The integrated multiple event representation framework (IMERF): a case study on critically-ill patients.

Area of Science:

• Computational biology
• Medical informatics
• Clinical data science

Background:

• Predicting multiple clinical events is complex.
• Existing models often struggle with overlapping risks.
• Understanding patient risk profiles is crucial for effective treatment.

Purpose of the Study:

• Introduce the Integrated Multiple Event Representation Framework (IMERF).
• Develop a novel approach for multiple clinical event risk prediction.
• Enable simultaneous prediction and characterization of overlapping events.

Main Methods:

• Utilized a two-stage process: multi-task learning and dimensionality reduction.
• Implemented convolutional neural networks in Stage 1 for superior performance.
• Employed generative topographic mapping (GTM) for data visualization and clustering in Stage 2.

Main Results:

• IMERF successfully modeled nine adverse events in intensive care unit (ICU) patients.
• Identified clear patterns of adverse event risk clusters using GTM.
• Highlighted distinct patient risk profiles and macro-clusters of event risk levels.

Conclusions:

• IMERF offers an interpretable framework for understanding complex, overlapping clinical event patterns.
• This advancement facilitates simultaneous prediction and characterization of multiple events.
• Potential for broader clinical applications beyond ICUs.