What question did this study set out to answer?

To develop and validate machine learning models tailored for in-hospital mortality prediction in cardiac surgery patients while addressing severe class imbalance.

August 15, 2026DiagnosticsOpen Access

Optimizing In-Hospital Mortality Prediction After Cardiac Surgery: A Machine Learning Approach Using Feature Engineering for Imbalanced Data

View Full Paper

Ask AI

Bookmark

Authors

PKPo-Cheng KaoChung Shan Medical University Hospital CWChih-Cheng WuNational Chung Hsing University JYJung‐Chun YehChung Shan Medical University Hospital

Discussion

Member takes

Implication

Retrospective study reveals superior mortality prediction with a 17-variable neural network in cardiac surgery patients, highlighting improved postoperative risk stratification over standard tools.

Key Points

To develop and validate machine learning models tailored for in-hospital mortality prediction in cardiac surgery patients while addressing severe class imbalance.
Extracted 68 first-day ICU variables from 6,941 open-heart surgery cases in the MIMIC-IV database (76 deaths, 1.095% mortality rate).
Trained logistic regression, random forest, XGBoost, and multilayer perceptron models using SMOTE for class imbalance and stratified 10-fold cross-validation.
Developed a streamlined 17-variable model and benchmarked performance against established ICU risk scores (SOFA and OASIS).
Among 68-variable models, random forest achieved the highest AUROC at 0.915 (95% CI, 0.855–0.966), whereas the streamlined 17-variable multilayer perceptron reached an AUROC of 0.920 (95% CI, 0.865–0.964), outperforming SOFA (0.688) and OASIS (0.690).
The 17-variable multilayer perceptron achieved an AUCPR of 0.203 (95% CI, 0.060–0.389) compared with 0.161 for SOFA and 0.046 for OASIS.
SHAP analysis identified bicarbonate levels, mechanical ventilation, and mean pulmonary arterial pressure as the top predictive clinical features.

Cite This Study

Kao et al. (2026) studied this question.

synapsesocial.com/papers/6a8019fe75c2e31742c865d1 https://doi.org/10.3390/diagnostics16162557

View Full Paper

Ask AI

Bookmark

Implication

Retrospective study reveals superior mortality prediction with a 17-variable neural network in cardiac surgery patients, highlighting improved postoperative risk stratification over standard tools.