A cluster-based approach to predict serious adverse events in surgery

Managing risks related to the actions and conditions of the various elements that make up an operating room is a major concern during surgery. Determining alert thresholds is one of the main challenges. In this document, we propose to focus on the causes that lead to incidents as well as their prediction, which are essential elements in the determination of alerts. For that purpose, we have designed an architecture that couples a Multi-Agent System (MAS) with Case-Based Reasoning (CBR). The ability to emulate a large number of situations thanks to MAS, combined with analytical data management thanks to CBR is an efficient way of analyzing the state of the system and predicting its evolution. Beyond this architecture, decision support tools have been integrated in order to classify the behavior of entities and predict their evolution. This paper presents and analyzes the performance of our original cluster-based method (similVar) dedicated to the determination of unpredefined alert thresholds and risk prediction in surgery rooms. The obtained results prove the ability of our approach to analyze and predict the evolution of variables as disparate as the constants of a patient (''CAPNIA'', ``HYPOTHERMIA'', ``FeCO2'', ``SpO2'' etc.) or human fatigue.