Development and validation of an interpretable machine learning model with multicenter routine tests to predict major adverse cardiac and cerebrovascular events after cardiac surgery

Q: What is the clinical evidence from this study?

Study design: Cohort. Population: Patients undergoing cardiac surgery with cardiopulmonary bypass (CPB) (n=27847). Intervention: Explainable Boosting Machine (EBM) predictive model based on baseline and perioperative inflammation-coagulation routine laboratory parameters vs. Logistic Regression, eXtreme Gradient Boosting (XGBoost), and Random Forest models. Primary outcome: In-hospital major adverse cardiovascular and cerebrovascular events (MACCE: composite of death, nonfatal stroke, nonfatal myocardial infarction, or refractory heart failure requiring ECMO or IABP) (AUC 0.789, 95% CI 0.724–0.847).

Q: What does this research mean for the field?

An Explainable Boosting Machine model utilizing routine perioperative inflammation and coagulation parameters effectively predicts major adverse cardiovascular and cerebrovascular events following cardiac surgery under cardiopulmonary bypass. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Jia Tan; Hao Yang; Yishun Wang; Wei Xinguang; Chen Chen; Kang Du; Gaofeng Meng; Yan Liu; Chengbin Zhou; Lei Du

doi:10.1186/s12872-026-06741-5

November 3, 2026BMC Cardiovascular DisordersOpen Access

Development and validation of an interpretable machine learning model with multicenter routine tests to predict major adverse cardiac and cerebrovascular events after cardiac surgery

View Full Paper

Ask AI

Bookmark

Share

Key result

Explainable Boosting Machine predicts post-cardiac surgery MACCE with an AUC of ~0.79.

AUC 0.789
95% CI 0.724–0.847
n=27,847

Why the study?

Systemic inflammatory response and coagulation disorder triggered by cardiac surgery under cardiopulmonary bypass may increase the risk of postoperative MACCE, motivating the development and validation of an interpretable machine learning model using routine laboratory data from the inflammation-coagulation axis.

Does an Explainable Boosting Machine (EBM) model incorporating routine perioperative blood counts and coagulation parameters accurately predict in-hospital major adverse cardiovascular and cerebrovascular events (MACCE) in patients undergoing cardiac surgery with cardiopulmonary bypass?

Key Points

To develop and validate an interpretable machine learning model utilizing routine perioperative inflammation and coagulation laboratory data to predict major adverse cardiac and cerebrovascular events after cardiac surgery under cardiopulmonary bypass.
Multicenter retrospective cohort study of 27,847 patients undergoing cardiac surgery with cardiopulmonary bypass across four tertiary cardiac centers in China from 2019 to 2023 (26,687 in the training cohort and 1,160 in the validation cohort).
Candidate predictors included baseline characteristics, perioperative blood counts, and routine coagulation parameters, with multiple machine learning algorithms trained and the Explainable Boosting Machine (EBM) selected for interpretability.
The overall incidence of MACCE was 12.7% (13.0% in the training cohort and 5.3% in the validation cohort).
The EBM model achieved the highest discriminative performance in the validation cohort with an AUC of 0.789 (95% CI, 0.724–0.847), accuracy of 0.802 (95% CI, 0.779–0.825), sensitivity of 0.613 (95% CI, 0.492–0.729), and specificity of 0.812 (95% CI, 0.790–0.836).
Top model predictors included postoperative lymphocyte percentage, preoperative thrombin time, postoperative international normalized ratio, lymphocyte count, red blood cell distribution width–standard deviation, fibrinogen, prothrombin time, and plateletcrit.

Study Design

Type

Cohort (n=27,847)

Multicenter

Yes

Structured PICO

Does an Explainable Boosting Machine (EBM) model incorporating routine perioperative blood counts and coagulation parameters accurately predict in-hospital major adverse cardiovascular and cerebrovascular events (MACCE) in patients undergoing cardiac surgery with cardiopulmonary bypass?

P

Population

27,847 patients undergoing on-pump cardiac surgery across four tertiary centers in China (26,687 training, 1,160 validation), followed until hospital discharge.

E

Exposure

Explainable Boosting Machine (EBM) machine learning model using 3 baseline variables, 18 complete blood count (CBC) parameters, and 6 coagulation parameters (both preoperative and postoperative measurements).

C

Comparator

Other machine learning models (Random Forest [RF], eXtreme Gradient Boosting [XGBoost], Logistic Regression [LR]) and perioperative clinical characteristics alone.

O

Outcome

In-hospital major adverse cardiovascular and cerebrovascular events (MACCE), a composite of death, nonfatal stroke, nonfatal myocardial infarction, or refractory heart failure requiring extracorporeal membrane oxygenation (ECMO) or intra-aortic balloon pump (IABP) support from cardiac surgery to hospital discharge.composite

Main Result

Effect estimate: AUC 0.789 (95% CI 0.724–0.847)

An interpretable Explainable Boosting Machine model utilizing routine perioperative inflammation and coagulation laboratory tests effectively predicts in-hospital MACCE after cardiopulmonary bypass cardiac surgery (AUC 0.789), offering a feasible early postoperative risk-stratification tool.

Limitations

Retrospective design is susceptible to residual confounding, selection bias, heterogeneity in blood sampling times, and incomplete clinical data collection.
More direct inflammatory markers (e.g., C-reactive protein, interleukin-6) were not included due to data unavailability.
Validation cohort was a within-system geographical holdout rather than rigorous external validation across independent healthcare systems.
MACCE was driven primarily by postoperative myocardial infarction rather than harder clinical endpoints such as death, stroke, or mechanical circulatory support.
Lower MACCE incidence in the validation cohort (5.3% vs 13.0%) and low positive predictive value (0.156) limit direct treatment decision-making without recalibration.
Retrospective design susceptible to residual confounding, selection bias, heterogeneity in blood sampling times, and incomplete clinical data collection
More direct inflammatory markers (e.g., C-reactive protein, interleukin-6) were not included due to data unavailability
Validation cohort was a within-system geographical holdout sharing common healthcare system management frameworks rather than a rigorous external validation
MACCE was driven primarily by postoperative myocardial infarction, limiting interpretation for harder clinical endpoints like death, stroke, or mechanical circulatory support
Lower MACCE incidence in the validation cohort (5.3%) compared to the training cohort (13.0%) may affect transportability and require recalibration, and the positive predictive value of the model remained low (0.156)

Ask about this paper

Answers are grounded in this paper and cite their sources.

Implication

May support postoperative MACCE risk stratification with routine labs; leaves open prospective validation before clinical adoption.

Abstract

Systemic inflammatory response and coagulation disorder triggered by cardiac surgery under cardiopulmonary bypass (CPB) may increase the risk of major adverse cardiovascular and cerebrovascular events (MACCE). This study aims to develop and validate an interpretable machine learning model based on the inflammation–coagulation axis using routine laboratory data to predict postoperative MACCE. This multicenter retrospective cohort study included patients undergoing CPB cardiac surgery at four tertiary cardiac centers in China between 2019 and 2023. Predictors included patients’ baseline characteristics, perioperative blood counts and routine coagulation parameters. Multiple machine learning models were developed and evaluated, with the Explainable Boosting Machine (EBM) selected for global and local interpretability analyses and sensitivity analyses. A total of 27,847 patients were included, with 26,687 assigned to the training set and 1,160 to the validation set. The EBM model achieved the best performance among all models (AUC 0.789; 95% CI, 0.724–0.847). Top important predictors included postoperative lymphocyte percentage, preoperative thrombin time, postoperative international normalized ratio, lymphocyte count, red blood cell distribution width–standard deviation, fibrinogen, prothrombin time, and plateletcrit. EBM effectively predicts MACCE after cardiac surgery and identifies key biomarkers related to the inflammation–coagulation axis as an early postoperative risk-stratification tool. Further validation in larger multicenter prospective studies is warranted.

Discussion

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

Cite This Study

Tan et al. (2026) conducted a cohort in Patients undergoing cardiac surgery with cardiopulmonary bypass (CPB) (n=27,847). Explainable Boosting Machine (EBM) predictive model based on baseline and perioperative inflammation-coagulation routine laboratory parameters vs. Logistic Regression, eXtreme Gradient Boosting (XGBoost), and Random Forest models was evaluated on In-hospital major adverse cardiovascular and cerebrovascular events (MACCE: composite of death, nonfatal stroke, nonfatal myocardial infarction, or refractory heart failure requiring ECMO or IABP) (AUC 0.789, 95% CI 0.724–0.847). The Explainable Boosting Machine model using routine inflammation-coagulation biomarkers predicted in-hospital MACCE after on-pump cardiac surgery with an AUC of 0.789 (95% CI 0.724–0.847).

synapsesocial.com/papers/6aebc8cf3e63ca1b6c315f00 https://doi.org/10.1186/s12872-026-06741-5

View Full Paper

Ask AI

Bookmark

Share

Development and validation of an interpretable machine learning model with multicenter routine tests to predict major adverse cardiac and cerebrovascular events after cardiac surgery

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Development and validation of an interpretable machine learning model with multicenter routine tests to predict major adverse cardiac and cerebrovascular events after cardiac surgery

Key Points

Study Design

Structured PICO

Main Result

Limitations

Ask about this paper

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading