Multicenter Sepsis Database(MSD):
Based on Real-world Clinical Data

The construction process of the sepsis knowledge graph primarily utilizes an ontology-based approach, involving key steps such as data preprocessing, analysis, ontology construction, knowledge extraction , fusion , and storage. Diverging from traditional knowledge graph construction methods, this study leverages LLMs knowledge extract to conduct initial data mining on raw data, enhancing the accuracy of the resulting knowledge graph. Following this, the research delves into mining and analyzing implicit knowledge within the data based on two-dimensional relationships. The input sources for graph construction encompass the following:

>The “top-down” approach involves defining a pattern layer based on logical relationships and hierarchical structures, followed by mapping data entities to this schema. Conversely, the “bottom-up” approach entails extracting entities and attributes from diverse data sources into the data layer of the knowledge graph, consolidating these extracted entities and attributes, and optimizing the schema layer of the knowledge graph for iterative updates to the ontology model.

The investigation of complications associated with sepsis holds paramount significance in clinical research.

Sepsis is a severe systemic inflammatory response often triggered by infection. In the knowledge graph, entering the following code reveals the relationships between the most commonly used medications for sepsis and their associated complications

Sepsis is defined as a comprehensive systemic inflammatory response, and its multifaceted nature is revealed through the statistical analysis of the top 25 symptoms observed in patients..