A Machine Learning Approach to Predict Mortality in Patients Enrolled in Cardiac Rehabilitation After Percutaneous Coronary Intervention for Acute Coronary Syndrome

Q: What is the clinical evidence from this study?

Study design: Cohort. Population: Acute coronary syndrome (STEMI, NSTEMI, or unstable angina) treated with percutaneous coronary intervention (n=1261). Intervention: LightGBM machine learning mortality prediction model vs. Logistic regression and other machine learning algorithms (Random Forest, SVM, Decision Tree, Fuzzy Decision Tree, Neural Network). Primary outcome: Prediction of all-cause mortality (Mean AUC 0.88 (Balanced accuracy 0.87, Sensitivity 0.86, Specificity 0.88)).

Q: What does this research mean for the field?

A LightGBM machine learning model predicts all-cause mortality in acute coronary syndrome patients post-PCI with high accuracy (mean AUC 0.88, balanced accuracy 0.87), identifying exercise capacity and daily sitting duration alongside clinical factors as key predictors. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Dimitrios A. Gatsios; Manuela Sestayo‐Fernández; Nikos Pappas; Ed P. de Kluiver; Arnoud W.J. van ’t Hof; Frank Christoph Mooren; Carlos Peña; Boris Schmitz; Boris Schmitz

doi:10.3390/ai7110450

November 2, 2026AIOpen Access

A Machine Learning Approach to Predict Mortality in Patients Enrolled in Cardiac Rehabilitation After Percutaneous Coronary Intervention for Acute Coronary Syndrome

View Full Paper

Ask AI

Bookmark

Share

Key result

LightGBM model predicts 1-year post-PCI mortality in ACS with ~0.88 AUC.

Mean AUC 0.88 (Balanced accuracy 0.87, Sensitivity 0.86, Specificity 0.88)
n=1,261

Why the study?

Patients with ACS treated with PCI face residual risk of adverse events and all-cause mortality, and traditional static risk assessment models are not designed to capture the complex, dynamic nature of post-intervention conditions.

Does a machine learning-based prediction model (LightGBM) accurately predict 1-year all-cause mortality compared to logistic regression in patients with acute coronary syndrome enrolled in cardiac rehabilitation after percutaneous coronary intervention?

Key Points

To develop and validate an explainable machine learning model predicting all-cause mortality in acute coronary syndrome patients participating in cardiac rehabilitation after percutaneous coronary intervention.
Analyzed clinical data from 1,261 coronary artery disease patients (mean age 61 ± 11.1 years, 16.6% women) who underwent percutaneous coronary intervention for acute coronary syndrome.
Trained and evaluated seven machine learning algorithms, validating the leading LightGBM model externally on the Eu-CARE dataset of 1,386 patients.
Applied model-agnostic SHapley Additive exPlanation (SHAP) methodology to determine individual feature importance and model explainability.
LightGBM demonstrated superior performance with a balanced accuracy of 0.87, sensitivity of 0.86, specificity of 0.88, and a mean AUC of 0.88 (std AUC 0.06).
External validation in a comparable cohort confirmed reliable true-positive detection with a sensitivity of 0.78 and a specificity of 0.73.
The strongest predictive impact came from daily sitting duration of 9 hours or more, physical exercise capacity below 6 METs, diabetes, statin non-use, and elevated serum creatinine.

Study Design

Type

Cohort (n=1,261)

Multicenter

Yes

Structured PICO

Does a machine learning-based prediction model (LightGBM) accurately predict 1-year all-cause mortality compared to logistic regression in patients with acute coronary syndrome enrolled in cardiac rehabilitation after percutaneous coronary intervention?

P

Population

1,261 patients (mean age 61 years, 16.6% female) with acute coronary syndrome treated with PCI and enrolled in cardiac rehabilitation, followed for a mean of 12.0 months (externally validated in 1,386 European patients).

E

Exposure

Machine learning risk prediction models (specifically LightGBM, Random Forest, Support Vector Machine, Decision Tree, Fuzzy Decision Tree, and Neural Network) utilizing clinical, demographic, laboratory, medication, lifestyle, and cardiopulmonary exercise testing (CPET) data to predict all-cause mortality, explained via SHapley Additive exPlanation (SHAP) methodology.

C

Comparator

Traditional logistic regression model (and comparison among the seven evaluated machine learning algorithms).

O

Outcome

All-cause mortality at approximately 1-year follow-up (mean 12.0 ± 20.9 months in the training cohort and 11.2 ± 3.7 months in the validation cohort), documented via central electronic health records and local clinic databases.hard clinical

Main Result

Effect estimate: Mean AUC 0.88 (Balanced accuracy 0.87, Sensitivity 0.86, Specificity 0.88)

Limitations

Potential risk of bias from handling missing data
Medium-sized training dataset with a low number of mortality events (28 deaths, 2.2%), creating a potential overfitting risk
Critical variables (number of affected coronary vessels and IPAQ physical activity data) were unavailable in the external Eu-CARE validation dataset
Trained model is not currently available as an online or mobile user-friendly calculator
Low events per variable (EPV) with only 28 deaths (2.2%) in the training set and a relatively large number of predictors, creating a potential overfitting risk.
Trained model is not available as an online or mobile user-friendly calculator.
Calibration intercept deviates moderately from 0, and calibration slope deviates moderately from 1.
Reduced specificity (0.73 in PCI cohort, 0.60 in broader CAD cohort) during external validation.

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

An explainable LightGBM machine learning model incorporating functional capacity (peak METs), lifestyle factors (sitting duration), and clinical data accurately predicts 1-year all-cause mortality in post-PCI ACS patients enrolled in cardiac rehabilitation, outperforming traditional logistic regression.

Abstract

Background and aims: Patients with acute coronary syndrome (ACS) treated with percutaneous coronary intervention (PCI) have a residual risk of adverse events and all-cause mortality. Traditional static risk assessment models are not designed to incorporate the complex and dynamic nature of post-intervention conditions which may be overcome by machine learning (ML) approaches. This study aimed to develop and validate a model for the prediction of all-cause mortality of ACS patients post-PCI using advanced ML. Methods and Results: Data from 1,261 CAD patients (mean age 61 ± 11.1 years, 16.6 % women) who underwent PCI during an episode of ACS was analysed using ML techniques. External validation was performed using the Eu-CARE dataset of 1,386 CAD patients and predictions were explained using the model-agnostic SHapley Additive exPlanation (SHAP) methodology. Three models, LightGBM, Random Forest and SVM, outperformed logistic regression in terms of accuracy (mean AUC > 0.8). LightGBM was identified as the most potent prediction model in terms of balanced accuracy (0.87), with a sensitivity of 0.86 and a specificity of 0.88 indicating strong predictive power. External validation confirmed reliable detection of true positives, although specificity dropped to 0.7. The highest SHAP values for impact on the model were detected for the variables “diabetes”, “physical exercise capacity (METs achieved during stress test)”, “average daily sitting duration”, “statin use”, “history of family heart disease”, “number of affected vessels”, “revascularization history”, “serum creatinine levels”, and “chronic hepatopathy”. The model was rated as “recommended” using the independent validation screening tool. Conclusion: We present a ML-based model with high predictive capacity to identify CAD patients post PCI with higher mortality risk. The model could serve in clinical practice as a decision support tool to identify high-risk patients who could benefit from intensified personalized treatment and lifestyle interventions.

Discussion

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

Cite This Study

Gatsios et al. (2026) conducted a cohort in Acute coronary syndrome (STEMI, NSTEMI, or unstable angina) treated with percutaneous coronary intervention (n=1,261). LightGBM machine learning mortality prediction model vs. Logistic regression and other machine learning algorithms (Random Forest, SVM, Decision Tree, Fuzzy Decision Tree, Neural Network) was evaluated on Prediction of all-cause mortality (Mean AUC 0.88 (Balanced accuracy 0.87, Sensitivity 0.86, Specificity 0.88)). The LightGBM machine learning model predicted one-year all-cause mortality in ACS patients post-PCI with a mean AUC of 0.88, balanced accuracy of 0.87, sensitivity of 0.86, and specificity of 0.88.

synapsesocial.com/papers/6aebc8ba3e63ca1b6c315de9 https://doi.org/10.3390/ai7110450

View Full Paper

Ask AI

Bookmark

Share

A Machine Learning Approach to Predict Mortality in Patients Enrolled in Cardiac Rehabilitation After Percutaneous Coronary Intervention for Acute Coronary Syndrome

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

A Machine Learning Approach to Predict Mortality in Patients Enrolled in Cardiac Rehabilitation After Percutaneous Coronary Intervention for Acute Coronary Syndrome

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading