A neuro-fuzzy based alarm system for septic shock patients with a comparison to medical scores

Abstract

During the last years we collected data of abdominal septic shock patients from clinics all over Germany. The mortality of septic shock is about 50%. Septic shock is related to immune system reactions and unusual measurements. Septic shock patients are intensely medicated during their stay at the intensive care unit. To help physicians recognizing the critical states of their patients as early as possible, we built a rule based alarm system based on a neuro-fuzzy inference machine. Analysing the patient data in a time window, we show the time dependency of the classification results. We give detailed classification results and explanation by rules. The results are compared to results obtained by using the most common scores in intensive care medicine. We discuss the advantages of the paradigms “neural networks” and “scores”, and we answer the important question: Is a neural network more performant than scores for abdominal septic shock patient data?.