatterjee, S.; Ghanta, R.K. Development of a Machine Learning Model to Predict Outcomes and Cost After Cardiac Surgery. Ann. Thorac. Surg. 2023, 115, 1533–1542. [Google Scholar] [CrossRef]
  72. Khalaji, A.; Behnoush, A.H.; Jameie, M.; Sharifi, A.; Sheikhy, A.; Fallahzadeh, A.; Sadeghian, S.; Pashang, M.; Bagheri, J.; Ahmadi Tafti, S.H.; et al. Machine learning algorithms for predicting mortality after coronary artery bypass grafting. Front. Cardiovasc. Med. 2022, 9, 977747. [Google Scholar] [CrossRef] [PubMed]
  73. Jia, T.; Xu, K.; Bai, Y.; Lv, M.; Shan, L.; Li, W.; Zhang, X.; Li, Z.; Wang, Z.; Zhao, X.; et al. Machine-learning predictions for acute kidney injuries after coronary artery bypass grafting: A real-life muticenter retrospective cohort study. BMC Med. Inform. Decis. Mak. 2023, 23, 270. [Google Scholar] [CrossRef] [PubMed]
  74. Kilic, A.; Goyal, A.; Miller, J.K.; Gleason, T.G.; Dubrawksi, A. Performance of a machine learning algorithm in predicting outcomes of aortic valve replacement. Ann. Thorac. Surg. 2021, 111, 503–510. [Google Scholar] [CrossRef]
  75. Chong, C.F.; Li, Y.C.; Wang, T.L.; Chang, H. Stratification of adverse outcomes by preoperative risk factors in coronary artery bypass graft patients: An artificial neural network prediction model. AMIA Annu. Symp. Proc. 2003, 2003, 160–164. [Google Scholar] [PubMed]
  76. Benjamin, M.M.; Rabbat, M.G. Artificial Intelligence in Transcatheter Aortic Valve Replacement: Its Current Role and Ongoing Challenges. Diagnostics 2024, 14, 261. [Google Scholar] [CrossRef]
  77. Paranjape, K.; Schinkel, M.; Nannan Panday, R.; Car, J.; Nanayakkara, P. Introducing Artificial Intelligence Training in Medical Education. JMIR Med. Educ. 2019, 5, e16048. [Google Scholar] [CrossRef]
  78. Lee, S.; Park, J.S.; Hong, J.H.; Woo, H.; Lee, C.; Yoon, J.H.; Lee, K.-B.; Chung, S.; Yoon, D.S.; Lee, J.H. Artificial Intelligence in Bacterial Diagnostics and Antimicrobial Susceptibility Testing: Current Advances and Future Prospects. Biosens. Bioelectron. 2025, 280, 117399. [Google Scholar] [CrossRef]
  79. Lessmann, N.; van Ginneken, B.; Zreik, M.; de Jong, P.A.; de Vos, B.D.; Viergever, M.A.; Isgum, I. Automatic Calcium Scoring in Low-Dose Chest CT Using Deep Neural Networks With Dilated Convolutions. IEEE Trans. Med. Imaging 2018, 37, 615–625. [Google Scholar] [CrossRef]
  80. Gouravani, M.; Shahrabi Farahani, M.; Salehi, M.A.; Shojaei, S.; Mirakhori, S.; Harandi, H.; Mohammadi, S.; Saleh, R.R. Diagnostic performance of artificial intelligence in detection of renal cell carcinoma: A systematic review and meta-analysis. BMC Cancer 2025, 25, 155. [Google Scholar] [CrossRef]
  81. Munafò, R.; Saitta, S.; Ingallina, G.; Denti, P.; Maisano, F.; Agricola, E.; Redaelli, A.; Votta, E. A Deep Learning-Based Fully Automated Pipeline for Regurgitant Mitral Valve Anatomy Analysis From 3D Echocardiography. IEEE Access 2024, 12, 5295–5308. [Google Scholar] [CrossRef]
  82. Kim, J.-C.; Chung, K. Recurrent neural network-based multimodal deep learning for estimating missing values in healthcare. Appl. Sci. 2022, 12, 7477. [Google Scholar] [CrossRef]
  83. Lee, S.; Lee, M.; Kim, S.-H.; Woo, J. Intraoperative hypotension prediction model based on systematic feature engineering and machine learning. Sensors 2022, 22, 3108. [Google Scholar] [CrossRef] [PubMed]
  84. Boehm, K.M.; Aherne, E.A.; Ellenson, L.; Nikolovski, I.; Alghamdi, M.; Vázquez-García, I.; Zamarin, D.; Long Roche, K.; Liu, Y.; Patel, D.; et al. Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer. Nat. Cancer 2022, 3, 723–733. [Google Scholar] [CrossRef] [PubMed]
  85. Jaltotage, B.; Lu, J.; Dwivedi, G. Use of Artificial Intelligence Including Multimodal Systems to Improve the Management of Cardiovascular Disease. Can. J. Cardiol. 2024, 40, 1804–1812. [Google Scholar] [CrossRef]
  86. Abimannan, S.; El-Alfy, E.-S.; Chang, Y.-S.; Hussain, S.; Shukla, S.; Satheesh, D. Ensemble Multifeatured Deep Learning Models and Applications: A Survey. IEEE Access 2023, 11, 107194–107217. [Google Scholar] [CrossRef]
  87. Ozturk, C.; Pak, D.H.; Rosalia, L.; Goswami, D.; Robakowski, M.E.; McKay, R.; Nguyen, C.T.; Duncan, J.S.; Roche, E.T. AI-Powered Multimodal Modeling of Personalized Hemodynamics in Aortic Stenosis. Adv. Sci. 2025, 12, 2404755. [Google Scholar] [CrossRef]
  88. Li, Y.; El Habib Daho, M.; Conze, P.-H.; Zeghlache, R.; Le Boité, H.; Tadayoni, R.; Cochener, B.; Lamard, M.; Quellec, G. A review of deep learning-based information fusion techniques for multimodal medical image classification. Comput. Biol. Med. 2024, 177, 108635. [Google Scholar] [CrossRef]
  89. Hosseinpour Aghaei, B.; Taiefi Nasrabadi, N.; Pirali, Y.; Shojaei, S.S.R. The Prevalence of Fasciola (Digenea: Fasciolidae) Species in Livestock and Humans in Iran, A Systematic Review. Arch. Razi Inst. 2025, 80, 263–269. [Google Scholar]
  90. Ali, S.; Akhlaq, F.; Imran, A.S.; Kastrati, Z.; Daudpota, S.M.; Moosa, M. The enlightening role of explainable artificial intelligence in medical & healthcare domains: A systematic literature review. Comput. Biol. Med. 2023, 166, 107555. [Google Scholar] [CrossRef]
  91. Jacob, C.; Brasier, N.; Laurenzi, E.; Heuss, S.; Mougiakakou, S.G.; Cöltekin, A.; Peter, M.K. AI for IMPACTS Framework for Evaluating the Long-Term Real-World Impacts of AI-Powered Clinician Tools: Systematic Review and Narrative Synthesis. J. Med. Internet Res. 2025, 27, e67485. [Google Scholar] [CrossRef] [PubMed]
  92. Ong Ly, C.; Unnikrishnan, B.; Tadic, T.; Patel, T.; Duhamel, J.; Kandel, S.; Moayedi, Y.; Brudno, M.; Hope, A.; Ross, H.; et al. Shortcut learning in medical AI hinders generalization: Method for estimating AI model generalization without external data. NPJ Digit. Med. 2024, 7, 124. [Google Scholar] [CrossRef] [PubMed]
  93. Bertl, M.; Lamo, Y.; Leucker, M.; Margaria, T.; Mohammadi, E.; Mukhiya, S.K.; Pechmann, L.; Piho, G.; Rabbi, F. Challenges for AI in Healthcare Systems. In Bridging the Gap between AI and Reality, Proceedings of the International Conference on Bridging the Gap between AI and Reality, Crete, Greece, 23–28 October 2023; Springer Nature: Cham, Switzerland, 2023. [Google Scholar]
Jpm 15 00302 g001
Figure 1. PRISMA.
Figure 1. PRISMA.
Jpm 15 00302 g001
Jpm 15 00302 g002
Figure 2. Overall AUCs for (a) all-cause mortality [10,20,21,22,26,31,35,37,38,39,41,42,44,46,47,48,49,50,51,52,53,54,56,57,58,59] (b) new PPI/LBBB [19,27,28,33,34,37,40,45,46] (c) valve-related dysfunction [23,33,37,46] (d) MACE [24,32,36,43,55] (e) stroke/TIA [18,37,46] (f) HF-related hospitalization [22,25,30].
Figure 2. Overall AUCs for (a) all-cause mortality [10,20,21,22,26,31,35,37,38,39,41,42,44,46,47,48,49,50,51,52,53,54,56,57,58,59] (b) new PPI/LBBB [19,27,28,33,34,37,40,45,46] (c) valve-related dysfunction [23,33,37,46] (d) MACE [24,32,36,43,55] (e) stroke/TIA [18,37,46] (f) HF-related hospitalization [22,25,30].
Jpm 15 00302 g002
Jpm 15 00302 g003
Figure 3. Overall accuracy for (a) all-cause mortality [20,35,37,46,50,53] (b) new PPI/LBBB [19,27,29,37,45,46] (c) valve-related dysfunction [23,37,46].
Figure 3. Overall accuracy for (a) all-cause mortality [20,35,37,46,50,53] (b) new PPI/LBBB [19,27,29,37,45,46] (c) valve-related dysfunction [23,37,46].
Jpm 15 00302 g003
Jpm 15 00302 g004
Figure 4. Overall recall for (a) all-cause mortality [20,35,37,42,46,47,49,53,56] (b) new PPI/LBBB [19,33,37,40,45,46] (c) valve-related dysfunction [23,33,37,46].
Figure 4. Overall recall for (a) all-cause mortality [20,35,37,42,46,47,49,53,56] (b) new PPI/LBBB [19,33,37,40,45,46] (c) valve-related dysfunction [23,33,37,46].
Jpm 15 00302 g004
Table 1. Characteristics of included studies.
Table 1. Characteristics of included studies.
Author YearCountryStudy DesignOverall Dataset Size% of Male ParticipantsMean ± SD Age of ParticipantsOutcomesAlgorithmArchitectureModel DevelopmentValidation Method
Asif 2024 [37]USRetrospective cohort834N/AN/AValve-related dysfunctionMLPInitial architecture search performed using Hyperopt, a Python library built for automatic model selection and hyperparameter optimizationBaseline and clinical dataN/A
New PPI/LBBB
Stroke
All-cause mortality
Barbieri 2024 [39]AustriaPost hoc analysis of a retrospective study307952.19N/AAll-cause mortalityABC-AS scoreN/ABaseline and clinical data + biomarkersN/A
Bruggemann 2024 [41]SwitzerlandRetrospective analysis144952.42N/AAll-cause mortalityDNNCT images 3D deep neural networkImaging dataCross-validation
Erck 2024 [42]NetherlandsRetrospective cohort119947.080 ± 7All-cause mortalityAdjusted model (intermuscular adipose tissue with deep learning model)CT imagesBaseline and clinical data + imaging data + procedural dataN/A
Erdogan 2024 [43]TurkeyRetrospective cohort45340.876.1 ± 6.6MACEXGBoostN/ABaseline and clinical data + imaging data + biomarkersN/A
Ouahidi 2024 [19]FranceRetrospective cohort52051.884.3 ± 5.5New PPI/LBBBSVMN/ABaseline and clinical data + imaging data + procedural dataCross-validation
Shi 2024 [23]ChinaRetrospective cohort23455.674.34 ± 7.62Valve-related dysfunctionLASSON/ABaseline and clinical data + imaging dataCross-validation
Tremamunno 2024 [55]USARetrospective cohort64858.977 ± 9.3MACEcVAECT imagesImaging dataN/A
Yordanov 2024 [59]NetherlandsRetrospective cohort16,66150.879.6All-cause mortalityCentralN/ABaseline and clinical data + biomarkers + procedural dataCross-validation
Zahid 2024 [30]USRetrospective cohort92,363N/AN/AHF-related hospitalizationLRN/ABaseline and clinical dataN/A
Zisiopoulou 2024 [31]GermanyProspective cohort28451.7681.03 ± 4.75All-cause mortalityLRN/AN/AN/A
Abdelkhalek 2023 [33]CanadaRetrospective cohort13357.981.33 ± 7.49New PPI/LBBBMMLRCT imagesImaging dataN/A
Valve-related dysfunction
Agasthi 2023 [34]USRetrospective cohort65742.680.7 ± 8.2New PPI/LBBBGBMN/ABaseline and clinical data + imaging dataCross-validation
Alhwiti 2023 [35]USRetrospective cohort54,73953.979.65 ± 8.5All-cause mortalityGBMN/ABaseline and clinical dataCross-validation
Barrett 2023 [40]USRetrospective cohort606N/AN/ANew PPI/LBBBPRIMEN/ABaseline and clinical data + procedural dataN/A
Chen 2023 [56]UKRetrospective cohort45051.082.43 ± 5.21All-cause mortalityGBSTN/AN/ACross-validation
Kwiecinski 2023 [49].MultinationalRetrospective cohort82346.082 ± 5All-cause mortalityXGBoostN/ABaseline and clinical data + imaging data + biomarkers + procedural dataCross-validation
Leha 2023 [58]GermanyRetrospective cohort28,14746.881 ± 6.1All-cause mortalityRFN/AProcedural dataCross-validation
Pollari 2023 [21]GermanyRetrospective cohort62945.081.9 (53.8–94.5)All-cause mortalityBayesN/ABaseline and clinical data + imaging data + biomarkersCross-validation
Savitz 2023 [22]USRetrospective cohort156556.681 ± 8.2HF-related hospitalizationGBMN/ABaseline and clinical dataCross-validation
Stan 2023 [24]RomaniaRetrospective cohort33860.376 (71–80)MACEXGBoostN/ABaseline and clinical dataCross-validation
Theis 2023 [26]GermanyRetrospective cohort76051.081 ± 6All-cause mortalityCNNN/AImaging dataCross-validation
Aquino 2022 [36]USRetrospective cohort19643.975 ± 11MACECT-FFR with CCTAN/AImaging dataN/A
Bansal 2022 [38]USRetrospective cohort49960.778.8 ± 9.9All-cause mortalityRFN/ABaseline and clinical dataCross-validation
Evertz 2022 [44]GermanyProspective cohort14262.080 (74–83)All-cause mortalityFully automated assessment of the volumetric parametersCommercially available AI software provided by Neosoft (suiteHEART,)Imaging dataN/A
Lertsanguansinchai 2022 [50]ThailandRetrospective cohort17843.881.6 ± 8.3All-cause mortalityDTN/ABaseline and clinical data + imaging dataCross-validation
Sulaiman 2022 [25]USRetrospective cohort117,39854.879.5 ± 8.4HF-related hospitalizationLASSON/ABaseline and clinical data + procedural
data		Random split
Agasthi 2021 [10]USRetrospective cohort105558.280.9 ± 7.9All-cause mortalityGBMN/ABaseline and clinical dataCross-validation
Galli 2021 [45]MultinationalRetrospective cohort151N/AN/ANew PPI/LBBBK-nearest neighbors ML modelMulti-slice CTImaging data+ procedural dataCross-validation
Lopes 2021 [51]NetherlandsRetrospective cohort179155.66N/AAll-cause mortalityXGBoostN/ABaseline and clinical data + imaging data + biomarkersCross-validation
Mamprin 2021 (1) [57]NetherlandsInter center Cross-validation study193148.05N/AAll-cause mortalityCatBoostN/ABaseline and clinical data + imaging data + biomarkersCross-validation
Mamprin 2021 (2) [53].NetherlandsRetrospective analysis27052.080.7 ± 6.2All-cause mortalityCatBoostN/ABaseline and clinical data + imaging data + biomarkers + procedural dataCross-validation
Okuno 2021 [18]FranceProspective cohort227952.083.2 years (interquartile range [IQR] 79.4–86)MACEEncoder–Decoder NNN/AImaging dataRandom split
Penso 2021 [20]ItalyRetrospective cohort47136.381 ± 6All-cause mortalityMLPN/ABaseline and clinical data + imaging dataCross-validation
Truong 2021 [27]USRetrospective cohort55752.080 ± 9New PPI/LBBBRFN/ABaseline and clinical data + imaging dataRandom split
Gomes 2020 [46]GermanyRetrospective analysis451N/AN/AValve-related dysfunctionXGBoostN/ABaseline and clinical data + procedural dataCross-validation
All-cause mortality
Stroke/TIA
New PPI/LBBBSVM
Abdul Ghffar 2020 [54]USRetrospective cohort14350.079.39
(75.07, 84.36)		All-cause mortalityN/AN/ABaseline and clinical dataCross-validation
Tsushima 2020 [28]USRetrospective cohort888N/AN/ANew PPI/LBBBLRN/ABaseline and clinical data + procedural dataRandom split
Hernandez-Suarez 2019 [47]USRetrospective cohort10,88352.381 ± 8.5All-cause mortalityLRN/ABaseline and clinical data + procedural dataN/A
Hoffmann 2020 [48]GermanyProspective cohort12958.982.67 ± 5.25All-cause mortalityGradient-boosted trees (linear predictor score)N/ABaseline and clinical data + biomarkersN/A
Lopes 2019 [52]NetherlandsRetrospective analysis147845.082.43 ± 6.23All-cause mortalityRFN/ABaseline and clinical data + imaging data + biomarkersN/A
Vejpongsa 2018 [29]USRetrospective cohort18,400N/AN/ANew PPI/LBBBLRN/ABaseline and clinical data + procedural dataN/A
Zusman 2017 [32]IsraelRetrospective cohort43543.082.67 ± 5.21MACELRN/ABaseline and clinical data + imaging data + biomarkersCross-validation
Abbreviations: MLP—Multilayer Perceptron; DNN—Deep Neural Network; SVM—Support Vector Machine; LR—Logistic Regression; LASSO—Least Absolute Shrinkage and Selection Operator; GBM—Gradient Boosting Machine; XGBoost—Extreme Gradient Boosting; RF—Random Forest; DT—Decision Tree; CNN—Convolutional Neural Network; cVAE—Conditional Variational Autoencoder; CatBoost—Categorical Boosting; SHAP—SHapley Additive exPlanations; GBST—Gradient-Boosted Survival Trees; RFC—Random Forest Classifier; I2I—Image-to-Image network; PRIME—Predictive Risk Modeling Evaluation; Encoder–Decoder NN—Encoder–Decoder Neural Network; Bayes—Bayes Classifier; MMLR—Multinominal Mutilvariate Logistic Regression; MACE—Major Adverse Cardiac Event; HF-related hospitalization—Heart Failure-related hospitalization; New PPI—New Permanent Pacemaker Implantation; LBBB—Left Bundle Branch Block; N/A—Not Applicable.
Table 2. Outcomes.
Table 2. Outcomes.
Outcome CategoryAUCAccuracyRecall
Estimate (95% CI)Heterogeneity (I2)Estimate (95% CI)Heterogeneity (I2)Estimate (95% CI)Heterogeneity
(I2)
Clinical outcomesAll-Cause Mortality0.78
(0.74, 0.82)		98.1%0.81
(0.69, 0.89)		98.2%0.90
(0.70, 0.97)		99%
MACE0.79
(0.67, 0.92)		89.9%N/AN/AN/AN/A
Stroke/TIA0.73
(0.59, 0.88)		97.1%N/AN/AN/AN/A
Heart Failure-Related Hospitalization0.7
(0.60, 0.81)		83.3%N/AN/AN/AN/A
Procedural Pacemaker and Conduction Abnormalities0.75
(0.68, 0.82)		93.2%0.73
(0.59, 0.84)		99.1%0.87
(0.50, 0.98)		99.2%
Valve-Related Dysfunction0.73
(0.62, 0.84)		96%0.79
(0.57, 0.91)		98.7%0.54
(0.26, 0.80)		99%
Abbreviations: MACE—Major Adverse Cardiac Event; AUC—Area Under Curve; CI—Confidence Interval; TIA—Transient Ischemic Attack; N/A—Not Applicable.
Table 3. Subgroup analysis for outcomes.
Table 3. Subgroup analysis for outcomes.
Clinical OutcomesSubgroupPooled Estimate95% CIHeterogeneity (I2)Clinical OutcomesSubgroupPooled Estimate95% CIHeterogeneity (I2)
All-Cause Mortality (AUC)Baseline and Clinical Data0.77(0.69, 0.85)94.1%Pacemaker and Conduction Abnormalities (AUC)Baseline and Clinical Data0.61(0.56, 0.66)N/A
Imaging Data0.73(0.71, 0.74)0%Imaging Data0.75(0.67, 0.82)N/A
Procedural Data0.75(0.72, 0.78)N/ABaseline and Clinical Data + Imaging Data0.77(0.68, 0.85)90.1%
Baseline and Clinical Data + Imaging Data0.77(0.69, 0.84)48.7%Baseline and Clinical Data + Procedural Data0.75(0.62, 0.88)73%
Baseline and Clinical Data + Biomarkers0.91(0.88, 0.95)0%Imaging Data + Procedural Data0.84(0.71, 0.97)N/A
Baseline and Clinical Data + Procedural Data0.95(0.90, 1.00)88.3%Baseline and Clinical Data + Imaging Data + Procedural Data0.92(0.84, 1.00)N/A
Baseline and Clinical Data + Imaging Data + Biomarkers0.78(0.66, 0.89)17.4%Valve-Related DysfunctionImaging Data0.8(0.73, 0.87)N/A
Baseline and Clinical Data + Imaging Data + Procedural Data0.55(0.52, 0.58)N/ABaseline and Clinical Data + Imaging Data0.74(0.67, 0.80)N/A
Baseline and Clinical Data + Biomarkers + Procedural Data0.68(0.66, 0.70)N/ABaseline and Clinical Data + Procedural Data0.57(0.52, 0.62)N/A
Baseline and Clinical Data + Imaging Data + Biomarkers + Procedural Data0.78(0.70,0.87)95.1%Heart Failure-Related Hospitalization (AUC)Baseline and Clinical Data0.67(0.49, 0.86)91.2%
MACE (AUC)Imaging Data0.67(0.58, 0.76)0%Baseline and Clinical Data + Procedural Data0.74(0.70, 0.78)
Baseline and Clinical Data + Imaging Data + Biomarkers0.84(0.60, 1.00)91%
Baseline and Clinical Data0.92(0.85, 0.99)N/A
MACE—Major Adverse Cardiac Event; AUC—Area Under Curve; CI—Confidence Interval; TIA—Transient Ischemic Attack; N/A—Not Applicable.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Shojaei, S.; Mousavi, A.; Kazemian, S.; Armani, S.; Maleki, S.; Fallahtafti, P.; Arashlow, F.T.; Daryabari, Y.; Naderian, M.; Alkhouli, M.; et al. Artificial Intelligence in Risk Stratification and Outcome Prediction for Transcatheter Aortic Valve Replacement: A Systematic Review and Meta-Analysis. J. Pers. Med. 2025, 15, 302. https://doi.org/10.3390/jpm15070302

AMA Style

Shojaei S, Mousavi A, Kazemian S, Armani S, Maleki S, Fallahtafti P, Arashlow FT, Daryabari Y, Naderian M, Alkhouli M, et al. Artificial Intelligence in Risk Stratification and Outcome Prediction for Transcatheter Aortic Valve Replacement: A Systematic Review and Meta-Analysis. Journal of Personalized Medicine. 2025; 15(7):302. https://doi.org/10.3390/jpm15070302

Chicago/Turabian Style

Shojaei, Shayan, Asma Mousavi, Sina Kazemian, Shiva Armani, Saba Maleki, Parisa Fallahtafti, Farzin Tahmasbi Arashlow, Yasaman Daryabari, Mohammadreza Naderian, Mohamad Alkhouli, and et al. 2025. "Artificial Intelligence in Risk Stratification and Outcome Prediction for Transcatheter Aortic Valve Replacement: A Systematic Review and Meta-Analysis" Journal of Personalized Medicine 15, no. 7: 302. https://doi.org/10.3390/jpm15070302

APA Style

Shojaei, S., Mousavi, A., Kazemian, S., Armani, S., Maleki, S., Fallahtafti, P., Arashlow, F. T., Daryabari, Y., Naderian, M., Alkhouli, M., Rana, J. S., Mehrani, M., Jenab, Y., & Hosseini, K. (2025). Artificial Intelligence in Risk Stratification and Outcome Prediction for Transcatheter Aortic Valve Replacement: A Systematic Review and Meta-Analysis. Journal of Personalized Medicine, 15(7), 302. https://doi.org/10.3390/jpm15070302

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop