Development and validation of the Sriya Expert Index Plus (SXI++) framework for heart disease prediction and risk reduction through clinical decision support

Introduction

Background

The Sriya Expert Index (SXI) and its enhanced variant Sriya Expert Index Plus (SXI++) represent a novel, multi-step analytic framework that combines statistical methods with machine learning (ML) and deep neural network (DNN) techniques to create a single, interpretable score for heart disease prediction. Conceptually, SXI functions as a dynamic, weighted average of the most important features, serving as a “Super Feature” or a Heart Disease Risk Score assigned individually to each patient. This single index simplifies complex, high-dimensional clinical data into a unidimensional representation, offering a strong positive correlation (R=0.98) with actual heart disease outcomes. Unlike traditional ML or DNN models, which often function as “black boxes”, SXI and SXI++ prioritize explainability, enabling clinicians to see the top contributing features and their respective weights for every prediction. This correlation is further visualized through linear or non-linear regression curves, demonstrating that a reduction in SXI score correlates with a reduction in heart disease risk. Moreover, decision trees built from current and target SXI scores offer informative and actionable insights, identifying which features drive higher or lower risk and guiding medical professionals on targeted interventions. This framework enhances both individual-level prediction and population-level risk stratification, marking a significant step toward explainable, data-driven cardiovascular care.

Heart disease remains a significant global health challenge, emphasizing the need for accurate and early prediction models. The 2017 Global Burden of Disease Study highlighted that heart diseases contribute to approximately 43% of all fatalities (1,2). In high-income countries, key risk factors for heart disease include poor dietary habits, tobacco use, excessive sugar intake, and obesity or high body fat (3,4). Meanwhile, low- and middle-income countries (LMICs) are also experiencing a rising prevalence of chronic diseases (5). Between 2010 and 2015, the global economic cost of heart diseases was projected to be around USD 3.7 trillion (6,7).

Diagnostic tools such as electrocardiogram (ECG) and computed tomography (CT) scans, critical for identifying coronary heart disease, are often prohibitively expensive and inaccessible for many people, particularly in low- and middle-income regions. This lack of affordability and availability has contributed to the loss of 17 million lives (5). The SXI++ model offers a potential solution to this disparity by utilizing readily available clinical data—such as age, blood pressure, cholesterol levels, and lifestyle factors—that can be obtained without the need for advanced diagnostic equipment. This approach enables early risk assessment and intervention in resource-constrained settings, potentially improving outcomes and reducing mortality associated with coronary heart disease in LMICs. Moreover, cardiovascular diseases (CVDs) account for 25–30% of companies’ annual healthcare expenses for their employees (8). Early detection of heart disease is therefore critical to reducing its physical and financial toll on individuals and organizations. According to World Health Organization (WHO) projections, the global death toll from heart diseases is expected to rise to 23.6 million by 2030, with heart disease and stroke being the primary contributors (9). To mitigate this impact, leveraging ML techniques to predict heart disease risk is vital for saving lives and reducing societal costs.

In the healthcare sector, vast amounts of data are generated daily, and ML techniques can uncover hidden patterns useful for clinical diagnoses (10). The importance of ML in medicine has been demonstrated by research conducted over the past decades. Key factors such as diabetes, hypertension, high cholesterol, and abnormal pulse rates are essential considerations for predicting heart disease (11). However, incomplete medical datasets often pose challenges, affecting the accuracy of predictions.

ML has emerged as a transformative tool in healthcare, enabling the diagnosis, detection, and prediction of various diseases. In recent years, there has been a growing interest in applying ML methods to assess disease risks. While existing studies have utilized these techniques to predict disease outcomes, many have struggled to achieve precise results in forecasting disease progression (12).

Rationale and knowledge gap

Heart disease remains a significant global health challenge, emphasizing the need for accurate and early prediction models. Common diagnostic methods, such as ECG, exercise stress testing, and coronary angiography, often have limited sensitivity (50–75% for ECG and stress testing) and struggle to integrate diverse clinical data, reducing their effectiveness in routine screening (12-15). ML offers a promising alternative, capable of analysing heterogeneous datasets and uncovering hidden patterns for heart disease prediction. Challenges remain in addressing the variability of clinical datasets, the imbalanced nature of heart disease-related data, and the lack of standardized feature engineering approaches. While the Cleveland Heart Disease dataset has been widely utilized in ML research, most existing studies focus on applying traditional algorithms without addressing core limitations such as dataset imbalance, static feature interpretation, and lack of actionable outputs. The SXI++ model was developed to overcome these barriers by introducing a dynamic, composite scoring system that integrates multiple ML algorithms and a proprietary DNN for iterative calibration. This allows for robust prediction even in heterogeneous datasets, while offering interpretable decision tree pathways that translate predictive scores into meaningful clinical recommendations. By doing so, SXI++ bridges the gap between high-performance predictions and real-world applicability, positioning it as a valuable decision support tool in heart disease diagnosis and intervention.

A study demonstrated the effectiveness of ML for CVD prediction using the Cleveland Heart Disease dataset, which comprised 303 instances and 17 attributes. Various classification methods were compared, including naive Bayes, decision trees, random forests, and k-nearest neighbours (KNN). The findings highlighted KNN as the most effective model, achieving an accuracy of 90.8% (16).

Another study employed ML to identify critical risk factors for CVD among patients with metabolic-associated fatty liver disease (MAFLD). Using blood biochemical analyses and subclinical atherosclerosis assessments from 191 patients, logistic regression, univariate feature ranking, and principal component analysis (PCA) were applied. The study revealed hypercholesterolemia, plaque scores, and diabetes duration as significant risk factors. The ML model successfully identified 85.11% of high-risk and 79.17% of low-risk patients, with an area under the curve (AUC) of 0.87 (2).

Future research must prioritize creating adaptable ML frameworks capable of generalizing across heterogeneous datasets and patient demographics. Addressing these knowledge gaps will elevate the role of ML in heart disease management, improving patient outcomes and reducing healthcare burdens globally. By incorporating advanced techniques like dynamic scoring systems and integrating neural networks, algorithms such as SXI++ can create models that can achieve higher accuracy, enhanced adaptability, and actionable insights.

Objective

Previous studies have primarily utilized ML techniques to address the challenges of early heart disease diagnosis, especially in identifying high-risk patients within clinical settings. However, these studies often faced difficulties in interpreting and explaining the influence of various clinical and physiological factors on heart disease predictions, assessing improvements in early detection across short-, mid-, and long-term periods, or implementing a comprehensive model capable of predicting heart disease without extensive prior training.

To address these gaps, this study is dedicated to the development and validation of an advanced predictive framework for heart disease diagnosis by focusing on: (I) evaluating the performance of the SXI++ algorithm as a multivariate scoring system for predicting heart disease as a binary classification problem; (II) enhancing the SXI++ scoring methodology through the integration of a proprietary DNN algorithm and correlating it with heart disease prediction rates across different timeframes (immediate, mid-term, and long-term); and (III) employing a targeted decision tree framework to interpret the most effective pathways leading to early and advanced heart disease diagnoses. This framework identifies critical clinical factors such as serum cholesterol levels and exercise-induced angina, providing actionable insights for prevention and treatment strategies. By establishing similarities in predictive pathways for various forms of heart disease, the study aims to develop a unified, interpretable model that not only improves personalized treatment strategies but also advances long-term patient outcomes.

Methods

Methodology overview

Two datasets were used in this study. One was the Cleveland Heart Disease dataset, and the other was the Heart Failure Clinical Records dataset, both widely recognized benchmarks in cardiovascular research, hosted on the UC Irvine Machine Learning Repository. These datasets were chosen for their structured format and relevance to heart disease diagnosis, facilitating comparison with existing studies. The Cleveland dataset comprises 303 records and 14 features, and the Heart Disease Clinical dataset includes 299 records with 13 features. The datasets include key clinical parameters such as age, serum cholesterol, chest pain type, ST depression, and major vessel counts. The target variable in the Cleveland dataset, originally values 1, 2, 3, and 4 denoting ≥50% narrowing in one, two, three, or four major coronary arteries respectively, was binarized into absence of heart disease [0] and presence of heart disease [1–4] for this analysis. The target variable in the Heart Disease Clinical dataset is the death event, a Boolean variable (true or false). While both datasets are small and have relatively homogeneous patient populations, which limits the generalizability of the findings, the data and results serve as a reliable starting point for initial model development and evaluation. The SXI++ framework achieved high performance metrics on this dataset, including an accuracy of more than 98% and a precision greater than 97%. However, we acknowledge that these results may not fully extend to broader, more diverse clinical populations. To address this limitation, future research will prioritize validation on larger, heterogeneous datasets to evaluate the model’s robustness and applicability across varied patient demographics.

To ensure data quality, missing values were handled using mean imputation for numerical variables and mode imputation for categorical variables. The dataset was divided into training (70%), validation (20%), and test (10%) sets, with the test set consisting of 60 records from each dataset (20 positive cases of heart disease and 40 negative cases). The SXI++ algorithm was trained on the training set, and its performance was assessed on both the validation and test sets using standard metrics: accuracy, precision [positive predictive value (PPV)], recall [negative predictive value (NPV)], and AUC.

The SXI++ algorithm employs a multivariate scoring system designed to enhance predictive accuracy by dynamically adjusting feature weights through iterative calibration and integrating a DNN. This approach combines composite scores from multiple ML algorithms, enabling the model to capture complex, non-linear relationships between clinical features and heart disease outcomes. Unlike conventional methods, SXI++ addresses challenges such as dataset variability and feature interdependencies, improving both precision and adaptability in its predictions.

To improve interpretability, a decision tree framework built using Random Forest models, was applied to identify the most influential clinical factors driving heart disease outcomes. This framework produces clear, interpretable pathways that highlight critical thresholds, such as elevated ST depression levels or increased major vessel counts, associated with higher risk. These pathways offer clinicians evidence-based guidance for identifying high-risk patients and tailoring interventions, effectively translating predictive outputs into practical clinical strategies. The analysis pinpointed factors like chest pain type, ST depression, and major vessel counts, which align with established cardiovascular risk indicators in the literature. While these findings suggest potential utility for early intervention, their clinical relevance in diverse settings requires further validation through collaboration with medical practitioners and testing on real-world patient data.

Dataset description

The heart disease diagnosis dataset integrates multiple sources, including Cleveland, Hungarian, Switzerland, and VA Long Beach, with two datasets chosen for this study due to their extensive use in previous heart disease research. Both datasets are sourced from the UC Irvine Machine Learning Repository and include 13 independent variables (e.g., age, sex, chest pain type, serum cholesterol, fasting blood sugar, maximum heart rate, serum creatinine, etc.), selected for their established clinical relevance in predicting coronary artery disease, as demonstrated in prior studies (13,16). We note here that despite the similarity of the datasets in terms of number of records and features, the datasets have completely different features and are reported to have been collected independently of each other. These attributes capture key demographic, clinical, and physiological risk factors, making them suitable for both expert and non-expert audiences when contextualized. The Cleveland dataset consists of 303 records, each containing 14 features (Table 1), including the target variable. The target variable is represented as an integer value ranging from 0 to 4, which helps classify the presence or absence of heart disease. Specifically, values 1, 2, 3, and 4 indicate the presence of heart disease, while a value of 0 signifies its absence. This classification is based on angiographic disease status, where a value of 0 corresponds to less than 50% diameter narrowing of coronary arteries, and a value of 1 or higher indicates a narrowing of 50% or greater. Key attributes include Ca, representing the number of major vessels colored by fluoroscopy, and Thal, denoting thallium stress test results categorized as normal [0], fixed defect [1], or reversible defect [2], with missing values imputed using the mean and mode respectively. The distribution of categorical features across diagnosed and non-diagnosed groups is detailed in (Table 2), while the statistical summary of numerical features such as age, cholesterol, and maximum heart rate is presented in (Table 3). A similar analysis of the second dataset is omitted for brevity.

SXI++ framework

The SXI++ framework is a dynamic and highly interpretable ML-based predictive model developed by Sriya.AI for diagnosing heart disease. It enhances traditional risk assessment by integrating a multistage analytical pipeline that starts with preprocessing and normalization, followed by bivariate correlation analysis to assess feature interdependence. The core innovation lies in its iterative scoring mechanism, which combines multiple ML algorithms [such as XGBoost, least absolute shrinkage and selection operator (LASSO), PCA, and Naïve Bayes] with a DNN. This hybrid system dynamically adjusts feature weights based on their impact, using a custom kernel initializer that emphasizes the most influential features. Additionally, the SXI++ framework incorporates decision tree models to generate interpretable diagnostic pathways, enabling clinicians to understand the reasoning behind predictions. The model demonstrated high accuracy (98.33%) and precision (97.5%) on the Cleveland Heart Disease dataset and showed a strong correlation (R²=0.98) between SXI scores and heart disease presence. Through iterative weight calibrations and Bayesian optimization, it fine-tunes its predictions to offer actionable insights, reduce misdiagnoses, and support early intervention strategies, thus presenting a powerful tool for improving cardiovascular care (Figure 1).

Figure 1 SXI++ framework, score + deep neural network algorithm + correlate. This figure illustrates the SXI++ framework, which integrates a dynamic scoring system with a deep neural network to optimize heart disease prediction. By correlating critical features with outcomes, the framework iteratively improves accuracy and provides actionable insights. The “Improve” component emphasizes enhanced predictions, reduced heart disease prevalence, and targeted clinical interventions. SXI, Sriya Expert Index; w.r.t., with respect to.

For more detailed information on the working of the SXI++ framework, we refer the readers to (15,17-20).

Statistical analysis

The performance of the SXI++ model was assessed using multiple statistical evaluation metrics, including accuracy, precision (PPV), recall (sensitivity), specificity, and the AUC. Accuracy was calculated as the ratio of correctly predicted observations to the total observations, while precision measured the proportion of true positive predictions among all positive predictions. Recall represented the ability of the model to correctly identify all positive cases, and specificity reflected the proportion of true negatives correctly identified. The AUC score provided an aggregate measure of performance across all classification thresholds. Performance metrics were calculated for both validation and test sets, and 95% confidence intervals for accuracy were estimated using the Wilson score method to assess the robustness of the results.

Ethical statement

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Since the research involved secondary analysis of anonymized data and did not include direct interaction with or intervention in human subjects, an ethics board review was not required. The study did not involve sensitive or personal data that would necessitate informed consent.

Results

Detailed results for the Cleveland dataset follow. Summary results for the second dataset are provided at the end of the section. The SXI benchmark stands at 2.033 (see Figure 2), serving as the threshold score for evaluating heart disease risk. Among the overall population, 45.87% of patients are diagnosed with heart disease, while 54.13% do not have heart disease. Based on the benchmark SXI score, 80% of patients diagnosed with heart disease fall above 2.033, indicating a strong correlation between higher SXI scores and heart disease diagnosis. Conversely, 14.56% of patients with heart disease have SXI scores below 2.033, showing that lower SXI scores are less predictive of heart disease. This distribution highlights a significant propensity to categorize patients diagnosed with heart disease towards above SXI or SXI >2.033 and no heart disease patients as below SXI. Additionally, from the data, we can see that of asymptomatic patients, about 72% have heart disease, of those with atypical angina, 18% have heart disease, with non-anginal pain, it is about 21%, and of those with typical angina, about 30% have heart disease. Further, from the patients in the dataset, around 25% of females and 55% of males have heart disease. Certain features such as heart disease, chest pain, angina, patient sex, heart rate, and blocked coronary vessels have a low P value (<0.0005) indicating that they are significantly associated with the target variable.

Figure 2 SXI score distribution. SXI, Sriya Expert Index.

The SXI++ algorithm demonstrated improved performance in classifying heart disease diagnoses, achieving outstanding metrics across both the test and validation sets. In the test set (Figure 3A), with 60 records, the model correctly classified 20 patients with heart disease and 39 patients without heart disease. Only one patient without heart disease was misclassified as having heart disease. Table 4 shows an impressive accuracy of 98.33%, a precision or PPV of 97.50%, recall or NPV of 100% and an AUC of 0.989–1.0 reflecting the model’s high accuracy and precision in predicting heart disease diagnoses.

Figure 3 Confusion matrices. The algorithm performs well even on data it has not seen before. (A) Confusion matrix for the test set. (B) Confusion matrix for the validation set.

Similarly, in the validation set (Figure 3B) of 30 records, the SXI++ algorithm maintained its strong performance. It correctly classified 10 patients with heart disease and 19 patients without heart disease, with only one patient without heart disease misclassified as having heart disease. Table 4 shows an accuracy of 96.67%, a precision or PPV of 95%, recall or NPV of 100% and an AUC of 0.975 (Figure 4), showcasing the robustness of the SXI++ algorithm. These results highlight the algorithm’s effectiveness in accurately predicting heart disease diagnoses, establishing the SXI++ algorithm as a reliable tool for applications where precise and consistent classification is critical.

Figure 4 ROC graph: the test & validation sets. AUC, area under the curve; ROC, receiver operating characteristic.

Tables 5-7 provide a comparative summary of key demographic, clinical, and outcome variables across the training (n=209), validation (n=30), and test (n=60) sets used in the heart disease survival prediction study. The mean age across groups ranges from 59.1 to 62.0 years, with a male predominance in all sets (63–72%). Important clinical predictors such as follow-up time, serum creatinine, and sodium levels are consistently distributed, though slight variations exist. Lifestyle and lab-related variables such as smoking status, platelet count, and creatinine phosphokinase levels are also reported. Notably, the event rate for death is higher in the validation (43.3%) and test (40.0%) sets compared to the training set (28.2%), indicating potential class imbalance. This comparison helps evaluate the representativeness and balance of the data subsets, which is essential for model training and generalization.

The current decision tree (Figure 5) illustrates the existing diagnostic landscape for heart disease, offering insights into the clinical features associated with current outcomes. At a current SXI score of 2.03, 45.87% of patients are diagnosed with heart disease. Key pathways leading to this diagnosis include patients exhibiting low levels of ST depression during exercise relative to rest (exercise_ST_depression ≤0.5), a lower count of major vessels visualized via fluoroscopy (major_vessels_count ≤0.5), and being female (sex_male ≤0.5). Additional contributing factors include reduced maximum heart rate and thalassemia patterns not indicating a reversible defect. In contrast, patients classified as not having heart disease tend to have higher ST depression levels, greater major vessel counts, and thalassemia profiles suggesting normal blood flow. These decision paths clarify the important diagnostic features currently influencing heart disease classification and reflect the model’s ability to identify key clinical variables tied to disease prevalence.

Figure 5 Current decision tree.

The target decision tree (Figure 6) introduces recommendations aimed at reducing the prevalence of heart disease to 36.70%, representing a 20% reduction from the current rate. According to this decision tree, patients are more likely to be diagnosed with heart disease if ST depression levels induced by exercise relative to rest (exercise_ST_depression) are >1, if the condition involves exercise-induced angina, if chest pain type is not typical angina, if major vessel counts are >0.4 (i.e., one or more vessels). In contrast, patients without heart disease are characterized by having ST depression levels (exercise_ST_depression) of <1, thalassemia (thal) indicating blood flow that is not a reversible defect, a maximum heart rate achieved of ≥174.6 and being of female gender. These paths provide actionable insights to guide clinical interventions aimed at reducing the likelihood of heart disease diagnoses.

Figure 6 Target decision tree.

Among the 303 patients in the Cleveland Heart Disease dataset, 139 were diagnosed with heart disease, while 164 were not, indicating a prevalence of 45.87% for heart disease and 54.13% for non-heart disease cases. The comparison between the current and target decision trees, conducted as part of the analytical process by the research team, illustrates the potential for reducing this prevalence and identifies strategies to achieve this goal. The current decision tree, generated using Random Forest models, maps clinical factors—such as asymptomatic chest pain, ST depression >0.5, and major vessel counts >0.4—contributing to the observed 45.87% prevalence. In contrast, the target decision tree, developed through SXI++ optimization, proposes adjustments to these factors, achieving a modelled 20% reduction in diagnoses (to 36.70%). This comparison, performed post-model training, evaluates how modifying variables like exercise-induced angina, ST depression levels, and age-related risks could lower disease risk. The resulting recommendations provide clinicians with a scientifically grounded framework to prioritize targeted interventions, emphasizing the value of incremental improvements for patient-specific care.

The results of the SXI++ algorithm on the second dataset are similar, with an accuracy of 98.3%, a precision of 99%, and an AUC of 0.989.

Discussions

Key findings

The SXI++ algorithm demonstrated outstanding performance in predicting heart disease, offering significant advancements over traditional diagnostic approaches. The model achieved an accuracy of over 98%, a precision (PPV) greater than 97%, a recall (NPV) of 100%, and an AUC score of more than 0.98. These metrics highlight the model’s improved ability to correctly identify patients with heart disease while ensuring that no at-risk patients are overlooked.

The study revealed a strong positive correlation between the SXI score and heart disease diagnoses, with a correlation coefficient of 0.98 and an R² value of 98. This indicates that as the SXI score increases, the likelihood of heart disease diagnosis rises proportionally. These findings highlight the capacity of the SXI++ algorithm to effectively predict and mitigate heart disease risk.

The decision tree analysis highlighted critical pathways for heart disease and non-heart disease outcomes. With an SXI score of 2.03, 45.87% of patients were diagnosed with heart disease, influenced by factors like asymptomatic chest pain, exercise-induced ST depression ≥1, major vessel counts ≥1, and male gender. Conversely, non-heart disease outcomes were linked to non-asymptomatic chest pain, normal thalassemia, serum cholesterol ≤330 mg/dL, and major vessel counts <1.

These findings underscore the capability of SXI++ to refine diagnostic pathways and reduce the prevalence of heart disease through targeted interventions. The decision trees not only validate the model’s predictions but also offer practical, interpretable insights for clinical decision-making, paving the way for improved patient outcomes.

Strengths and limitations

Many existing ML models for heart disease prediction demonstrate strong initial performance but suffer from limitations in adaptability and clinical interpretability. Prior studies (e.g., references 2 and 13) employed algorithms like KNN, logistic regression, and PCA, showing reasonable accuracy (AUC of 0.87, 90.8% accuracy with KNN). However, these models often rely on static feature weights and lack mechanisms for recalibrating predictions across diverse or imbalanced datasets. They also struggle to generalize beyond the training environment and frequently act as black-box systems, offering limited transparency or actionable insights to clinicians. These shortcomings hinder their integration into real-world healthcare settings, where interpretability and adaptability to varied patient populations are essential for clinical decision-making.

The SXI++ framework was developed to address these gaps by incorporating a dynamic scoring system, iterative calibration, and a DNN that collectively enhance predictive performance. Unlike conventional models, SXI++ assigns evolving feature weights based on data patterns and integrates decision tree pathways for interpretability. It demonstrated high accuracy (above 98%) and strong correlation with disease presence (R²=0.98) across two independent datasets. These strengths enable precise, personalized predictions while supporting clinician understanding of risk pathways. Nonetheless, SXI++ has its limitations. The retrospective design and modest dataset sizes limit the generalizability of results. Moreover, external validation in diverse populations and real-time clinical environments is needed to confirm the model’s robustness. Future research should prioritize prospective trials and integration into electronic medical records (EMRs) to evaluate its real-world applicability.

Comparison with similar research

Previous studies have explored the application of ML to identify CVD risk factors in specific patient populations. For instance, a study on patients with MAFLD used blood biochemical analyses and subclinical atherosclerosis assessments to identify critical risk factors such as hypercholesterolemia, plaque scores, and diabetes duration. Using logistic regression, univariate feature ranking, and PCA, the ML model achieved an AUC of 0.87, correctly classifying 85.11% of high-risk and 79.17% of low-risk patients (2).

However, these approaches faced challenges like traditional methods, including handling imbalanced datasets, variability in clinical parameters, and the absence of standardized feature engineering practices. Unlike such models, the SXI++ framework in this study offers a more comprehensive approach by integrating diverse clinical parameters and leveraging its dynamic scoring mechanism. It provides actionable insights through decision tree pathways, enhancing interpretability and utility in clinical practice. This positions SXI++ as a robust and scalable alternative, addressing the limitations of earlier models and enabling more accurate and clinically relevant predictions.

Explanation of findings

The improved performance of the SXI++ algorithm in predicting heart disease stems from its scoring mechanism, which integrates multiple ML techniques into a dynamic and iterative framework. Central to this approach is a DNN designed to optimize feature weighting by iteratively analysing patterns in clinical data. This network adjusts the influence of features based on their relevance to heart disease outcomes, enabling the model to capture complex, non-linear relationships among variables such as chest pain type, ST depression levels, and major vessel counts. By balancing the contributions of these features, the algorithm distinguishes between heart disease and non-heart disease cases with high accuracy, providing a robust and adaptable solution for clinical prediction.

The strong correlation between SXI scores and heart disease diagnoses (R²=98) further validates the reliability of the model. As the SXI score increases, the proportional rise in heart disease likelihood underscores the model's ability to quantify risk accurately. The observed reductions in heart disease diagnoses with incremental decreases in the SXI score (e.g., a 2.6% decrease leading to a 20% reduction in cases) highlight the model’s potential to drive meaningful clinical improvements. This demonstrates the SXI++ algorithm’s capacity to not only predict but also guide interventions to reduce the prevalence of heart disease.

Decision tree analysis adds another layer of interpretability to the SXI++ framework by identifying key pathways that influence outcomes. Factors such as exercise-induced ST depression, asymptomatic chest pain, and major vessel counts were shown to contribute significantly to heart disease diagnoses, while lower ST depression levels, normal thalassemia, and higher maximum heart rates were protective. These insights enable healthcare providers to prioritize interventions for high-risk patients, emphasizing the model’s practical utility in clinical decision-making.

The ability of the SXI++ algorithm to combine predictive accuracy with actionable insights positions it as a transformative tool for improving cardiovascular care. Its focus on refining diagnostic pathways and reducing heart disease prevalence demonstrates its potential to enhance patient outcomes while optimizing resource allocation in healthcare settings.

Implications, recommendations, and user interaction

Implications

The SXI++ algorithm shows promise for improving cardiovascular care by supporting accurate heart disease prediction, as demonstrated on the Cleveland Heart Disease dataset. While further validation on diverse datasets is needed, its ability to integrate clinical features suggests potential for broader application in clinical settings, which could enhance patient outcomes and optimize resource allocation pending confirmation in varied environments.

Recommendations

To enhance clinical adoption, future research should:

• Validate SXI++ algorithm on external datasets, including global and outpatient populations.
• Integrate SXI++ algorithm into EMRs for real-time monitoring and prediction.
• Conduct prospective trials to evaluate its impact on patient outcomes, such as reducing the incidence of heart disease-related events.

User interaction and expertise

The SXI++ framework requires minimal user interaction, automating data preprocessing and feature selection. Users need only provide structured input data, such as vital signs and lab results, and define the target outcome. While no expertise in ML is necessary, familiarity with clinical data and predictive analytics can enhance its effective use. An intuitive interface tailored for clinicians would further streamline its adoption.

Future investigation

To enhance the SXI++ algorithm’s impact on heart disease prediction, future research should focus on integrating it into an interoperable platform that connects with existing EMR systems using Health Level Seven International Fast Healthcare Interoperability Resources (HL7 FHIR) standards for real-time risk stratification (21-23). Prospective, randomized trials in diverse healthcare settings are planned to validate the model’s performance across varied patient populations. Additionally, studies could explore SXI++’s application in predicting the effectiveness of lifestyle modifications or pharmacological interventions for high-risk patients. Further investigations will prioritize external validation and real-time testing to ensure the model’s robustness and clinical utility, positioning SXI++ as a cornerstone of precision medicine in cardiovascular care.

Conclusions

This study evaluated the SXI++ framework for heart disease prediction using two datasets (303 patients and 199 patients), achieving a high accuracy of greater than 98%, a precision higher than 97.5%, a recall of 100%, and an AUC exceeding 0.98. The model’s interpretable decision trees identified key risk factors, such as asymptomatic chest pain, ST depression >0.5, and major vessel counts >0.5, providing actionable clinical insights for targeted interventions. A strong correlation (R²=0.98) between SXI scores and heart disease diagnoses suggests potential for effective risk stratification. Additionally, the model’s ability to model a 20% reduction in heart disease prevalence (from 45.87% to 36.70%) by optimizing factors like exercise-induced angina highlights its promise for guiding preventive strategies. Given the study’s reliance on a small, single-source dataset without external validation, further research, including prospective trials and testing on diverse populations, is essential to confirm SXI++’s generalizability and clinical utility. These findings underscore the potential of SXI++ to support cardiovascular care by enhancing early diagnosis and personalized treatment, pending broader validation to ensure its effectiveness across varied clinical settings.

Acknowledgments

None.

Footnote

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-25-47/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-25-47/coif). R.K., P.Y., and S.K. are employed by Sriya.AI LLC. M.B. is employed by Clarkson University and is a consultant at Sriya.AI. Sriya.AL LLC, which has filed US provisional patents on the underlying core technology and its applications in the healthcare industry. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Since the research involved secondary analysis of anonymized data and did not include direct interaction with or intervention in human subjects, an ethics board review was not required. The study did not involve sensitive or personal data that would necessitate informed consent.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Estes C, Anstee QM, Arias-Loste MT, et al. Modeling NAFLD disease burden in China, France, Germany, Italy, Japan, Spain, United Kingdom, and United States for the period 2016-2030. J Hepatol 2018;69:896-904. [Crossref] [PubMed]
2. Drożdż K, Nabrdalik K, Kwiendacz H, et al. Risk factors for cardiovascular disease in patients with metabolic-associated fatty liver disease: a machine learning approach. Cardiovasc Diabetol 2022;21:240. [Crossref] [PubMed]
3. Murthy HSN, Meenakshi M. Dimensionality reduction using neuro-genetic approach for early prediction of coronary heart disease. In: Proceedings of the International Conference on Circuits, Communication, Control and Computing. Bangalore, India; 21–22 November. IEEE; 2014:329-32.
4. Benjamin EJ, Muntner P, Alonso A, et al. Heart Disease and Stroke Statistics-2019 Update: A Report From the American Heart Association. Circulation 2019;139:e56-528. [Crossref] [PubMed]
5. Shorewala V. Early detection of coronary heart disease using ensemble techniques. Inform. Med. Unlocked 2021;26:100655.
6. Mozaffarian D, Benjamin EJ, Go AS, et al. Heart disease and stroke statistics--2015 update: a report from the American Heart Association. Circulation 2015;131:e29-322. [Crossref] [PubMed]
7. Maiga J, Hungilo GG, Pranowo. Comparison of Machine Learning Models in Prediction of Cardiovascular Disease Using Health Record Data. In: Proceedings of the 2019 International Conference on Informatics, Multimedia, Cyber and Information System (ICIMCIS). Jakarta, Indonesia; 24–25 October. IEEE; 2019:45-8.
8. Li J, Loerbroks A, Bosma H, et al. Work stress and cardiovascular disease: a life course perspective. J Occup Health 2016;58:216-9. [Crossref] [PubMed]
9. Purushottam; Saxena K, Sharma R. Efficient Heart Disease Prediction System. Procedia Comput Sci 2016;85:962-9.
10. Soni J, Ansari U, Sharma D, et al. Predictive Data Mining for Medical Diagnosis: An Overview of Heart Disease Prediction. Int J Comput Appl 2011;17:43-8.
11. Mohan S, Thirumalai, C, Srivastava, G. Effective Heart Disease Prediction Using Hybrid Machine Learning Techniques. IEEE Access 2019;7:81542-54.
12. Waigi R, Choudhary S, Fulzele P, et al. Predicting the risk of heart disease using advanced machine learning approach. European Journal of Molecular & Clinical Medicine 2020;7:1638-45.
13. Detrano R, Janosi A, Steinbrunn W, et al. International application of a new probability algorithm for the diagnosis of coronary artery disease. Am J Cardiol 1989;64:304-10. [Crossref] [PubMed]
14. Chicco D, Jurman G. Machine learning can predict survival of patients with heart failure from serum creatinine and ejection fraction alone. BMC Med Inform Decis Mak 2020;20:16. [Crossref] [PubMed]
15. Mahto D, Yadav P, Joseph AT, et al. AI2-SXI algorithm enables predicting and reducing the risk of less than 30 days patient readmissions with 99% accuracy and precision. J Med Artif Intell 2025;8:10.
16. Shah D, Patel S, Bharti SK, et al. Heart Disease Prediction using Machine Learning Techniques. SN Comput Sci 2020;1:345.
17. Kilambi S. AI Square Enabled by Sriya Expert Index (SXI): Method of Determining and Use. U.S. Provisional Patent Application No. 63/549,553, 4 Feb 2024.
18. Kilambi S. AI Square Enabled by Sriya Expert Index (SXI) Plus Rein-forcement Learning. U.S. Provisional Patent Application No. 63/553,335, 14 Feb 2024.
19. Kilambi S. Processing of Large Numerical Models (LNM) by AI2 enabled SXI. U.S. Provisional Patent Application No. 63/575,991, 8 Apr 2024.
20. Kilambi S. AI Square Enabled by Sriya Expert Index (SXI): Generative AI. U.S. Provisional Patent Application No. 63/554,252, 16 Feb 2024.
21. Hemann BA, Bimson WF, Taylor AJ. The Framingham Risk Score: an appraisal of its benefits and limitations. Am Heart Hosp J 2007;5:91-6. [Crossref] [PubMed]
22. Walinjkar A, Woods J. Personalized wearable systems for real-time ECG classification and healthcare interoperability: Real-time ECG classification and FHIR interoperability. 2017 Internet Technologies and Applications (ITA); Wrexham, UK. IEEE; 2017:9-14.
23. Yang X, Miao Y. Distributed Agent Based Interoperable Virtual EMR System for Healthcare System Integration. J Med Syst 2011;35:309-19. [Crossref] [PubMed]