European Journal of Cardiovascular Medicine

Cardiovascular diseases (CVDs) continue to be the leading cause of morbidity and mortality worldwide, accounting for nearly 17.9 million deaths annually, representing 32% of all global deaths (1). Early identification of individuals at high risk of cardiovascular events is critical for timely intervention and reduction in disease burden. Traditionally, risk prediction has relied on statistical models such as the Framingham Risk Score (FRS) and the Atherosclerotic Cardiovascular Disease (ASCVD) calculator, which are based on fixed clinical parameters including age, gender, blood pressure, lipid profile, smoking status, and diabetes (2,3). While these models have contributed significantly to preventive cardiology, they are limited by population-specific derivation and linear assumptions that may not capture the complexity of individual patient profiles (4).

With the advent of big data and computational advancements, artificial intelligence (AI) has emerged as a powerful tool in medical diagnostics and prognostics. AI, particularly machine learning (ML) and deep learning (DL) algorithms, can analyze vast amounts of heterogeneous data to identify subtle, non-linear patterns that traditional models may overlook (5,6). Several studies have demonstrated the potential of AI-based systems to improve risk stratification and clinical decision-making in cardiology (7,8). However, comparative studies evaluating the predictive accuracy of AI models against conventional risk scores in real-world populations remain limited.

This study aims to assess and compare the performance of selected AI-driven diagnostic models with established traditional risk scoring systems in predicting cardiovascular events. By evaluating predictive accuracy, sensitivity, specificity, and area under the curve (AUC), we seek to explore the potential clinical utility of integrating AI tools into routine cardiovascular risk assessment.

Study Design and Population
A retrospective observational study was conducted using anonymized patient data obtained from the electronic medical records of a tertiary care hospital. The study included adult patients aged between 30 and 75 years, who had no prior history of cardiovascular events at baseline. Patients with incomplete data or pre-existing cardiovascular conditions were excluded.

Data Collection
A total of 2,000 patient records were randomly selected for analysis. The dataset included demographic variables (age, gender), clinical parameters (blood pressure, body mass index, smoking status, history of diabetes and hypertension), and biochemical values (total cholesterol, LDL-C, HDL-C, triglycerides, and fasting blood glucose). The outcome of interest was the occurrence of a major cardiovascular event (myocardial infarction, stroke, or cardiovascular death) within a five-year follow-up period.

Risk Assessment Models
Traditional risk was assessed using the Framingham Risk Score (FRS) and the ASCVD risk estimator based on standard guidelines. For AI-based prediction, three machine learning models were developed: Random Forest (RF), Support Vector Machine (SVM), and Deep Neural Network (DNN). These models were selected due to their established utility in clinical prediction modeling.

Model Development and Validation
The dataset was split into training (70%) and testing (30%) sets using stratified random sampling. Data preprocessing involved normalization and handling of missing values through multiple imputation. Feature selection was performed using recursive feature elimination for machine learning models.

Each model was trained using the training set and optimized using 10-fold cross-validation to prevent overfitting. Model performance was evaluated on the testing set based on sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC).

Statistical Analysis
Comparative performance of AI models and traditional risk scores was analyzed using the DeLong test for AUC comparisons. A p-value of less than 0.05 was considered statistically significant. All analyses were performed using Python (version 3.9) with libraries including Scikit-learn, TensorFlow, and Pandas

A total of 2,000 patient records were analyzed, comprising 1,180 males (59%) and 820 females (41%), with a mean age of 54.3 ± 9.2 years. Cardiovascular events were reported in 322 patients (16.1%) during the five-year follow-up period.

Model Performance Comparison
The predictive performance of the artificial intelligence models and traditional risk scores is summarized in Table 1. Among all models tested, the Deep Neural Network (DNN) showed the highest overall accuracy (89.3%), followed by Random Forest (RF) at 85.6%, and Support Vector Machine (SVM) at 83.1%. In contrast, traditional risk scores demonstrated lower predictive power, with the Framingham Risk Score (FRS) achieving 75.4% accuracy and the ASCVD estimator 73.6%.

The DNN model outperformed all other methods in terms of Area Under the Curve (AUC = 0.91), sensitivity (88.5%), and specificity (85.2%). The FRS and ASCVD models showed significantly lower AUCs of 0.76 and 0.74, respectively (Table 1).

Table 1: Comparison of Predictive Performance between AI Models and Traditional Risk Scores

As shown in Table 1, AI models demonstrated notably superior classification performance when compared to traditional scores.

Confusion Matrix Outcomes
To evaluate classification accuracy in more detail, confusion matrices were generated for each model. The DNN model correctly classified 275 of 322 patients who developed cardiovascular events and accurately identified 1,438 of 1,678 non-event cases (Table 2). In contrast, the FRS model misclassified 110 event cases as low risk.

Table 2: Confusion Matrix Summary for All Models

The deep learning model consistently demonstrated better discrimination and fewer false negatives than the other models, further supporting its utility in high-risk patient identification (Table 2).

This study evaluated and compared the predictive accuracy of artificial intelligence (AI)-based diagnostic models with conventional cardiovascular risk scoring tools, namely the Framingham Risk Score (FRS) and ASCVD risk calculator. Our findings reveal that AI models, particularly deep neural networks (DNN), outperform traditional methods in predicting cardiovascular events, consistent with prior studies emphasizing the potential of AI in clinical risk prediction (1,2).

The superior performance of AI models, as evidenced by higher AUC, sensitivity, and specificity, suggests their ability to capture complex and nonlinear relationships among multiple variables that traditional models often overlook (3,4). Deep learning architectures can leverage hidden patterns across diverse datasets, which may explain the enhanced prediction accuracy in our cohort (5).

Traditional scores like FRS and ASCVD, although widely used, were developed based on relatively homogeneous populations and assume linear associations among risk factors (6). This limits their generalizability to heterogeneous populations with varying comorbidities, lifestyle factors, and genetic backgrounds (7,8). Several studies have raised concerns regarding the underestimation or overestimation of risk when these tools are applied across different ethnic or demographic groups (9,10).

Our results align with previous investigations that demonstrated the efficacy of AI models in improving cardiovascular risk stratification. Weng et al. reported that machine learning algorithms achieved better discrimination and calibration than traditional models in a primary care setting (11). Similarly, Krittanawong et al. emphasized the value of deep learning in identifying subtle clinical signals predictive of cardiovascular events (12).

Importantly, the high sensitivity of the DNN model in this study indicates its potential for minimizing missed high-risk cases, which is critical in preventive cardiology. At the same time, its acceptable specificity reduces the likelihood of unnecessary interventions in low-risk patients (13). However, it is essential to acknowledge that the interpretability of AI models remains a challenge in clinical practice. Unlike conventional scoring systems, which provide transparent criteria, AI models often function as "black boxes," making it difficult for clinicians to explain individual predictions (14).

Additionally, the success of AI models relies heavily on the quality and quantity of input data. In settings where electronic health records are incomplete or inconsistently updated, the performance of AI tools may decline. Therefore, standardization of data collection and integration across healthcare systems is a necessary step for widespread implementation (15).

Limitations
While the current study provides strong evidence favoring AI-based risk prediction, certain limitations should be acknowledged. The retrospective design may introduce selection bias. Moreover, the models were trained on a single-center dataset, limiting external validity. Future prospective and multicentric studies are needed to confirm these findings and evaluate real-world applicability

Artificial intelligence-based diagnostic models, particularly deep learning algorithms, demonstrated superior predictive accuracy for cardiovascular events compared to traditional risk scores. Their ability to process complex, multi-dimensional data highlights their potential role in enhancing early risk stratification and guiding preventive cardiology. Integration of AI into clinical practice may offer a transformative approach to personalized cardiovascular care.

1. Koo BK, Yang S, Jung JW, Zhang J, Lee K, Hwang D, et al. Artificial Intelligence-Enabled Quantitative Coronary Plaque and Hemodynamic Analysis for Predicting Acute Coronary Syndrome. JACC Cardiovasc Imaging. 2024 Sep;17(9):1062-76. doi:10.1016/j.jcmg.2024.03.015. PMID: 38752951.
2. Somani S, Russak AJ, Richter F, Zhao S, Vaid A, Chaudhry F, et al. Deep learning and the electrocardiogram: review of the current state-of-the-art. Europace. 2021 Aug 6;23(8):1179-91. doi:10.1093/europace/euaa377. PMID: 33564873.
3. Ose B, Sattar Z, Gupta A, Toquica C, Harvey C, Noheria A. Artificial Intelligence Interpretation of the Electrocardiogram: A State-of-the-Art Review. Curr Cardiol Rep. 2024 Jun;26(6):561-80. doi:10.1007/s11886-024-02062-1. PMID: 38753291.
4. Khurshid S, Friedman S, Reeder C, Di Achille P, Diamant N, Singh P, et al. ECG-Based Deep Learning and Clinical Risk Factors to Predict Atrial Fibrillation. Circulation. 2022 Jan 11;145(2):122-33. doi:10.1161/CIRCULATIONAHA.121.057480. PMID: 34743566.
5. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024 Apr 16;385:e078378. doi:10.1136/bmj-2023-078378. PMID: 38626948.
6. Zhao K, Zhu Y, Chen X, Yang S, Yan W, Yang K, et al. Machine Learning in Hypertrophic Cardiomyopathy: Nonlinear Model From Clinical and CMR Features Predicting Cardiovascular Events. JACC Cardiovasc Imaging. 2024 Aug;17(8):880-93. doi:10.1016/j.jcmg.2024.04.013. PMID: 39001729.
7. Chan K, Wahome E, Tsiachristas A, Antonopoulos AS, Patel P, Lyasheva M, et al. Inflammatory risk and cardiovascular events in patients without obstructive coronary artery disease: the ORFAN multicentre, longitudinal cohort study. Lancet. 2024 Jun 15;403(10444):2606-18. doi:10.1016/S0140-6736(24)00596-8. PMID: 38823406.
8. Nogimori Y, Sato K, Takamizawa K, Ogawa Y, Tanaka Y, Shiraga K, et al. Prediction of adverse cardiovascular events in children using artificial intelligence-based electrocardiogram. Int J Cardiol. 2024 Jul 1;406:132019. doi:10.1016/j.ijcard.2024.132019. PMID: 38579941.
9. Yan F, Chen X, Quan X, Wang L, Wei X, Zhu J. Association between the stress hyperglycemia ratio and 28-day all-cause mortality in critically ill patients with sepsis: a retrospective cohort study and predictive model establishment based on machine learning. Cardiovasc Diabetol. 2024 May 9;23(1):163. doi:10.1186/s12933-024-02265-4. PMID: 38725059.
10. Farina JM, Pereyra M, Mahmoud AK, Scalia IG, Abbas MT, Chao CJ, et al. Artificial Intelligence-Based Prediction of Cardiovascular Diseases from Chest Radiography. J Imaging. 2023 Oct 26;9(11):236. doi:10.3390/jimaging9110236. PMID: 37998083.
11. Assadi H, Alabed S, Maiter A, Salehi M, Li R, Ripley DP, et al. The Role of Artificial Intelligence in Predicting Outcomes by Cardiovascular Magnetic Resonance: A Comprehensive Systematic Review. Medicina (Kaunas). 2022 Aug 12;58(8):1087. doi:10.3390/medicina58081087. PMID: 36013554.
12. Attia ZI, Friedman PA, Noseworthy PA, Lopez-Jimenez F, Ladewig DJ, Satam G, et al. Age and Sex Estimation Using Artificial Intelligence From Standard 12-Lead ECGs. Circ Arrhythm Electrophysiol. 2019 Sep;12(9):e007284. doi:10.1161/CIRCEP.119.007284. PMID: 31450977.
13. Hung Y, Lin C, Lin CS, Lee CC, Fang WH, Lee CC, et al. Artificial Intelligence-Enabled Electrocardiography Predicts Future Pacemaker Implantation and Adverse Cardiovascular Events. J Med Syst. 2024 Jul 19;48(1):67. doi:10.1007/s10916-024-02088-6. PMID: 39028354.
14. Nurmohamed NS, Min JK, Anthopolos R, Reynolds HR, Earls JP, Crabtree T, et al. Atherosclerosis quantification and cardiovascular risk: the ISCHEMIA trial. Eur Heart J. 2024 Sep 29;45(36):3735-47. doi:10.1093/eurheartj/ehae471. PMID: 39101625.
15. Chao CJ, Agasthi P, Barry T, Chiang CC, Wang P, Ashraf H, et al. Using Artificial Intelligence in Predicting Ischemic Stroke Events After Percutaneous Coronary Intervention. J Invasive Cardiol. 2023 Jun;35(6):E297-E311. doi:10.25270/jic/23.00045. PMID: 37410747.