[Paper Review] Deep Learning Approach for Predicting 30 Day Readmissions after Coronary Artery Bypass Graft Surgery

Ramesh B. Manyam, Yanqing Zhang+4 more|arXiv (Cornell University)|Dec 3, 2018

Cardiac, Anesthesia and Surgical OutcomesMedicine17 references3 citations

TL;DR

This study proposes an ensemble machine learning model combining Cox Proportional Hazards (CPH) regression and DeepSurv, a deep learning survival model, to predict 30-day all-cause readmissions after isolated coronary artery bypass graft (CABG) surgery using pre-discharge perioperative data. The model achieved a c-statistic of 0.712 on validation data after 25,000 training iterations, demonstrating improved predictive accuracy over traditional methods by leveraging time-varying clinical variables.

ABSTRACT

Hospital Readmissions within 30 days after discharge following Coronary Artery Bypass Graft (CABG) Surgery are substantial contributors to healthcare costs. Many predictive models were developed to identify risk factors for readmissions. However, majority of the existing models use statistical analysis techniques with data available at discharge. We propose an ensembled model to predict CABG readmissions using pre-discharge perioperative data and machine learning survival analysis techniques. Firstly, we applied fifty one potential readmission risk variables to Cox Proportional Hazard (CPH) survival regression univariate analysis. Fourteen of them turned out to be significant (with p value < 0.05), contributing to readmissions. Subsequently, we applied these 14 predictors to multivariate CPH model and Deep Learning Neural Network (NN) representation of the CPH model, DeepSurv. We validated this new ensembled model with 453 isolated adult CABG cases. Nine of the fourteen perioperative risk variables were identified as the most significant with Hazard Ratios (HR) of greater than 1.0. The concordance index metrics for CPH, DeepSurv, and ensembled models were then evaluated with training and validation datasets. Our ensembled model yielded promising results in terms of c-statistics, as we raised the the number of iterations and data set sizes. 30 day all-cause readmissions among isolated CABG patients can be predicted more effectively with perioperative pre-discharge data, using machine learning survival analysis techniques. Prediction accuracy levels could be improved further with deep learning algorithms.

Motivation & Objective

• To identify significant pre-discharge perioperative risk factors for 30-day all-cause readmissions after isolated CABG surgery.
• To improve prediction accuracy of readmission risk beyond traditional statistical models by incorporating time-varying clinical data.
• To develop an ensemble model integrating multivariate CPH and deep learning survival analysis (DeepSurv) for enhanced predictive performance.
• To support clinical decision-making by enabling early risk stratification and potential bedside application via a smart-app.

Proposed method

• Conducted univariate CPH analysis on 51 potential risk variables to identify 14 significant predictors (p < 0.05).
• Applied multivariate CPH and DeepSurv models to the 14 significant predictors using a de-identified dataset from the STS Adult Cardiac Surgery Database (2014–2017).
• Used the Lifelines and DeepSurv Python packages to implement and train the deep learning survival model with adaptive feature learning.
• Employed a 20% validation split of 2,293 isolated CABG cases, applying case deletion and conditional mean imputation for missing data.
• Evaluated model performance using the concordance index (c-statistic) across training and validation sets with increasing training iterations.
• Constructed an ensemble model by combining predictions from CPH and DeepSurv to improve generalization and predictive accuracy.

Experimental results

Research questions

• RQ1 Which pre-discharge perioperative variables are most predictive of 30-day all-cause readmission after isolated CABG surgery?
• RQ2 Can an ensemble model combining CPH regression and DeepSurv outperform individual models in predicting post-CABG readmissions?
• RQ3 How does the inclusion of time-varying clinical data (e.g., lab values, vitals, medications) improve readmission prediction accuracy compared to static discharge-time data?
• RQ4 To what extent does increasing training iterations and data volume enhance the c-statistic of the ensemble model?
• RQ5 Can the model support clinical decision-making by generating a real-time Risk of Readmission (RoR) score for post-CABG patients?

Key findings

• Fourteen perioperative variables were identified as significant predictors of 30-day readmission through univariate CPH analysis (p < 0.05), including gender, preoperative creatinine, and length of stay.
• Nine of the 14 significant predictors had hazard ratios (HR) > 1.0 in the multivariate CPH model, indicating increased risk of readmission.
• The ensemble model achieved a c-statistic of 0.712 on the validation set after 25,000 training epochs, outperforming both standalone CPH (0.631) and DeepSurv (0.571) models.
• The c-statistic for the ensemble model increased with more training iterations, indicating improved generalization and predictive stability.
• The model demonstrated improved performance with larger data volumes and higher iteration counts, suggesting scalability and potential for further accuracy gains.
• Survival curves and 95% confidence intervals confirmed the clinical significance of key predictors such as gender and prior myocardial infarction.