SepsisCalc: Integrating Clinical Calculators into Early Sepsis Prediction via Dynamic Temporal Graph Construction

TL;DR

SepsisCalc tackles early sepsis prediction by embedding clinical calculators into a dynamic temporal graph representation of EHR data. It dynamically estimates calculator scores and selectively integrates them into the graph, aligning AI predictions with clinician workflows, using a heterogeneous graph neural network to predict patient-level sepsis risk and organ dysfunction. Across the MIMIC-III, AmsterdamUMCdb, and OSUWMC datasets, SepsisCalc outperforms state-of-the-art baselines and provides interpretable, organ-specific risk assessments to enable actionable early interventions. The work also demonstrates a deployment-ready system integrated into an EHR, supporting effective human-AI collaboration in clinical decision making.

Abstract

Sepsis is an organ dysfunction caused by a deregulated immune response to an infection. Early sepsis prediction and identification allow for timely intervention, leading to improved clinical outcomes. Clinical calculators (e.g., the six-organ dysfunction assessment of SOFA) play a vital role in sepsis identification within clinicians' workflow, providing evidence-based risk assessments essential for sepsis diagnosis. However, artificial intelligence (AI) sepsis prediction models typically generate a single sepsis risk score without incorporating clinical calculators for assessing organ dysfunctions, making the models less convincing and transparent to clinicians. To bridge the gap, we propose to mimic clinicians' workflow with a novel framework SepsisCalc to integrate clinical calculators into the predictive model, yielding a clinically transparent and precise model for utilization in clinical settings. Practically, clinical calculators usually combine information from multiple component variables in Electronic Health Records (EHR), and might not be applicable when the variables are (partially) missing. We mitigate this issue by representing EHRs as temporal graphs and integrating a learning module to dynamically add the accurately estimated calculator to the graphs. Experimental results on real-world datasets show that the proposed model outperforms state-of-the-art methods on sepsis prediction tasks. Moreover, we developed a system to identify organ dysfunctions and potential sepsis risks, providing a human-AI interaction tool for deployment, which can help clinicians understand the prediction outputs and prepare timely interventions for the corresponding dysfunctions, paving the way for actionable clinical decision-making support for early intervention.

Figures (10)

• Figure 1: Workflows of clinicians and AI for sepsis identification. Clinicians examine sepsis by assessing organ dysfunctions with multiple clinical calculators as evidence, while AI workflow only gives an overall sepsis risk score.
• Figure 2: Different EHR representation methods. (A) An example of sequential representation. (B) Example of graph representation with temporal information of clinical observations. (C) The proposed dynamic temporal graph representation with clinical event interaction and clinical calculators. Note that only partial calculator and organ nodes and edges are plotted for graph illustration in subfigure C.
• Figure 3: Framework of SepsisCalc. (A) Dynamic temporal graph construction. (B) Sepsis prediction framework.
• Figure 4: Setting of sepsis onset prediction.
• Figure 5: Sepsis risk prediction performance in both full and missing observation settings.