Explainable AI Examines Sepsis Mortality Risk and Emergency Care Delays
Posted on 20 Nov 2026
Preventing sepsis-related deaths in emergency departments (EDs) remains challenging because patient deterioration can result from multiple rapidly evolving and interconnected factors. These include triage decisions, blood testing, clinical assessments, treatment timing, and staffing workflows, making it difficult for hospitals to identify the causes of delays in appropriate care. A new artificial intelligence (AI) approach aims to address this challenge by predicting mortality risk in patients with sepsis while explaining the factors contributing to its prediction.
Developed by researchers in the Causal Explainability Group at the Department of Informatics at King’s College London, the method analyzes the sequence of events during a patient’s ED visit, known as a patient trace. By examining the timing and relationships between clinical events, the system seeks to identify potential causes of adverse outcomes, including treatment delays and missed opportunities for timely intervention.
The method builds on predictive process monitoring, which analyzes patient event sequences to forecast outcomes such as hospital admission or discharge. Unlike conventional models that primarily identify statistical correlations, the new approach incorporates a causal explainability framework to examine dependencies between events over time. This allows the system to generate explanations for its predictions by identifying specific events and processes that may have contributed to a patient’s outcome.
The researchers evaluated the approach using real-world hospital data, demonstrating its ability to identify potential contributing factors in patient deaths. The findings suggest that such explanations could help hospital managers identify weaknesses in ED workflows and prioritize operational improvements to support timely care. Presented at the IEEE International Conference on Data Mining, the research also demonstrated that causal explanations could be generated without compromising the model’s computational efficiency.
“Correlation presents an incomplete picture in a hospital and is of limited use to the managers in charge of improving departments. They lack the rigor to point to specific causes of a patient's deterioration and the ability to explain to a doctor, who may disagree with its conclusions, what factors contributed to a patient's death,” said Dr. Tom Yaacov, research associate in the group and lead author of the paper.
Our methods are rooted in a formal and rigorous framework of causality, and they allow a much more accurate and fine-grained analysis of the causes of the delay in the patient's processing. They can also find the most influential causes—those that need to be addressed first to improve outcomes for a stretched hospital. Moreover, we can incorporate expert knowledge of medical professionals—something that the existing correlation-based methods cannot do,” said Dr. Yaacov.
Related Links
King’s College London