Interpretable Clustering in Healthcare
Interpretable clustering is crucial for understanding patient phenotypes, enabling the identification of clinically meaningful associations between patient profiles and adverse events. This method enhances patient safety and treatment efficacy by providing clear, actionable insights into complex patient data.
Patient Phenotyping Framework: The talk introduces an interpretable patient phenotyping framework using k-means clustering, statistical tests, visualizations, and interpretable machine learning. This approach helps identify distinct patient profiles (phenotypes) based on continuous demographic, medical history, and pre-operational variables.
Identification of High-Risk Clusters: The method identifies distinct clusters among patients. This finding highlights the potential of clustering to uncover clinically meaningful associations between patient phenotypes and adverse events.
Integration of SHAP Values: The innovation lies in integrating SHAP values with machine learning algorithms to interpret cluster memberships. This method balances model complexity with interpretability, providing detailed insights into the key features influencing patient outcomes, such as creatinine levels and annulus area.