Comparing machine learning models and traditional approaches for predicting obstructive coronary artery disease: a systematic review and meta-analysis

Q: What are the key findings of this study?

Machine learning models demonstrated higher predictive discrimination for obstructive coronary artery disease compared to traditional approaches, with a mean difference in AUC of 0.09.

Q: What is the clinical evidence from this study?

Study design: Meta-Analysis. Population: Obstructive coronary artery disease (oCAD) (n=89143). Intervention: Machine learning models vs. Traditional approaches (e.g., Framingham Risk Score, Diamond-Forrester model, CAD consortium clinical score). Primary outcome: Comparative discrimination (Area Under the Curve [AUC]) between machine learning models and traditional approaches (MD 0.09, 95% CI 0.05-0.13, p=<0.00001).

Q: What does this research mean for the field?

Machine learning models achieve superior predictive discrimination for obstructive coronary artery disease compared to traditional clinical risk scores. Novelty: ClaimNovelty.SYNTHESIS. Consensus alignment: ConsensusAlignment.NEUTRAL.

Q: What question did this study set out to answer?

To compare the diagnostic performance of machine learning models with traditional risk scores for predicting obstructive coronary artery disease and evaluate differences across algorithm types.

Jingjing Guo; Chenggong Bao; Lihong Gong; Zhe Zhang

doi:10.3389/fcvm.2026.1892900

July 30, 2026Frontiers in Cardiovascular MedicineOpen Access

Comparing machine learning models and traditional approaches for predicting obstructive coronary artery disease: a systematic review and meta-analysis

View Full Paper

Ask AI

Bookmark

Share

Key result

Machine learning improves obstructive CAD prediction discrimination over traditional scores by ~0.09 AUC.

MD 0.09
95% CI 0.05-0.13
P<0.00001
n=89,143

Figures

Why the study?

Do machine learning models improve the prediction of obstructive coronary artery disease compared to traditional clinical risk scores in patients evaluated for CAD?

Key Points

To compare the diagnostic performance of machine learning models with traditional risk scores for predicting obstructive coronary artery disease and evaluate differences across algorithm types.
Systematic search of EMBASE, Web of Science, Cochrane Library, Scopus, and PubMed from database inception to February 9, 2026, identifying 41 eligible studies.
Random-effects meta-analysis and subgroup analyses pooled area under the receiver operating characteristic curve (AUC) values and calculated weighted mean differences (WMDs) against traditional approaches.
Risk of bias across 86 machine learning models was evaluated using the Prediction Model Risk of Bias Assessment Tool (PROBAST).
Pooled mean AUCs across test and external validation sets were 0.85 (95% CI: 0.75, 0.95) for CatBoost, 0.82 (95% CI: 0.78, 0.86) for XGBoost, 0.81 (95% CI: 0.76, 0.87) for RF, and 0.78 (95% CI: 0.72, 0.83) for LR.
Machine learning models outperformed traditional scores, with pooled WMDs in AUC of 0.13 (95% CI: 0.07–0.19) for clinical features, 0.06 (95% CI: −0.04 to 0.15) for biological signals, and 0.04 (95% CI: 0.01–0.07) for imaging-based predictors.
PROBAST assessment revealed an overall high risk of bias in 38.4% (33 of 86) of machine learning models, with analysis methods being of concern in 74.42% (64 models).

Study Design

Type

Meta-Analysis (n=89,143)

Structured PICO

Do machine learning models improve the prediction of obstructive coronary artery disease compared to traditional clinical risk scores in patients evaluated for CAD?

P

Population

89,143 patients from 12 studies were included in a meta-analysis comparing the diagnostic performance of machine learning models versus traditional risk scores for predicting obstructive coronary artery disease.

I

Intervention

Machine learning (ML) models (including Logistic Regression, Random Forest, XGBoost, Deep Learning, K-Nearest Neighbors, LightGBM, Multilayer Perceptron, Support Vector Machine, and CatBoost) for predicting oCAD.

C

Comparator

Traditional clinical risk scores and approaches (e.g., Diamond-Forrester score, CAD consortium clinical score, Framingham Risk Score, Duke clinical score, AHA Pre-Test Probability Model).

O

Outcome

Comparative discrimination between ML models and traditional approaches in predicting oCAD, measured by Area Under the Curve (AUC).surrogate

Main Result

Mean Difference: 0.09 (95% CI 0.05–0.13)

p-value: p=<0.00001

Machine learning models demonstrate superior discrimination compared to traditional clinical risk scores for predicting obstructive coronary artery disease, though current evidence is limited by high heterogeneity and methodological biases.

Limitations

Did not conduct a calibration analysis because most studies failed to report calibration metrics or used inappropriate methods.
Focused exclusively on predictive accuracy and did not explore risk factors associated with oCAD.
Selecting the highest AUC model from each study may introduce selection bias, potentially overestimating machine learning performance.
Substantial heterogeneity across studies, primarily attributed to variations in predictors and diagnostic criteria.
High risk of bias in many included models due to poor reporting of discrimination and calibration statistics or inappropriate handling of missing data.
Sample-size calculations were not performed and some studies had insufficient sample sizes
Inadequate handling of missing data (e.g., simple deletion)
Feature selection based solely on univariate regression
Weights of features used in ML models were not reported
High between-study heterogeneity

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

ML models may enhance obstructive CAD prediction over traditional scores; extends evidence but high heterogeneity leaves open need for standardized validation.

Abstract

Background: Obstructive coronary artery disease (oCAD) is a major cause of cardiovascular morbidity and mortality, and accurate prediction is essential for guiding clinical decision-making and secondary prevention. However, the potential of machine learning (ML) models for predicting oCAD remains insufficiently explored; therefore, this study aimed to compare the performance of ML methods with traditional approaches and evaluate differences among ML algorithm classes. Methods: A systematic search of EMBASE, Web of Science, Cochrane Library, Scopus, and PubMed was conducted from inception to February 9, 2026. Risk of bias was assessed using the Prediction Model Risk of Bias Assessment Tool. Random-effects meta-analysis was performed to pool AUCs of ML algorithms and compare the predictive performance of ML models with traditional approaches for oCAD. Meta-regression and subgroup analyses were used to explore heterogeneity. Heterogeneity in AUC was mainly associated with differences in diagnostic methods and predictors for oCAD. Among the algorithms evaluated, LR, RF, and XGBoost were the most commonly employed, suggesting their potential in clinical practice. Conclusions: ML shows promising discrimination relative to traditional scores. Future studies should emphasize adequate sample-size calculation, appropriate feature-selection strategies, and standardized handling of missing data and focus on developing and validating ML models based on accessible and noninvasive data sources.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Guo et al. (2026) conducted a meta-analysis in Obstructive coronary artery disease (oCAD) (n=89,143). Machine learning models vs. Traditional approaches (e.g., Framingham Risk Score, Diamond-Forrester model, CAD consortium clinical score) was evaluated on Comparative discrimination (Area Under the Curve [AUC]) between machine learning models and traditional approaches (MD 0.09, 95% CI 0.05-0.13, p=<0.00001). Machine learning models demonstrated higher predictive discrimination for obstructive coronary artery disease compared to traditional approaches, with a mean difference in AUC of 0.09.

synapsesocial.com/papers/6a81f1e59048143b53776ddf https://doi.org/10.3389/fcvm.2026.1892900

View Full Paper

Ask AI

Bookmark

Share

Comparing machine learning models and traditional approaches for predicting obstructive coronary artery disease: a systematic review and meta-analysis

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Comparing machine learning models and traditional approaches for predicting obstructive coronary artery disease: a systematic review and meta-analysis

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading