Description:

Quantifying Clinical Relationships and Patient Care Activities to Predict Patient Outcomes: An Edge-Weighted Multilayer Network Approach

Multilayer networks (MLNs)—networks which contain multiple types of relationships, data types, or other additional features of complexity and connectivity—are particularly useful for understanding interconnected complex systems represented by heterogeneous data. A MLN structure can be developed that describes dynamic and varying types of connections between different types of data, which more accurately mirrors the interactions among the real-life system's components than simple networks. Patient care activity logs, produced by clinician interaction with the electronic health record (EHR), is an exemplar of an interconnected complex system described by high dimensional heterogeneous data. This has inspired the development of an MLN model which analyzes patient care activity logs, with the aim of evaluating clinical processes by identifying activities in treatment when differences in clinical relationships are systematically predictive of patient outcomes. In this talk I will outline how an MLN framework was applied to hospital activity logs to, first, identify significant clinician relationships using risk-adjusted patient outcomes, and then, weight patient care activities performed by those clinicians to determine which are most predictive of outcomes. Additional evidence demonstrating the clinical relevance of this method will include an example of how the MLN model output can be used in hospital quality improvement initiatives.