A Concept for Graph-Based Temporal Similarity of Patient Data

Area of Science:

• Medical Informatics
• Computational Biology
• Data Science

Background:

• Computer-based decision support systems (CDSS) are utilized for dedicated tasks like sepsis detection.
• Current CDSS for sepsis detection report suboptimal positive predictive values, often around 46%.

Purpose of the Study:

• To introduce a novel approach for sepsis detection using temporal electronic patient record data.
• To leverage case-based reasoning and graph theory for enhanced similarity measures in patient data analysis.

Main Methods:

• Organizing temporal patient data into a time-graph structure.
• Employing graph theory-based approaches to quantify similarity between patient cases.
• Utilizing the open MIMIC III dataset for the development and evaluation of the time-graph similarity framework.

Main Results:

• The developed time-graph similarity framework demonstrates a novel method for analyzing temporal patient data.
• The approach aims to improve upon the limited positive predictive values of existing sepsis detection systems.

Conclusions:

• The novel time-graph similarity approach shows promise for improving sepsis detection accuracy.
• Future work includes extending this concept to other diseases beyond sepsis.