Supervised Nonnegative Matrix Factorization to Predict ICU Mortality Risk.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Critical Care Medicine

Background:

• Accurate ICU mortality risk prediction is crucial for timely clinical intervention but faces challenges in feature interpretability and prediction accuracy.
• Existing methods often lack interpretability, hindering clinical trust and application.
• Subgraph Augmented Nonnegative Matrix Factorization (SANMF) shows promise for interpretable time-series analysis.

Purpose of the Study:

• To develop an interpretable and accurate method for ICU mortality risk prediction.
• To enhance the predictive performance of SANMF by incorporating supervised learning.
• To address the limitations of unsupervised SANMF in clinical prediction tasks.

Main Methods:

• Proposed a supervised SANMF algorithm by integrating logistic regression loss into the NMF framework.
• Employed an alternating optimization procedure to solve the proposed model.
• Validated the method using simulation data before applying it to real-world ICU patient data.

Main Results:

• The supervised SANMF method demonstrated superior performance compared to conventional supervised NMF methods in ICU mortality risk prediction.
• The approach successfully balanced the trade-off between prediction accuracy and feature interpretability.
• Effectiveness was verified through simulation studies and application to ICU data.

Conclusions:

• The proposed supervised SANMF algorithm offers a robust and interpretable solution for ICU mortality risk prediction.
• This method enhances clinical decision-making by providing accurate and understandable risk assessments.
• The findings suggest a promising direction for applying advanced machine learning techniques in critical care settings.