Development of a Machine Learning-Based Predictive Model and Clinically Oriented Web Application for 30-Day Mortality
Telmo Miguel-Medina1, Susel Góngora Alonso1, Isabel de la Torre Díez1
1eHealth and Telemedicine Group (GTe), University of Valladolid, 47011 Valladolid, Spain.
Sensors (Basel, Switzerland)
|March 14, 2026
Summary
A new machine learning model accurately predicts 30-day mortality in cardiac surgery patients. A web application allows real-time risk assessment, aiding clinical decision-making.
Area of Science:
- Cardiovascular Surgery
- Medical Informatics
- Machine Learning
Background:
- Cardiac surgery carries significant 30-day mortality risks.
- Accurate preoperative risk assessment is crucial for patient management.
- Existing prediction models may lack real-time clinical integration.
Purpose of the Study:
- To develop and validate a machine learning (ML) model for predicting 30-day mortality in cardiac surgery.
- To create a clinician-oriented web application for real-time model implementation.
- To enhance preoperative risk assessment and clinical decision support.
Main Methods:
- Retrospective analysis of 325 cardiac surgery patients.
- Supervised machine learning, including XGBoost model training and cross-validation.
- Development of a StreamLit-based web application with SHAP explainability.
Main Results:
- XGBoost model achieved high performance: AUC-ROC of 0.968, recall of 0.800, Brier score of 0.058.
- Web application provides real-time mortality predictions with model transparency.
- Clinician feedback indicated the tool is intuitive and valuable for risk assessment.
Conclusions:
- A robust ML model integrated with a functional web application offers a practical tool for cardiac surgery decision-making.
- The combined approach improves accuracy and accessibility of risk prediction.
- Further multicentre validation and user-centered refinement are planned.